Menu
▾
▴
oprofile-commits — Diffs of cvs commits (occassionally high-traffic)
You can subscribe to this list here.
| 2001 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
(7) |
Aug
(50) |
Sep
(210) |
Oct
(93) |
Nov
(99) |
Dec
(101) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2002 |
Jan
(283) |
Feb
(42) |
Mar
(148) |
Apr
(94) |
May
(256) |
Jun
(206) |
Jul
(101) |
Aug
(28) |
Sep
(159) |
Oct
(102) |
Nov
(216) |
Dec
(159) |
| 2003 |
Jan
(112) |
Feb
(149) |
Mar
(332) |
Apr
(261) |
May
(528) |
Jun
(270) |
Jul
(231) |
Aug
(231) |
Sep
(338) |
Oct
(445) |
Nov
(212) |
Dec
(26) |
| 2004 |
Jan
(179) |
Feb
(109) |
Mar
(20) |
Apr
(20) |
May
(55) |
Jun
(11) |
Jul
(38) |
Aug
(13) |
Sep
(28) |
Oct
(13) |
Nov
(41) |
Dec
(30) |
| 2005 |
Jan
(17) |
Feb
(4) |
Mar
(62) |
Apr
(176) |
May
(44) |
Jun
(13) |
Jul
(16) |
Aug
(154) |
Sep
(6) |
Oct
(13) |
Nov
(4) |
Dec
(6) |
| 2006 |
Jan
(12) |
Feb
(11) |
Mar
(9) |
Apr
(1) |
May
(10) |
Jun
(6) |
Jul
(10) |
Aug
(10) |
Sep
(27) |
Oct
(25) |
Nov
(40) |
Dec
(18) |
| 2007 |
Jan
(13) |
Feb
(25) |
Mar
(6) |
Apr
(14) |
May
(52) |
Jun
(22) |
Jul
(20) |
Aug
(8) |
Sep
(2) |
Oct
(138) |
Nov
(152) |
Dec
(73) |
| 2008 |
Jan
(52) |
Feb
(31) |
Mar
(5) |
Apr
(48) |
May
(20) |
Jun
(14) |
Jul
(36) |
Aug
(16) |
Sep
(1) |
Oct
(19) |
Nov
(13) |
Dec
(4) |
| 2009 |
Jan
(2) |
Feb
(13) |
Mar
(10) |
Apr
(28) |
May
(46) |
Jun
(21) |
Jul
(21) |
Aug
(22) |
Sep
(8) |
Oct
(25) |
Nov
(15) |
Dec
(2) |
| 2010 |
Jan
(12) |
Feb
|
Mar
(8) |
Apr
(3) |
May
(2) |
Jun
(5) |
Jul
(11) |
Aug
(17) |
Sep
|
Oct
(12) |
Nov
(14) |
Dec
(10) |
| 2011 |
Jan
(6) |
Feb
|
Mar
(8) |
Apr
(6) |
May
(17) |
Jun
(8) |
Jul
(4) |
Aug
(9) |
Sep
(2) |
Oct
(3) |
Nov
(4) |
Dec
(2) |
| 2012 |
Jan
(4) |
Feb
(7) |
Mar
(16) |
Apr
(9) |
May
(15) |
Jun
(22) |
Jul
(30) |
Aug
(36) |
Sep
(6) |
Oct
(5) |
Nov
(9) |
Dec
(11) |
| 2013 |
Jan
(17) |
Feb
(11) |
Mar
(7) |
Apr
(8) |
May
(15) |
Jun
(19) |
Jul
(27) |
Aug
(3) |
Sep
(3) |
Oct
(8) |
Nov
(6) |
Dec
(10) |
| 2014 |
Jan
(15) |
Feb
(16) |
Mar
(4) |
Apr
(3) |
May
(10) |
Jun
(9) |
Jul
(9) |
Aug
(31) |
Sep
(11) |
Oct
(6) |
Nov
(4) |
Dec
(3) |
| 2015 |
Jan
(2) |
Feb
(1) |
Mar
(2) |
Apr
(3) |
May
(4) |
Jun
(5) |
Jul
(14) |
Aug
(2) |
Sep
(2) |
Oct
(1) |
Nov
(4) |
Dec
(4) |
| 2016 |
Jan
|
Feb
(1) |
Mar
(1) |
Apr
(4) |
May
(2) |
Jun
|
Jul
(2) |
Aug
(1) |
Sep
(1) |
Oct
(3) |
Nov
(1) |
Dec
|
| 2017 |
Jan
(4) |
Feb
|
Mar
(1) |
Apr
(3) |
May
|
Jun
(9) |
Jul
(11) |
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
| 2018 |
Jan
(1) |
Feb
|
Mar
(1) |
Apr
|
May
|
Jun
(3) |
Jul
(10) |
Aug
(1) |
Sep
|
Oct
|
Nov
|
Dec
|
| 2019 |
Jan
|
Feb
|
Mar
(1) |
Apr
(1) |
May
(6) |
Jun
|
Jul
(2) |
Aug
(2) |
Sep
(1) |
Oct
|
Nov
|
Dec
(1) |
| 2020 |
Jan
|
Feb
|
Mar
(2) |
Apr
|
May
|
Jun
|
Jul
(4) |
Aug
|
Sep
(1) |
Oct
|
Nov
|
Dec
|
| 2021 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
(1) |
Dec
|
| 2024 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
(2) |
Dec
|
|
From: John L. <mov...@us...> - 2001-09-02 00:32:32
|
Update of /cvsroot/oprofile/oprofile
In directory usw-pr-cvs1:/tmp/cvs-serv14034
Modified Files:
ChangeLog oprofile.c
Log Message:
philippe's patch
Index: ChangeLog
===================================================================
RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v
retrieving revision 1.92
retrieving revision 1.93
diff -u -d -r1.92 -r1.93
--- ChangeLog 2001/09/01 02:03:34 1.92
+++ ChangeLog 2001/09/02 00:32:28 1.93
@@ -1,3 +1,10 @@
+2001-09-01 Philippe Elie <ph...@cl...>
+
+ * oprofile.c: fix pmc_setup()
+
+ * pp/Makefile.in:
+ * dae/Makefile.in: fix uninstall problem
+
2001-08-31 Philippe Elie <ph...@cl...>
* oprofile.h:
Index: oprofile.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/oprofile.c,v
retrieving revision 1.76
retrieving revision 1.77
diff -u -d -r1.76 -r1.77
--- oprofile.c 2001/09/01 02:03:34 1.76
+++ oprofile.c 2001/09/02 00:32:28 1.77
@@ -441,10 +441,13 @@
/* Stop and clear all counter: IA32 use bit 22 of eventsel_msr0 to
* enable/disable all counter, AMD use separate bit 22 in each msr,
- * all other bits are cleared except the reserved bits 21 */
+ * all bits are cleared except the reserved bits 21 */
for (i = 0 ; i < op_nr_counters ; ++i) {
rdmsr(eventsel_msr[i], low, high);
- wrmsr(eventsel_msr[i], low & ~(3 << 22), high);
+ wrmsr(eventsel_msr[i], low & (1 << 22), high);
+
+ /* avoid a false detection of ctr overflow in NMI handler */
+ wrmsr(perfctr_msr[i], -1, -1);
}
/* setup each counter */
@@ -501,7 +504,8 @@
/* this is pretty bogus really. especially as we don't re-enable it.
* Instead, save state set up, and restore with pmc_unsetup or similar */
#if !defined(CONFIG_X86_UP_APIC) || !defined(OP_EXPORTED_DO_NMI)
- wrmsr(eventsel_msr[1], low, high);
+ /* PHE FIXME: on my config this start counter 1 with the new code */
+// wrmsr(eventsel_msr[1], low, high);
#endif
}
|
|
From: Dave J. <da...@us...> - 2001-09-01 21:50:38
|
Update of /cvsroot/oprofile/oprofile In directory usw-pr-cvs1:/tmp/cvs-serv5883 Modified Files: .cvsignore Log Message: Update ignorance list. Index: .cvsignore =================================================================== RCS file: /cvsroot/oprofile/oprofile/.cvsignore,v retrieving revision 1.2 retrieving revision 1.3 diff -u -d -r1.2 -r1.3 --- .cvsignore 2000/08/18 01:24:37 1.2 +++ .cvsignore 2001/09/01 21:50:35 1.3 @@ -3,6 +3,7 @@ version.h a.diff oprofile.s +confdefs.h config.log config.cache config.status |
|
From: John L. <mov...@us...> - 2001-09-01 02:03:43
|
Update of /cvsroot/oprofile/oprofile
In directory usw-pr-cvs1:/tmp/cvs-serv19169
Modified Files:
ChangeLog Makefile.in op_events.c op_user.h oprofile.c
oprofile.h
Log Message:
philippe's huge patch
Index: ChangeLog
===================================================================
RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v
retrieving revision 1.91
retrieving revision 1.92
diff -u -d -r1.91 -r1.92
--- ChangeLog 2001/08/31 17:16:35 1.91
+++ ChangeLog 2001/09/01 02:03:34 1.92
@@ -1,5 +1,63 @@
-2001-08-31 John Levon <mo...@co...>
+2001-08-31 Philippe Elie <ph...@cl...>
+
+ * oprofile.h:
+ * dae/oprofiled.h:
+ * op_user.h:
+ * oprofpp.h:
+ * dae/opd_util.h: share commmon declarations, put it in
+ op_user.h, include op_user.h in other .h when necessary.
+ * op_events.c:
+ * all Makefile.in: update dependancies.
+
+2001-08-30 Philippe Elie <ph...@cl...>
+
+ * oprofile.c: use loop instead of fixed counter number
+
+2001-08-26 Philippe Elie <ph...@cl...>
+
+ * op_events.c: add option to op_help for the gui
+ * gui/oprofile: use op_help to build description
+ of events, accept Athlon (only 2 counters)
+
+2001-08-25 Philippe Elie <ph...@cl...>
+
+ * pp/oprofpp.c:
+ * dae/op_start: use loop instead of fixed number of counter.
+
+2001-08-24 Philippe Elie <ph...@cl...>
+
+ * pp/oprof_convert.c: add v4 --> v5 conversion
+
+2001-08-23 Philippe Elie <ph...@cl...>
+
+ * oprofile.c: correct some comment.
+
+2001-08-22 Philippe Elie <ph...@cl...>
+
+ * op_user.h:
+ * dae/opd_proc.c:
+ * dae/opd_util.c:
+ * dae/opd_util.h:
+ * dae/oprofiled.h:
+ * dae/oprofiled.c:
+ * dae/oprofiled.c: row to column file format, lot of changes.
+
+2001-08-21 Philippe Elie <ph...@cl...>
+
+ * op_events.c: Add Athlon string/unit maks description
+
+2001-08-20 Philippe Elie <ph...@cl...>
+ * op_events.c: change in handling of search/check events
+ * dae/oprofiled.c:
+ * dae/oprofiled.h:
+ * oprofile.c:
+ * oprofile.h:
+ * pp/oprofpp.c:
+ * pp/oprofpp.h: reflects caller/prototype, lot of changes.
+
+2001-08-31 John Levon <mo...@co...>
+
* Makefile.in:
* op_x86.c:
* oprofile.h:
Index: Makefile.in
===================================================================
RCS file: /cvsroot/oprofile/oprofile/Makefile.in,v
retrieving revision 1.24
retrieving revision 1.25
diff -u -d -r1.24 -r1.25
--- Makefile.in 2001/08/31 17:16:35 1.24
+++ Makefile.in 2001/09/01 02:03:34 1.25
@@ -94,7 +94,7 @@
oprofile.s: oprofile.c oprofile.h
$(CC) $(KCFLAGS) -S $<
-op_events.o: op_events.c
+op_events.o: op_events.c op_user.h
$(CC) $(KCFLAGS) -c -o $@ $<
oprofile_nmi.o: oprofile_nmi.S
Index: op_events.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/op_events.c,v
retrieving revision 1.20
retrieving revision 1.21
diff -u -d -r1.20 -r1.21
--- op_events.c 2001/08/19 20:09:17 1.20
+++ op_events.c 2001/09/01 02:03:34 1.21
@@ -19,15 +19,6 @@
/* See IA32 Vol. 3 Appendix A */
-/* for allowed */
-#define OP_0_ONLY 0
-#define OP_1_ONLY 1
-#define OP_ANY 2
-#define OP_PII_PIII 3
-#define OP_PII_ONLY 4
-#define OP_PIII_ONLY 5
-#define OP_ATHLON_ONLY 6
-
#ifdef __KERNEL__
#include <linux/string.h>
#define strcmp(a,b) strnicmp((a),(b),strlen((b)))
@@ -35,10 +26,10 @@
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
+#include <limits.h>
#endif
-#define u8 unsigned char
-#define uint unsigned int
+#include "op_user.h"
enum unit_mask_type {
/* useless but required by the hardware */
@@ -50,186 +41,186 @@
};
struct op_event {
- uint allowed;
- u8 val; /* event number */
- u8 unit; /* which unit mask if any allowed */
+ uint counter_mask; /* bitmask of allowed counter */
+ u16 cpu_mask; /* bitmask of allowed cpu_type */
+ u8 val; /* event number */
+ u8 unit; /* which unit mask if any allowed */
const char *name;
- int min_count; /* minimum counter value allowed */
+ int min_count; /* minimum counter value allowed */
};
struct op_unit_mask {
uint num; /* number of possible unit masks */
enum unit_mask_type unit_type_mask;
+ /* only the gui use it */
+ u8 default_mask;
/* up to seven allowed unit masks */
u8 um[7];
};
-int op_check_events_str(char *ctr0_type, char *ctr1_type, u8 ctr0_um, u8 ctr1_um, int p2, u8 *ctr0_t, u8 *ctr1_t);
-int op_check_events(u8 ctr0_type, u8 ctr1_type, u8 ctr0_um, u8 ctr1_um, int proc);
-int op_min_count(u8 ctr_type);
-#ifdef OP_EVENTS_DESC
-void op_get_event_desc(u8 type, u8 um, char **typenamep, char **typedescp, char **umdescp);
-#endif
-
static struct op_unit_mask op_unit_masks[] = {
- /* not used */
- { 1, utm_mandatory, { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 }, },
+ /* use by the gui */
+ { 1, utm_mandatory, 0x0f, { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 }, },
/* MESI counters */
- { 5, utm_bitmask, { 0x1, 0x2, 0x4, 0x8, 0xf, 0x0, 0x0 }, },
- /* EBL self/any */
- { 2, utm_exclusive, { 0x0, 0x20, 0x0, 0x0, 0x0, 0x0, 0x0 }, },
+ { 5, utm_bitmask, 0x0f, { 0x8, 0x4, 0x2, 0x1, 0xf, 0x0, 0x0 }, },
+ /* EBL self/any default to any transitions */
+ { 2, utm_exclusive, 0x20, { 0x0, 0x20, 0x0, 0x0, 0x0, 0x0, 0x0 }, },
/* MMX PII events */
- { 1, utm_mandatory, { 0xf, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 }, },
- { 7, utm_bitmask, { 0x1, 0x2, 0x4, 0x8, 0x10, 0x20, 0x3f }, },
- { 2, utm_exclusive, { 0x0, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0 }, },
- { 5, utm_bitmask, { 0x1, 0x2, 0x4, 0x8, 0xf, 0x0, 0x0 }, },
+ { 1, utm_mandatory, 0xf, { 0xf, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 }, },
+ { 7, utm_bitmask, 0x3f, { 0x1, 0x2, 0x4, 0x8, 0x10, 0x20, 0x3f }, },
+ { 2, utm_exclusive, 0x0, { 0x0, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0 }, },
+ { 5, utm_bitmask, 0x0f, { 0x1, 0x2, 0x4, 0x8, 0xf, 0x0, 0x0 }, },
/* KNI PIII events */
- { 4, utm_exclusive, { 0x0, 0x1, 0x2, 0x3, 0x0, 0x0, 0x0 }, },
- { 2, utm_bitmask, { 0x0, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0 }, },
+ { 4, utm_exclusive, 0x0, { 0x0, 0x1, 0x2, 0x3, 0x0, 0x0, 0x0 }, },
+ { 2, utm_bitmask, 0x1, { 0x0, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0 }, },
+ /* Athlon MOESI cache events */
+ { 6, utm_bitmask, 0x1f, { 0x10, 0x8, 0x4, 0x2, 0x1, 0x1f }, }
};
+/* the following are just short cut for filling the table of event */
+#define OP_ATHLON (1 << CPU_ATHLON)
+#define OP_PPRO (1 << CPU_PPRO)
+#define OP_PII (1 << CPU_PII)
+#define OP_PIII (1 << CPU_PIII)
+#define OP_PII_PIII (OP_PII | OP_PIII)
+#define OP_IA_ALL (OP_PII_PIII | OP_PPRO)
+
+#define CTR_ALL (~0u)
+#define CTR_0 (1 << 0)
+#define CTR_1 (1 << 1)
+
/* Allowed, Event #, unit mask, name, minimum event value */
static struct op_event op_events[] = {
+ /* Clocks */
+ { CTR_ALL, OP_IA_ALL, 0x79, 0, "CPU_CLK_UNHALTED", 6000 },
/* Data Cache Unit (DCU) */
- {OP_ATHLON_ONLY,0x40,0,"DATA_CACHE_ACCESSES", 500,},
- {OP_ATHLON_ONLY,0x41,0,"DATA_CACHE_MISSES", 500,},
- {OP_ATHLON_ONLY,0x42,0,"DATA_CACHE_REFILLS_FROM_L2", 500,},
- {OP_ATHLON_ONLY,0x43,0,"DATA_CACHE_REFILLS_FROM_SYSTEM", 500,},
- {OP_ATHLON_ONLY,0x44,0,"DATA_CACHE_WRITEBACKS", 500,},
- {OP_ATHLON_ONLY,0x45,0,"L1_DTLB_MISSES_L2_DTLD_HITS", 500,},
- {OP_ATHLON_ONLY,0x46,0,"L1_AND_L2_DTLB_MISSES", 500,},
- {OP_ATHLON_ONLY,0x47,0,"MISALIGNED_DATA_REFS", 500,},
- {OP_ATHLON_ONLY,0x80,0,"ICACHE_FETCHES", 500,},
- {OP_ATHLON_ONLY,0x81,0,"ICACHE_MISSES", 500,},
- {OP_ATHLON_ONLY,0x84,0,"L1_ITLB_MISSES_L2_ITLB_HITS", 500,},
- {OP_ATHLON_ONLY,0x85,0,"L1_AND_L2_ITLB_MISSES", 500,},
- {OP_ATHLON_ONLY,0xc0,0,"RETIRED_INSNS", 500,},
- {OP_ATHLON_ONLY,0xc1,0,"RETIRED_OPS", 500,},
- {OP_ATHLON_ONLY,0xc2,0,"RETIRED_BRANCHES", 500,},
- {OP_ATHLON_ONLY,0xc3,0,"RETIRED_BRANCHES_MISPREDICTED", 500,},
- {OP_ATHLON_ONLY,0xc4,0,"RETIRED_TAKEN_BRANCHES", 500,},
- {OP_ATHLON_ONLY,0xc5,0,"RETIRED_TAKEN_BRANCHES_MISPREDICTED", 500,},
- {OP_ATHLON_ONLY,0xc6,0,"RETIRED_FAR_CONTROL_TRANSFERS", 500,},
- {OP_ATHLON_ONLY,0xc7,0,"RETIRED_RESYNC_BRANCHES", 500,},
- {OP_ATHLON_ONLY,0xcd,0,"INTERRUPTS_MASKED", 500,},
- {OP_ATHLON_ONLY,0xce,0,"INTERRUPTS_MASKED_PENDING", 500,},
- {OP_ATHLON_ONLY,0xcf,0,"HARDWARE_INTERRUPTS", 500,},
-
- {OP_ANY,0x43,0,"DATA_MEM_REFS", 500 },
- {OP_ANY,0x45,0,"DCU_LINES_IN", 500 },
- {OP_ANY,0x46,0,"DCU_M_LINES_IN", 500 },
- {OP_ANY,0x47,0,"DCU_M_LINES_OUT", 500},
- {OP_ANY,0x48,0,"DCU_MISS_OUTSTANDING", 500 },
-
+ { CTR_ALL, OP_IA_ALL, 0x43, 0, "DATA_MEM_REFS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x45, 0, "DCU_LINES_IN", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x46, 0, "DCU_M_LINES_IN", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x47, 0, "DCU_M_LINES_OUT", 500},
+ { CTR_ALL, OP_IA_ALL, 0x48, 0, "DCU_MISS_OUTSTANDING", 500 },
/* Intruction Fetch Unit (IFU) */
- {OP_ANY,0x80,0,"IFU_IFETCH", 500 },
- {OP_ANY,0x81,0,"IFU_IFETCH_MISS", 500 },
- {OP_ANY,0x85,0,"ITLB_MISS", 500},
- {OP_ANY,0x86,0,"IFU_MEM_STALL", 500 },
- {OP_ANY,0x87,0,"ILD_STALL", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x80, 0, "IFU_IFETCH", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x81, 0, "IFU_IFETCH_MISS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x85, 0, "ITLB_MISS", 500},
+ { CTR_ALL, OP_IA_ALL, 0x86, 0, "IFU_MEM_STALL", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x87, 0, "ILD_STALL", 500 },
/* L2 Cache */
- {OP_ANY,0x28,1,"L2_IFETCH", 500 },
- {OP_ANY,0x29,1,"L2_LD", 500 },
- {OP_ANY,0x2a,1,"L2_ST", 500 },
- {OP_ANY,0x24,0,"L2_LINES_IN", 500 },
- {OP_ANY,0x26,0,"L2_LINES_OUT", 500 },
- {OP_ANY,0x25,0,"L2_M_LINES_INM", 500 },
- {OP_ANY,0x27,0,"L2_M_LINES_OUTM", 500 },
- {OP_ANY,0x2e,1,"L2_RQSTS", 500 },
- {OP_ANY,0x21,0,"L2_ADS", 500 },
- {OP_ANY,0x22,0,"L2_DBUS_BUSY", 500 },
- {OP_ANY,0x23,0,"L2_DMUS_BUSY_RD", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x28, 1, "L2_IFETCH", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x29, 1, "L2_LD", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x2a, 1, "L2_ST", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x24, 0, "L2_LINES_IN", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x26, 0, "L2_LINES_OUT", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x25, 0, "L2_M_LINES_INM", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x27, 0, "L2_M_LINES_OUTM", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x2e, 1, "L2_RQSTS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x21, 0, "L2_ADS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x22, 0, "L2_DBUS_BUSY", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x23, 0, "L2_DMUS_BUSY_RD", 500 },
/* External Bus Logic (EBL) */
- {OP_ANY,0x62,2,"BUS_DRDY_CLOCKS", 500 },
- {OP_ANY,0x63,2,"BUS_LOCK_CLOCKS", 500 },
- {OP_ANY,0x60,0,"BUS_REQ_OUTSTANDING", 500 },
- {OP_ANY,0x65,2,"BUS_TRAN_BRD", 500 },
- {OP_ANY,0x66,2,"BUS_TRAN_RFO", 500 },
- {OP_ANY,0x67,2,"BUS_TRANS_WB", 500 },
- {OP_ANY,0x68,2,"BUS_TRAN_IFETCH", 500 },
- {OP_ANY,0x69,2,"BUS_TRAN_INVAL", 500 },
- {OP_ANY,0x6a,2,"BUS_TRAN_PWR", 500 },
- {OP_ANY,0x6b,2,"BUS_TRANS_P", 500 },
- {OP_ANY,0x6c,2,"BUS_TRANS_IO", 500 },
- {OP_ANY,0x6d,2,"BUS_TRANS_DEF", 500 },
- {OP_ANY,0x6e,2,"BUS_TRAN_BURST", 500 },
- {OP_ANY,0x70,2,"BUS_TRAN_ANY", 500 },
- {OP_ANY,0x6f,2,"BUS_TRAN_MEM", 500 },
- {OP_ANY,0x64,0,"BUS_DATA_RCV", 500 },
- {OP_ANY,0x61,0,"BUS_BNR_DRV", 500 },
- {OP_ANY,0x7a,0,"BUS_HIT_DRV", 500 },
- {OP_ANY,0x7b,0,"BUS_HITM_DRV", 500 },
- {OP_ANY,0x7e,0,"BUS_SNOOP_STALL", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x62, 2, "BUS_DRDY_CLOCKS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x63, 2, "BUS_LOCK_CLOCKS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x60, 0, "BUS_REQ_OUTSTANDING", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x65, 2, "BUS_TRAN_BRD", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x66, 2, "BUS_TRAN_RFO", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x67, 2, "BUS_TRANS_WB", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x68, 2, "BUS_TRAN_IFETCH", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x69, 2, "BUS_TRAN_INVAL", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x6a, 2, "BUS_TRAN_PWR", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x6b, 2, "BUS_TRANS_P", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x6c, 2, "BUS_TRANS_IO", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x6d, 2, "BUS_TRANS_DEF", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x6e, 2, "BUS_TRAN_BURST", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x70, 2, "BUS_TRAN_ANY", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x6f, 2, "BUS_TRAN_MEM", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x64, 0, "BUS_DATA_RCV", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x61, 0, "BUS_BNR_DRV", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x7a, 0, "BUS_HIT_DRV", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x7b, 0, "BUS_HITM_DRV", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x7e, 0, "BUS_SNOOP_STALL", 500 },
/* Floating Point Unit (FPU) */
- {OP_0_ONLY,0xc1,0,"COMP_FLOP_RET", 3000 },
- {OP_0_ONLY,0x10,0,"FLOPS", 3000 },
- {OP_1_ONLY,0x11,0,"FP_ASSIST", 500 },
- {OP_1_ONLY,0x12,0,"MUL", 1000 },
- {OP_1_ONLY,0x13,0,"DIV", 500 },
- {OP_0_ONLY,0x14,0,"CYCLES_DIV_BUSY", 1000 },
+ { CTR_0, OP_IA_ALL, 0xc1, 0, "COMP_FLOP_RET", 3000 },
+ { CTR_0, OP_IA_ALL, 0x10, 0, "FLOPS", 3000 },
+ { CTR_1, OP_IA_ALL, 0x11, 0, "FP_ASSIST", 500 },
+ { CTR_1, OP_IA_ALL, 0x12, 0, "MUL", 1000 },
+ { CTR_1, OP_IA_ALL, 0x13, 0, "DIV", 500 },
+ { CTR_0, OP_IA_ALL, 0x14, 0, "CYCLES_DIV_BUSY", 1000 },
/* Memory Ordering */
- {OP_ANY,0x03,0,"LD_BLOCKS", 500 },
- {OP_ANY,0x04,0,"SB_DRAINS", 500 },
- {OP_ANY,0x05,0,"MISALIGN_MEM_REF", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x03, 0, "LD_BLOCKS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x04, 0, "SB_DRAINS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0x05, 0, "MISALIGN_MEM_REF", 500 },
/* PIII KNI */
- {OP_PIII_ONLY,0x07,7,"EMON_KNI_PREF_DISPATCHED", 500 },
- {OP_PIII_ONLY,0x4b,7,"EMON_KNI_PREF_MISS", 500 },
+ { CTR_ALL, OP_PIII, 0x07, 7, "EMON_KNI_PREF_DISPATCHED", 500 },
+ { CTR_ALL, OP_PIII, 0x4b, 7, "EMON_KNI_PREF_MISS", 500 },
/* Instruction Decoding and Retirement */
- {OP_ANY,0xc0,0,"INST_RETIRED", 6000 },
- {OP_ANY,0xc2,0,"UOPS_RETIRED", 6000 },
- {OP_ANY,0xd0,0,"INST_DECODED", 6000 },
+ { CTR_ALL, OP_IA_ALL, 0xc0, 0, "INST_RETIRED", 6000 },
+ { CTR_ALL, OP_IA_ALL, 0xc2, 0, "UOPS_RETIRED", 6000 },
+ { CTR_ALL, OP_IA_ALL, 0xd0, 0, "INST_DECODED", 6000 },
/* PIII KNI */
- {OP_PIII_ONLY,0xd8,8,"EMON_KNI_INST_RETIRED", 3000 },
- {OP_PIII_ONLY,0xd9,8,"EMON_KNI_COMP_INST_RET", 3000 },
+ { CTR_ALL, OP_PIII, 0xd8, 8, "EMON_KNI_INST_RETIRED", 3000 },
+ { CTR_ALL, OP_PIII, 0xd9, 8, "EMON_KNI_COMP_INST_RET", 3000 },
/* Interrupts */
- {OP_ANY,0xc8,0,"HW_INT_RX", 500 },
- {OP_ANY,0xc6,0,"CYCLES_INT_MASKED", 500 },
- {OP_ANY,0xc7,0,"CYCLES_INT_PENDING_AND_MASKED", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xc8, 0, "HW_INT_RX", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xc6, 0, "CYCLES_INT_MASKED", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xc7, 0, "CYCLES_INT_PENDING_AND_MASKED", 500 },
/* Branches */
- {OP_ANY,0xc4,0,"BR_INST_RETIRED", 500 },
- {OP_ANY,0xc5,0,"BR_MISS_PRED_RETIRED", 500 },
- {OP_ANY,0xc9,0,"BR_TAKEN_RETIRED", 500 },
- {OP_ANY,0xca,0,"BR_MISS_PRED_TAKEN_RET", 500 },
- {OP_ANY,0xe0,0,"BR_INST_DECODED", 500 },
- {OP_ANY,0xe2,0,"BTB_MISSES", 500 },
- {OP_ANY,0xe4,0,"BR_BOGUS", 500 },
- {OP_ANY,0xe6,0,"BACLEARS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xc4, 0, "BR_INST_RETIRED", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xc5, 0, "BR_MISS_PRED_RETIRED", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xc9, 0, "BR_TAKEN_RETIRED", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xca, 0, "BR_MISS_PRED_TAKEN_RET", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xe0, 0, "BR_INST_DECODED", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xe2, 0, "BTB_MISSES", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xe4, 0, "BR_BOGUS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xe6, 0, "BACLEARS", 500 },
/* Stalls */
- {OP_ANY,0xa2,0,"RESOURCE_STALLS", 500 },
- {OP_ANY,0xd2,0,"PARTIAL_RAT_STALLS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xa2, 0, "RESOURCE_STALLS", 500 },
+ { CTR_ALL, OP_IA_ALL, 0xd2, 0, "PARTIAL_RAT_STALLS", 500 },
/* Segment Register Loads */
- {OP_ANY,0x06,0,"SEGMENT_REG_LOADS", 500 },
- /* Clocks */
- {OP_ANY,0x79,0,"CPU_CLK_UNHALTED", 6000 },
+ { CTR_ALL, OP_IA_ALL, 0x06, 0, "SEGMENT_REG_LOADS", 500 },
/* MMX (Pentium II only) */
- {OP_PII_ONLY,0xb0,0,"MMX_INSTR_EXEC", 3000 },
- {OP_PII_PIII,0xb1,0,"MMX_SAT_INSTR_EXEC", 3000 },
- {OP_PII_PIII,0xb2,3,"MMX_UOPS_EXEC", 3000 },
- {OP_PII_PIII,0xb3,4,"MMX_INSTR_TYPE_EXEC", 3000 },
- {OP_PII_PIII,0xcc,5,"FP_MMX_TRANS", 3000 },
- {OP_PII_PIII,0xcd,0,"MMX_ASSIST", 500 },
- {OP_PII_ONLY,0xce,0,"MMX_INSTR_RET", 3000 },
+ { CTR_ALL, OP_PII, 0xb0, 0, "MMX_INSTR_EXEC", 3000 },
+ { CTR_ALL, OP_PII_PIII, 0xb1, 0, "MMX_SAT_INSTR_EXEC", 3000 },
+ { CTR_ALL, OP_PII_PIII, 0xb2, 3, "MMX_UOPS_EXEC", 3000 },
+ { CTR_ALL, OP_PII_PIII, 0xb3, 4, "MMX_INSTR_TYPE_EXEC", 3000 },
+ { CTR_ALL, OP_PII_PIII, 0xcc, 5, "FP_MMX_TRANS", 3000 },
+ { CTR_ALL, OP_PII_PIII, 0xcd, 0, "MMX_ASSIST", 500 },
+ { CTR_ALL, OP_PII_PIII, 0xce, 0, "MMX_INSTR_RET", 3000 },
/* segment renaming (Pentium II only) */
- {OP_PII_PIII,0xd4,6,"SEG_RENAME_STALLS", 500 },
- {OP_PII_PIII,0xd5,6,"SEG_REG_RENAMES", 500 },
- {OP_PII_PIII,0xd6,0,"RET_SEG_RENAMES", 500 },
-};
+ { CTR_ALL, OP_PII, 0xd4, 6, "SEG_RENAME_STALLS", 500 },
+ { CTR_ALL, OP_PII, 0xd5, 6, "SEG_REG_RENAMES", 500 },
+ { CTR_ALL, OP_PII, 0xd6, 0, "RET_SEG_RENAMES", 500 },
-uint op_nr_events = sizeof(op_events)/sizeof(struct op_event);
+ /* Data Cache Unit (DCU) */
+ /* PHE FIXME : for athlon check the op_min count */
+ /* PHE FIXME: Athlon put the most used entry at begin and update the
+ * description string, this is needed for the gui to display the most
+ * common used event first. */
+ { CTR_ALL, OP_ATHLON, 0x40, 0, "DATA_CACHE_ACCESSES", 500,},
+ { CTR_ALL, OP_ATHLON, 0x41, 0, "DATA_CACHE_MISSES", 500,},
+ { CTR_ALL, OP_ATHLON, 0x42, 9, "DATA_CACHE_REFILLS_FROM_L2", 500,},
+ { CTR_ALL, OP_ATHLON, 0x43, 9, "DATA_CACHE_REFILLS_FROM_SYSTEM", 500,},
+ { CTR_ALL, OP_ATHLON, 0x44, 9, "DATA_CACHE_WRITEBACKS", 500,},
+ { CTR_ALL, OP_ATHLON, 0x45, 0, "L1_DTLB_MISSES_L2_DTLD_HITS", 500,},
+ { CTR_ALL, OP_ATHLON, 0x46, 0, "L1_AND_L2_DTLB_MISSES", 500,},
+ { CTR_ALL, OP_ATHLON, 0x47, 0, "MISALIGNED_DATA_REFS", 500,},
+ { CTR_ALL, OP_ATHLON, 0x80, 0, "ICACHE_FETCHES", 500,},
+ { CTR_ALL, OP_ATHLON, 0x81, 0, "ICACHE_MISSES", 500,},
+ { CTR_ALL, OP_ATHLON, 0x84, 0, "L1_ITLB_MISSES_L2_ITLB_HITS", 500,},
+ { CTR_ALL, OP_ATHLON, 0x85, 0, "L1_AND_L2_ITLB_MISSES", 500,},
+ { CTR_ALL, OP_ATHLON, 0xc0, 0, "RETIRED_INSNS", 500,},
+ { CTR_ALL, OP_ATHLON, 0xc1, 0, "RETIRED_OPS", 500,},
+ { CTR_ALL, OP_ATHLON, 0xc2, 0, "RETIRED_BRANCHES", 500,},
+ { CTR_ALL, OP_ATHLON, 0xc3, 0, "RETIRED_BRANCHES_MISPREDICTED", 500,},
+ { CTR_ALL, OP_ATHLON, 0xc4, 0, "RETIRED_TAKEN_BRANCHES", 500,},
+ { CTR_ALL, OP_ATHLON, 0xc5, 0, "RETIRED_TAKEN_BRANCHES_MISPREDICTED", 500,},
+ { CTR_ALL, OP_ATHLON, 0xc6, 0, "RETIRED_FAR_CONTROL_TRANSFERS", 500,},
+ { CTR_ALL, OP_ATHLON, 0xc7, 0, "RETIRED_RESYNC_BRANCHES", 500,},
+ { CTR_ALL, OP_ATHLON, 0xcd, 0, "INTERRUPTS_MASKED", 500,},
+ { CTR_ALL, OP_ATHLON, 0xce, 0, "INTERRUPTS_MASKED_PENDING", 500,},
+ { CTR_ALL, OP_ATHLON, 0xcf, 0, "HARDWARE_INTERRUPTS", 500,},
+};
-#define OP_EVENTS_OK 0x0
-#define OP_CTR0_NOT_FOUND 0x1
-#define OP_CTR1_NOT_FOUND 0x2
-#define OP_CTR0_NO_UM 0x4
-#define OP_CTR1_NO_UM 0x8
-#define OP_CTR0_NOT_ALLOWED 0x10
-#define OP_CTR1_NOT_ALLOWED 0x20
-#define OP_CTR0_PII_EVENT 0x40
-#define OP_CTR1_PII_EVENT 0x80
-#define OP_CTR0_PIII_EVENT 0x100
-#define OP_CTR1_PIII_EVENT 0x200
-#define OP_CTR0_ATHLON_EVENT 0x400
-#define OP_CTR1_ATHLON_EVENT 0x800
+#define op_nr_events (sizeof(op_events)/sizeof(op_events[0]))
/**
* op_check_unit_mask - sanity check unit mask value
@@ -279,24 +270,32 @@
return -1;
}
-
-
/**
* op_min_count - get the minimum count value.
* @ctr_type: event value
+ * @cpu_type:
+ *
+ * 0 Pentium Pro
+ *
+ * 1 Pentium II
+ *
+ * 2 Pentium III
*
+ * 3 AMD Athlon
+ *
* The function returns > 0 if the event is found
* 0 otherwise
*/
-int op_min_count(u8 ctr_type) {
+int op_min_count(u8 ctr_type, int cpu_type)
+{
int ret = 0;
- int ctr_e = 0;
uint i;
+ int cpu_mask = 1 << cpu_type;
- for (i = 0; i < op_nr_events && !ctr_e; i++) {
- if (op_events[i].val == ctr_type) {
+ for (i = 0; i < op_nr_events; i++) {
+ if (op_events[i].val == ctr_type && (op_events[i].cpu_mask & cpu_mask)) {
ret = op_events[i].min_count;
- ctr_e = 1;
+ break;
}
}
@@ -305,14 +304,13 @@
/**
* op_check_events - sanity check event values
- * @ctr0_type: event value for counter 0
- * @ctr1_type: event value for counter 1
- * @ctr0_um: unit mask for counter 0
- * @ctr1_um: unit mask for counter 1
- * @proc: processor type
+ * @ctr: counter number
+ * @ctr_type: event value for counter 0
+ * @ctr_um: unit mask for counter 0
+ * @cpu_type: processor type
*
* Check that the counter event and unit mask values
- * are allowed. @proc should be set as follows :
+ * are allowed. @cpu_type should be set as follows :
*
* 0 Pentium Pro
*
@@ -322,116 +320,48 @@
*
* 3 AMD Athlon
*
- * Use 0 values for @ctr0_type and @ctr1_type if the
- * counter is not used.
+ * Don't fail if ctr_type == 0.
*
- * The function returns 1 if the values are allowed,
+ * The function returns bitmask of failure cause
* 0 otherwise
*/
-int op_check_events(u8 ctr0_type, u8 ctr1_type, u8 ctr0_um, u8 ctr1_um, int proc)
+int op_check_events(int ctr, u8 ctr_type, u8 ctr_um, int cpu_type)
{
int ret = 0x0;
- uint i;
- int ctr0_e=0,ctr1_e=0;
-
- if (ctr0_type) {
- for (i=0; i < op_nr_events && !ctr0_e; i++) {
- if (op_events[i].val == ctr0_type) {
- switch (op_events[i].allowed) {
- case OP_1_ONLY:
- ret |= OP_CTR0_NOT_ALLOWED;
- break;
-
- case OP_PII_ONLY:
- if (proc != 1)
- ret |= OP_CTR0_PII_EVENT;
- break;
-
- case OP_PIII_ONLY:
- if (proc != 2)
- ret |= OP_CTR0_PIII_EVENT;
- break;
+ uint i = 0;
+ uint cpu_mask = 1 << cpu_type;
+ uint ctr_mask = 1 << ctr;
- case OP_PII_PIII:
- if (!proc)
- ret |= OP_CTR0_PII_EVENT;
- break;
+ if (ctr_type != 0) {
+ for ( ; i < op_nr_events; i++) {
+ if (op_events[i].val == ctr_type && (op_events[i].cpu_mask & cpu_mask)) {
+ if ((op_events[i].counter_mask & ctr_mask) == 0)
+ ret |= OP_EVT_CTR_NOT_ALLOWED;
- case OP_ATHLON_ONLY:
- if (proc != 3)
- ret |= OP_CTR0_ATHLON_EVENT;
- break;
- default:
- break;
- }
if (op_events[i].unit &&
- op_check_unit_mask(&op_unit_masks[op_events[i].unit], ctr0_um) < 0)
- ret |= OP_CTR0_NO_UM;
- ctr0_e=1;
+ op_check_unit_mask(&op_unit_masks[op_events[i].unit], ctr_um) < 0)
+ ret |= OP_EVT_NO_UM;
break;
}
}
}
- if (ctr1_type) {
- for (i=0; i<op_nr_events && !ctr1_e; i++) {
- if (op_events[i].val == ctr1_type) {
- switch (op_events[i].allowed) {
- case OP_0_ONLY:
- ret |= OP_CTR1_NOT_ALLOWED;
- break;
-
- case OP_PII_ONLY:
- if (proc != 1)
- ret |= OP_CTR1_PII_EVENT;
- break;
-
- case OP_PIII_ONLY:
- if (proc != 2)
- ret |= OP_CTR1_PIII_EVENT;
- break;
-
- case OP_PII_PIII:
- if (!proc)
- ret |= OP_CTR1_PII_EVENT;
- break;
-
- case OP_ATHLON_ONLY:
- if (proc != 3)
- ret |= OP_CTR1_ATHLON_EVENT;
- break;
-
- default:
- break;
- }
- if (op_events[i].unit &&
- op_check_unit_mask(&op_unit_masks[op_events[i].unit], ctr1_um) < 0)
- ret |= OP_CTR1_NO_UM;
- ctr1_e=1;
- }
- }
- }
-
- if (!ctr0_e && ctr0_type)
- ret |= OP_CTR0_NOT_FOUND;
- if (!ctr1_e && ctr1_type)
- ret |= OP_CTR1_NOT_FOUND;
+ if (i == op_nr_events)
+ ret |= OP_EVT_NOT_FOUND;
return ret;
}
/**
* op_check_events_str - sanity check event strings and unit masks
- * @ctr0_type: event name for counter 0
- * @ctr1_type: event name for counter 1
- * @ctr0_um: unit mask for counter 0
- * @ctr1_um: unit mask for counter 1
- * @p2: processor type
- * @ctr0_t: event value for counter 0
- * @ctr1_t: event value for counter 1
+ * @ctr: ctr number
+ * @ctr_type: event name for counter
+ * @ctr_um: unit mask for counter
+ * @cpu_type: processor type
+ * @ctr_t: event value for counter
*
* Check that the counter event and unit mask values
- * are allowed. @p2 should be set as follows :
+ * are allowed. @cpu_type should be set as follows :
*
* 0 Pentium Pro
*
@@ -441,54 +371,38 @@
*
* 3 AMD Athlon
*
- * Use "" strings for @ctr0_type and @ctr1_type if the
- * counter is not used.
+ * Use "" strings for @ctr_type if the counter is not used.
*
- * On successful return, @ctr0_t and @ctr1_t will contain
- * the event values for counters 0 and 1 respectively
+ * On successful return, @ctr_t will contain
+ * the event values for counters ctr
*
* The function returns 1 if the values are allowed,
* 0 otherwise
*/
-int op_check_events_str(char *ctr0_type, char *ctr1_type, u8 ctr0_um, u8 ctr1_um, int p2, u8 *ctr0_t, u8 *ctr1_t)
+int op_check_events_str(int ctr, char *ctr_type, u8 ctr_um, int cpu_type, u8 *ctr_t)
{
uint i;
- int ctr0_e=0;
- int ctr1_e=0;
-
- if (!ctr0_type)
- ctr0_type="";
-
- if (!ctr1_type)
- ctr1_type="";
+ int ctr_found = 0;
+ uint cpu_mask = 1 << cpu_type;
- if (strcmp(ctr0_type,"")) {
- for (i=0; i < op_nr_events && !ctr0_e; i++) {
- if (!strcmp(ctr0_type, op_events[i].name)) {
- ctr0_e = 1;
- *ctr0_t = op_events[i].val;
- break;
- }
- }
- }
+ if (!ctr_type)
+ ctr_type="";
- if (strcmp(ctr1_type,"")) {
- for (i=0; i < op_nr_events && !ctr1_e; i++) {
- if (!strcmp(ctr1_type, op_events[i].name)) {
- ctr1_e = 1;
- *ctr1_t = op_events[i].val;
+ if (strcmp(ctr_type,"")) {
+ for (i=0; i < op_nr_events; i++) {
+ if (!strcmp(ctr_type, op_events[i].name) &&
+ (op_events[i].cpu_mask & cpu_mask)) {
+ *ctr_t = op_events[i].val;
+ ctr_found = 1;
break;
}
}
}
-
- if (strcmp(ctr0_type,"") && !ctr0_e)
- return OP_CTR0_NOT_FOUND;
- if (strcmp(ctr1_type,"") && !ctr1_e)
- return OP_CTR1_NOT_FOUND;
+ if (strcmp(ctr_type,"") && !ctr_found)
+ return OP_EVT_NOT_FOUND;
- return op_check_events(*ctr0_t, *ctr1_t, ctr0_um, ctr1_um, p2);
+ return op_check_events(ctr, *ctr_t, ctr_um, cpu_type);
}
#ifdef OP_EVENTS_DESC
@@ -498,11 +412,11 @@
static struct op_unit_desc op_unit_descs[] = {
{ { NULL, NULL, NULL, NULL, NULL, NULL, NULL, }, },
- { { "(I)nvalid cache state",
- "(S)hared cache state",
+ { { "(M)odified cache state",
"(E)xclusive cache state",
- "(M)odified cache state",
- "MESI cache state", NULL, NULL, }, },
+ "(S)hared cache state",
+ "(I)nvalid cache state",
+ "all MESI cache state", NULL, NULL, }, },
{ { "self-generated transactions",
"any transactions", NULL, NULL, NULL, NULL, NULL, }, },
{ { "mandatory", NULL, NULL, NULL, NULL, NULL, NULL, }, },
@@ -528,10 +442,16 @@
"prefetch T2",
"weakly ordered stores", NULL, NULL, NULL, }, },
{ { "packed and scalar", "packed", NULL, NULL, NULL, NULL, NULL, }, },
-
+ { { "(M)odified cache state",
+ "(O)wner cache state",
+ "(E)xclusive cache state",
+ "(S)hared cache state",
+ "(I)nvalid cache state",
+ "all MOESI cache state", NULL, }, },
};
static char *op_event_descs[] = {
+ "clocks processor is not halted",
/* Data Cache Unit (DCU) */
"all memory references, cachable and non",
"total lines allocated in the DCU",
@@ -616,7 +536,6 @@
"cycles or events for partial stalls",
/* Segment Register Loads */
"number of segment register loads",
- "clocks processor is not halted",
/* MMX (Pentium II only) */
"number of MMX instructions executed",
"number of MMX saturating instructions executed",
@@ -629,6 +548,31 @@
"number of segment register renaming stalls",
"number of segment register renames",
"number of segment register rename events retired",
+ /* Athlon/Duron */
+ /* Cache */
+ "Data cache accesses",
+ "Data cache misses",
+ "Data cache refills from L2",
+ "Data cache refills from system",
+ "Data cache write backs",
+ "L1 DTLB misses and L2 DTLB hits",
+ "L1 and L2 DTLB misses",
+ "Misaligned data references",
+ "Instruction cache fetches)",
+ "Instruction cache misses)",
+ "L1 ITLB misses (and L2 ITLB hits)",
+ "(L1 and) L2 ITLB misses",
+ "Retired instructions (includes exceptions, interrupts, resyncs)",
+ "Retired Ops",
+ "Retired branches (conditional, unconditional, exceptions, interrupts)",
+ "Retired branches mispredicted",
+ "Retired taken branches",
+ "Retired taken branches mispredicted",
+ "Retired far control transfers",
+ "Retired resync branches (only non-control transfer branches counted)",
+ "Interrupts masked cycles (IF=0)",
+ "Interrupts masked while pending cycles (INTR while IF=0)",
+ "Number of taken hardware interrupts",
};
/**
@@ -658,9 +602,7 @@
if (um_mask_desc_index == -1)
return NULL;
else if (um_mask_desc_index == 0) {
- /* FIXME: Perhaps needs to combine the different string? (this
- * needs dynamic alloc and updating the code/comment of callers)
- */
+ /* avoid dynamic alloc to simplify caller's life */
return "set with multiple units, check the documentation";
}
@@ -669,6 +611,7 @@
/**
* op_get_event_desc - get event name and description
+ * @cpu_type: the cpu_type
* @type: event value
* @um: unit mask
* @typenamep: returned event name string
@@ -676,7 +619,7 @@
* @umdescp: returned unit mask description string
*
* Get the associated event name and descriptions given
- * the event value and unit mask value. It is a fatal error
+ * the cpu type, event value and unit mask value. It is a fatal error
* to supply a non-valid @type value, but an invalid @um
* will not exit.
*
@@ -685,14 +628,15 @@
* NULL when @um is invalid for the given @type value.
* These strings are in text section so should not be freed.
*/
-void op_get_event_desc(u8 type, u8 um, char **typenamep, char **typedescp, char **umdescp)
+void op_get_event_desc(int cpu_type, u8 type, u8 um, char **typenamep, char **typedescp, char **umdescp)
{
uint i;
+ int cpu_mask = 1 << cpu_type;
*typenamep = *typedescp = *umdescp = NULL;
for (i=0; i < op_nr_events; i++) {
- if (op_events[i].val == type) {
+ if (op_events[i].val == type && (op_events[i].cpu_mask & cpu_mask)) {
*typenamep = (char *)op_events[i].name;
*typedescp = op_event_descs[i];
@@ -713,49 +657,171 @@
#include "version.h"
-int main(int argc, char *argv[])
+static int cpu_type = DEFAULT_CPU_TYPE;
+
+static const char* cpu_type_str[MAX_CPU_TYPE] = {
+ "Pentium Pro",
+ "PII",
+ "PIII",
+ "Athlon"
+};
+
+/**
+ * help_for_event - output event name and description
+ * @i: event number
+ *
+ * output an help string for the event @i
+ */
+static void help_for_event(int i)
{
- uint i;
- uint j;
+ uint k, j;
+ uint mask;
- if (argc>1) {
- if ((!strcmp(argv[1],"--version") || !strcmp(argv[1],"-v"))) {
- printf(VERSION_STRING " compiled on " __DATE__ " " __TIME__ "\n");
- exit(0);
+ printf("%s", op_events[i].name);
+
+ printf(": (counter: ");
+ if (op_events[i].counter_mask == CTR_ALL) {
+ printf("all");
+ } else {
+ mask = op_events[i].counter_mask;
+ for (k = 0; k < CHAR_BIT * sizeof(op_events[i].counter_mask); ++k) {
+ if (mask & (1 << k)) {
+ printf("%d", k);
+ mask &= ~(1 << k);
+ if (mask)
+ printf(", ");
+ }
}
+ }
+ printf(")");
- /* interpret given string */
+ printf(" (supported cpu: ");
+ mask = op_events[i].cpu_mask;
+ for (k = 0; k < MAX_CPU_TYPE; ++k) {
+ if (mask & (1 << k)) {
+ printf("%s", cpu_type_str[k]);
+ mask &= ~(1 << k);
+ if (mask)
+ printf(", ");
+ }
+ }
- for (i=0; i < op_nr_events; i++) {
- if (!strcmp(op_events[i].name, argv[1]))
- printf("%d\n", op_events[i].val);
+ printf(")\n\t%s (min count: %d)\n", op_event_descs[i], op_events[i].min_count);
+
+ if (op_events[i].unit) {
+ int unit_idx = op_events[i].unit;
+
+ printf("\tUnit masks\n");
+ printf("\t----------\n");
+
+ for (j=0; j < op_unit_masks[unit_idx].num; j++) {
+ printf("\t%.2x: %s\n",
+ op_unit_masks[unit_idx].um[j],
+ op_unit_descs[unit_idx].desc[j]);
}
- return 0;
}
+}
- printf("oprofile: available events\n");
- printf("--------------------------\n\n");
- printf("See Intel Architecture Developer's Manual\nVol. 3, Appendix A\n\n");
+/**
+ * help_event_for_gui - output event name and description
+ * @i: event number
+ *
+ * output an help string for the event @i, in
+ * a manner adapted for use by the oprofile gui
+ */
+static void help_event_for_gui(int i)
+{
+ uint j;
- for (i=0; i < op_nr_events; i++) {
- printf("%s:", op_events[i].name);
- printf("\n\t%s ", op_event_descs[i]);
- switch (op_events[i].allowed) {
- case 0: printf("- counter 0 only\n"); break;
- case 1: printf("- counter 1 only\n"); break;
- case 3: printf("- Pentium II/III only\n"); break;
- case 4: printf("- Pentium II only\n"); break;
- case 5: printf("- Pentium III only\n"); break;
- case 6: printf("- AMD Athlon only\n"); break;
- default: printf("\n"); break;
+ printf("{ %s", op_events[i].name);
+
+ printf(" %d %d", op_events[i].cpu_mask, op_events[i].counter_mask);
+
+ printf(" \"%s\" %d", op_event_descs[i], op_events[i].min_count);
+
+ if (op_events[i].unit) {
+ int unit_idx = op_events[i].unit;
+
+ printf("\n");
+
+ switch (op_unit_masks[unit_idx].unit_type_mask) {
+ case utm_mandatory:
+ printf("\tmandatory ");
+ break;
+ case utm_exclusive:
+ printf("\texclusive ");
+ break;
+ case utm_bitmask:
+ printf("\tbitmask ");
+ break;
}
- if (op_events[i].unit) {
- printf("\tUnit masks\n");
- printf("\t----------\n");
- for (j=0; j < op_unit_masks[op_events[i].unit].num; j++) {
- printf("\t%.2x: %s\n",
- op_unit_masks[op_events[i].unit].um[j],
- op_unit_descs[op_events[i].unit].desc[j]);
+
+ printf("%d\n", op_unit_masks[unit_idx].default_mask);
+ printf("\t{\n");
+
+ for (j=0; j < op_unit_masks[unit_idx].num; j++) {
+ printf("\t { %d \"%s\" }\n",
+ op_unit_masks[unit_idx].um[j],
+ op_unit_descs[unit_idx].desc[j]);
+ }
+
+ printf("\t}\n");
+ }
+
+ printf(" }\n");
+}
+
+int main(int argc, char *argv[])
+{
+ int i;
+ uint j;
+ int cpu_type_mask;
+ int for_gui;
+
+ for_gui = 0;
+ for (i = 1 ; i < argc ; ++i) {
+ if (!strcmp(argv[i], "--version")) {
+ printf(VERSION_STRING " compiled on " __DATE__ " " __TIME__ "\n");
+ return 0;
+ } else if (!strcmp(argv[i], "--help")) {
+ printf("op_help [--version|--cpu-type] event_name");
+
+ return 0;
+ } else if (!strncmp(argv[i], "--cpu-type=", 11)) {
+ sscanf(argv[i] + 11, "%d", &cpu_type);
+ if (cpu_type < 0 || cpu_type >= MAX_CPU_TYPE) {
+ fprintf(stderr, "invalid cpu type %d, default to to %s\n", cpu_type, cpu_type_str[DEFAULT_CPU_TYPE]);
+ cpu_type = DEFAULT_CPU_TYPE;
+ }
+ } else if (!strcmp(argv[1], "--gui-description")) {
+ for_gui = 1;
+ } else {
+ cpu_type_mask = 1 << cpu_type;
+ for (j=0; j < op_nr_events; j++) {
+ if (!strcmp(op_events[j].name, argv[i]) &&
+ (op_events[j].cpu_mask & cpu_type_mask)) {
+ printf("%d\n", op_events[j].val);
+ }
+ }
+ return 0;
+ }
+ }
+
+ if (for_gui == 0) {
+ printf("oprofile: available events\n");
+ printf("--------------------------\n\n");
+ printf("See Intel Architecture Developer's Manual\nVol. 3, Appendix A\n\n");
+ } else {
+ printf("{ DISABLED -1 -1 \"Select it to disable this counter\" 0 }\n");
+ }
+
+ cpu_type_mask = 1 << cpu_type;
+ for (j=0; j < op_nr_events; j++) {
+ if ((op_events[j].cpu_mask & cpu_type_mask) != 0) {
+ if (for_gui) {
+ help_event_for_gui(j);
+ } else {
+ help_for_event(j);
}
}
}
Index: op_user.h
===================================================================
RCS file: /cvsroot/oprofile/oprofile/op_user.h,v
retrieving revision 1.3
retrieving revision 1.4
diff -u -d -r1.3 -r1.4
--- op_user.h 2001/06/22 03:16:24 1.3
+++ op_user.h 2001/09/01 02:03:34 1.4
@@ -22,6 +22,31 @@
#include "version.h"
+/* PHE FIXME: op_config.h, typedef it and prefix by op_ ? */
+#define u8 unsigned char
+#define u16 unsigned short
+#define u32 unsigned int
+#define uint unsigned int
+#define ulong unsigned long
+#define fd_t int
+
+/* event check returns */
+#define OP_EVENTS_OK 0x0
+#define OP_EVT_NOT_FOUND 0x1
+#define OP_EVT_NO_UM 0x2
+#define OP_EVT_CTR_NOT_ALLOWED 0x4
+
+/* supported cpu type */
+#define CPU_NO_GOOD -1
+#define CPU_PPRO 0
+#define CPU_PII 1
+#define CPU_PIII 2
+#define CPU_ATHLON 3
+#define MAX_CPU_TYPE 4
+
+/* default value must be set by configure */
+#define DEFAULT_CPU_TYPE CPU_PII
+
#ifndef NR_CPUS
#define NR_CPUS 32
#endif
@@ -34,15 +59,47 @@
/* 65536 * 32 = 2097152 bytes default */
#define OP_DEFAULT_HASH_SIZE 65536
-#define OP_BITS 2
+/* kernel image entries are offset by this many entries */
+#define OPD_KERNEL_OFFSET 524288
+
+/* FIXME ATHLON: four next #define can be tuned for full support to Athlon */
+/* PHE: I don't fix the problem "is these should be autoconf and/or runtime
+ * settings" so hack it at hand for now */
+/* maximum nr. of counters, up to 4 for Athlon (18 for P4). The primary use
+ * of this variable is for static/local array dimension. Never use it in loop
+ * or in array index acccess/index checking. Don't change it without updating
+ * OP_BITS_CTR! */
+#define OP_MAX_COUNTERS 2
+
+/* The real number of counters supported for the running cpu, I define it
+ * separately to allow using a variable rather a constant so distribution can
+ * provide a generic binary for oprofile. This need also a few additional
+ * fixes */
+#define op_nr_counters 2
+
+/* IA32 have only one bit to start/stop all counter, AMD have separate bit
+ * inside each event select MSR */
+#define separate_running_bit 0
+
+/* the number of bits neccessary to store OP_MAX_COUNTERS values */
+#define OP_BITS_CTR 1
+
+/* the number of reserved bits in count, + 1 is for the notification bit */
+#define OP_BITS (OP_BITS_CTR + 1)
+
+/* The number of bits available to store count */
+#define OP_BITS_COUNT (16 - OP_BITS)
+
+/* counter nr mask */
+#define OP_CTR_MASK ((~0U << OP_BITS_COUNT) >> 1)
/* top OP_BITS bits of count are used as follows: */
/* is this actually a notification ? */
-#define OP_NOTE (1U<<15)
+#define OP_NOTE (1U << (OP_BITS_COUNT + OP_BITS_CTR))
/* which perf counter the sample is from */
-#define OP_COUNTER (1U<<14)
+#define OP_COUNTER(x) (((x) & OP_CTR_MASK) >> OP_BITS_COUNT)
-#define OP_COUNT_MASK ((1U << (16 - OP_BITS)) - 1U)
+#define OP_COUNT_MASK ((1U << OP_BITS_COUNT) - 1U)
/* mapping notification types */
/* fork(),vfork(),clone() */
@@ -103,5 +160,12 @@
/* size of hash map in bytes */
#define OP_HASH_MAP_SIZE (OP_HASH_MAP_NR * sizeof(struct op_hash))
+
+/* op_events.c */
+int op_min_count(u8 ctr_type, int cpu_type);
+int op_check_events(int ctr, u8 ctr_type, u8 ctr_um, int cpu_type);
+/* not used currently */
+int op_check_events_str(int ctr, char *ctr_type, u8 ctr_um, int cpu_type, u8 *ctr_t);
+void op_get_event_desc(int cpu_type, u8 type, u8 um, char **typenamep, char **typedescp, char **umdescp);
#endif /* OP_USER_H */
Index: oprofile.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/oprofile.c,v
retrieving revision 1.75
retrieving revision 1.76
diff -u -d -r1.75 -r1.76
--- oprofile.c 2001/08/31 17:16:35 1.75
+++ oprofile.c 2001/09/01 02:03:34 1.76
@@ -39,14 +39,8 @@
pid_t pgrp_filter;
/* the MSRs we need */
-static u32 ctrlreg0 = MSR_IA32_EVNTSEL0;
-static u32 ctrlreg1 = MSR_IA32_EVNTSEL1;
-static u32 ctrlreg2 = MSR_K7_PERFCTL2;
-static u32 ctrlreg3 = MSR_K7_PERFCTL3;
-static u32 ctrreg0 = MSR_IA32_PERFCTR0;
-static u32 ctrreg1 = MSR_IA32_PERFCTR1;
-static u32 ctrreg2 = MSR_K7_PERFCTR2;
-static u32 ctrreg3 = MSR_K7_PERFCTR3;
+static uint perfctr_msr[OP_MAX_COUNTERS];
+static uint eventsel_msr[OP_MAX_COUNTERS];
static u32 prof_on __cacheline_aligned;
@@ -80,7 +74,7 @@
{
ops->eip = regs->eip;
ops->pid = current->pid;
- ops->count = OP_COUNTER*ctr + 1;
+ ops->count = (1U << OP_BITS_COUNT)*ctr + 1;
}
inline static void op_do_profile(struct _oprof_data *data, struct pt_regs *regs, int ctr)
@@ -123,16 +117,16 @@
asmlinkage void op_do_nmi(struct pt_regs *regs)
{
struct _oprof_data *data = &oprof_data[smp_processor_id()];
+ int i;
if (pid_filter && current->pid != pid_filter)
return;
if (pgrp_filter && current->pgrp != pgrp_filter)
return;
- if (data->ctrs & OP_CTR_0)
- op_check_ctr(data, regs, 0);
- if (data->ctrs & OP_CTR_1)
- op_check_ctr(data, regs, 1);
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ op_check_ctr(data, regs, i);
+ }
}
/* ---------------- NMI handler setup ------------ */
@@ -202,7 +196,7 @@
/* PHE I think when the doc says : -if you disable the apic the bits
* of LVT cannot be reset- it talk about the SW disable through bit 8
* of SPIV see 7.4.14 (7.5.14) not the hardware disable, so it is ok
- * but I perhaps we need a software disable of the APIC at the end
+ * but perhaps we need a software disable of the APIC at the end
*/
/* first disable via MSR */
/* IA32 V3, 7.4.2 */
@@ -296,12 +290,6 @@
printk(KERN_INFO "oprofile: setting up APIC.\n");
/* ugly hack */
- /* PHE : the memory must be uncachable! this is perhaps more a
- * problem for Athlon than for PII.
- * Also this stuff probably mishandled SMP ?
- * The _PAGE_GLOBAL attribute should be probably avoided but make
- * this later
- */
my_set_fixmap();
/* FIXME: NMI delivery for SMP ? */
@@ -419,80 +407,150 @@
static void pmc_setup(void *dummy)
{
uint low, high;
+ int i;
// first, let's use the right MSRs
switch (cpu_type) {
case CPU_ATHLON:
- ctrlreg0 = MSR_K7_PERFCTL0;
- ctrlreg1 = MSR_K7_PERFCTL1;
- ctrreg0 = MSR_K7_PERFCTR0;
- ctrreg1 = MSR_K7_PERFCTR1;
+ eventsel_msr[0] = MSR_K7_PERFCTL0;
+ eventsel_msr[1] = MSR_K7_PERFCTL1;
+/* FIXME ATHLON enable only with OP_MAX_COUNTERS 4 and an athlon to test this
+ eventsel_msr[2] = MSR_K7_PERFCTL2;
+ eventsel_msr[3] = MSR_K7_PERFCTL3;
+*/
+ perfctr_msr[0] = MSR_K7_PERFCTR0;
+ perfctr_msr[1] = MSR_K7_PERFCTR1;
+/* FIXME ATHLON ditto
+ perfctr_msr[2] = MSR_K7_PERFCTR2;
+ perfctr_msr[3] = MSR_K7_PERFCTR3;
+*/
break;
- default:;
+ default:
+ eventsel_msr[0] = MSR_IA32_EVNTSEL0;
+ eventsel_msr[1] = MSR_IA32_EVNTSEL1;
+ perfctr_msr[0] = MSR_IA32_PERFCTR0;
+ perfctr_msr[1] = MSR_IA32_PERFCTR1;
+ break;
}
-
- rdmsr(ctrlreg0, low, high);
- // FIXME: enable bit is per-counter on athlon
- wrmsr(ctrlreg0, low & ~(1<<22), high);
+ /* FIXME ATHLON: use #if 0 to get the old code, which would work on
+ * athlon. The new code would work but I had get a painfull debug so
+ * I prefer to keep the old code until Athlon tests are available. */
+#if 1
/* IA Vol. 3 Figure 15-3 */
- rdmsr(ctrlreg0, low, high);
+ /* Stop and clear all counter: IA32 use bit 22 of eventsel_msr0 to
+ * enable/disable all counter, AMD use separate bit 22 in each msr,
+ * all other bits are cleared except the reserved bits 21 */
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ rdmsr(eventsel_msr[i], low, high);
+ wrmsr(eventsel_msr[i], low & ~(3 << 22), high);
+ }
+
+ /* setup each counter */
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ if (op_ctr_val[i]) {
+ rdmsr(eventsel_msr[i], low, high);
+
+ low &= 1 << 21; /* do not touch the reserved bit */
+ set_perfctr(op_ctr_count[i], i);
+
+ pmc_fill_in(&low, op_ctr_kernel[i], op_ctr_user[i],
+ op_ctr_val[i], op_ctr_um[i]);
+
+ wrmsr(eventsel_msr[i], low, high);
+ }
+ }
+
+ /* Here all setup is made except the start/stop bit 21), counter
+ * disabled contains zeros in the eventsel msr except the reserved bit
+ * 21 */
+#else
+ rdmsr(eventsel_msr[0], low, high);
+ // FIXME: enable bit is per-counter on athlon
+ wrmsr(eventsel_msr[0], low & ~(1<<22), high);
/* clear */
low &= (1<<21);
+ /* IA Vol. 3 Figure 15-3 */
+
if (op_ctr_val[0]) {
set_perfctr(op_ctr_count[0], 0);
pmc_fill_in(&low, op_ctr_kernel[0], op_ctr_user[0], op_ctr_val[0], op_ctr_um[0]);
}
- wrmsr(ctrlreg0, low, 0);
+ wrmsr(eventsel_msr[0], low, 0);
- rdmsr(ctrlreg1, low, high);
+ rdmsr(eventsel_msr[1], low, high);
/* clear */
low &= (3<<21);
if (op_ctr_val[1]) {
set_perfctr(op_ctr_count[1], 1);
pmc_fill_in(&low, op_ctr_kernel[1], op_ctr_user[1], op_ctr_val[1], op_ctr_um[1]);
- wrmsr(ctrlreg1, low, high);
+
+ wrmsr(eventsel_msr[1], low, high);
}
+#endif
/* disable ctr1 if the UP oopser might be on, but we can't do anything
* interesting with the NMIs
*/
- /* PHE FIXME Have always a meaning ? */
+ /* PHE FIXME ATHLON bugged if running 3/4 counters ?, I really don't
+ * understand when and why this is executed */
/* this is pretty bogus really. especially as we don't re-enable it.
* Instead, save state set up, and restore with pmc_unsetup or similar */
#if !defined(CONFIG_X86_UP_APIC) || !defined(OP_EXPORTED_DO_NMI)
- wrmsr(ctrlreg1, low, high);
+ wrmsr(eventsel_msr[1], low, high);
#endif
}
static void pmc_start(void *info)
{
uint low,high;
+ int i;
if (info && (*((uint *)info) != smp_processor_id()))
return;
- /* enable counters */
- rdmsr(ctrlreg0, low, high);
- // FIXME: bit 22 per-counter on athlon
- wrmsr(ctrlreg0, low | (1<<22), high);
+ /* assert: all enable counter are setup except the bit start/stop,
+ * all counter disable contains zeroes (except perhaps the reserved
+ * bit 21) */
+
+ /* enable all needed counter */
+ if (separate_running_bit == 0) {
+ rdmsr(eventsel_msr[0], low, high);
+ wrmsr(eventsel_msr[0], low | (1 << 22), high);
+ } else {
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ if (op_ctr_count[i]) {
+ rdmsr(eventsel_msr[i], low, high);
+ wrmsr(eventsel_msr[i], low | (1 << 22), high);
+ }
+ }
+ }
}
static void pmc_stop(void *info)
{
uint low,high;
+ int i;
if (info && (*((uint *)info) != smp_processor_id()))
return;
/* disable counters */
- rdmsr(ctrlreg0, low, high);
- // FIXME: bit 22 per-counter on athlon
- wrmsr(ctrlreg0, low & ~(1<<22), high);
+ if (separate_running_bit == 0) {
+ rdmsr(eventsel_msr[0], low, high);
+ wrmsr(eventsel_msr[0], low & ~(1 << 22), high);
+ } else {
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ if (op_ctr_count[i]) {
+ rdmsr(eventsel_msr[i], low, high);
+ wrmsr(eventsel_msr[i], low & ~(1 << 22), high);
+ }
+ }
+ }
}
inline static void pmc_select_start(uint cpu)
@@ -784,6 +842,11 @@
return -EFAULT;
}
+ /* PHE FIXME: this is inefficient if only few counter are
+ * enable. Some other place in the code must be fixed if we
+ * allocate only on demand, take care than note currently are
+ * always put in data[0]->buffer even if the counter 0 is
+ * disable */
data->buffer = vmalloc(buf_size);
if (!data->buffer) {
printk(KERN_ERR "oprofile: failed to allocate eviction buffer of %lu bytes\n",buf_size);
@@ -807,61 +870,71 @@
int ret;
int i;
uint cpu;
+ int ok;
struct _oprof_data *data;
+ int enabled;
op_check_range(op_hash_size, 256, 262144, "op_hash_size value %d not in range (%d %d)\n");
op_check_range(op_buf_size, 1024, 1048576, "op_buf_size value %d not in range (%d %d)\n");
- /* FIXME: hardcoded 2 counters */
- for (i = 0; i < 2; i++) {
+ enabled = 0;
+ for (i = 0; i < op_nr_counters ; i++) {
if (op_ctr_on[i]) {
- int min_count = op_min_count(op_ctr_val[i]);
+ int min_count = op_min_count(op_ctr_val[i], cpu_type);
if (!op_ctr_user[i] && !op_ctr_kernel[i]) {
printk(KERN_ERR "oprofile: neither kernel nor user set for counter %d\n", i);
return 0;
}
op_check_range(op_ctr_count[i], min_count, OP_MAX_PERF_COUNT, "ctr count value %d not in range (%d %ld)\n");
+
+ enabled = 1;
}
}
- /* FIXME: below needs generalising for multiple counters */
-
- if (!op_ctr_on[0] && !op_ctr_on[1]) {
+ if (!enabled) {
printk(KERN_ERR "oprofile: neither counter enabled.\n");
return 0;
}
/* hw_ok() has set cpu_type */
- ret = op_check_events(op_ctr_val[0], op_ctr_val[1], op_ctr_um[0], op_ctr_um[1], cpu_type);
+ ok = 1;
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ ret = op_check_events(i, op_ctr_val[i], op_ctr_um[i], cpu_type);
- if (ret & OP_CTR0_NOT_FOUND) printk(KERN_ERR "oprofile: ctr0: no such event\n");
- if (ret & OP_CTR1_NOT_FOUND) printk(KERN_ERR "oprofile: ctr1: no such event\n");
- if (ret & OP_CTR0_NO_UM) printk(KERN_ERR "oprofile: ctr0: invalid unit mask\n");
- if (ret & OP_CTR1_NO_UM) printk(KERN_ERR "oprofile: ctr1: invalid unit mask\n");
- if (ret & OP_CTR0_NOT_ALLOWED) printk(KERN_ERR "oprofile: ctr0 can't count this event\n");
- if (ret & OP_CTR1_NOT_ALLOWED) printk(KERN_ERR "oprofile: ctr1 can't count this event\n");
- if (ret & OP_CTR0_PII_EVENT) printk(KERN_ERR "oprofile: ctr0: event only available on PII\n");
- if (ret & OP_CTR1_PII_EVENT) printk(KERN_ERR "oprofile: ctr1: event only available on PII\n");
- if (ret & OP_CTR0_PIII_EVENT) printk(KERN_ERR "oprofile: ctr0: event only available on PIII\n");
- if (ret & OP_CTR1_PIII_EVENT) printk(KERN_ERR "oprofile: ctr1: event only available on PIII\n");
- if (ret & OP_CTR0_ATHLON_EVENT) printk(KERN_ERR "oprofile: ctr1: event only available on Athlon\n");
- if (ret & OP_CTR1_ATHLON_EVENT) printk(KERN_ERR "oprofile: ctr1: event only available on Athlon\n");
+ if (ret & OP_EVT_NOT_FOUND)
+ printk(KERN_ERR "oprofile: ctr%d: %d: no such event for cpu %d\n",
+ i, op_ctr_val[i], cpu_type);
- if (ret)
+ if (ret & OP_EVT_NO_UM)
+ printk(KERN_ERR "oprofile: ctr%d: 0x%.2x: invalid unit mask for cpu %d\n",
+ i, op_ctr_um[i], cpu_type);
+
+ if (ret & OP_EVT_CTR_NOT_ALLOWED)
+ printk(KERN_ERR "oprofile: ctr%d: %d: can't count event for this counter\n",
+ i, op_ctr_val[i]);
+
+ if (ret != OP_EVENTS_OK)
+ ok = 0;
+ }
+
+ if (!ok)
return 0;
for (cpu=0; cpu < smp_num_cpus; cpu++) {
data = &oprof_data[cpu];
- data->ctrs |= OP_CTR_0 * (!!op_ctr_on[0]);
- data->ctrs |= OP_CTR_1 * (!!op_ctr_on[1]);
/* make sure the buffer and hash table have been set up */
if (!data->buffer || !data->entries)
return 0;
- data->ctr_count[0] = op_ctr_count[0];
- data->ctr_count[1] = op_ctr_count[1];
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ if (op_ctr_on[i]) {
+ data->ctr_count[i] = op_ctr_count[i];
+ } else {
+ data->ctr_count[i] = 0;
+ }
+ }
}
return 1;
@@ -1055,7 +1128,7 @@
return err;
}
-static int nr_oprof_static = 6;
+static int nr_oprof_static = 7;
static ctl_table oprof_table[] = {
{ 1, "bufsize", &op_buf_size, sizeof(int), 0600, NULL, &lproc_dointvec, NULL, },
@@ -1064,6 +1137,7 @@
{ 1, "kernel_only", &kernel_only, sizeof(int), 0600, NULL, &lproc_dointvec, NULL, },
{ 1, "pid_filter", &pid_filter, sizeof(pid_t), 0600, NULL, &lproc_dointvec, NULL, },
{ 1, "pgrp_filter", &pgrp_filter, sizeof(pid_t), 0600, NULL, &lproc_dointvec, NULL, },
+ { 1, "cpu_type", &cpu_type, sizeof(int), 0400, NULL, &lproc_dointvec, NULL, },
{ 0, }, { 0, }, { 0, }, { 0, },
{ 0, },
};
@@ -1088,14 +1162,13 @@
static int __init init_sysctl(void)
{
ctl_table *next = &oprof_table[nr_oprof_static];
- ctl_table *counter_table[OP_MAX_COUNTERS + 1];
+ ctl_table *counter_table[op_nr_counters + 1];
ctl_table *tab;
int i,j;
/* FIXME: no proper numbers, or verifiers (where possible) */
- /* for each counter - FIXME: hardcoded to 2 counters currently */
- for (i=0; i < 2; i++) {
+ for (i=0; i < op_nr_counters; i++) {
next->ctl_name = 1;
next->procname = names[i];
next->mode = 0700;
Index: oprofile.h
===================================================================
RCS file: /cvsroot/oprofile/oprofile/oprofile.h,v
retrieving revision 1.47
retrieving revision 1.48
diff -u -d -r1.47 -r1.48
--- oprofile.h 2001/08/31 17:16:35 1.47
+++ oprofile.h 2001/09/01 02:03:34 1.48
@@ -47,21 +47,11 @@
struct op_sample *buffer; /* eviction buffer */
uint hash_size; /* nr. in hash table */
uint buf_size; /* nr. in buffer */
- uint ctr_count[2]; /* reset counter values */
uint nextbuf; /* next in buffer (atomic) */
- u16 next; /* next sample in entry */
- u16 ctrs; /* which counters are set */
+ uint next; /* next sample in entry */
+ uint ctr_count[OP_MAX_COUNTERS]; /* reset counter values */
} __attribute__((__aligned__(SMP_CACHE_BYTES)));
-#define OP_CTR_0 0x1
-#define OP_CTR_1 0x2
-
-#define CPU_NO_GOOD -1
-#define CPU_PPRO 0
-#define CPU_PII 1
-#define CPU_PIII 2
-#define CPU_ATHLON 3
-
/* MSRs */
#ifndef MSR_IA32_PERFCTR0
#define MSR_IA32_PERFCTR0 0xc1
@@ -109,9 +99,6 @@
#define streqn(a, b, len) (!strncmp((a), (b), (len)))
-/* maximum nr. of counters, up to 4 for Athlon (18 for P4) */
-#define OP_MAX_COUNTERS 4
-
/* oprof_data->ready will be set this many samples
* before the end of the eviction buffer.
* The purpose of this is to avoid overflowing the sample
@@ -146,25 +133,11 @@
((((((eip&0xff000)>>3) ^ eip) ^ (pid&0xff)) ^ (eip<<9)) \
^ (ctr<<8)) & (data->hash_size - 1)
-/* relying on MSR numbers being neighbours */
-#define get_perfctr(l,h,c) do { rdmsr(ctrreg0 + c, (l), (h)); } while (0)
-#define set_perfctr(l,c) do { wrmsr(ctrreg0 + c, -(u32)(l), -1); } while (0)
+/* do not rely on MSR numbers being neighbours */
+#define get_perfctr(l,h,c) do { rdmsr(perfctr_msr[(c)], (l), (h)); } while (0)
+#define set_perfctr(l,c) do { wrmsr(perfctr_msr[(c)], -(u32)(l), -1); } while (0)
#define ctr_overflowed(n) (!((n) & (1U<<31)))
-#define OP_EVENTS_OK 0x0
-#define OP_CTR0_NOT_FOUND 0x1
-#define OP_CTR1_NOT_FOUND 0x2
-#define OP_CTR0_NO_UM 0x4
-#define OP_CTR1_NO_UM 0x8
-#define OP_CTR0_NOT_ALLOWED 0x10
-#define OP_CTR1_NOT_ALLOWED 0x20
-#define OP_CTR0_PII_EVENT 0x40
-#define OP_CTR1_PII_EVENT 0x80
-#define OP_CTR0_PIII_EVENT 0x100
-#define OP_CTR1_PIII_EVENT 0x200
-#define OP_CTR0_ATHLON_EVENT 0x400
-#define OP_CTR1_ATHLON_EVENT 0x800
-
#define op_check_range(val,l,h,str) do { \
if ((val) < (l) || (val) > (h)) { \
printk(str, (val), (l), (h)); \
@@ -218,22 +191,14 @@
#include <linux/completion.h>
#endif
-// FIXME: these names are too easy to mis-type !
-extern uint ctrlreg0;
-extern uint ctrlreg1;
-extern uint ctrlreg2;
-extern uint ctrlreg3;
-extern uint ctrreg0;
-extern uint ctrreg1;
-extern uint ctrreg2;
-extern uint ctrreg3;
+/* These arrays are filled by hw_ok() */
+extern uint perfctr_msr[OP_MAX_COUNTERS];
+extern uint eventsel_msr[OP_MAX_COUNTERS];
int oprof_init(void);
int find_intel_smp(void);
void oprof_exit(void);
void my_set_fixmap(void);
-int op_min_count(u8 ctr_type);
-int op_check_events(u8 ctr0_type, u8 ctr1_type, u8 ctr0_um, u8 ctr1_um, int proc);
void op_intercept_syscalls(void);
void op_replace_syscalls(void);
void op_save_syscalls(void);
|
Update of /cvsroot/oprofile/oprofile/pp
In directory usw-pr-cvs1:/tmp/cvs-serv19169/pp
Modified Files:
Makefile.in op_to_source opf_container.cpp opf_filter.cpp
opf_filter.h oprof_convert.c oprofpp.c oprofpp.h
Log Message:
philippe's huge patch
Index: Makefile.in
===================================================================
RCS file: /cvsroot/oprofile/oprofile/pp/Makefile.in,v
retrieving revision 1.14
retrieving revision 1.15
diff -u -d -r1.14 -r1.15
--- Makefile.in 2001/08/18 16:11:17 1.14
+++ Makefile.in 2001/09/01 02:03:34 1.15
@@ -15,19 +15,19 @@
install: $(INSTALL_LIST)
-$(MKDIR_P) $(BINDIR)
- @for f in $(INSTALL_LIST); do \
+ for f in $(INSTALL_LIST); do \
cp $$f $(BINDIR)/$$f && chmod 755 $(BINDIR)/$$f; \
done
uninstall:
- @for f in $(INSTALL_LIST); do \
- rm -f $(BINDIR)/$$f; \
+ for f in $(INSTALL_LIST); do \
+ -rm -f $(BINDIR)/$$f; \
done
CFLAGS=-Wall -Wstrict-prototypes -W -pipe -DOP_EVENTS_DESC -L/usr/lib @CFLAGS@
LIBS=-lpopt -lbfd -liberty -ldl
INCLUDES=@extra_includes@
-HEADERS=oprofpp.h ../dae/opd_util.h
+HEADERS=oprofpp.h ../dae/opd_util.h ../op_user.h
MY_SOURCES=oprofpp.c
SOURCES=$(MY_SOURCES) ../dae/opd_util.c ../op_events.c
@@ -40,17 +40,17 @@
$(CC) $(CFLAGS) $(INCLUDES) -o $@ $(SOURCES) $(LIBS)
oprof_convert: oprof_convert.c ../dae/opd_util.c ../dae/opd_util.h
- $(CC) oprof_convert.c ../dae/opd_util.c -Wall -o $@
+ $(CC) -Wall oprof_convert.c ../dae/opd_util.c -o $@
#CXX_DEBUG_LIBS = -L/usr/src/STLport-4.0/lib -lm -lstlport_gcc_stldebug
#CXXFLAGS_DEBUG_STL = -D__STL_DEBUG -I/usr/src/STLport-4.0/stlport -Wno-long-long
OPF_CXXFLAGS=-Wall -pedantic -O2 -g $(CXXFLAGS_DEBUG_STL)
-opf_container.o: opf_container.cpp opf_filter.h
+opf_container.o: opf_container.cpp opf_filter.h ../op_user.h
g++ -c $(OPF_CXXFLAGS) $< -o $@
-opf_filter.o : opf_filter.cpp opf_filter.h
+opf_filter.o : opf_filter.cpp opf_filter.h ../op_user.h
g++ -c $(OPF_CXXFLAGS) $< -o $@
opf_filter: opf_filter.o opf_container.o
Index: op_to_source
===================================================================
RCS file: /cvsroot/oprofile/oprofile/pp/op_to_source,v
retrieving revision 1.4
retrieving revision 1.5
diff -u -d -r1.4 -r1.5
--- op_to_source 2001/07/25 16:43:30 1.4
+++ op_to_source 2001/09/01 02:03:34 1.5
@@ -59,7 +59,6 @@
echo "see $0 --help"
exit 1
fi
- oprofpp_option="$oprofpp_option -f $1"
sample_file=$1
;;
-i|--image-file)
@@ -69,7 +68,6 @@
echo "see $0 --help"
exit 1
fi
- oprofpp_option="$oprofpp_option -i $1"
object_file=$1
;;
-h|--help)
@@ -103,6 +101,9 @@
shift
done
+# PHE FIXME : this checking need to be remade with the new samples files naming
+# scheme
+
# Check than enough option has been given.
if [ "$object_file" == "" -a "$sample_file" == "" ]; then
echo "You need to give at least an object-file or a sample-file"
@@ -115,9 +116,19 @@
exit 1
fi
-if [ "$sample_file" != "" -a ! -f "$sample_file" ]; then
- echo "sample file $sample_file does not exist"
- exit 1
+# PHE FIXME: bogus here is case of passing sample_filename without the #x
+# prefix. Should be rewrite perhaps.
+#if [ "$sample_file" != "" -a ! -f "$sample_file" ]; then
+# echo "sample file $sample_file does not exist"
+# exit 1
+#fi
+
+if [ "$sample_file" != "" ]; then
+ oprofpp_option="$oprofpp_option -f $sample_file"
+fi
+
+if [ "$object_file" != "" ]; then
+ oprofpp_option="$oprofpp_option -i $object_file"
fi
#echo "option : $op_filter_option"
Index: opf_container.cpp
===================================================================
RCS file: /cvsroot/oprofile/oprofile/pp/opf_container.cpp,v
retrieving revision 1.3
retrieving revision 1.4
diff -u -d -r1.3 -r1.4
--- opf_container.cpp 2001/07/22 13:51:06 1.3
+++ opf_container.cpp 2001/09/01 02:03:34 1.4
@@ -41,7 +41,7 @@
// Functors used as predicate for less than comparison of counter_array_t. There is no
// multi-sort choice, ie the the counter number used is statically set at ctr time.
struct less_sample_entry_by_samples_nr {
- // Precondition: index < max_counter_number. Used also as default ctr.
+ // Precondition: index < op_nr_counters. Used also as default ctr.
less_sample_entry_by_samples_nr(size_t index_ = 0) : index(index_) {}
bool operator()(const sample_entry * lhs, const sample_entry * rhs) const {
@@ -120,7 +120,7 @@
// the nth first symbols.
// the nth first symbols that accumulate a certain amount of samples.
std::multiset<const symbol_entry *, less_sample_entry_by_samples_nr>
- symbol_entry_by_samples_nr[max_counter_number];
+ symbol_entry_by_samples_nr[OP_MAX_COUNTERS];
std::set<const symbol_entry *, less_by_file_loc> symbol_entry_by_file_loc;
};
@@ -128,7 +128,7 @@
symbol_container_impl::symbol_container_impl() {
// symbol_entry_by_samples_nr has been setup by the default ctr, we need to
// rebuild it. do not assume than index 0 is correctly build
- for (size_t i = 0 ; i < max_counter_number ; ++i) {
+ for (size_t i = 0 ; i < op_nr_counters ; ++i) {
less_sample_entry_by_samples_nr compare(i);
symbol_entry_by_samples_nr[i] =
multiset<const symbol_entry *, less_sample_entry_by_samples_nr>(compare);
@@ -172,7 +172,7 @@
// Update the sets of symbols entries sorted by samples count and the set of
// symbol entries sorted by file location.
for (size_t i = 0 ; i < v.size() ; ++i) {
- for (size_t counter = 0 ; counter < max_counter_number ; ++counter)
+ for (size_t counter = 0 ; counter < op_nr_counters ; ++counter)
symbol_entry_by_samples_nr[counter].insert(&v[i]);
symbol_entry_by_file_loc.insert(&v[i]);
@@ -204,7 +204,7 @@
// get a vector of symbols sorted by increased count.
void symbol_container_impl::get_symbols_by_count(size_t counter, vector<const symbol_entry*>& v) const {
- if (counter >= max_counter_number) {
+ if (counter >= op_nr_counters) {
throw "symbol_container_impl::get_symbols_by_count() : invalid counter number";
}
@@ -321,7 +321,7 @@
counter += (*it1)->counter;
}
- for (size_t i = 0 ; i < max_counter_number ; ++i)
+ for (size_t i = 0 ; i < op_nr_counters ; ++i)
if (counter[i] != 0)
return true;
Index: opf_filter.cpp
===================================================================
RCS file: /cvsroot/oprofile/oprofile/pp/opf_filter.cpp,v
retrieving revision 1.3
retrieving revision 1.4
diff -u -d -r1.3 -r1.4
--- opf_filter.cpp 2001/07/22 13:51:06 1.3
+++ opf_filter.cpp 2001/09/01 02:03:34 1.4
@@ -155,7 +155,7 @@
// oprofpp give some info on the setting of the counters. These
// are stored here.
- counter_setup counter_info[max_counter_number];
+ counter_setup counter_info[OP_MAX_COUNTERS];
// TODO : begin_comment, end_comment must be based on the current
// extension and must be properties of source_file.
@@ -223,13 +223,13 @@
counter_array_t::counter_array_t()
{
- for (size_t i = 0 ; i < max_counter_number ; ++i)
+ for (size_t i = 0 ; i < op_nr_counters ; ++i)
value[i] = 0;
}
counter_array_t & counter_array_t::operator+=(const counter_array_t & rhs)
{
- for (size_t i = 0 ; i < max_counter_number ; ++i)
+ for (size_t i = 0 ; i < op_nr_counters ; ++i)
value[i] += rhs.value[i];
return *this;
@@ -243,7 +243,7 @@
out << hex << vma << dec << " ";
- for (size_t i = 0 ; i < max_counter_number ; ++i)
+ for (size_t i = 0 ; i < op_nr_counters ; ++i)
out << counter[i] << " ";
}
@@ -265,7 +265,7 @@
sscanf(str.c_str() + pos, "%lx%n", &vma, &number_of_char_read);
- for (size_t i = 0 ; i < max_counter_number ; ++i) {
+ for (size_t i = 0 ; i < op_nr_counters ; ++i) {
int temp;
sscanf(str.c_str() + pos + number_of_char_read, "%u%n", &counter[i], &temp);
@@ -352,7 +352,7 @@
out << "total samples :";
- for (size_t i = 0 ; i < max_counter_number ; ++i)
+ for (size_t i = 0 ; i < op_nr_counters ; ++i)
cout << " " << counter_info[i].total_samples;
cout << endl;
@@ -382,7 +382,7 @@
if (pos == 0) {
size_t counter_number;
if (sscanf(str.c_str(), "Counter %u", &counter_number) == 1 &&
- counter_number < max_counter_number) {
+ counter_number < op_nr_counters) {
counter_info[counter_number].enabled = true;
size_t pos = str.find("counted ");
@@ -439,11 +439,11 @@
bool output::calc_total_samples() {
- for (size_t i = 0 ; i < max_counter_number ; ++i)
+ for (size_t i = 0 ; i < op_nr_counters ; ++i)
counter_info[i].total_samples = 0;
for (size_t i = 0 ; i < symbols.size() ; ++i) {
- for (size_t j = 0 ; j < max_counter_number ; ++j)
+ for (size_t j = 0 ; j < op_nr_counters ; ++j)
counter_info[j].total_samples += symbols[i].sample.counter[j];
}
@@ -454,18 +454,19 @@
total_counter += samples[i].counter;
}
- for (size_t i = 0 ; i < max_counter_number ; ++i) {
+ for (size_t i = 0 ; i < op_nr_counters ; ++i) {
if (total_counter[i] != counter_info[i].total_samples) {
cerr << "output::calc_total_samples() : "
<< "bad counter accumulation"
<< " " << total_counter[i]
- << " " << counter_info[i].total_samples;
+ << " " << counter_info[i].total_samples
+ << endl;
exit(1);
}
}
}
- for (size_t i = 0 ; i < max_counter_number ; ++i)
+ for (size_t i = 0 ; i < op_nr_counters ; ++i)
if (counter_info[i].total_samples != 0)
return true;
@@ -491,7 +492,7 @@
if (prefix.length())
out << " " << prefix;
- for (size_t i = 0 ; i < max_counter_number ; ++i)
+ for (size_t i = 0 ; i < op_nr_counters ; ++i)
if (counter_info[i].enabled)
output_one_counter(counter[i], counter_info[i].total_samples);
@@ -754,18 +755,18 @@
size_t output::get_sort_counter_nr() const {
size_t index = sort_by_counter;
- if (index >= max_counter_number || counter_info[index].enabled == false) {
- cerr << "sort_by_counter invalid or counter[sort_by_counter] disabled : switch "
- << "to the first valid counter" << endl;
-
- for (index = 0 ; index < max_counter_number ; ++index) {
+ if (index >= size_t(op_nr_counters) || counter_info[index].enabled == false) {
+ for (index = 0 ; index < op_nr_counters ; ++index) {
if (counter_info[index].enabled)
break;
}
+
+ cerr << "sort_by_counter invalid or counter[sort_by_counter] disabled : switch "
+ << "to the first valid counter " << index << endl;
}
// paranoid checking.
- if (index == max_counter_number)
+ if (index == op_nr_counters)
throw "output::output_source(input &) cannot find a counter enabled";
return index;
@@ -925,7 +926,7 @@
}
bool have_counter_info = false;
- for (size_t i = 0 ; i < max_counter_number && !have_counter_info ; ++i) {
+ for (size_t i = 0 ; i < op_nr_counters && !have_counter_info ; ++i) {
if (counter_info[i].enabled)
have_counter_info = true;
}
@@ -938,6 +939,8 @@
read_input(in);
+// debug_dump_vector(out);
+
if (calc_total_samples() == false) {
cerr << "The input contains zero samples" << endl;
@@ -948,7 +951,7 @@
output_command_line();
- for (size_t i = 0 ; i < max_counter_number ; ++i) {
+ for (size_t i = 0 ; i < op_nr_counters ; ++i) {
if (i != 0)
out << endl;
@@ -1002,7 +1005,6 @@
poptContext optcon;
char c;
- // The cast to (char **) is neccessary on some version of popt.
optcon = poptGetContext(NULL, argc, argv, options, 0);
c=poptGetNextOpt(optcon);
Index: opf_filter.h
===================================================================
RCS file: /cvsroot/oprofile/oprofile/pp/opf_filter.h,v
retrieving revision 1.4
retrieving revision 1.5
diff -u -d -r1.4 -r1.5
--- opf_filter.h 2001/07/22 13:51:06 1.4
+++ opf_filter.h 2001/09/01 02:03:34 1.5
@@ -5,8 +5,7 @@
#include <iostream>
#include <string>
-// The maximum number of counter. Would be a parameter coming from configure.
-static const size_t max_counter_number = 2;
+#include "../op_user.h"
// Add run-time checking if true.
static const bool sanity_check = true;
@@ -30,7 +29,7 @@
counter_array_t & operator+=(const counter_array_t &);
private:
- size_t value[max_counter_number];
+ size_t value[OP_MAX_COUNTERS];
};
//---------------------------------------------------------------------------
Index: oprof_convert.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/pp/oprof_convert.c,v
retrieving revision 1.8
retrieving revision 1.9
diff -u -d -r1.8 -r1.9
--- oprof_convert.c 2001/08/20 19:49:54 1.8
+++ oprof_convert.c 2001/09/01 02:03:34 1.9
@@ -14,19 +14,25 @@
* Place - Suite 330, Boston, MA 02111-1307, USA.
*/
+/* Do not touch this file, the v5 samples files format make it easier to
+ * convert but the method used here is not adapated. This utilities can convert
+ * from version 2,3,4 to 5 newer version should use a new stuff */
+
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
+#include <sys/mman.h>
#include "../version.h"
#include "../dae/opd_util.h"
static char * filename;
-#define OPD_MANGLE_CHAR '}'
-
-#define OPD_MAGIC 0xdeb6
+/* magic number <= v4 */
+#define OPD_MAGIC_V4 0xdeb6
+#define OPD_MAGIC_V5 "DAE\n"
+
/* at the end of the sample files */
struct opd_footer_v2 {
u16 magic;
@@ -60,6 +66,23 @@
u32 reserved[31];
};
+/* At the begin of samples files */
+struct opd_footer_v5 {
+ u8 magic[4];
+ u32 version;
+ u8 is_kernel;
+ u32 ctr_event;
+ u32 ctr_um;
+ /* ctr number, used for sanity checking */
+ u32 ctr;
+ u32 cpu_type;
+ u32 ctr_count;
+ double cpu_speed;
+ time_t mtime;
+ /* binary compatibility reserve */
+ u32 reserved2[21];
+};
+
/*
* Precondition : fp is seeked on the begin of an opd_footer_vn, it is always ok to
* read an opd_footer_v2 at the begin of function but the version number field is not
@@ -92,6 +115,15 @@
fwrite(&footer_v3, sizeof(footer_v3), 1, fp);
}
+/* this function is used only between conversion from v3 to v4, (ChangeLog
+ * 2001-07-25), but backup of samples file has been added at 2001-07-15 (no
+ * release between this two date). This means than cvs between this can have
+ * backup and this files are mishandled by this function du the "-%d" suffix
+ * It is a don't take care case.
+ *
+ * Symptom is mtime checking loss (mtime is set to 0), people who have
+ * this problem have tried to use a cvs alpha version as a production tool
+ * let's fix themself the problem */
static char * get_binary_name(void)
{
char *file;
@@ -149,72 +181,234 @@
fwrite(&footer_v4, sizeof(footer_v4), 1, fp);
}
-/* provide a default conversion function which just increase the version number, It may
- * be used when a reserved field is used without changing the total sizeof of opd_footer
- * *and* the zero value in the reserve work correctly with your new code */
-/* Not used for now so no static to avoid warning. This code has not been tested */
-/*static*/ void bump_version(FILE* fp)
+/* the "old" samples file format */
+struct old_opd_fentry {
+ u32 count0;
+ u32 count1;
+};
+
+static int cpu_type = -1;
+
+/* PHE FIXME: really a fucking function, if you have problem with it flame
+ * me at <ph...@cl...> */
+/* create one row file during translate from colum to row format v4 --> v5 */
+static void do_mapping_transfer(uint nr_samples, int counter,
+ const char* filename,
+ const struct opd_footer_v4* footer_v4,
+ const struct old_opd_fentry* old_samples,
+ int session_number)
{
- struct opd_footer_v2 footer_v2;
+ char* out_filename;
+ int out_fd;
+ struct opd_footer_v5* footer;
+ u32* samples;
+ uint k;
+ u32 size;
+ int dirty;
- fread(&footer_v2, sizeof(footer_v2), 1, fp);
+ if (cpu_type == -1 || cpu_type < 0 || cpu_type > 3) {
+ fprintf(stderr, "converting %s to new file format require option --cpu-type=[0|1|2|3]\n", filename);
+ return;
+ }
- footer_v2.version++;
+ out_filename = opd_malloc(strlen(filename) + 32);
+ strcpy(out_filename, filename);
+ sprintf(out_filename + strlen(out_filename), "#%d", counter);
- fseek(fp, -sizeof(footer_v2), SEEK_CUR);
+ size = (nr_samples * sizeof(u32)) + sizeof(struct opd_footer_v5);
- fwrite(&footer_v2, sizeof(footer_v2), 1, fp);
-}
+ /* New samples files are lazilly created, respect this behavior
+ * by not creating file with no valid ctr_type_val */
+ dirty = 0;
-static struct converter converter_array[] = {
- { v2_to_v3, sizeof(struct opd_footer_v2), 2 },
- { v3_to_v4, sizeof(struct opd_footer_v3), 3 },
-};
+ if ((counter == 0 && footer_v4->v2.ctr0_type_val) ||
+ (counter == 1 && footer_v4->v2.ctr1_type_val)) {
+ if (session_number != -1)
+ sprintf(out_filename + strlen(filename), "-%d",
+ session_number);
-/* This acts as a samples of how to add new conversion from version N to M :
+ unlink(out_filename);
- define a :
+ out_fd = open(out_filename, O_CREAT|O_RDWR, 0644);
+ if (out_fd == -1) {
+ fprintf(stderr,"oprofiled: open of image sample file \"%s\" failed: %s\n", out_filename, strerror(errno));
+ goto err1;
+ }
-struct opd_footer_vM
- {
- struct opd_footer_v2 v2;
- // additionnal field
+ /* PHE FIXME: same issue in opd_proc.c with more comments */
+#if 0
+ /* ugly: mmap needs than fd size is sufficient. */
+ if (lseek(out_fd, size - 1, SEEK_SET) == -1) {
+ fprintf(stderr, "oprofiled: seek failed for \"%s\". %s\n", out_filename, strerror(errno));
+ goto err2;
+ }
- };
+ /* PHE FIXME: this unsparse the file by one memory page
+ * size at the end file :( An another way to grow a file ? */
+ if (write(out_fd, "", 1) != 1) {
+ perror("oprof_convert: cannot grow sample file. ");
+ goto err2;
+ }
+#else
+ if (ftruncate(out_fd, size) == -1) {
+ fprintf(stderr, "oprof_convert: ftruncate failed for \"%s\". %s\n", out_filename, strerror(errno));
+ goto err2;
+ }
+#endif
- define a function :
-void vN_to_vM(FILE* fp) {
- // get inspiration from v2_to_v3 code ...
+ footer = mmap(0, size, PROT_READ|PROT_WRITE, MAP_SHARED, out_fd, 0);
+
+ if (footer == (void *)-1) {
+ fprintf(stderr,"oprof_convert: mmap of output sample file \"%s\" failed: %s\n", out_filename, strerror(errno));
+ goto err2;
+ }
+
+ samples = (void *)(footer + 1);
+
+ memset(footer, '\0', sizeof(struct opd_footer_v5));
+ memcpy(footer->magic, OPD_MAGIC_V5, sizeof(footer->magic));
+ footer->version = 5;
+ footer->is_kernel = footer_v4->v2.is_kernel;
+ footer->ctr_event = (counter == 0)
+ ? footer_v4->v2.ctr0_type_val
+ : footer_v4->v2.ctr1_type_val;
+
+ footer->ctr_um = (counter == 0)
+ ? footer_v4->v2.ctr0_um
+ : footer_v4->v2.ctr1_um;
+
+ footer->ctr_count = (counter == 0)
+ ? footer_v4->ctr0_count
+ : footer_v4->ctr1_count;
+
+ footer->ctr = counter;
+
+ footer->cpu_type = cpu_type;
+ footer->cpu_speed = footer_v4->cpu_speed;
+ footer->mtime = footer_v4->mtime;
+
+ for (k = 0 ; k < nr_samples ; ++k) {
+ u32 sample = counter
+ ? old_samples[k].count1
+ : old_samples[k].count0;
+
+ /* keep the sparse of the file */
+ if (sample) {
+ samples[k] = sample;
+ dirty = 1;
+ }
+ }
+
+ munmap(footer, size);
+
+err2:
+ close(out_fd);
+err1:
+ }
+
+ /* respect the new lazilly file creation behavior of the daemon */
+ if (dirty == 0) {
+ unlink(out_filename);
+ }
+
+ opd_free(out_filename);
}
- add the following entry at the end of converter_array:
+/* PHE FIXME: really a fucking function, if you have problem with it flame
+ * me at <ph...@cl...> */
+static void v4_to_v5(FILE* fp)
+{
+ struct opd_footer_v4 footer_v4;
+ char *out_filename;
+ char *session_str;
+ int session_number;
+ int ok;
+ int counter;
+ int old_size, old_fd, nr_samples;
+ size_t len_filename;
+ struct old_opd_fentry *old_samples;
- { vN_to_vM, sizeof(struct opd_footer_vN), N },
+ fread(&footer_v4, sizeof(footer_v4), 1, fp);
- Note than if you use the reserved field to add additionnal field *and* the 0 binary
- value of the reserve acts as a default value for the new code which use this new
- field then you do not have to write a converter, simply add the following line to the
- conversion array:
- { bump_version, sizeof(struct opd_footer_vN), N },
+ ok = 1;
+ session_number = -1;
- PHIL 20010623 : these features has not been tested.
-*/
+ session_str = strrchr(filename, '-');
+ if (session_str) {
+ int read = 0;
+ if (sscanf(session_str + 1, "%d%n", &session_number, &read) == 1) {
+
+ /* Dangerous if the bin filename end with "-%d" I
+ * prefer reject this corner case */
+ ok = session_str[read] && !session_str[read + 1];
+ }
+ }
+
+ if (ok == 0) {
+ /* PHE FIXME: probably needs to remove this warning */
+// fprintf(stderr, "Bad file naming scheme %s\n", filename);
+ session_number = -1;
+ }
+
+ old_fd = open(filename, O_RDONLY);
+ if (old_fd == -1) {
+ fprintf(stderr, "oprof_convert: Opening %s failed. %s\n", filename, strerror(errno));
+ goto out;
+ }
+
+ old_size = opd_get_fsize(filename, 1);
+ if (old_size < sizeof(struct opd_footer_v4)) {
+ fprintf(stderr, "oprof_convert: sample file %s not enough big %d, expected %d\n", filename, old_size, sizeof(struct opd_footer_v4));
+ goto err1;
+ }
+
+ old_samples = mmap(0, old_size, PROT_READ, MAP_PRIVATE, old_fd, 0);
+ if (old_samples == (void *)-1) {
+ fprintf(stderr, "oprof_convert: mmap of %s failed. %s\n", filename, strerror(errno));
+ goto err1;
+ }
+
+ nr_samples = (old_size - sizeof(struct opd_footer_v4)) / sizeof(struct old_opd_fentry);
+
+ len_filename = strlen(filename);
+ out_filename = opd_malloc(len_filename + 32);
+ strcpy(out_filename, filename);
+
+ for (counter = 0; counter < 2 ; ++counter) {
+ do_mapping_transfer(nr_samples, counter, filename,
+ &footer_v4, old_samples,
+ session_number);
+ }
+
+ munmap(old_samples, old_size);
+
+err1:
+ close(old_fd);
+out:
+ opd_free(out_filename);
+}
+static struct converter converter_array[] = {
+ { v2_to_v3, sizeof(struct opd_footer_v2), 2 },
+ { v3_to_v4, sizeof(struct opd_footer_v3), 3 },
+ { v4_to_v5, sizeof(struct opd_footer_v4), 4 },
+};
+
#define nr_converter (signed int)((sizeof(converter_array) / sizeof(converter_array[0])))
/* return -1 if no converter are available, else return the index of the first
* converter */
-static int get_converter_index(FILE* fp)
+static int get_converter_index(FILE* fp, int last_index)
{
struct opd_footer_v2 footer_v2;
int i;
- for (i = 0 ; i < nr_converter ; ++i) {
+ for (i = last_index ; i < nr_converter ; ++i) {
fseek(fp, -converter_array[i].sizeof_struct, SEEK_END);
fread(&footer_v2, sizeof(footer_v2), 1, fp);
- if (footer_v2.magic == OPD_MAGIC &&
+ if (footer_v2.magic == OPD_MAGIC_V4 &&
footer_v2.version == converter_array[i].version) {
fseek(fp, -converter_array[i].sizeof_struct, SEEK_END);
@@ -253,7 +447,15 @@
exit(EXIT_SUCCESS);
}
- for (i = 1 ; i < argc ; ++i) {
+ i = 1;
+
+ /* for v4 --> v5 */
+ if (strncmp(argv[i], "--cpu-type", strlen("--cpu-type")) == 0) {
+ sscanf(argv[i] + strlen("--cpu-type="), "%d", &cpu_type);
+ ++i;
+ }
+
+ for ( ; i < argc ; ++i) {
// used in v3 conversion
filename = argv[i];
@@ -265,13 +467,16 @@
}
converted = 0;
- while ((converter_index = get_converter_index(fp)) != -1) {
+ converter_index = 0;
+ while ((converter_index = get_converter_index(fp, converter_index)) != -1) {
converter_array[converter_index].convert(fp);
+ converter_index++;
converted = 1;
}
if (converted == 0) {
fprintf(stderr, "no converter found for %s (file already converted ?)\n", argv[i]);
+
err++;
}
Index: oprofpp.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/pp/oprofpp.c,v
retrieving revision 1.36
retrieving revision 1.37
diff -u -d -r1.36 -r1.37
--- oprofpp.c 2001/08/20 19:46:13 1.36
+++ oprofpp.c 2001/09/01 02:03:34 1.37
@@ -40,16 +40,20 @@
static int list_all_symbols_details;
static int output_linenr_info;
-static struct opd_fentry *samples;
+/* PHE FIXME would be an array of struct ? */
+/* indexed by samples[counter_nr][offset] */
+static struct opd_fentry *samples[OP_MAX_COUNTERS];
+static struct opd_footer *footer[OP_MAX_COUNTERS];
+static char *ctr_name[OP_MAX_COUNTERS];
+static char *ctr_desc[OP_MAX_COUNTERS];
+static char *ctr_um_desc[OP_MAX_COUNTERS];
static uint nr_samples;
-static struct opd_footer footer;
-static char *ctr0_name;
-static char *ctr0_desc;
-static char *ctr0_um_desc;
-static char *ctr1_name;
-static char *ctr1_desc;
-static char *ctr1_um_desc;
-static int ctr;
+/* if != -1 appended to samples files with "-%d" format */
+static int session_number = -1;
+/* PHE FIXME: check the logic and this comment */
+/* counter k is selected if (ctr == -1 || ctr == k), if ctr == -1 counter 0
+ * is used for sort purpose */
+static int ctr = -1;
static u32 sect_offset;
static struct poptOption options[] = {
@@ -65,6 +69,8 @@
{ "base-dir", 'b', POPT_ARG_STRING, &basedir, 0, "base directory of profile daemon", NULL, },
{ "list-all-symbols-details", 'L', POPT_ARG_NONE, &list_all_symbols_details, 0, "list samples for all symbols", NULL, },
{ "output-linenr-info", 'o', POPT_ARG_NONE, &output_linenr_info, 0, "output filename:linenr info", NULL },
+ /* PHE FIXME document later */
+ { "session-number", 'S', POPT_ARG_INT, &session_number, 0, "which session use", "session number" },
POPT_AUTOHELP
{ NULL, 0, 0, NULL, 0, NULL, NULL, },
};
@@ -109,6 +115,9 @@
poptContext optcon;
char c;
const char *file;
+ const char *file_session_str;
+ char *file_ctr_str;
+ int counter;
/* Some old version of popt needs the cast to char ** */
optcon = poptGetContext(NULL, argc, (char **)argv, options, 0);
@@ -164,7 +173,55 @@
/* we'll "leak" this memory */
samplefile = remangle(imagefile);
}
- }
+ }
+
+ /* We must get account 2 var : ctr, mangled filename, if ctr is not
+ * provided it can be provided inside the filename itself by the #%d
+ * terminateur, if ctr is given we check an eventual conflict */
+ /* PHE FIXME: can be simplified probably ? */
+ file_ctr_str = strrchr(samplefile, '#');
+ counter = -1;
+ if (file_ctr_str) {
+ sscanf(file_ctr_str + 1, "%d", &counter);
+ }
+
+ if (ctr != counter) {
+ /* PHE FIXME documentation of this behavior */
+ /* a --counter=x have priority on the # suffixe of filename */
+ if (ctr != -1 && counter != -1) {
+ /* conflict between #%d and --ctr option */
+ fprintf(stderr, "oprofpp: conflict between %s filename counter nr and --ctr %d option, using --ctr option %d\n", file_ctr_str, ctr, ctr);
+ }
+
+ }
+
+ counter = ctr;
+
+ if (ctr == -1) {
+ /* list_all_symbols_details always output all counter and do
+ * not made any sort, it is the responsability of the backend
+ * (op_to_source) to treat this */
+ if (!list_all_symbols_details)
+ ctr = 0;
+ }
+
+ /* Now record the session number */
+ if (file_ctr_str) {
+ file_session_str = strchr(file_ctr_str, '-');
+ if (file_session_str) {
+ sscanf(file_session_str + 1, "%d", &session_number);
+ }
+ }
+
+ if (file_ctr_str)
+ file_ctr_str[0] = '\0';
+
+ if (ctr != -1 && (ctr < 0 || ctr >= op_nr_counters)) {
+ ctr = 0;
+
+ fprintf(stderr, "oprofpp: invalid counter number, using %d\n", ctr);
+
+ }
}
/**
@@ -259,7 +316,6 @@
}
newmtime = opd_get_mtime(file);
-
if (newmtime != mtime) {
fprintf(stderr, "oprofpp: the last modified time of the binary file %s does not match "
"that of the sample file. Either this is the wrong binary or the binary "
@@ -282,7 +338,9 @@
if (!imagefile)
free(file);
- if (footer.is_kernel) {
+ /* PHE FIXME: this is broken, should search for a opended sample file
+ * */
+ if (footer[ctr == -1 ? 0 : ctr]->is_kernel) {
asection *sect;
sect = bfd_get_section_by_name(ibfd, ".text");
@@ -438,24 +496,20 @@
struct opp_count {
asymbol *sym;
- u32 count0;
- u32 count1;
+ u32 count[OP_MAX_COUNTERS];
};
-
+
int countcomp(const void *a, const void *b)
{
struct opp_count *ca= (struct opp_count *)a;
struct opp_count *cb= (struct opp_count *)b;
- if (ctr) {
- if (ca->count1 < cb->count1)
- return -1;
- return (ca->count1 > cb->count1);
- } else {
- if (ca->count0 < cb->count0)
- return -1;
- return (ca->count0 > cb->count0);
- }
+ /* PHE FIXME: I think it is correct... */
+ int counter = ctr == -1 ? 0 : ctr;
+
+ if (ca->count[counter] < cb->count[counter])
+ return -1;
+ return (ca->count[counter] > cb->count[counter]);
}
/**
@@ -465,26 +519,24 @@
* Lists all the symbols from the image specified by samplefile,
* in decreasing sample count order, to standard out.
*/
-void do_list_symbols(bfd * ibfd)
+void do_list_symbols(asymbol **syms, uint num)
{
- asymbol **syms;
struct opp_count *scounts;
u32 start, end;
- uint num, tot0 = 0, tot1 = 0,i,j;
-
- num = get_symbols(ibfd, &syms);
+ uint tot[OP_MAX_COUNTERS],i,j;
+ int found_samples, k;
- if (!num) {
- fprintf(stderr, "oprofpp: couldn't get any symbols from image file.\n");
- exit(EXIT_FAILURE);
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ tot[i] = 0;
}
-
scounts = opd_calloc0(num,sizeof(struct opp_count));
for (i=0; i < num; i++) {
scounts[i].sym = syms[i];
get_symbol_range(syms[i], (i == num-1) ? NULL : syms[i+1], &start, &end);
+ /* FIXME PHE: was in old code, seems bad at my eyes (compare
+ * an adress with a number of samples) */
if (start >= nr_samples) {
fprintf(stderr,"oprofpp: start 0x%x out of range (max 0x%x)\n", start, nr_samples);
exit(EXIT_FAILURE);
@@ -495,34 +547,39 @@
}
for (j = start; j < end; j++) {
- if (samples[j].count0)
- verbprintf("Adding %u 0-samples for symbol $%s$ at pos j 0x%x\n", samples[j].count0, syms[i]->name, j);
- else if (samples[j].count1)
- verbprintf("Adding %u 1-samples for symbol $%s$ at pos j 0x%x\n", samples[j].count1, syms[i]->name, j);
- scounts[i].count0 += samples[j].count0;
- scounts[i].count1 += samples[j].count1;
- tot0 += samples[j].count0;
- tot1 += samples[j].count1;
+ /* PHE FIXME too imbrication level */
+ for (k = 0 ; k < op_nr_counters ; ++k) {
+ if (samples[k] && samples[k][j].count) {
+ verbprintf("Adding %u 0-samples for symbol $%s$ at pos j 0x%x\n",
+ samples[k][j].count, syms[i]->name, j);
+ scounts[i].count[k] += samples[k][j].count;
+ tot[k] += samples[k][j].count;
+ }
+ }
}
}
-
+
qsort(scounts, num, sizeof(struct opp_count), countcomp);
for (i=0; i < num; i++) {
printf_symbol(scounts[i].sym->name);
- if (ctr && scounts[i].count1) {
- printf("[0x%.8lx]: %2.4f%% (%u samples)\n", scounts[i].sym->value+scounts[i].sym->section->vma,
- (((double)scounts[i].count1) / tot1)*100.0, scounts[i].count1);
- } else if (!ctr && scounts[i].count0) {
- printf("[0x%.8lx]: %2.4f%% (%u samples)\n", scounts[i].sym->value+scounts[i].sym->section->vma,
- (((double)scounts[i].count0) / tot0)*100.0, scounts[i].count0);
- } else if (!streq("", scounts[i].sym->name))
+
+ /* PHE FIXME too imbrication level */
+ found_samples = 0;
+ for (k = 0; k < op_nr_counters ; ++k) {
+ if (scounts[i].count[k]) {
+ printf("[0x%.8lx]: %2.4f%% (%u samples)\n", scounts[i].sym->value+scounts[i].sym->section->vma,
+ (((double)scounts[i].count[k]) / tot[k])*100.0, scounts[i].count[k]);
+ found_samples = 1;
+ }
+
+ }
+ if (found_samples == 0 && !streq("", scounts[i].sym->name)) {
printf(" (0 samples)\n");
+ }
}
opd_free(scounts);
- opd_free(syms);
- bfd_close(ibfd);
}
/**
@@ -532,18 +589,11 @@
* Lists all the samples for the symbol specified by symbol, from the image
* specified by @samplefile, in decreasing sample count order, to standard out.
*/
-void do_list_symbol(bfd * ibfd)
+void do_list_symbol(bfd * ibfd, asymbol **syms, uint num)
{
- asymbol **syms;
u32 start, end;
- u32 num,i,j;
-
- num = get_symbols(ibfd,&syms);
-
- if (!num) {
- fprintf(stderr, "oprofpp: couldn't get any symbols from image file.\n");
- exit(EXIT_FAILURE);
- }
+ u32 i, j;
+ int k;
for (i=0; i < num; i++) {
if (streq(syms[i]->name, symbol))
@@ -556,14 +606,26 @@
printf("Samples for symbol \"%s\" in image %s\n", symbol, ibfd->filename);
get_symbol_range(syms[i], (i == num-1) ? NULL : syms[i+1], &start, &end);
for (j=start; j < end; j++) {
- if (samples[j].count0 || samples[j].count1) {
- printf("%s+%x/%x:\t%u \t%u\n", symbol, sym_offset(syms[i], j), end-start,
- samples[j].count0, samples[j].count1);
+ for (k = 0 ; k < op_nr_counters ; ++k) {
+ if (samples[k] && samples[k][j].count) {
+ printf("%s+%x/%x:", symbol, sym_offset(syms[i], j), end-start);
+ break;
+ }
+ }
+
+ /* k != op_nr_counters <==> one of the counter is not empty */
+ if (k != op_nr_counters) {
+ for (k = 0 ; k < op_nr_counters ; ++k) {
+ if (samples[k])
+ /* PHE: please keep this space a few
+ * time, I can't live without it ;) */
+ printf("\t%u ", samples[k][j].count);
+ else
+ printf("\t%u", 0U);
+ }
+ printf("\n");
}
}
-
- opd_free(syms);
- bfd_close(ibfd);
}
#define GMON_VERSION 1
@@ -583,25 +645,18 @@
* Dump gprof-format samples for the image specified by samplefile to
* the file specified by gproffile.
*/
-void do_dump_gprof(bfd * ibfd)
+/* PHE: this fun has not been tested */
+void do_dump_gprof(asymbol **syms, uint num)
{
static struct gmon_hdr hdr = { "gmon", GMON_VERSION, {0,0,0,},};
FILE *fp;
- asymbol **syms;
u32 start, end;
- uint num,i,j;
+ uint i, j;
u32 low_pc = (u32)-1;
u32 high_pc = 0;
u16 * hist;
u32 histsize;
- num = get_symbols(ibfd,&syms);
-
- if (!num) {
- fprintf(stderr, "oprofpp: couldn't get any symbols from image file.\n");
- exit(EXIT_FAILURE);
- }
-
fp=opd_open_file(gproffile, "w");
opd_write_file(fp,&hdr, sizeof(struct gmon_hdr));
@@ -642,14 +697,12 @@
for (i=0; i < num; i++) {
get_symbol_range(syms[i], (i == num-1) ? NULL : syms[i+1], &start, &end);
for (j=start; j < end; j++) {
- u32 count;
+ u32 count = 0;
u32 pos;
pos = (sym_offset(syms[i],j) + syms[i]->value + syms[i]->section->vma - low_pc) / MULTIPLIER;
- if (ctr)
- count = samples[j].count1;
- else
- count = samples[j].count0;
+ /* get_options have set ctr to one value != -1 */
+ count = samples[ctr][j].count;
if (count > (u16)-1) {
printf("Warning: capping sample count !\n");
@@ -665,8 +718,6 @@
opd_write_file(fp, hist, histsize * sizeof(u16));
opd_close_file(fp);
- opd_free(syms);
- bfd_close(ibfd);
}
/**
@@ -681,7 +732,7 @@
* that work with output from oprofpp.
*/
static void
-translate_address (bfd* ibfd, asymbol** syms, asection* section, bfd_vma pc)
+translate_address (bfd* ibfd, asymbol **syms, asection *section, bfd_vma pc)
{
int found;
bfd_vma vma;
@@ -709,6 +760,33 @@
}
/**
+ * accumulate_samples - lookup and output linenr info from a vma address
+ * in a given section to standard output.
+ * @counter: where to accumulate the samples, can be %NULL
+ * @index: index number of the samples.
+ *
+ * if @counter != %NULL accumulate samples count in @counter
+ * else just return 0/1 if they exist samples at @index
+ *
+ * return non-zero if samples has been found
+ */
+static int accumulate_samples(u32 counter[OP_MAX_COUNTERS], int index)
+{
+ int k;
+ int found_samples = 0;
+
+ for (k = 0; k < op_nr_counters; ++k) {
+ if (samples[k] && samples[k][index].count) {
+ found_samples = 1;
+ if (counter)
+ counter[k] += samples[k][index].count;
+ }
+ }
+
+ return found_samples;
+}
+
+/**
* do_list_all_symbols_details - list all samples for all symbols.
* @ibfd: the bfd
*
@@ -718,32 +796,30 @@
* Do not change output format without changing the corresponding tools
* that work with output from oprofpp
*/
-void do_list_all_symbols_details(bfd* ibfd) {
- uint num, i, j;
+void do_list_all_symbols_details(bfd *ibfd, asymbol **syms, uint num)
+{
+ uint i, j;
u32 start, end;
- asymbol **syms;
- num = get_symbols(ibfd, &syms);
-
- if (!num) {
- fprintf(stderr, "oprofpp: couldn't get any symbols from image file.\n");
- exit(EXIT_FAILURE);
- }
-
for (i = 0 ; i < num ; ++i) {
- u32 ctr0, ctr1;
+ u32 counter[OP_MAX_COUNTERS];
+ int k;
+ int found_samples;
get_symbol_range(syms[i], (i == num-1) ? NULL : syms[i+1],
&start, &end);
+ for (k = 0; k < op_nr_counters; ++k) {
+ counter[k] = 0;
+ }
+
/* To avoid outputing 0 samples symbols */
- ctr0 = ctr1 = 0;
- for (j=start; j < end; j++) {
- ctr0 += samples[j].count0;
- ctr1 += samples[j].count1;
+ found_samples = 0;
+ for (j = start; j < end; ++j) {
+ found_samples |= accumulate_samples(counter, j);
}
- if (ctr0 || ctr1) {
+ if (found_samples) {
bfd_vma vma, base_vma;
base_vma = syms[i]->value + syms[i]->section->vma;
@@ -757,12 +833,22 @@
printf(" ");
}
- printf("%.8lx %u %u ", base_vma, ctr0, ctr1);
+ printf("%.8lx ", base_vma);
+ for (k = 0 ; k < op_nr_counters ; ++k) {
+ printf("%u ", counter[k]);
+ }
printf_symbol(syms[i]->name);
printf("\n");
- for (j=start; j < end; j++) {
- if (samples[j].count0 || samples[j].count1) {
+ for (j = start; j < end; j++) {
+
+ for (k = 0; k < op_nr_counters; ++k) {
+ counter[k] = 0;
+ }
+
+ found_samples = accumulate_samples(counter, j);
+
+ if (found_samples) {
vma = sym_offset(syms[i], j) +
base_vma;
@@ -773,99 +859,280 @@
syms[i]->section,
vma);
}
- printf(" %.8lx %u %u\n",
- vma, samples[j].count0,
- samples[j].count1);
+
+ printf(" %.8lx", vma);
+
+ for (k = 0; k < op_nr_counters; ++k) {
+ printf(" %u", counter[k]);
+ }
+
+ printf("\n");
}
}
}
}
-
- opd_free(syms);
- bfd_close(ibfd);
}
-int main(int argc, char const *argv[])
+/**
+ * open_samples_file - open a samples file
+ * @counter: the counter number
+ * @size: where to return the mapped size, this do NOT
+ * include the footer size
+ * @can_fail: allow to fail gracefully
+ *
+ * open and mmap the given samples files, return -1
+ * the global var samples[@counter] and footer[@counter]
+ * are updated in case of success else a fatal error occur
+ * The following field of the footer are checked
+ * u8 magic[4];
+ * u16 version;
+ * u8 ctr_event;
+ * u8 ctr_um;
+ * u8 ctr;
+ * u8 cpu_type;
+ *
+ * return the file descriptor (can be -1 if @can_fail) else
+ * produces a fatal error
+ */
+static fd_t open_samples_file(u32 counter, size_t *size,
+ int can_fail)
{
fd_t fd;
- FILE *fp;
- size_t size;
- bfd *ibfd;
-
- get_options(argc, argv);
+ char* temp;
- fp = fopen(samplefile,"r");
- if (!fp) {
- fprintf(stderr, "oprofpp: fopen of %s failed. %s\n", samplefile, strerror(errno));
- exit(EXIT_FAILURE);
+ temp = opd_malloc(strlen(samplefile) + 32);
+
+ strcpy(temp, samplefile);
+
+ sprintf(temp + strlen(temp), "#%d", counter);
+ if (session_number != -1)
+ sprintf(temp + strlen(temp), "-%d", session_number);
+
+ fd = open(temp, O_RDONLY);
+ if (fd == -1) {
+ if (!can_fail) {
+ fprintf(stderr, "oprofpp: Opening %s failed. %s\n", temp, strerror(errno));
+ exit(EXIT_FAILURE);
+ }
+
+ goto err1;
}
-
- if (fseek(fp, -sizeof(struct opd_footer), SEEK_END) == -1) {
- fprintf(stderr, "oprofpp: fseek of %s failed. %s\n", samplefile, strerror(errno));
- exit(EXIT_FAILURE);
+
+ *size = opd_get_fsize(temp, 1) - sizeof(struct opd_footer);
+ if (*size < sizeof(struct opd_footer)) {
+ if (!can_fail) {
+ fprintf(stderr, "oprofpp: sample file %s not enough big %d, expected %d\n", temp, *size, sizeof(struct opd_footer));
+ exit(EXIT_FAILURE);
+ }
+
+ goto err2;
}
-
- if (fread(&footer, sizeof(struct opd_footer), 1, fp) != 1) {
- fprintf(stderr, "oprofpp: fread of %s failed. %s\n", samplefile, strerror(errno));
+
+ footer[counter] = mmap(0, *size + sizeof(struct opd_footer), PROT_READ,
+ MAP_PRIVATE, fd, 0);
+ if (footer[counter] == (void *)-1) {
+ if (!can_fail) {
+ fprintf(stderr, "oprofpp: mmap of %s failed. %s\n", temp, strerror(errno));
+ exit(EXIT_FAILURE);
+ }
+ goto err2;
+ }
+
+ /* need a cast here :( */
+ samples[counter] = (struct opd_fentry *)(footer[counter] + 1);
+
+ if (memcmp(footer[counter]->magic, OPD_MAGIC, sizeof(footer[0]->magic))) {
+ if (!can_fail) {
+ /* FIXME: is 4.4 ok : there is no zero terminator */
+ fprintf(stderr, "oprofpp: wrong magic %4.4s, expected %s.\n", footer[counter]->magic, OPD_MAGIC);
+ exit(EXIT_FAILURE);
+ }
+ goto err3;
+ }
+
+ if (footer[counter]->version != OPD_VERSION) {
+ if (!can_fail) {
+ fprintf(stderr, "oprofpp: wrong version 0x%x, expected 0x%x.\n", footer[counter]->version, OPD_VERSION);
+ exit(EXIT_FAILURE);
+ }
+ goto err3;
+ }
+
+ /* This should be warrented by the daemon */
+ if (footer[counter]->ctr != counter) {
+ if (!can_fail) {
+ fprintf(stderr, "oprofpp: sanity check counter number fail %d, expect %d.\n", footer[counter]->ctr, counter);
+ exit(EXIT_FAILURE);
+ }
+ goto err3;
+ }
+
+ opd_free(temp);
+
+ return fd;
+
+err3:
+ munmap(footer[counter], *size + sizeof(struct opd_footer));
+err2:
+ close(fd);
+err1:
+ fd = -1;
+ footer[counter] = NULL;
+ samples[counter] = NULL;
+ *size = 0;
+ return fd;
+}
+
+/**
+ * check_and_output_event - translate event number, um mask
+ * to string and output them.
+ * @counter: counter number
+ *
+ * if event is valid output to stdout the description
+ * of event.
+ */
+static void check_and_output_event(int i)
+{
+ op_get_event_desc(footer[i]->cpu_type, footer[i]->ctr_event,
+ footer[i]->ctr_um, &ctr_name[i], &ctr_desc[i],
+ &ctr_um_desc[i]);
+
+ printf("Counter %d counted %s events (%s) with a unit mask of "
+ "0x%.2x (%s) count %u\n",
+ i, ctr_name[i], ctr_desc[i], footer[i]->ctr_um,
+ ctr_um_desc[i] ? ctr_um_desc[i] : "Not set",
+ footer[i]->ctr_count);
+}
+
+/**
+ * check_footers - check coherence between two footers.
+ * @f1: first footer
+ * @f2: second footer
+ *
+ * verify than footer @f1 and @f2 are coherent.
+ * fatal error ocur if the footers are not coherent.
+ */
+static void check_footers(struct opd_footer* f1, struct opd_footer* f2)
+{
+ if (f1->mtime != f2->mtime) {
+ fprintf(stderr, "oprofpp: footer time stamp are different (%ld, %ld)\n", f1->mtime, f2->mtime);
exit(EXIT_FAILURE);
}
- fclose(fp);
- if (footer.magic != OPD_MAGIC) {
- fprintf(stderr, "oprofpp: wrong magic 0x%x, expected 0x%x.\n", footer.magic, OPD_MAGIC);
+ if (f1->is_kernel != f2->is_kernel) {
+ fprintf(stderr, "oprofpp: footer is_kernel are different\n");
exit(EXIT_FAILURE);
}
-
- if (footer.version != OPD_VERSION) {
- fprintf(stderr, "oprofpp: wrong version 0x%x, expected 0x%x.\n", footer.version, OPD_VERSION);
+
+ if (f1->cpu_speed != f2->cpu_speed) {
+ fprintf(stderr, "oprofpp: footer cpu speed are different (%f, %f)",
+ f2->cpu_speed, f2->cpu_speed);
exit(EXIT_FAILURE);
}
+}
+
+int main(int argc, char const *argv[])
+{
+ bfd *ibfd;
+ asymbol **syms;
+ int i, j;
+ fd_t fd[OP_MAX_COUNTERS];
+ size_t size[OP_MAX_COUNTERS];
+ uint num;
+ time_t mtime = 0;
- if (footer.ctr0_type_val) {
- op_get_event_desc(footer.ctr0_type_val, footer.ctr0_um, &ctr0_name, &ctr0_desc, &ctr0_um_desc);
- printf("Counter 0 counted %s events (%s) with a unit mask of 0x%.2x (%s) count %u\n",ctr0_name, ctr0_desc,
- footer.ctr0_um, ctr0_um_desc ? ctr0_um_desc : "Not set", footer.ctr0_count);
+ get_options(argc, argv);
+
+ for (i = 0; i < op_nr_counters ; ++i) {
+ fd[i] = -1;
+ size[i] = 0;
+ samples[i] = NULL;
+ footer[i] = NULL;
}
- if (footer.ctr1_type_val) {
- op_get_event_desc(footer.ctr1_type_val, footer.ctr1_um, &ctr1_name, &ctr1_desc, &ctr1_um_desc);
- printf("Counter 1 counted %s events (%s) with a unit mask of 0x%.2x (%s) count %u\n",ctr1_name, ctr1_desc,
- footer.ctr1_um, ctr1_um_desc ? ctr1_um_desc : "Not set", footer.ctr1_count);
+ for (i = 0; i < op_nr_counters ; ++i) {
+ if (ctr == -1 || ctr == i)
+ /* PHE FIXME: in some case we can pass 0 as can_fail
+ * to get better error message */
+ fd[i] = open_samples_file(i, &size[i], 1);
}
- printf("Cpu speed was (MHz estimation) : %f\n", footer.cpu_speed);
-
- fd = open(samplefile, O_RDONLY);
+ for (i = 0; i < op_nr_counters ; ++i) {
+ if (fd[i] != -1)
+ break;
+ }
- if (fd == -1) {
- fprintf(stderr, "oprofpp: Opening %s failed. %s\n", samplefile, strerror(errno));
+ if (i == op_nr_counters) {
+ fprintf(stderr, "Can not open any samples files for %s last error %s\n", samplefile, strerror(errno));
exit(EXIT_FAILURE);
}
- size = opd_get_fsize(samplefile, 1) - sizeof(struct opd_footer);
- samples = (struct opd_fentry *)mmap(0, size, PROT_READ, MAP_PRIVATE, fd, 0);
+ /* sanity check between the different samples files */
+ for (i = 0 ; i < op_nr_counters; ++i) {
+ if (fd[i] != -1)
+ break;
+ }
- if (samples == (void *)-1) {
- fprintf(stderr, "oprofpp: mmap of %s failed. %s\n", samplefile, strerror(errno));
+ for (j = i + 1; j < op_nr_counters ; ++j) {
+ if (fd[j] != -1) {
+ if (size[i] != size[j]) {
+ fprintf(stderr, "oprofpp: mapping file size "
+ "for ctr (%d, %d) are different "
+ "(%d, %d)\n", i, j, size[i], size[j]);
+
+ exit(EXIT_FAILURE);
+ }
+
+ check_footers(footer[i], footer[j]);
+ }
+ }
+
+ /* output and sanity check on ctr_um, ctr_event and cpu_type */
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ if (fd[i] != -1) {
+ check_and_output_event(i);
+
+ /* redundant set but correct and simplify the code */
+ nr_samples = size[i] / sizeof(struct opd_fentry);
+ mtime = footer[i]->mtime;
+ }
+ }
+
+ verbprintf("nr_samples %d\n", nr_samples);
+
+ ibfd = open_image_file(samplefile, mtime);
+
+ num = get_symbols(ibfd, &syms);
+
+ verbprintf("nr symbols %u\n", num);
+
+ if (!num) {
+ fprintf(stderr, "oprofpp: couldn't get any symbols from image file.\n");
exit(EXIT_FAILURE);
}
- nr_samples = size/sizeof(struct opd_fentry);
+ printf("Cpu speed was (MHz estimation) : %f\n",
+ footer[ctr == -1 ? 0 : ctr]->cpu_speed);
- ibfd = open_image_file(samplefile, footer.mtime);
-
if (list_symbols) {
- do_list_symbols(ibfd);
+ do_list_symbols(syms, num);
} else if (symbol) {
- do_list_symbol(ibfd);
+ do_list_symbol(ibfd, syms, num);
} else if (gproffile) {
- do_dump_gprof(ibfd);
+ do_dump_gprof(syms, num);
} else if (list_all_symbols_details) {
- do_list_all_symbols_details(ibfd);
+ do_list_all_symbols_details(ibfd, syms, num);
}
-
- munmap(samples, size);
- close(fd);
+
+ opd_free(syms);
+ bfd_close(ibfd);
+
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ if (footer[i]) {
+ munmap(footer[i], size[i] + sizeof(struct opd_footer));
+ close(fd[i]);
+ }
+ }
return 0;
}
Index: oprofpp.h
===================================================================
RCS file: /cvsroot/oprofile/oprofile/pp/oprofpp.h,v
retrieving revision 1.14
retrieving revision 1.15
diff -u -d -r1.14 -r1.15
--- oprofpp.h 2001/07/25 16:43:30 1.14
+++ oprofpp.h 2001/09/01 02:03:34 1.15
@@ -30,6 +30,7 @@
#include <sys/mman.h>
#include "../dae/opd_util.h"
+#include "../op_user.h"
#include "../version.h"
/* missing from libiberty.h */
@@ -40,18 +41,6 @@
# define DMGL_ANSI (1 << 1) /* Include const, volatile, etc */
#endif
char *cplus_demangle (const char *mangled, int options);
-
-void op_get_event_desc(u8 type, u8 um, char **typenamep, char **typedescp, char **umdescp);
-
-
-#define FALSE 0
-#define TRUE 1
-#define uint unsigned int
-#define ulong unsigned long
-#define u8 unsigned char
-#define u16 u_int16_t
-#define u32 u_int32_t
-#define fd_t int
#define verbprintf(args...) \
do { \
@@ -59,36 +48,5 @@
printf(args); \
} while (0)
-/* kernel image entries are offset by this many entries */
-#define OPD_KERNEL_OFFSET 524288
-
-/* this char replaces '/' in sample filenames */
-#define OPD_MANGLE_CHAR '}'
-
-struct opd_fentry {
- u32 count0;
- u32 count1;
-};
-/* FIXME : Carefull these are also present in dae/oprofiled.h */
-#define OPD_MAGIC 0xdeb6
-#define OPD_VERSION 0x4
-/* at the end of the sample files */
-struct opd_footer {
- u16 magic;
- u16 version;
- u8 is_kernel;
- u8 ctr0_type_val;
- u8 ctr1_type_val;
- u8 ctr0_um;
- u8 ctr1_um;
- u8 reserved1[16];
- u32 ctr0_count;
- u32 ctr1_count;
- /* Set to 0.0 if not available */
- double cpu_speed;
- time_t mtime;
- /* binary compatibility reserve */
- u32 reserved2[31];
-};
|
|
From: John L. <mov...@us...> - 2001-09-01 02:03:38
|
Update of /cvsroot/oprofile/oprofile/dae
In directory usw-pr-cvs1:/tmp/cvs-serv19169/dae
Modified Files:
Makefile.in op_start opd_proc.c opd_util.c opd_util.h
oprofiled.c oprofiled.h
Log Message:
philippe's huge patch
Index: Makefile.in
===================================================================
RCS file: /cvsroot/oprofile/oprofile/dae/Makefile.in,v
retrieving revision 1.14
retrieving revision 1.15
diff -u -d -r1.14 -r1.15
--- Makefile.in 2001/08/18 16:11:17 1.14
+++ Makefile.in 2001/09/01 02:03:34 1.15
@@ -14,13 +14,13 @@
install: $(INSTALL_LIST)
-$(MKDIR_P) $(BINDIR)
- @for f in $(INSTALL_LIST); do \
+ for f in $(INSTALL_LIST); do \
cp $$f $(BINDIR)/$$f && chmod 755 $(BINDIR)/$$f; \
done
uninstall:
- @for f in $(INSTALL_LIST); do \
- rm -f $(BINDIR)/$$f; \
+ for f in $(INSTALL_LIST); do \
+ -rm -f $(BINDIR)/$$f; \
done
clean:
Index: op_start
===================================================================
RCS file: /cvsroot/oprofile/oprofile/dae/op_start,v
retrieving revision 1.36
retrieving revision 1.37
diff -u -d -r1.36 -r1.37
--- op_start 2001/08/19 20:23:02 1.36
+++ op_start 2001/09/01 02:03:34 1.37
@@ -23,17 +23,7 @@
BUF_SIZE=262144
HASH_SIZE=65536
-CTR0_EVENT=
-CTR1_EVENT=
KERNEL_ONLY=0
-CTR0_UM=0
-CTR1_UM=0
-CTR0_COUNT=600000
-CTR1_COUNT=600000
-CTR0_USER=1
-CTR0_KERNEL=1
-CTR1_USER=1
-CTR1_KERNEL=1
IGNORE_MYSELF=0
DIR="/var/opd"
MAP_FILE=
@@ -42,6 +32,18 @@
PGRP_FILTER=0
VERBOSE=0
+# PHE FIXME: too many changes needs a redesign. Proposition:
+# 1) check for a cpu_type switch,
+# 2) stop the daemon
+# 3) modprobe oprofile [cpu_type=xx]
+# 4) if --cpu-type not given get it from /proc/sys/dev/oprofile/cpu_type
+# 5) default init
+# 6) parse parameter
+# 7) launch the daemon
+# this avoid passing cpu_type in the common case and allow in case or module
+# misdetect the used core to overwrite it through an init parameters to the
+# module
+
if test -f /proc/cpuinfo; then
modelname=`cat /proc/cpuinfo | grep "model\ name\ :" | sed -e 's/ //g' | cut -d':' -f2`
case $modelname in
@@ -74,29 +76,41 @@
CPUTYPE=0
fi
-# FIXME: needs to be generalised for more counters (e.g. athlon)
+case $CPUTYPE in
+ 0|1|2)
+ OP_MAX_COUNTERS=2
+ ;;
+ 3)
+ OP_MAX_COUNTERS=4
+ ;;
+esac
+
+# we can now default define individual counter setup variable.
+f=0
+while (( $f < $OP_MAX_COUNTERS )); do
+ CTR_COUNT[$f]=600000
+ CTR_USER[$f]=1
+ CTR_KERNEL[$f]=1
+ f=$(($f+1))
+done
do_help() {
-echo 'op_start: usage:
+echo "op_start: usage:
Module options
--buffer-size=num number of samples in kernel buffer
--hash-table-size=num number of entries in kernel hash table
--kernel-only=[0|1] profile only the kernel
- --ctr0-event=name symbolic event name for ctr0
- --ctr1-event=name symbolic event name for ctr1
- --ctr0-count=val number of events between samples for ctr0
- --ctr1-count=val number of events between samples for ctr1
- --ctr0-unit-mask=val unit mask for ctr0
- --ctr1-unit-mask=val unit mask for ctr1
- --ctr0-kernel=[0|1] whether to count kernel events for ctr 0
- --ctr0-user=[0|1] whether to count user events for ctr 0
- --ctr1-kernel=[0|1] whether to count kernel events for ctr 1
- --ctr1-user=[0|1] whether to count user events for ctr1
- --pid-filter=pid Only profile process pid (if compiled in)
- --pgrp-filter=pgrp Only profile process group pgrp (if compiled in)
+ --ctrN-event=name symbolic event name for ctr N
+ --ctrN-count=val number of events between samples for ctr N
+ --ctrN-unit-mask=val unit mask for ctr N
+ --ctrN-kernel=[0|1] whether to count kernel events for ctr N
+ --ctrN-user=[0|1] whether to count user events for ctr N
+Allowed range for N is [0-$OP_MAX_COUNTERS[
+ --pid-filter=pid Only profile process pid
+ --pgrp-filter=pgrp Only profile process group pgrp
Daemon options
- --use-cpu=[0|1|2|3] 0 or PPro, 1 for PII, 2 for PIII, 3 for AMD Athlon.
+ --use-cpu=[0|1|2|3] 0: PPro, 1: PII, 2: PIII, 3: AMD Athlon.
--ignore-myself=[0|1] ignore samples for oprofiled
--log-file=file log file
--base-dir=dir base directory of daemon
@@ -109,7 +123,7 @@
General options
--list-events list event types and unit masks
- --help this message'
+ --help this message"
}
# utilities functions.
@@ -135,10 +149,17 @@
echo "/proc/sys/$dev_name does not exist or is not a regular file" >&2
exit 1
fi
-
echo $val > /proc/sys/$dev_name
}
+# PHE FIXME: assume ctr < 9, how from an awk expr extract the int field?
+# probably needs to use bash string manipulation.
+# extract the field N from --ctrN-xxxxxx
+extract_int()
+{
+ echo $1 | awk '{print substr($0, 6, 1)}'
+}
+
LOG_FILE=
SAMPLES_DIR=
DEVICE_FILE=
@@ -150,6 +171,10 @@
arg=`echo $1 | awk -F= '{print $1}'`
val=`echo $1 | awk -F= '{print $2}'`
case "$arg" in
+ # PHE FIXME: this work actually but if the forced cpu type have
+ # more counter than OP_MAX_COUNTERS only OP_NR_MAX_COUNTERS
+ # can be used. if less the script do not work and you can
+ # get many trouble.
--use-cpu)
if [ "$val" != "" ]; then
CPUTYPE=$val
@@ -175,35 +200,21 @@
KERNEL_ONLY=1
fi
;;
- --ctr0-unit-mask)
- CTR0_UM=$val
- ;;
- --ctr1-unit-mask)
- CTR1_UM=$val
- ;;
- --ctr0-event)
- CTR0_EVENT=$val
- ;;
- --ctr1-event)
- CTR1_EVENT=$val
- ;;
- --ctr0-count)
- CTR0_COUNT=$val
- ;;
- --ctr1-count)
- CTR1_COUNT=$val
+ # PHE FIXME: see comment at extract_int
+ --ctr*-unit-mask)
+ CTR_UM[`extract_int $1`]=$val
;;
- --ctr0-user)
- CTR0_USER=$val
+ --ctr*-event)
+ CTR_EVENT[`extract_int $1`]=$val
;;
- --ctr0-kernel)
- CTR0_KERNEL=$val
+ --ctr*-count)
+ CTR_COUNT[`extract_int $1`]=$val
;;
- --ctr1-user)
- CTR1_USER=$val
+ --ctr*-user)
+ CTR_USER[`extract_int $1`]=$val
;;
- --ctr1-kernel)
- CTR1_KERNEL=$val
+ --ctr*-kernel)
+ CTR_KERNEL[`extract_int $1`]=$val
;;
--base-dir)
DIR=$val
@@ -251,33 +262,24 @@
shift
done
-if [ -z "$CTR0_EVENT" -a -z "$CTR1_EVENT" ]; then
+if [[ ${#CTR_EVENT[*]} == 0 ]]; then
echo "You haven't specified what events you would like to count, e.g."
echo "op_start ... --ctr0-event=CPU_CLK_UNHALTED --ctr0-count=600000"
echo "Enter op_start --help for full options"
exit 1
fi
-
-
-CTR0_EVENT_VAL=
-if [ ! -z "$CTR0_EVENT" ]; then
- CTR0_EVENT_VAL=`op_help $CTR0_EVENT`
-fi
-
-CTR1_EVENT_VAL=
-if [ ! -z "$CTR1_EVENT" ]; then
- CTR1_EVENT_VAL=`op_help $CTR1_EVENT`
-fi
-
-if [ -z "$CTR0_EVENT_VAL" -a ! -z "$CTR0_EVENT" ]; then
- echo "Unknown event \"$CTR0_EVENT\""
- exit 1
-fi
-if [ -z "$CTR1_EVENT_VAL" -a ! -z "$CTR1_EVENT" ]; then
- echo "Unknown event \"$CTR1_EVENT\""
- exit 1
-fi
+f=0
+while (( $f < $OP_MAX_COUNTERS )); do
+ if [[ ${#CTR_EVENT[$f]} != 0 ]]; then
+ CTR_EVENT_VAL[$f]=`op_help --cpu-type=$CPU_TYPE ${CTR_EVENT[$f]}`
+ if [ -z "${CTR_EVENT_VAL[$f]}" -a ! -z "${CTR_EVENT[$f]}" ]; then
+ echo "Unknown event \"${CTR_EVENT[$f]}\""
+ exit 1
+ fi
+ fi
+ f=$(($f+1))
+done
# stop any existing daemon
@@ -326,14 +328,15 @@
echo "Parameters used:"
echo "BUF_SIZE $BUF_SIZE"
echo "HASH_SIZE $HASH_SIZE"
- echo "CTR0_EVENT $CTR0_EVENT"
- echo "CTR1_EVENT $CTR1_EVENT"
- echo "CTR0_UM $CTR0_UM"
- echo "CTR1_UM $CTR1_UM"
- echo "CTR0_COUNT $CTR0_COUNT"
- echo "CTR1_COUNT $CTR1_COUNT"
- echo "CTR0_OSUSR $CTR0_OSUSR"
- echo "CTR1_OSUSR $CTR1_OSUSR"
+ f=0
+ while (( $f < $OP_MAX_COUNTERS )); do
+ echo "CTR${f}_EVENT ${CTR_EVENT[$f]}"
+ echo "CTR${f}_COUNT ${CTR_COUNT[$f]}"
+ echo "CTR${f}_UM ${CTR_UM[$f]}"
+ echo "CTR${f}_USER ${CTR_USER[$f]}"
+ echo "CTR${f}_KERNEL ${CTR_KERNEL[$f]}"
+ f=$(($f+1))
+ done
echo "CPUTYPE $CPUTYPE"
echo "IGNORE_MYSELF $IGNORE_MYSELF"
echo "DIR $DIR"
@@ -365,8 +368,8 @@
exit 1
fi
-if [ ! -f "$MAP_FILE" ]; then
- echo "The specified map file \"$MAP_FILE\" doesn't exist."
+if [ ! -f "$VMLINUX" ]; then
+ echo "The specified vmlinux file \"$VMLINUX\" doesn't exist."
exit 1
fi
@@ -419,36 +422,32 @@
chmod 700 "$HASH_MAP_DEVICE_FILE"
# Necessary in this case :
-# op_start ctr0-on ctr1-on
-# op_start ctr0-on
-$SYSCTL -w dev.oprofile.0.enabled=0 >/dev/null
-$SYSCTL -w dev.oprofile.1.enabled=0 >/dev/null
-$SYSCTL -w dev.oprofile.0.event=0 >/dev/null
-$SYSCTL -w dev.oprofile.1.event=0 >/dev/null
+# op_start ctr0-on ctr1-on then op_start ctr0-on
+f=0
+while (( $f < $OP_MAX_COUNTERS )); do
+ $SYSCTL -w dev.oprofile.$f.enabled=0 >/dev/null
+ $SYSCTL -w dev.oprofile.$f.event=0 >/dev/null
+ f=$(($f+1))
+done
-$SYSCTL -w dev.oprofile.hashsize=$HASH_SIZE >/dev/null
-$SYSCTL -w dev.oprofile.bufsize=$BUF_SIZE >/dev/null
-$SYSCTL -w dev.oprofile.kernel_only=$KERNEL_ONLY >/dev/null
-$SYSCTL -w dev.oprofile.pid_filter=$PID_FILTER >/dev/null
-$SYSCTL -w dev.oprofile.pgrp_filter=$PGRP_FILTER >/dev/null
+$SYSCTL -w dev.oprofile.hashsize=$HASH_SIZE
+$SYSCTL -w dev.oprofile.bufsize=$BUF_SIZE
+$SYSCTL -w dev.oprofile.kernel_only=$KERNEL_ONLY
+$SYSCTL -w dev.oprofile.pid_filter=$PID_FILTER
+$SYSCTL -w dev.oprofile.pgrp_filter=$PGRP_FILTER
-# FIXME do something for all CPUs
-if [ "$CTR0_EVENT" != "" ]; then
- $SYSCTL -w dev.oprofile.0.enabled=1 >/dev/null
- $SYSCTL -w dev.oprofile.0.count=$CTR0_COUNT >/dev/null
- $SYSCTL -w dev.oprofile.0.kernel=$CTR0_KERNEL >/dev/null
- $SYSCTL -w dev.oprofile.0.user=$CTR0_USER >/dev/null
- $SYSCTL -w dev.oprofile.0.unit_mask=$CTR0_UM >/dev/null
- $SYSCTL -w dev.oprofile.0.event=$CTR0_EVENT_VAL >/dev/null
-fi
-if [ "$CTR1_EVENT" != "" ]; then
- $SYSCTL -w dev.oprofile.1.enabled=1 >/dev/null
- $SYSCTL -w dev.oprofile.1.count=$CTR1_COUNT >/dev/null
- $SYSCTL -w dev.oprofile.1.kernel=$CTR1_KERNEL >/dev/null
- $SYSCTL -w dev.oprofile.1.user=$CTR1_USER >/dev/null
- $SYSCTL -w dev.oprofile.1.unit_mask=$CTR1_UM >/dev/null
- $SYSCTL -w dev.oprofile.1.event=$CTR1_EVENT_VAL >/dev/null
-fi
+f=0
+while (( $f < $OP_MAX_COUNTERS )); do
+ if [ "${CTR_EVENT[$f]}" != "" ]; then
+ $SYSCTL -w dev.oprofile.$f.enabled=1
+ $SYSCTL -w dev.oprofile.$f.count=${CTR_COUNT[$f]}
+ $SYSCTL -w dev.oprofile.$f.kernel=${CTR_KERNEL[$f]}
+ $SYSCTL -w dev.oprofile.$f.user=${CTR_USER[$f]}
+ $SYSCTL -w dev.oprofile.$f.unit_mask=${CTR_UM[$f]}
+ $SYSCTL -w dev.oprofile.$f.event=${CTR_EVENT_VAL[$f]}
+ fi
+ f=$(($f+1))
+done
OPD_ARGS="--buffer-size=$BUF_SIZE --use-cpu=$CPUTYPE --ignore-myself=$IGNORE_MYSELF \
--log-file=$LOG_FILE --base-dir=$DIR --samples-dir=$SAMPLES_DIR \
Index: opd_proc.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/dae/opd_proc.c,v
retrieving revision 1.64
retrieving revision 1.65
diff -u -d -r1.64 -r1.65
--- opd_proc.c 2001/08/18 15:58:31 1.64
+++ opd_proc.c 2001/09/01 02:03:34 1.65
@@ -23,9 +23,6 @@
#define OPD_DEFAULT_MAPS 16
#define OPD_MAP_INC 8
-/* kernel image entries are offset by this many entries */
-#define OPD_KERNEL_OFFSET 524288
-
extern int kernel_only;
extern int verbose;
extern unsigned long opd_stats[];
@@ -36,7 +33,11 @@
/* hash of process lists */
static struct opd_proc *opd_procs[OPD_MAX_PROC_HASH];
-struct opd_footer footer;
+u32 ctr_count[OP_MAX_COUNTERS];
+u8 ctr_event[OP_MAX_COUNTERS];
+u8 ctr_um[OP_MAX_COUNTERS];
+extern u32 cpu_type;
+double cpu_speed;
/* image structure */
static struct opd_image *opd_images;
@@ -52,7 +53,8 @@
static struct opd_proc *opd_add_proc(u16 pid);
static void opd_grow_images(void);
-static void opd_open_image(struct opd_image *image);
+static void opd_open_sample_file(struct opd_image *image, int counter);
+static void opd_open_image(struct opd_image *image, const char *name, int kernl);
static int opd_find_image(const char *name);
static int opd_add_image(const char *name, int kernel);
static int opd_get_image(const char *name, int kernel);
@@ -62,7 +64,8 @@
static void opd_put_mapping(struct opd_proc *proc, int image_nr, u32 start, u32 offset, u32 end);
static struct opd_proc *opd_get_proc(u16 pid);
static void opd_delete_proc(struct opd_proc *proc);
-static void opd_handle_old_sample_file(char * mangled, time_t mtime);
+static void opd_handle_old_sample_file(int counter, const char * mangled, time_t mtime);
+static void opd_handle_old_sample_files(const char * mangled, time_t mtime);
/* every so many minutes, clean up old procs, msync mmaps, and
report stats */
@@ -71,10 +74,14 @@
struct opd_proc *proc;
struct opd_proc *next;
uint i;
+ int j;
- for (i=0; i < nr_images; i++) {
- if (opd_images[i].fd != -1)
- msync(opd_images[i].start, opd_images[i].len, MS_ASYNC);
+ for (i = 0; i < nr_images; i++) {
+ struct opd_image* image = &opd_images[i];
+ for (j = 0 ; j < op_nr_counters ; ++j) {
+ if (image->sample_files[j].fd > 1)
+ msync(image->sample_files[i].start, image->len + sizeof(struct opd_footer), MS_ASYNC);
+ }
}
for (i=0; i < OPD_MAX_PROC_HASH; i++) {
@@ -169,19 +176,13 @@
opd_close_file(fp);
}
-struct opd_fentry {
- u32 count0;
- u32 count1;
-};
-
-
/**
* opd_save_old_sample_file - back up the sample file
* @file: the file name of the sample
*
* Back up a sample file.
*/
-static void opd_save_old_sample_file(char * file)
+static void opd_save_old_sample_file(const char *file)
{
char * savename;
int gen = 0;
@@ -199,8 +200,7 @@
opd_free(savename);
}
-
-
+
/**
* opd_handle_old_sample_file - deal with old sample file
* @mangled: the sample file name
@@ -209,11 +209,11 @@
* If an old sample file exists, verify it is usable.
* If not, move or delete it.
*/
-static void opd_handle_old_sample_file(char * mangled, time_t mtime)
+static void opd_handle_old_sample_file(int counter, const char * mangled, time_t mtime)
{
struct opd_footer oldfooter;
FILE * fp;
-
+
if (!opd_get_fsize(mangled, 0))
return;
@@ -221,25 +221,20 @@
if (!fp)
goto del;
- if (fseek(fp, -sizeof(struct opd_footer), SEEK_END) == -1)
- goto closedel;
-
if (fread(&oldfooter, sizeof(struct opd_footer), 1, fp) != 1)
goto closedel;
- if (oldfooter.magic != OPD_MAGIC || oldfooter.version != OPD_VERSION)
- goto closedel;
+ if (memcmp(&oldfooter.magic, OPD_MAGIC, sizeof(oldfooter.magic)) || oldfooter.version != OPD_VERSION)
+ goto closedel;
if (difftime(mtime, oldfooter.mtime))
goto closedel;
/* versions match, but we might be using different values */
- if (oldfooter.ctr0_type_val != footer.ctr0_type_val ||
- oldfooter.ctr1_type_val != footer.ctr1_type_val ||
- oldfooter.ctr0_um != footer.ctr0_um ||
- oldfooter.ctr1_um != footer.ctr1_um ||
- oldfooter.ctr0_count != footer.ctr0_count ||
- oldfooter.ctr1_count != footer.ctr1_count) {
+ if (oldfooter.ctr_event != ctr_event[counter] ||
+ oldfooter.ctr_um != ctr_um[counter] ||
+ oldfooter.ctr_count != ctr_count[counter] ||
+ oldfooter.cpu_type != cpu_type) {
fclose(fp);
opd_save_old_sample_file(mangled);
return;
@@ -255,91 +250,86 @@
verbprintf("Deleting old sample file \"%s\".\n", mangled);
unlink(mangled);
}
+
+
+/**
+ * opd_handle_old_sample_files - deal with old sample file
+ * @image_name: image to open file for
+ * @mtime: the new mtime of the binary
+ *
+ * to simplify admin of sample file we rename or remove sample
+ * files for each counter.
+ *
+ * If an old sample file exists, verify it is usable.
+ * If not, move or delete it.
+ */
+static void opd_handle_old_sample_files(const char *image_name, time_t mtime)
+{
+ int i;
+ char *mangled;
+
+ mangled = opd_mangle_filename(smpdir, image_name);
+
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ sprintf(mangled + strlen(mangled), "#%d", i);
+
+ opd_handle_old_sample_file(i, mangled, mtime);
+ }
+
+ free(mangled);
+}
/**
* opd_open_image - open an image sample file
* @image: image to open file for
+ * @name: name of the image to add
+ * @kernel: is the image a kernel/module image
*
- * Open and initialise an image sample file for
- * the image @image and set up memory mappings for
- * it. image->kernel and image->name must have meaningful
- * values.
+ * @image at funtion entry is uninitialised
+ * @name is copied i.e. should be GC'd separately from the
+ * image structure if appropriate.
+ *
+ * Initialise an opd_image struct for the image @image
+ * without opening the associated samples files. At return
+ * the @image is fully initialized.
*/
-static void opd_open_image(struct opd_image *image)
+static void opd_open_image(struct opd_image *image, const char *name, int kernel)
{
- char *mangled;
- char *c;
- char *c2;
+ int i;
- mangled = opd_malloc(strlen(smpdir) + 2 + strlen(image->name));
- strcpy(mangled, smpdir);
- strcat(mangled, "/");
- c = mangled + strlen(smpdir) + 1;
- c2 = image->name;
- do {
- if (*c2 == '/')
- *c++ = OPD_MANGLE_CHAR;
- else
- *c++ = *c2;
- } while (*++c2);
- *c = '\0';
+ /* PHE FIXME : store the mangled name ? */
+ image->name = opd_strdup(name);
+ image->kernel = kernel;
+ image->len = 0;
- verbprintf("Statting $%s$\n", image->name);
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ image->sample_files[i].fd = -1;
+ image->sample_files[i].start = (void *)-1;
+ image->sample_files[i].footer = (void *)-1;
+ }
- /* for each byte in original, two u32 counters */
- image->len = opd_get_fsize(image->name, 0) * sizeof(u32) * 2;
+ verbprintf("Statting \"%s\"\n", name);
+
+ /* for each byte in original one counter */
+ image->len = opd_get_fsize(name, 0) * sizeof(struct opd_fentry);
if (!image->len) {
- fprintf(stderr, "stat failed for %s\n", image->name);
- image->fd = -1;
- opd_free(mangled);
+ fprintf(stderr, "stat failed for %s\n", name);
return;
}
- footer.mtime = opd_get_mtime(image->name);
+ image->mtime = opd_get_mtime(name);
/* give space for "negative" entries. This is because we
* don't know about kernel/module sections other than .text so
* a sample could be before our nominal start of image, or
* after the start */
if (image->kernel)
- image->len += OPD_KERNEL_OFFSET * sizeof(struct opd_fentry) * 2;
-
- opd_handle_old_sample_file(mangled, footer.mtime);
-
- verbprintf("Opening $%s$\n", mangled);
-
- image->fd = open(mangled, O_CREAT|O_RDWR, 0644);
- if (image->fd == -1) {
- fprintf(stderr,"oprofiled: open of image sample file \"%s\" failed: %s\n", mangled, strerror(errno));
- goto out;
- }
-
- if (lseek(image->fd, image->len, SEEK_SET) == -1) {
- fprintf(stderr, "oprofiled: seek failed for \"%s\". %s\n", mangled, strerror(errno));
- goto err;
- }
-
- footer.is_kernel = image->kernel;
-
- if ((write(image->fd, &footer, sizeof(struct opd_footer))) < (int)sizeof(struct opd_footer)) {
- perror("oprofiled: wrote less than expected opd_footer. ");
- goto err;
- }
+ image->len += OPD_KERNEL_OFFSET * sizeof(struct opd_fentry);
- image->start = mmap(0, image->len, PROT_READ|PROT_WRITE, MAP_SHARED, image->fd, 0);
+ opd_handle_old_sample_files(name, image->mtime);
- if (image->start == (void *)-1) {
- perror("oprofiled: mmap() failed. ");
- goto err;
- }
-out:
- opd_free(mangled);
- return;
-err:
- close(image->fd);
- image->fd=-1;
- goto out;
+ /* samples files are lazilly openeded */
}
/**
@@ -361,9 +351,103 @@
*/
inline static u16 opd_get_counter(const u16 count)
{
- return (count & OP_COUNTER);
+ return OP_COUNTER(count);
}
+/*
+ * opd_open_sample_file - open an image sample file
+ * @image: image to open file for
+ * @counter: counter number
+ *
+ * Open image sample file for the image @image, counter
+ * @counter and set up memory mappings for it.
+ * image->kernel and image->name must have meaningful
+ * values.
+ */
+static void opd_open_sample_file(struct opd_image *image, int counter)
+{
+ char* mangled;
+ struct opd_sample_file *sample_file;
+
+ sample_file = &image->sample_files[counter];
+
+ /* avoid flood of error messages */
+ if (sample_file->fd == -2)
+ return;
+
+ mangled = opd_mangle_filename(smpdir, image->name);
+
+ sprintf(mangled + strlen(mangled), "#%d", counter);
+
+ verbprintf("Opening \"%s\"\n", mangled);
+
+ sample_file->fd = open(mangled, O_CREAT|O_RDWR, 0644);
+ if (sample_file->fd == -1) {
+ fprintf(stderr,"oprofiled: open of image sample file \"%s\" failed: %s\n", mangled, strerror(errno));
+ goto err1;
+ }
+
+ /* PHE FIXME: keep it, I need opinion on the two manner to do the job
+ * I am unsure if there is no a thrid way to make this */
+#if 0
+ /* ugly: mmap needs than fd size is sufficient, bugs in this area
+ * is difficult to understand, mmap sucess but the returned pointer
+ * cause a segfault, then the daemon is killed without coredump nor
+ * message error. PHE FIXME: why this mmap behavior, and why daemon die
+ * silently ? */
+ if (lseek(sample_file->fd, image->len + sizeof(struct opd_footer) - 1, SEEK_SET) == -1) {
+ fprintf(stderr, "oprofiled: seek failed for \"%s\". %s\n", mangled, strerror(errno));
+ goto err2;
+ }
+
+ /* PHE FIXME: this unsparse the file by one memory page size at the
+ * end of file :( An another way to grow the file ?, 0 write size do
+ * not work. try a write from the zero page? */
+ if (write(sample_file->fd, "", 1) != 1) {
+ perror("oprofiled: cannot grow sample file. ");
+ goto err2;
+ }
+#else
+ /* PHE FIXME: perhaps this is equivalent to the code above other #if
+ * alternative */
+ if (ftruncate(sample_file->fd, image->len + sizeof(struct opd_footer) - 1) == -1) {
+ fprintf(stderr, "oprofiled: ftruncate failed for \"%s\". %s\n", mangled, strerror(errno));
+ goto err2;
+ }
+#endif
+
+ sample_file->footer = mmap(0, image->len + sizeof(struct opd_footer),
+ PROT_READ|PROT_WRITE, MAP_SHARED, sample_file->fd, 0);
+
+ if (sample_file->footer == (void *)-1) {
+ fprintf(stderr,"oprofiled: mmap of image sample file \"%s\" failed: %s\n", mangled, strerror(errno));
+ goto err2;
+ }
+
+ sample_file->start = sample_file->footer + 1;
+
+ memset(sample_file->footer, '\0', sizeof(struct opd_footer));
+ sample_file->footer->version = OPD_VERSION;
+ memcpy(&sample_file->footer->magic, OPD_MAGIC, sizeof(sample_file->footer->magic));
+ sample_file->footer->is_kernel = image->kernel;
+ sample_file->footer->ctr_event = ctr_event[counter];
+ sample_file->footer->ctr_um = ctr_um[counter];
+ sample_file->footer->ctr = counter;
+ sample_file->footer->cpu_type = cpu_type;
+ sample_file->footer->ctr_count = ctr_count[counter];
+ sample_file->footer->cpu_speed = cpu_speed;
+ sample_file->footer->mtime = image->mtime;
+out:
+ opd_free(mangled);
+ return;
+err2:
+ close(sample_file->fd);
+err1:
+ /* avoid flood of error messages */
+ sample_file->fd = -2;
+ goto out;
+}
+
/**
* opd_put_image_sample - write sample to file
* @image: image for sample
@@ -382,35 +466,36 @@
inline static void opd_put_image_sample(struct opd_image *image, u32 offset, u16 count)
{
struct opd_fentry *fentry;
+ struct opd_sample_file* sample_file;
+ int counter;
- if (image->fd < 1) {
- verbprintf("Trying to write to non-opened image %s\n", image->name);
- return;
+ counter = opd_get_counter(count);
+ sample_file = &image->sample_files[counter];
+
+ if (sample_file->fd < 1) {
+ opd_open_sample_file(image, counter);
+ if (sample_file->fd < 1) {
+ /* opd_open_sample_file output an error message */
+ return;
+ }
}
- fentry = image->start + (offset*sizeof(struct opd_fentry));
+ fentry = sample_file->start + (offset*sizeof(struct opd_fentry));
if (image->kernel)
fentry += OPD_KERNEL_OFFSET;
- if (((u32)fentry) > ((u32)(image->start + image->len))) {
+ if (((u32)fentry) > ((u32)(sample_file->start + image->len))) {
fprintf(stderr, "fentry %p out of bounds for \"%s\" (start %p, len 0x%.8lx, "
- "end %p, orig offset 0x%.8x, kernel %d)\n", fentry, image->name, image->start,
- image->len, image->start + image->len, offset, image->kernel);
+ "end %p, orig offset 0x%.8x, kernel %d)\n", fentry, image->name, sample_file->start,
+ image->len, sample_file->start + image->len, offset, image->kernel);
return;
}
- if (opd_get_counter(count)) {
- if (fentry->count1 + opd_get_count(count) < fentry->count1)
- fentry->count1 = (u32)-1;
- else
- fentry->count1 += opd_get_count(count);
- } else {
- if (fentry->count0 + opd_get_count(count) < fentry->count0)
- fentry->count0 = (u32)-1;
- else
- fentry->count0 += opd_get_count(count);
- }
+ if (fentry->count + opd_get_count(count) < fentry->count)
+ fentry->count = (u32)-1;
+ else
+ fentry->count += opd_get_count(count);
}
@@ -424,11 +509,11 @@
{
/* 0 is reserved for the kernel image */
opd_images = opd_calloc0(sizeof(struct opd_image), OPD_DEFAULT_IMAGES);
- opd_images[0].name = opd_malloc(strlen(vmlinux) + 1);
- strncpy(opd_images[0].name,vmlinux,strlen(vmlinux) + 1);
-
+
+ opd_images[0].name = opd_strdup(vmlinux);
opd_images[0].kernel = 1;
- opd_open_image(&opd_images[0]);
+
+ opd_open_image(&opd_images[0], vmlinux, 1);
nr_images = 1;
}
@@ -502,11 +587,7 @@
*/
static int opd_add_image(const char *name, int kernel)
{
- opd_images[nr_images].name = opd_malloc(strlen(name)+1);
- strncpy(opd_images[nr_images].name, name, strlen(name)+1);
-
- opd_images[nr_images].kernel = kernel;
- opd_open_image(&opd_images[nr_images]);
+ opd_open_image(&opd_images[nr_images], name, kernel);
nr_images++;
if (nr_images == max_nr_images)
@@ -1050,7 +1131,7 @@
return;
printf("DO_PUT_SAMPLE (LAST_MAP): calc offset 0x%.8x, map start 0x%.8x,"
- " end 0x%.8x, offset 0x%.8x, name $%s$\n",
+ " end 0x%.8x, offset 0x%.8x, name \"%s\"\n",
offset, map->start, map->end, map->offset, opd_images[map->image].name);
}
@@ -1145,7 +1226,7 @@
*/
void opd_put_mapping(struct opd_proc *proc, int image_nr, u32 start, u32 offset, u32 end)
{
- verbprintf("Placing mapping for process %d: 0x%.8x-0x%.8x, off 0x%.8x, $%s$\n",
+ verbprintf("Placing mapping for process %d: 0x%.8x-0x%.8x, off 0x%.8x, \"%s\"\n",
proc->pid, start, end, offset, opd_images[image_nr].name);
proc->maps[proc->nr_maps].image = image_nr;
Index: opd_util.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/dae/opd_util.c,v
retrieving revision 1.18
retrieving revision 1.19
diff -u -d -r1.18 -r1.19
--- opd_util.c 2001/07/27 00:28:44 1.18
+++ opd_util.c 2001/09/01 02:03:34 1.19
@@ -143,7 +143,7 @@
*
* Returns a char* pointer to the duplicated string.
*/
-char *opd_strdup(const char* str)
+char *opd_strdup(const char *str)
{
char *temp = opd_malloc(strlen(str) + 1);
memcpy(temp, str, strlen(str) + 1);
@@ -151,6 +151,44 @@
return temp;
}
+/**
+ * opd_mangle_filename - mangle a file filename
+ * @smpdir: base directory name
+ * @image_name: a path name to the image file
+ * @extra_size: an extra size to allocate
+ *
+ * allocate memory of size strlen(@image_name) +
+ * strlen(@smpdir) + 2 + 32 then concat
+ * @smpdir and the mangled name of @filename
+ * the 32 bytes added are assumed to concat
+ * something like "-%d"
+ *
+ * Returns a char* pointer to the mangled string.
+ *
+ */
+char* opd_mangle_filename(const char *smpdir, const char* image_name)
+{
+ char *mangled;
+ char *c;
+ const char *c2;
+
+ mangled = opd_malloc(strlen(smpdir) + 2 + strlen(image_name) + 32);
+ strcpy(mangled, smpdir);
+ strcat(mangled, "/");
+ c = mangled + strlen(smpdir) + 1;
+ c2 = image_name;
+ do {
+ if (*c2 == '/')
+ *c++ = OPD_MANGLE_CHAR;
+ else
+ *c++ = *c2;
+ } while (*++c2);
+ *c = '\0';
+
+ return mangled;
+}
+
+
/* remove_component_p() and op_simlify_pathname() comes from the gcc
preprocessor */
@@ -582,13 +620,15 @@
struct stat st;
if (stat(file, &st)) {
- if (!fatal)
+ if (!fatal)
return 0;
fprintf(stderr,"opd_get_fsize: stat failed\n");
exit(1);
}
+ /* PHE FIXME caller can not make any difference between failure and
+ * zero file size when fatal != 0 */
return st.st_size;
}
Index: opd_util.h
===================================================================
RCS file: /cvsroot/oprofile/oprofile/dae/opd_util.h,v
retrieving revision 1.13
retrieving revision 1.14
diff -u -d -r1.13 -r1.14
--- opd_util.h 2001/07/27 00:28:44 1.13
+++ opd_util.h 2001/09/01 02:03:34 1.14
@@ -27,15 +27,40 @@
#include <time.h>
#include <fcntl.h>
+#include "../op_user.h"
+
+/* this char replaces '/' in sample filenames */
+#define OPD_MANGLE_CHAR '}'
+
#define FALSE 0
#define TRUE 1
-#define u8 unsigned char
-#define u16 u_int16_t
-#define u32 u_int32_t
-#define fd_t int
#define streq(a,b) (!strcmp((a), (b)))
+#define OPD_MAGIC "DAE\n"
+#define OPD_VERSION 0x5
+
+/* header of the sample files */
+struct opd_footer {
+ u8 magic[4];
+ u32 version;
+ u8 is_kernel;
+ u32 ctr_event;
+ u32 ctr_um;
+ /* ctr number, used for sanity checking */
+ u32 ctr;
+ u32 cpu_type;
+ u32 ctr_count;
+ double cpu_speed;
+ time_t mtime;
+ /* binary compatibility reserve */
+ u32 reserved2[21];
+};
+
+struct opd_fentry {
+ u32 count;
+};
+
/* utility functions */
#define opd_calloc(memb, size) opd_malloc(memb*size)
#define opd_calloc0(memb, size) opd_malloc0(memb*size)
@@ -50,6 +75,7 @@
void *opd_realloc(void *buf, size_t size);
void opd_free(void *p);
char *opd_strdup(const char* str);
+char* opd_mangle_filename(const char *smpdir, const char* filename);
char *opd_simplify_pathname(char *path);
char *opd_relative_to_absolute_path(const char *path, const char *base_dir);
Index: oprofiled.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/dae/oprofiled.c,v
retrieving revision 1.39
retrieving revision 1.40
diff -u -d -r1.39 -r1.40
--- oprofiled.c 2001/08/19 20:09:17 1.39
+++ oprofiled.c 2001/09/01 02:03:34 1.40
@@ -17,7 +17,10 @@
#include "oprofiled.h"
-extern struct opd_footer footer;
+extern double cpu_speed;
+extern u32 ctr_count[OP_MAX_COUNTERS];
+extern u8 ctr_event[OP_MAX_COUNTERS];
+extern u8 ctr_um[OP_MAX_COUNTERS];
static int showvers;
int verbose;
@@ -26,7 +29,7 @@
/* Unfortunately popt does not have, on many versions, the POPT_ARG_DOUBLE type
* so I must first store it as a string. */
static const char *cpu_speed_str;
-static int cpu_type;
+u32 cpu_type;
static int ignore_myself;
static int opd_buf_size=OP_DEFAULT_BUF_SIZE;
static char *opd_dir="/var/opd/";
@@ -48,7 +51,7 @@
static struct poptOption options[] = {
{ "buffer-size", 'b', POPT_ARG_INT, &opd_buf_size, 0, "nr. of entries in kernel buffer", "num", },
- { "use-cpu", 'p', POPT_ARG_INT, &cpu_type, 0, "0 for PPro, 1 for PII, 2 for PIII", "[0|1|2]" },
+ { "use-cpu", 'p', POPT_ARG_INT, &cpu_type, 0, "0 for PPro, 1 for PII, 2 for PIII, 3 for Athlon", "[0|1|2|3]" },
{ "ignore-myself", 'm', POPT_ARG_INT, &ignore_myself, 0, "ignore samples of oprofile driver", "[0|1]"},
{ "log-file", 'l', POPT_ARG_STRING, &logfilename, 0, "log file", "file", },
{ "base-dir", 'd', POPT_ARG_STRING, &opd_dir, 0, "base directory of daemon", "dir", },
@@ -152,7 +155,10 @@
{
poptContext optcon;
int ret;
+ int i, ok;
char c;
+ /* should be sufficient to hold /proc/sys/dev/oprofile/%d/yyyy */
+ char filename[PATH_MAX + 1];
/* Some old version of popt need the cast to char ** */
optcon = poptGetContext(NULL, argc, (char **)argv, options, 0);
@@ -184,38 +190,49 @@
exit(1);
}
- footer.magic = OPD_MAGIC;
- footer.version = OPD_VERSION;
- footer.ctr0_type_val = opd_read_int_from_file("/proc/sys/dev/oprofile/0/event");
- footer.ctr0_um = (u8) opd_read_int_from_file("/proc/sys/dev/oprofile/0/unit_mask");
- footer.ctr1_type_val = opd_read_int_from_file("/proc/sys/dev/oprofile/1/event");
- footer.ctr1_um = (u8) opd_read_int_from_file("/proc/sys/dev/oprofile/1/unit_mask");
-
- footer.ctr0_count = opd_read_int_from_file("/proc/sys/dev/oprofile/0/count");
- footer.ctr1_count = opd_read_int_from_file("/proc/sys/dev/oprofile/1/count");
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ sprintf(filename, "/proc/sys/dev/oprofile/%d/event", i);
+ ctr_event[i]= opd_read_int_from_file(filename);
- ret = op_check_events(footer.ctr0_type_val, footer.ctr1_type_val, footer.ctr0_um, footer.ctr1_um, cpu_type);
+ sprintf(filename, "/proc/sys/dev/oprofile/%d/count", i);
+ ctr_count[i]= opd_read_int_from_file(filename);
- if (ret & OP_CTR0_NOT_FOUND) fprintf(stderr, "oprofiled: ctr0: %d: no such event\n",footer. ctr0_type_val);
- if (ret & OP_CTR1_NOT_FOUND) fprintf(stderr, "oprofiled: ctr1: %d: no such event\n", footer.ctr1_type_val);
- if (ret & OP_CTR0_NO_UM) fprintf(stderr, "oprofiled: ctr0: 0x%.2x: invalid unit mask\n", footer.ctr0_um);
- if (ret & OP_CTR1_NO_UM) fprintf(stderr, "oprofiled: ctr1: 0x%.2x: invalid unit mask\n", footer.ctr1_um);
- if (ret & OP_CTR0_NOT_ALLOWED) fprintf(stderr, "oprofiled: ctr0: %d: can't count event\n", footer.ctr0_type_val);
- if (ret & OP_CTR1_NOT_ALLOWED) fprintf(stderr, "oprofiled: ctr1: %d: can't count event\n", footer.ctr1_type_val);
- if (ret & OP_CTR0_PII_EVENT) fprintf(stderr, "oprofiled: ctr0: %d: event only available on PII\n", footer.ctr0_type_val);
- if (ret & OP_CTR1_PII_EVENT) fprintf(stderr, "oprofiled: ctr1: %d: event only available on PII\n", footer.ctr1_type_val);
- if (ret & OP_CTR0_PIII_EVENT) fprintf(stderr, "oprofiled: ctr0: %d: event only available on PIII\n", footer.ctr0_type_val);
- if (ret & OP_CTR1_PIII_EVENT) fprintf(stderr, "oprofiled: ctr1: %d: event only available on PIII\n", footer.ctr1_type_val);
- if (ret & OP_CTR0_ATHLON_EVENT) fprintf(stderr, "oprofiled: ctr0: %d: event only available on AMD Athlon\n", footer.ctr0_type_val);
- if (ret & OP_CTR1_ATHLON_EVENT) fprintf(stderr, "oprofiled: ctr1: %d: event only available on AMD Athlon\n", footer.ctr1_type_val);
+ sprintf(filename, "/proc/sys/dev/oprofile/%d/unit_mask", i);
+ ctr_um[i]= opd_read_int_from_file(filename);
+ }
- if (ret != OP_EVENTS_OK) {
+ cpu_type = opd_read_int_from_file("/proc/sys/dev/oprofile/cpu_type");
+
+
+ ok = 1;
+ /* PHE FIXME : We need to translate cpu_type to a string, see
+ * cpu_typ_str in op_events.c */
+ for (i = 0 ; i < op_nr_counters ; ++i) {
+ ret = op_check_events(i, ctr_event[i], ctr_um[i], cpu_type);
+
+ if (ret & OP_EVT_NOT_FOUND)
+ fprintf(stderr, "oprofiled: ctr%d: %d: no such event for cpu %d\n",
+ i, ctr_event[i], cpu_type);
+
+ if (ret & OP_EVT_NO_UM)
+ fprintf(stderr, "oprofiled: ctr%d: 0x%.2x: invalid unit mask for cpu %d\n",
+ i, ctr_um[i], cpu_type);
+
+ if (ret & OP_EVT_CTR_NOT_ALLOWED)
+ fprintf(stderr, "oprofiled: ctr%d: %d: can't count event for this counter\n",
+ i, ctr_count[i]);
+
+ if (ret != OP_EVENTS_OK)
+ ok = 0;
+ }
+
+ if (!ok) {
poptPrintHelp(optcon, stderr, 0);
exit(1);
}
if (cpu_speed_str && strlen(cpu_speed_str)) {
- sscanf(cpu_speed_str, "%lf", &footer.cpu_speed);
+ sscanf(cpu_speed_str, "%lf", &cpu_speed);
}
}
@@ -339,7 +356,7 @@
struct op_mapping mapping;
/* prevent signals from messing us up */
- sigprocmask(SIG_BLOCK, &maskset, NULL);
+ sigprocmask(SIG_UNBLOCK, &maskset, NULL);
opd_stats[OPD_DUMP_COUNT]++;
Index: oprofiled.h
===================================================================
RCS file: /cvsroot/oprofile/oprofile/dae/oprofiled.h,v
retrieving revision 1.30
retrieving revision 1.31
diff -u -d -r1.30 -r1.31
--- oprofiled.h 2001/08/19 20:09:17 1.30
+++ oprofiled.h 2001/09/01 02:03:34 1.31
@@ -61,9 +61,6 @@
#define streq(a,b) (!strcmp((a), (b)))
#define streqn(a,b,n) (!strncmp((a), (b), (n)))
-/* this char replaces '/' in sample filenames */
-#define OPD_MANGLE_CHAR '}'
-
/* maximum nr. of kernel modules */
#define OPD_MAX_MODULES 64
@@ -84,48 +81,20 @@
OPD_MAX_STATS /* end of stats */
};
-/* event check returns */
-#define OP_EVENTS_OK 0x0
-#define OP_CTR0_NOT_FOUND 0x1
-#define OP_CTR1_NOT_FOUND 0x2
-#define OP_CTR0_NO_UM 0x4
-#define OP_CTR1_NO_UM 0x8
-#define OP_CTR0_NOT_ALLOWED 0x10
-#define OP_CTR1_NOT_ALLOWED 0x20
-#define OP_CTR0_PII_EVENT 0x40
-#define OP_CTR1_PII_EVENT 0x80
-#define OP_CTR0_PIII_EVENT 0x100
-#define OP_CTR1_PIII_EVENT 0x200
-#define OP_CTR0_ATHLON_EVENT 0x400
-#define OP_CTR1_ATHLON_EVENT 0x800
-
-/* FIXME : Carefull these are also present in pp/oprofpp.h */
-#define OPD_MAGIC 0xdeb6
-#define OPD_VERSION 0x4
-
-/* at the end of the sample files */
-struct opd_footer {
- u16 magic;
- u16 version;
- u8 is_kernel;
- u8 ctr0_type_val;
- u8 ctr1_type_val;
- u8 ctr0_um;
- u8 ctr1_um;
- u8 reserved1[16];
- u32 ctr0_count;
- u32 ctr1_count;
- /* Set to 0.0 if not available */
- double cpu_speed;
- time_t mtime;
- /* binary compatibility reserve */
- u32 reserved2[31];
+struct opd_sample_file {
+ fd_t fd;
+ /* mapped memory begin here */
+ struct opd_footer *footer;
+ /* start + sizeof(footer) ie. begin of map of samples */
+ void *start;
+ /* the size of mapped memory comes from the opd_image */
};
struct opd_image {
- fd_t fd;
- void *start;
+ struct opd_sample_file sample_files[OP_MAX_COUNTERS];
+ /* NOT counted the size of footer, to allow quick access check */
off_t len;
+ time_t mtime; /* image file mtime */
u8 kernel;
char *name;
};
@@ -155,8 +124,6 @@
struct opd_proc *prev;
struct opd_proc *next;
};
-
-int op_check_events(u8 ctr0_type, u8 ctr1_type, u8 ctr0_um, u8 ctr1_um, int proc);
void opd_get_ascii_procs(void);
void opd_init_images(void);
|
|
From: John L. <mov...@us...> - 2001-09-01 02:03:38
|
Update of /cvsroot/oprofile/oprofile/gui
In directory usw-pr-cvs1:/tmp/cvs-serv19169/gui
Modified Files:
Makefile.in oprofile
Log Message:
philippe's huge patch
Index: Makefile.in
===================================================================
RCS file: /cvsroot/oprofile/oprofile/gui/Makefile.in,v
retrieving revision 1.3
retrieving revision 1.4
diff -u -d -r1.3 -r1.4
--- Makefile.in 2001/08/18 16:11:17 1.3
+++ Makefile.in 2001/09/01 02:03:34 1.4
@@ -15,7 +15,7 @@
cd $(BINDIR) && chmod 755 oprofile
uninstall:
- rm -f $(BINDIR)/oprofile
+ -rm -f $(BINDIR)/oprofile
@echo "---------------------------------------------------------------"
@echo "The uninstall target does not remove your ~/.oprofile directory"
@echo "---------------------------------------------------------------"
Index: oprofile
===================================================================
RCS file: /cvsroot/oprofile/oprofile/gui/oprofile,v
retrieving revision 1.10
retrieving revision 1.11
diff -u -d -r1.10 -r1.11
--- oprofile 2001/08/16 02:23:01 1.10
+++ oprofile 2001/09/01 02:03:34 1.11
@@ -18,20 +18,55 @@
# To manipulate the global var needed by each counter.
set counter_setup_var { event_type counter unit_mask usr_count kernel_count }
-# This list is here to avoid maintaining separate list for each processor
-# type / counter number. It is the main data structure which describe all
-# counter event. When adding new processor/event/counter only this data
-# structure need update (adding counter require also work in the code).
+set processor_type -1
+set title ""
-# TODO : this must be shared in some way with op_events.c
+# Fixme : Celeron Copermine string ?, how to != between a celeron with
+# a PII core and a PIII core ? TODO look into the core source file.
+
+set cpuinfo_file [open /proc/cpuinfo r]
+while { [gets $cpuinfo_file var] != -1 } {
+
+ switch -regexp "$var" {
+ ".*Pentium Pro.*|.*PentiumPro.*" { set processor_type 0; set title "Pentium Pro" }
+ ".*Athlon.*|.*Duron.*|.*K7.*" { set processor_type 3; set title "K7/TB core" }
+ ".*Pentium III.*" { set processor_type 2; set title "Pentium III" }
+ ".*Coppermine.*" { set processor_type 2; set title "Pentium III (Coppermine)" }
+ ".*Celeron.*" { set processsor_type 2; set title "Pentium III (Coppermine)" }
+ ".*Pentium II.*" { set processor_type 1; set title "Pentium II" }
+ ".*Celeron.*" { set processor_type 1; set title "Celeron" }
+ }
+}
+
+# Add this line later when the profiler core support the P4.
+# ".*Pentium IV.*" { set processor_type 4; set title "Pentium IV" }
+
+close $cpuinfo_file
+
+if { [expr $processor_type == -1] } then {
+ set answer [tk_messageBox \
+ -message "Cannot detect processor : default to P2" -type okcancel]
+ switch $answer {
+ cancel { exit 1 }
+ }
+ set processor_type 1
+ set title "Pentium II"
+}
+# This list is the main data structure which describe all events available
+# for this cpu_type. The list is build from the output of:
+# op_help --gui-description
+# When adding new processor/event/counter only data struct in op_events.c
+# needs to be updated.
+#
# list entry item :
# 1 event type/name
# 2 bit mask
# 1 : avail for PPro
# 2 : avail for P2
# 4 : avail for P3
-# 8 : avail for P4 currently not supported.
+# 8 : AMD K7/Duron
+# 16 : avail for P4 currently not supported.
# 3 bit mask
# 1 : available for counter 0
# 2 : available for counter 1
@@ -45,291 +80,13 @@
# exclusive means than only one of the value can be passed to the profiler.
# 7 the default value
# 8 a list of : { unit_bit_mask_or_value string }+
-set event_type_list {
- { DISABLED -1 -1 "Select it to disable this counter" 0 }
- { CPU_CLK_UNHALTED 7 3 "Clock processor not halted" 6000 }
- { DATA_MEM_REFS 7 3 "All memory reference, cachable and non" 500 }
- { DCU_LINES_IN 7 3 "Total line allocated in DCU" 500 }
- { DCU_M_LINES_IN 7 3 "Number of M state lines allocated in DCU" 500 }
- { DCU_M_LINES_OUT 7 3 "Number of M lines evicted from the DCU" 500 }
- { DCU_MISS_OUTSTANDING 7 3 "Number of cycles while DCU miss outstanding" 500 }
- { IFU_IFETCH 7 3 "Number of non/cachable instruction fetches" 500 }
- { IFU_IFETCH_MISS 7 3 "Number of instruction fetch misses" 500 }
- { ITLB_MISS 7 3 "Number of ITLB misses" 500 }
- { IFU_MEM_STALL 7 3 "Cycles instruction fetch pipe is stalled" 500 }
- { ILD_STALL 7 3 "Cycles instruction length decoder is stalled" 500 }
- { L2_IFETCH 7 3 "Number of L2 instruction fetches" 500
- bit_mask 15
- {
- { 1 "(I)nvalid cache state" }
- { 2 "(S)hared cache state" }
- { 4 "(E)xclusive cache state" }
- { 8 "(M)odified cache state" }
- { 15 "MESI cache state" }
- }
- }
- { L2_LD 7 3 "Number of L2 data load" 500
- bit_mask 15
- {
- { 1 "(I)nvalid cache state" }
- { 2 "(S)hared cache state" }
- { 4 "(E)xclusive cache state" }
- { 8 "(M)odified cache state" }
- { 15 "MESI cache state" }
- }
- }
- { L2_ST 7 3 "Number of L2 data stores" 500
- bit_mask 15
- {
- { 1 "(I)nvalid cache state" }
- { 2 "(S)hared cache state" }
- { 4 "(E)xclusive cache state" }
- { 8 "(M)odified cache state" }
- { 15 "MESI cache state" }
- }
- }
- { L2_LINES_IN 7 3 "Number of allocated lines in L2" 500 }
- { L2_LINES_OUT 7 3 "Number of recovered lines from L2" 500 }
- { L2_M_LINES_INM 7 3 "Number of modified lines allocated in L2" 500 }
- { L2_M_LINES_OUTM 7 3 "Number of modified lines removed from L2" 500 }
- { L2_RQSTS 7 3 "Number of L2 requests" 500
- bit_mask 15
- {
- { 1 "(I)nvalid cache state" }
- { 2 "(S)hared cache state" }
- { 4 "(E)xclusive cache state" }
- { 8 "(M)odified cache state" }
- { 15 "MESI cache state" }
- }
- }
- { L2_ADS 7 3 "Number of L2 address strobes" 500 }
- { L2_DBUS_BUSY 7 3 "Number of cycles data bus was busy" 500 }
- { L2_DMUS_BUSY_RD 7 3 "Cycles data bus was busy in xfer from L2 to cpu" 500 }
- { BUS_DRDY_CLOCKS 7 3 "Number of clocks DRDY is asserted" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_LOCK_CLOCKS 7 3 "Number of clocks LOCK is asserted" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_REQ_OUTSTANDING 7 3 "Number of outstanding bus request" 500 }
- { BUS_TRAN_BRD 7 3 "Number of burst read transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_TRAN_RFO 7 3 "Number of read for ownership transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_TRANS_WB 7 3 "Number of write back transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_TRAN_IFETCH 7 3 "Number of instruction fetch transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_TRAN_INVAL 7 3 "Number of invalid transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_TRAN_PWR 7 3 "Number of partial write transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_TRANS_P 7 3 "Number of partial transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_TRANS_IO 7 3 "Number of I/O transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_TRANS_DEF 7 3 "Number of deferred transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_TRAN_BURST 7 3 "Number of burst transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_TRAN_ANY 7 3 "Number of all transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_TRAN_MEM 7 3 "Number of memory transactions" 500
- exclusive 32
- {
- { 0 "self-generated transactions" }
- { 32 "any transitions" }
- }
- }
- { BUS_DATA_RCV 7 3 "Bus cycles this processor is receiving data" 500 }
- { BUS_BNR_DRV 7 3 "bus cycles this processor is driving BNR pin" 500 }
- { BUS_HIT_DRV 7 3 "bus cycles this processor is driving HIT pin" 500 }
- { BUS_HITM_DRV 7 3 "bus cycles this processor is driving HITM pin" 500 }
- { BUS_SNOOP_STALL 7 3 "cycles during bus snoop stall" 500 }
- { COMP_FLOP_RET 7 1
- "Number of computational FP operations retired - Counter 0 only" 3000 }
- { FLOPS 7 1
- "Number of computational FP operations executed - Counter 0 only" 3000 }
- { FP_ASSIST 7 2
- "Number of exceptions handled by microcode - Counter 1 only" 500 }
- { MUL 7 2 "Number of multiplies - Counter 1 only" 1000 }
- { DIV 7 2 "Number of divides - Counter 1 only" 500 }
- { CYCLES_DIV_BUSY 7 1 "Cycles divider is busy - Counter 0 only" 1000 }
- { LD_BLOCKS 7 3 "Number of store blocks" 500 }
- { SB_DRAINS 7 3 "Number of store buffer drain cycles" 500 }
- { MISALIGN_MEM_REF 7 3 "Number of misaligned data memory references" 500 }
- { EMON_KNI_PREF_DISPATCHED 4 3
- "Number KNI pre-fetch/weakly ordered insns dispatch - P3 only" 500
- exclusive 0
- {
- { 0 "Prefecth NTA" }
- { 1 "Prefetch T1" }
- { 2 "Prefetch T2" }
- { 3 "weakly ordered stores" }
- }
- }
- { EMON_KNI_PREF_MISS 4 3
- "Number KNI pre-fetch/weakly ordered insns that miss all cache - P3 only" 500
- exclusive 0
- {
- { 0 "Prefecth NTA" }
- { 1 "Prefetch T1" }
- { 2 "Prefetch T2" }
- { 3 "weakly ordered stores" }
- }
- }
- { INST_RETIRED 7 3 "Number of instructions retired" 6000 }
- { UOPS_RETIRED 7 3 "Number of UOPs retired" 6000 }
- { INST_DECODED 7 3 "Number of instructions decoded" 6000 }
- { EMON_KNI_INST_RETIRED 4 3 "Number of KNI instructions retired - P3 only" 3000
- exclusive 0
- {
- { 0 "Packed and scalar" }
- { 1 "Packed" }
- }
- }
- { EMON_KNI_COMP_INST_RET 4 3
- "Number of KNI instructions computation retired - P3 only" 3000
- exclusive 0
- {
- { 0 "Packed and scalar" }
- { 1 "Packed" }
- }
- }
- { HW_INT_RX 7 3 "Number of hardware interrupts received" 500 }
- { CYCLES_INT_MASKED 7 3 "Cycles interrupt are disabled" 500 }
- { CYCLES_INT_PENDING_AND_MASKED 7 3
- "Cycles interrupts are disabled with pending interrupts" 500 }
- { BR_INST_RETIRED 7 3 "Number of branch instructions retired" 500 }
- { BR_MISS_PRED_RETIRED 7 3 "Number of mispredicted branches retired" 500 }
- { BR_TAKEN_RETIRED 7 3 "Number of taken branches retired" 500 }
- { BR_MISS_PRED_TAKEN_RET 7 3
- "Number of taken mispredictions branches retired" 500 }
- { BR_INST_DECODED 7 3 "Number of branches instructions decoded" 500 }
- { BTB_MISSES 7 3 "Number of branches that miss the BTB" 500 }
- { BR_BOGUS 7 3 "Number of bogus branches" 500 }
- { BACLEARS 7 3 "Number of times BACLEAR is asserted" 500 }
- { RESOURCE_STALLS 7 3 "Cycles during resource related stalls" 500 }
- { PARTIAL_RAT_STALLS 7 3 "cycles or event for partial stalls" 500 }
- { SEGMENT_REG_LOADS 7 3 "Number of segment register loads" 500 }
- { MMX_INSTR_EXEC 2 3 "Number of MMX instructions executed - P2 only" 3000 }
- { MMX_SAT_INSTR_EXEC 6 3
- "Number of MMX saturating instructions executed - P2/P3 only" 3000 }
- { MMX_UOPS_EXEC 6 3 "Number of MMX UOPs executed - P2/P3 only" 3000
- mandatory 15
- {
- { 15 "Mandatory" }
- }
- }
- { MMX_INSTR_TYPE_EXEC 6 3 "Number of MMX packing instructions - P2/P3 only" 3000
- bit_mask 63
- {
- { 1 "MMX packed multiplies" }
- { 2 "MMX packed shifts" }
- { 4 "MMX pack operations" }
- { 8 "MMX unpack operations" }
- { 16 "MMX packed logical" }
- { 32 "MMX packed arithmetic" }
- { 63 "MMX pack/unpack and all operations" }
- }
- }
- { FP_MMX_TRANS 6 3 "MMX-floating point transitions - P2/P3 only" 3000
- exclusive 0
- {
- { 0 "Transitions from MMX to floating point" }
- { 1 "Transitions from floating point to MMX" }
- }
- }
- { MMX_ASSIST 6 3 "Number of EMMS instructions executed - P2/P3 only" 500 }
- { MMX_INSTR_RET 2 3 "Number of MMX instructions retired - P2 only" 3000 }
- { SEG_RENAME_STALLS 6 3
- "Number of segment register renaming stalls - P2/P3 only" 3000
- bit_mask 15
- {
- { 1 "ES register" }
- { 2 "DS register" }
- { 4 "FS register" }
- { 8 "GS register" }
- { 15 "ES, DS, FS, GS register" }
- }
- }
- { SEG_REG_RENAMES 6 3 "Number of segment register renames - P2/P3 only" 3000
- bit_mask 15
- {
- { 1 "ES register" }
- { 2 "DS register" }
- { 4 "FS register" }
- { 8 "GS register" }
- { 15 "ES, DS, FS, GS register" }
- }
- }
- { RET_SEG_RENAMES 6 3
- "Number of segment register rename events retired - P2/P3 only" 3000 }
-}
+#
+# use op_help --gui-description to see the exact format
+#set processor_type 3
+
+set event_type_list [exec op_help --gui-description --cpu-type=$processor_type]
+
# This is defined in oprofile.h
set OP_MAX_PERF_COUNT 2147483647
@@ -358,8 +115,6 @@
}
}
-set event_type_0 "CPU_CLK_UNHALTED"
-
# In some situation it is incorrect to get the System.map and kernel
# filename from here but this provide a correct default value in many case.
set build_dir /lib/modules/[exec uname -r]/build
@@ -1408,7 +1163,6 @@
# The main procedure.
proc setup_event_frame { } {
-
create_event_frame .f_event_type 0
create_event_frame .f_event_type 1
@@ -1439,6 +1193,10 @@
}
}
+ set width [expr ([winfo reqw $f.f] + [winfo reqw $f.sb1] + 4) * 2]
+ set height [expr [winfo screenh .] * 3 / 4]
+ wm geometry . ${width}x${height}
+
# see comment on wm withdraw and update idletasks
wm deiconify .
}
@@ -1452,9 +1210,6 @@
# TODO : put all this code in setup_event_frame
-# TODO : how to get dynamically the correct width ?
-wm geometry . 576x600
-
frame .f_config_and_start_stop
set f .f_config_and_start_stop
@@ -1466,7 +1221,7 @@
frame .f_unit_mask -borderwidth 2 -relief sunken
create_text . status_bar 1 1
-pack .status_bar -side bottom -anchor w -fill y -expand yes
+pack .status_bar -side bottom -anchor w -fill both -expand yes
bind . <Motion> { onMouseMotion %W }
@@ -1489,38 +1244,6 @@
pack $f.b_advanced_setup $f.b_default_setup $f.b_quit_and_save_setup \
-side left -expand yes
-
-set processor_type -1
-set title ""
-
-# Fixme : Celeron Copermine string ?, how to != between a celeron with
-# a PII core and a PIII core ? TODO look into the core source file.
-
-set cpuinfo_file [open /proc/cpuinfo r]
-while { [gets $cpuinfo_file var] != -1 } {
-
- switch -regexp "$var" {
- ".*Pentium Pro .*" { set processor_type 0; set title "Pentium Pro" }
- ".*Pentium II .*" { set processor_type 1; set title "Pentium II" }
- ".*Celeron .*" { set processor_type 1; set title "Celeron" }
- ".*Pentium III .*" { set processor_type 2; set title "Pentium III" }
- }
-}
-
-# Add this line later when the profiler core support the P4.
-# ".*Pentium IV .*" { set processor_type 3; set title "Pentium IV" }
-
-close $cpuinfo_file
-
-if { [expr $processor_type == -1] } then {
- set answer [tk_messageBox \
- -message "Cannot detect processor : default to P2" -type okcancel]
- switch $answer {
- cancel { exit 1 }
- }
- set processor_type 1
- set title "Pentium II"
-}
wm title . "Profiling option for $title processor"
|
|
From: John L. <mov...@us...> - 2001-08-31 17:16:38
|
Update of /cvsroot/oprofile/oprofile
In directory usw-pr-cvs1:/tmp/cvs-serv5151
Modified Files:
ChangeLog Makefile.in oprofile.c oprofile.h
Added Files:
op_x86.c
Log Message:
mptable parse
--- NEW FILE: op_x86.c ---
/*
* op_x86.c
*
* A variety of Intel hardware grubbery.
*
* Based in part on arch/i386/kernel/mpparse.c
*/
#include <linux/mm.h>
#include <linux/init.h>
#include <linux/config.h>
#include <asm/smp.h>
#include <asm/mpspec.h>
#include <asm/io.h>
static int __init mpf_checksum(unsigned char *mp, int len)
{
int sum = 0;
while (len--)
sum += *mp++;
return sum & 0xFF;
}
static int __init mpf_table_ok(struct intel_mp_floating * mpf, unsigned long *bp)
{
if (*bp != SMP_MAGIC_IDENT)
return 0;
if (mpf->mpf_length != 1)
return 0;
if (mpf_checksum((unsigned char *)bp, 16))
return 0;
return (mpf->mpf_specification == 1 || mpf->mpf_specification == 4);
}
static int __init smp_scan_config (unsigned long base, unsigned long length)
{
unsigned long *bp = phys_to_virt(base);
struct intel_mp_floating *mpf;
while (length > 0) {
mpf = (struct intel_mp_floating *)bp;
if (mpf_table_ok(mpf, bp))
return 1;
bp += 4;
length -= 16;
}
return 0;
}
int __init find_intel_smp (void)
{
unsigned int address;
/*
* FIXME: Linux assumes you have 640K of base ram..
* this continues the error...
*
* 1) Scan the bottom 1K for a signature
* 2) Scan the top 1K of base RAM
* 3) Scan the 64K of bios
*/
if (smp_scan_config(0x0,0x400) ||
smp_scan_config(639*0x400,0x400) ||
smp_scan_config(0xF0000,0x10000))
return 1;
/*
* If it is an SMP machine we should know now, unless the
* configuration is in an EISA/MCA bus machine with an
* extended bios data area.
*
* there is a real-mode segmented pointer pointing to the
* 4K EBDA area at 0x40E, calculate and scan it here.
*
* NOTE! There are Linux loaders that will corrupt the EBDA
* area, and as such this kind of SMP config may be less
* trustworthy, simply because the SMP table may have been
* stomped on during early boot. These loaders are buggy and
* should be fixed.
*/
address = *(unsigned short *)phys_to_virt(0x40E);
address <<= 4;
return smp_scan_config(address, 0x1000);
}
Index: ChangeLog
===================================================================
RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v
retrieving revision 1.90
retrieving revision 1.91
diff -u -d -r1.90 -r1.91
--- ChangeLog 2001/08/20 20:05:19 1.90
+++ ChangeLog 2001/08/31 17:16:35 1.91
@@ -1,3 +1,11 @@
+2001-08-31 John Levon <mo...@co...>
+
+ * Makefile.in:
+ * op_x86.c:
+ * oprofile.h:
+ * oprofile.c: a prototype detector for SMP hardware,
+ some small cleanup
+
2001-08-20 Dave Jones <da...@su...>
* doc/oprofile.sgml:
Index: Makefile.in
===================================================================
RCS file: /cvsroot/oprofile/oprofile/Makefile.in,v
retrieving revision 1.23
retrieving revision 1.24
diff -u -d -r1.23 -r1.24
--- Makefile.in 2001/08/18 16:11:17 1.23
+++ Makefile.in 2001/08/31 17:16:35 1.24
@@ -74,8 +74,8 @@
KCFLAGS := $(BKCFLAGS) -march=i686
ASMFLAGS := -D__ASSEMBLY__ -DMODULE -D__KERNEL__ -traditional
-oprofile.o: op_init.o oprofile_c.o oprofile_nmi.o op_syscalls.o op_events.o
- ld -r -o $@ op_init.o oprofile_c.o oprofile_nmi.o op_syscalls.o op_events.o
+oprofile.o: op_init.o oprofile_c.o oprofile_nmi.o op_syscalls.o op_events.o op_x86.o
+ ld -r -o $@ op_init.o oprofile_c.o oprofile_nmi.o op_syscalls.o op_events.o op_x86.o
oprofile.h: op_user.h version.h
@@ -83,6 +83,9 @@
$(CC) $(BKCFLAGS) -c -o $@ $<
oprofile_c.o: oprofile.c oprofile.h
+ $(CC) $(KCFLAGS) -c -o $@ $<
+
+op_x86.o: op_x86.c
$(CC) $(KCFLAGS) -c -o $@ $<
op_syscalls.o: op_syscalls.c oprofile.h
Index: oprofile.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/oprofile.c,v
retrieving revision 1.74
retrieving revision 1.75
diff -u -d -r1.74 -r1.75
--- oprofile.c 2001/08/20 19:46:13 1.74
+++ oprofile.c 2001/08/31 17:16:35 1.75
@@ -50,6 +50,7 @@
static u32 prof_on __cacheline_aligned;
+static int smp_hardware;
static int op_major;
int cpu_type;
@@ -261,48 +262,15 @@
apic_write(APIC_LVTPC, old_lvtpc[smp_processor_id()]);
}
-/* PHE bit 8 of APIC_BASE_MSR is the BSP flags, it is set by harware at boot
- * up sequence and would be keeped ok by software to allow #INIT or #ERR, else
- * all the processor on the next MP sequence fall in shutdown.
- *
- * On UP architecture it should always be zero (ok for my PII) 7.4.8 (7.5.8),
- * I am unsure but I have presumption than kernel cannot boot properly if bsp
- * flags is set on UP machine.
- *
- * I think it is ok in all case, perhaps corner case on SMP with only one
- * processor plugged ?
- */
-inline static int is_bsp_proc(void)
-{
- uint l;
- uint h;
- rdmsr(MSR_IA32_APICBASE, l, h);
- wrmsr(MSR_IA32_APICBASE, l & ~(1<<11), h);
-
- return (h & (1 << 8)) ? 1 : 0;
-}
-
static int __init apic_needs_setup(void)
{
- /* FIXME: we need to detect UP kernel, SMP hardware, and
- * return 0 here, to avoid screwing up the APIC
- */
- /* PHE: for UP kernel SMP machine this processor should have the bsp
- * flags set, something like:
-
-#ifndef X86_UP_APIC
- if (smp_num_cpus == 1) {
- return !is_bsp_proc();
- }
-#endif
- return 0;
-
- would work
- */
return
/* if enabled, the kernel has already set it up */
#ifdef CONFIG_X86_UP_APIC
0 &&
+#else
+/* otherwise, we detect SMP hardware via the MP table */
+ !smp_hardware &&
#endif
smp_num_cpus == 1;
}
@@ -320,10 +288,12 @@
uint val;
if (!apic_needs_setup()) {
+ printk(KERN_INFO "oprofile: no APIC setup needed.\n");
lvtpc_apic_setup(NULL);
return 0;
}
+ printk(KERN_INFO "oprofile: setting up APIC.\n");
/* ugly hack */
/* PHE : the memory must be uncachable! this is perhaps more a
@@ -555,8 +525,7 @@
{
int i;
- threadpid = current->pid;
-
+ // FIXME: kernel lock ?
daemonize();
sprintf(current->comm, "oprof-thread");
siginitsetinv(¤t->blocked, sigmask(SIGKILL));
@@ -575,6 +544,7 @@
/* FIXME: determine best value here */
schedule_timeout(HZ/10);
+ // FIXME: signal pending
if (diethreaddie)
break;
}
@@ -587,7 +557,7 @@
{
init_completion(&threadstop);
diethreaddie = 0;
- if (kernel_thread(oprof_thread, NULL, CLONE_FS|CLONE_FILES|CLONE_SIGHAND)<0)
+ if ((threadpid = kernel_thread(oprof_thread, NULL, CLONE_FS|CLONE_FILES|CLONE_SIGHAND)) < 0)
printk(KERN_ERR "oprofile: couldn't spawn wakeup thread.\n");
}
@@ -1186,7 +1156,9 @@
{
int err;
- printk(KERN_INFO "%s\n",op_version);
+ printk(KERN_INFO "%s\n", op_version);
+
+ smp_hardware = find_intel_smp();
if ((err = apic_setup()))
return err;
Index: oprofile.h
===================================================================
RCS file: /cvsroot/oprofile/oprofile/oprofile.h,v
retrieving revision 1.46
retrieving revision 1.47
diff -u -d -r1.46 -r1.47
--- oprofile.h 2001/08/19 20:09:17 1.46
+++ oprofile.h 2001/08/31 17:16:35 1.47
@@ -32,6 +32,9 @@
/* userspace/module interface */
#include "op_user.h"
+#undef min
+#undef max
+
#define OP_NR_ENTRY (SMP_CACHE_BYTES/sizeof(struct op_sample))
struct op_entry {
@@ -226,6 +229,7 @@
extern uint ctrreg3;
int oprof_init(void);
+int find_intel_smp(void);
void oprof_exit(void);
void my_set_fixmap(void);
int op_min_count(u8 ctr_type);
|
|
From: Dave J. <da...@us...> - 2001-08-20 20:05:22
|
Update of /cvsroot/oprofile/oprofile/doc In directory usw-pr-cvs1:/tmp/cvs-serv31594/doc Modified Files: oprofile.sgml Log Message: Film at 11: Programmer updates documentation. Index: oprofile.sgml =================================================================== RCS file: /cvsroot/oprofile/oprofile/doc/oprofile.sgml,v retrieving revision 1.20 retrieving revision 1.21 diff -u -d -r1.20 -r1.21 --- oprofile.sgml 2001/08/18 16:11:17 1.20 +++ oprofile.sgml 2001/08/20 20:05:19 1.21 @@ -53,11 +53,15 @@ </para></listitem> </varlistentry> <varlistentry> - <term>Intel P6 processor</term> + <term>Intel P6 processor or AMD Athlon/Duron.</term> <listitem><para> - An Intel P6 processor is required. In marketing terms this translates to anything between a Pentium Pro - (NOT Pentium Classics) and a Pentium III. Specifically, Pentium IVs are not yet supported due to different - hardware. Also note that Mobile P6 processors lack the necessary CPU features and are also not supported. + A CPU with a P6 generation core is required. In marketing terms this translates to anything between an Intel + Pentium Pro (NOT Pentium Classics) and a Pentium III. Specifically, Pentium IVs are not yet supported due to + different hardware. + Also note that Mobile P6 processors lack the necessary CPU features and are also not supported. + The AMD Athlon & Duron CPUs are also supported, as their <acronym>APIC</acronym> and performance counters + are very similar. These CPUs offer 4 seperate counters as opposed to the 2 provided by the Intel P6 family. + OProfile currently only supports using 2 of these available counters. </para></listitem> </varlistentry> <varlistentry> @@ -334,8 +338,8 @@ <varlistentry> <term><option>--use-cpu</option></term> <listitem><para> - Specify 0 for Pentium Pros, 1 for Pentium II and 2 for Pentium III. This is only needed if you use - an event not available on the Pentium Pro. + Specify 0 for Pentium Pros, 1 for Pentium II, 2 for Pentium III and 3 for AMD Athlon/Duron. + This is only needed if you use an event not available on the Pentium Pro. </para></listitem> </varlistentry> <varlistentry> @@ -379,9 +383,11 @@ <title>Intel P6 Performance Counters</title> <para> The hardware performance counters are detailed in the Intel IA-32 Architecture Manual, Volume 3, available -from <ulink url="http://developer.intel.com/">http://developer.intel.com/</ulink>. P6-core processors are capable -of delivering an interrupt to the local <acronym>APIC</acronym> <acronym>LVTPC</acronym> vector -when a counter overflows. This is the basic mechanism on which OProfile is based. The kernel module +from <ulink url="http://developer.intel.com/">http://developer.intel.com/</ulink>. The AMD Athlon/Duron +implementation is detailed in <ulink url="http://www.amd.com/products/cpg/athlon/techdocs/pdf/22007.pdf"> +http://www.amd.com/products/cpg/athlon/techdocs/pdf/22007.pdf</ulink>. +P6-core processors are capable of delivering an interrupt to the local <acronym>APIC</acronym> <acronym>LVTPC</acronym> +vector when a counter overflows. This is the basic mechanism on which OProfile is based. The kernel module installs an interrupt handler for this vector. The delivery mode is set to <acronym>NMI</acronym> so that blocking interrupts in the kernel does not prevent profiling. When the interrupt handler is called, the current <acronym>EIP</acronym> <acronym>PC</acronym> value, process id, and counter (there are only |
|
From: Dave J. <da...@us...> - 2001-08-20 20:05:22
|
Update of /cvsroot/oprofile/oprofile In directory usw-pr-cvs1:/tmp/cvs-serv31594 Modified Files: ChangeLog Log Message: Film at 11: Programmer updates documentation. Index: ChangeLog =================================================================== RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v retrieving revision 1.89 retrieving revision 1.90 diff -u -d -r1.89 -r1.90 --- ChangeLog 2001/08/20 19:49:54 1.89 +++ ChangeLog 2001/08/20 20:05:19 1.90 @@ -1,5 +1,7 @@ 2001-08-20 Dave Jones <da...@su...> + * doc/oprofile.sgml: + Updated to reflect new Athlon/Duron capabilities. * pp/oprof_convert.c: * pp/oprofpp.c: use EXIT_FAILURE | EXIT_SUCCESS in exit() calls. |
|
From: Dave J. <da...@us...> - 2001-08-20 19:49:57
|
Update of /cvsroot/oprofile/oprofile/pp
In directory usw-pr-cvs1:/tmp/cvs-serv24720/pp
Modified Files:
oprof_convert.c
Log Message:
Oops. Missed these ones.
More EXIT_FAILURE | EXIT_SUCCESS changes.
Index: oprof_convert.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/pp/oprof_convert.c,v
retrieving revision 1.7
retrieving revision 1.8
diff -u -d -r1.7 -r1.8
--- oprof_convert.c 2001/07/27 23:22:39 1.7
+++ oprof_convert.c 2001/08/20 19:49:54 1.8
@@ -237,20 +237,20 @@
if (argc <= 1) {
fprintf(stderr, "Syntax: %s filename [filenames]\n", argv[0]);
- exit(1);
+ exit(EXIT_FAILURE);
}
/* Should use popt in future if new options are added */
if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-v") == 0) {
printf("%s : " VERSION_STRING " compiled on " __DATE__ " " __TIME__ "\n", argv[0]);
- exit(0);
+ exit(EXIT_SUCCESS);
}
if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-h") == 0) {
printf("Syntax : %s filename [filenames]\n", argv[0]);
- exit(0);
+ exit(EXIT_SUCCESS);
}
for (i = 1 ; i < argc ; ++i) {
|
|
From: Dave J. <da...@us...> - 2001-08-20 19:49:57
|
Update of /cvsroot/oprofile/oprofile In directory usw-pr-cvs1:/tmp/cvs-serv24720 Modified Files: ChangeLog Log Message: Oops. Missed these ones. More EXIT_FAILURE | EXIT_SUCCESS changes. Index: ChangeLog =================================================================== RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v retrieving revision 1.88 retrieving revision 1.89 diff -u -d -r1.88 -r1.89 --- ChangeLog 2001/08/20 19:46:13 1.88 +++ ChangeLog 2001/08/20 19:49:54 1.89 @@ -1,5 +1,6 @@ 2001-08-20 Dave Jones <da...@su...> + * pp/oprof_convert.c: * pp/oprofpp.c: use EXIT_FAILURE | EXIT_SUCCESS in exit() calls. * oprofile.c: Make non-exported functions static |
|
From: Dave J. <da...@us...> - 2001-08-20 19:46:17
|
Update of /cvsroot/oprofile/oprofile
In directory usw-pr-cvs1:/tmp/cvs-serv23028
Modified Files:
ChangeLog oprofile.c
Log Message:
* pp/oprofpp.c: use EXIT_FAILURE | EXIT_SUCCESS in exit() calls.
* oprofile.c: Make non-exported functions static and remove some
duplicate definitions.
Index: ChangeLog
===================================================================
RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v
retrieving revision 1.87
retrieving revision 1.88
diff -u -d -r1.87 -r1.88
--- ChangeLog 2001/08/19 20:23:02 1.87
+++ ChangeLog 2001/08/20 19:46:13 1.88
@@ -1,3 +1,10 @@
+2001-08-20 Dave Jones <da...@su...>
+
+ * pp/oprofpp.c: use EXIT_FAILURE | EXIT_SUCCESS
+ in exit() calls.
+ * oprofile.c: Make non-exported functions static
+ and remove duplicate definitions.
+
2001-08-19 John Levon <mo...@co...>
* dae/op_start: handle celeries in cpu type
Index: oprofile.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/oprofile.c,v
retrieving revision 1.73
retrieving revision 1.74
diff -u -d -r1.73 -r1.74
--- oprofile.c 2001/08/19 20:09:17 1.73
+++ oprofile.c 2001/08/20 19:46:13 1.74
@@ -39,16 +39,16 @@
pid_t pgrp_filter;
/* the MSRs we need */
-uint ctrlreg0 = MSR_IA32_EVNTSEL0;
-uint ctrlreg1 = MSR_IA32_EVNTSEL1;
-uint ctrlreg2 = MSR_K7_PERFCTL2;
-uint ctrlreg3 = MSR_K7_PERFCTL3;
-uint ctrreg0 = MSR_IA32_PERFCTR0;
-uint ctrreg1 = MSR_IA32_PERFCTR1;
-uint ctrreg2 = MSR_K7_PERFCTR2;
-uint ctrreg3 = MSR_K7_PERFCTR3;
+static u32 ctrlreg0 = MSR_IA32_EVNTSEL0;
+static u32 ctrlreg1 = MSR_IA32_EVNTSEL1;
+static u32 ctrlreg2 = MSR_K7_PERFCTL2;
+static u32 ctrlreg3 = MSR_K7_PERFCTL3;
+static u32 ctrreg0 = MSR_IA32_PERFCTR0;
+static u32 ctrreg1 = MSR_IA32_PERFCTR1;
+static u32 ctrreg2 = MSR_K7_PERFCTR2;
+static u32 ctrreg3 = MSR_K7_PERFCTR3;
-u32 prof_on __cacheline_aligned;
+static u32 prof_on __cacheline_aligned;
static int op_major;
int cpu_type;
@@ -61,15 +61,6 @@
extern spinlock_t map_lock;
-extern unsigned int ctrlreg0;
-extern unsigned int ctrlreg1;
-extern unsigned int ctrlreg2;
-extern unsigned int ctrlreg3;
-extern unsigned int ctrreg0;
-extern unsigned int ctrreg1;
-extern unsigned int ctrreg2;
-extern unsigned int ctrreg3;
-
/* ---------------- NMI handler ------------------ */
/* FIXME: this whole handler would probably be better in straight asm */
@@ -152,14 +143,14 @@
/* this masking code is unsafe and nasty but might deal with the small
* race when installing the NMI entry into the IDT
*/
-void mask_lvtpc(void * e)
+static void mask_lvtpc(void * e)
{
ulong v = apic_read(APIC_LVTPC);
lvtpc_masked = v & APIC_LVT_MASKED;
apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
}
-void unmask_lvtpc(void * e)
+static void unmask_lvtpc(void * e)
{
if (!lvtpc_masked)
apic_write(APIC_LVTPC, apic_read(APIC_LVTPC) & ~APIC_LVT_MASKED);
@@ -560,7 +551,7 @@
/* we have to have another thread because we can't
* do wake_up() from NMI due to no locking
*/
-int oprof_thread(void *arg)
+static int oprof_thread(void *arg)
{
int i;
@@ -592,7 +583,7 @@
return 0;
}
-void oprof_start_thread(void)
+static void oprof_start_thread(void)
{
init_completion(&threadstop);
diethreaddie = 0;
@@ -600,7 +591,7 @@
printk(KERN_ERR "oprofile: couldn't spawn wakeup thread.\n");
}
-void oprof_stop_thread(void)
+static void oprof_stop_thread(void)
{
diethreaddie = 1;
kill_proc(SIGKILL, threadpid, 1);
@@ -1124,7 +1115,7 @@
/* NOTE: we do *not* support sysctl() syscall */
-int __init init_sysctl(void)
+static int __init init_sysctl(void)
{
ctl_table *next = &oprof_table[nr_oprof_static];
ctl_table *counter_table[OP_MAX_COUNTERS + 1];
@@ -1166,7 +1157,7 @@
return -EFAULT;
}
-void __exit cleanup_sysctl(void)
+static void __exit cleanup_sysctl(void)
{
int i;
ctl_table *next = &oprof_table[nr_oprof_static];
|
|
From: Dave J. <da...@us...> - 2001-08-20 19:46:17
|
Update of /cvsroot/oprofile/oprofile/pp
In directory usw-pr-cvs1:/tmp/cvs-serv23028/pp
Modified Files:
oprofpp.c
Log Message:
* pp/oprofpp.c: use EXIT_FAILURE | EXIT_SUCCESS in exit() calls.
* oprofile.c: Make non-exported functions static and remove some
duplicate definitions.
Index: oprofpp.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/pp/oprofpp.c,v
retrieving revision 1.35
retrieving revision 1.36
diff -u -d -r1.35 -r1.36
--- oprofpp.c 2001/07/25 03:51:33 1.35
+++ oprofpp.c 2001/08/20 19:46:13 1.36
@@ -120,19 +120,19 @@
poptBadOption(optcon, POPT_BADOPTION_NOALIAS),
poptStrerror(c));
poptPrintHelp(optcon, stderr, 0);
- exit(1);
+ exit(EXIT_FAILURE);
}
if (showvers) {
printf(VERSION_STRING " compiled on " __DATE__ " " __TIME__ "\n");
- exit(0);
+ exit(EXIT_SUCCESS);
}
if (!list_all_symbols_details && !list_symbols &&
!gproffile && !symbol) {
fprintf(stderr, "oprofpp: no mode specified. What do you want from me ?\n");
poptPrintHelp(optcon, stderr, 0);
- exit(1);
+ exit(EXIT_FAILURE);
}
/* non-option file, either a sample or binary image file */
@@ -159,7 +159,7 @@
if (!imagefile) {
fprintf(stderr, "oprofpp: no samples file specified.\n");
poptPrintHelp(optcon, stderr, 0);
- exit(1);
+ exit(EXIT_FAILURE);
} else {
/* we'll "leak" this memory */
samplefile = remangle(imagefile);
@@ -264,19 +264,19 @@
fprintf(stderr, "oprofpp: the last modified time of the binary file %s does not match "
"that of the sample file. Either this is the wrong binary or the binary "
"has been modified since the sample file was created.\n", file);
- exit(1);
+ exit(EXIT_FAILURE);
}
ibfd = bfd_openr(file, NULL);
if (!ibfd) {
fprintf(stderr,"oprofpp: bfd_openr of %s failed.\n", file);
- exit(1);
+ exit(EXIT_FAILURE);
}
if (!bfd_check_format_matches(ibfd, bfd_object, &matching)) {
fprintf(stderr,"oprofpp: BFD format failure for %s.\n", file);
- exit(1);
+ exit(EXIT_FAILURE);
}
if (!imagefile)
@@ -305,7 +305,7 @@
{
if (num - sect_offset > num) {
fprintf(stderr,"oprofpp: less than zero offset ? \n");
- exit(1);
+ exit(EXIT_FAILURE);
}
/* adjust for kernel images */
@@ -476,7 +476,7 @@
if (!num) {
fprintf(stderr, "oprofpp: couldn't get any symbols from image file.\n");
- exit(1);
+ exit(EXIT_FAILURE);
}
@@ -487,11 +487,11 @@
get_symbol_range(syms[i], (i == num-1) ? NULL : syms[i+1], &start, &end);
if (start >= nr_samples) {
fprintf(stderr,"oprofpp: start 0x%x out of range (max 0x%x)\n", start, nr_samples);
- exit(1);
+ exit(EXIT_FAILURE);
}
if (end > nr_samples) {
fprintf(stderr,"oprofpp: end 0x%x out of range (max 0x%x)\n", end, nr_samples);
- exit(1);
+ exit(EXIT_FAILURE);
}
for (j = start; j < end; j++) {
@@ -542,7 +542,7 @@
if (!num) {
fprintf(stderr, "oprofpp: couldn't get any symbols from image file.\n");
- exit(1);
+ exit(EXIT_FAILURE);
}
for (i=0; i < num; i++) {
@@ -599,7 +599,7 @@
if (!num) {
fprintf(stderr, "oprofpp: couldn't get any symbols from image file.\n");
- exit(1);
+ exit(EXIT_FAILURE);
}
fp=opd_open_file(gproffile, "w");
@@ -727,7 +727,7 @@
if (!num) {
fprintf(stderr, "oprofpp: couldn't get any symbols from image file.\n");
- exit(1);
+ exit(EXIT_FAILURE);
}
for (i = 0 ; i < num ; ++i) {
@@ -797,28 +797,28 @@
fp = fopen(samplefile,"r");
if (!fp) {
fprintf(stderr, "oprofpp: fopen of %s failed. %s\n", samplefile, strerror(errno));
- exit(1);
+ exit(EXIT_FAILURE);
}
if (fseek(fp, -sizeof(struct opd_footer), SEEK_END) == -1) {
fprintf(stderr, "oprofpp: fseek of %s failed. %s\n", samplefile, strerror(errno));
- exit(1);
+ exit(EXIT_FAILURE);
}
if (fread(&footer, sizeof(struct opd_footer), 1, fp) != 1) {
fprintf(stderr, "oprofpp: fread of %s failed. %s\n", samplefile, strerror(errno));
- exit(1);
+ exit(EXIT_FAILURE);
}
fclose(fp);
if (footer.magic != OPD_MAGIC) {
fprintf(stderr, "oprofpp: wrong magic 0x%x, expected 0x%x.\n", footer.magic, OPD_MAGIC);
- exit(1);
+ exit(EXIT_FAILURE);
}
if (footer.version != OPD_VERSION) {
fprintf(stderr, "oprofpp: wrong version 0x%x, expected 0x%x.\n", footer.version, OPD_VERSION);
- exit(1);
+ exit(EXIT_FAILURE);
}
if (footer.ctr0_type_val) {
@@ -839,7 +839,7 @@
if (fd == -1) {
fprintf(stderr, "oprofpp: Opening %s failed. %s\n", samplefile, strerror(errno));
- exit(1);
+ exit(EXIT_FAILURE);
}
size = opd_get_fsize(samplefile, 1) - sizeof(struct opd_footer);
@@ -847,7 +847,7 @@
if (samples == (void *)-1) {
fprintf(stderr, "oprofpp: mmap of %s failed. %s\n", samplefile, strerror(errno));
- exit(1);
+ exit(EXIT_FAILURE);
}
nr_samples = size/sizeof(struct opd_fentry);
|
|
From: John L. <mov...@us...> - 2001-08-19 20:23:05
|
Update of /cvsroot/oprofile/oprofile/dae In directory usw-pr-cvs1:/tmp/cvs-serv6244/dae Modified Files: op_start Log Message: recognise celerons in cpu type Index: op_start =================================================================== RCS file: /cvsroot/oprofile/oprofile/dae/op_start,v retrieving revision 1.35 retrieving revision 1.36 diff -u -d -r1.35 -r1.36 --- op_start 2001/08/19 20:09:17 1.35 +++ op_start 2001/08/19 20:23:02 1.36 @@ -57,7 +57,15 @@ *Athlon* | *Duron* | *K7*) CPUTYPE=3 ;; + # FIXME: I'm not sure this is right ... + *Coppermine*) + CPUTYPE=2 + ;; + *Celeron*) + CPUTYPE=1 + ;; *) + echo "Unknown CPU type. Please send /proc/cpuinfo to mo...@co..." CPUTYPE=0 ;; esac |
|
From: John L. <mov...@us...> - 2001-08-19 20:23:05
|
Update of /cvsroot/oprofile/oprofile In directory usw-pr-cvs1:/tmp/cvs-serv6244 Modified Files: ChangeLog Log Message: recognise celerons in cpu type Index: ChangeLog =================================================================== RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v retrieving revision 1.86 retrieving revision 1.87 diff -u -d -r1.86 -r1.87 --- ChangeLog 2001/08/19 20:09:17 1.86 +++ ChangeLog 2001/08/19 20:23:02 1.87 @@ -1,5 +1,9 @@ 2001-08-19 John Levon <mo...@co...> + * dae/op_start: handle celeries in cpu type + +2001-08-19 John Levon <mo...@co...> + * op_init.c: * oprofile.h: * oprofile.c: small tidies of the Athlon support |
|
From: John L. <mov...@us...> - 2001-08-19 20:09:20
|
Update of /cvsroot/oprofile/oprofile/dae
In directory usw-pr-cvs1:/tmp/cvs-serv3584/dae
Modified Files:
op_start oprofiled.c oprofiled.h
Log Message:
the athlon patch. more work needed, but it works. Thanks Dave !!
Index: op_start
===================================================================
RCS file: /cvsroot/oprofile/oprofile/dae/op_start,v
retrieving revision 1.34
retrieving revision 1.35
diff -u -d -r1.34 -r1.35
--- op_start 2001/08/16 02:23:01 1.34
+++ op_start 2001/08/19 20:09:17 1.35
@@ -34,7 +34,6 @@
CTR0_KERNEL=1
CTR1_USER=1
CTR1_KERNEL=1
-USE_PII=0
IGNORE_MYSELF=0
DIR="/var/opd"
MAP_FILE=
@@ -43,6 +42,30 @@
PGRP_FILTER=0
VERBOSE=0
+if test -f /proc/cpuinfo; then
+ modelname=`cat /proc/cpuinfo | grep "model\ name\ :" | sed -e 's/ //g' | cut -d':' -f2`
+ case $modelname in
+ *PentiumPro*)
+ CPUTYPE=0
+ ;;
+ *PentiumIII*)
+ CPUTYPE=2
+ ;;
+ *PentiumII*)
+ CPUTYPE=1
+ ;;
+ *Athlon* | *Duron* | *K7*)
+ CPUTYPE=3
+ ;;
+ *)
+ CPUTYPE=0
+ ;;
+ esac
+
+else
+ CPUTYPE=0
+fi
+
# FIXME: needs to be generalised for more counters (e.g. athlon)
do_help() {
@@ -65,7 +88,7 @@
--pgrp-filter=pgrp Only profile process group pgrp (if compiled in)
Daemon options
- --use-cpu=[0|1|2] 0 or PPro, 1 for PII, 2 for PIII
+ --use-cpu=[0|1|2|3] 0 or PPro, 1 for PII, 2 for PIII, 3 for AMD Athlon.
--ignore-myself=[0|1] ignore samples for oprofiled
--log-file=file log file
--base-dir=dir base directory of daemon
@@ -121,9 +144,7 @@
case "$arg" in
--use-cpu)
if [ "$val" != "" ]; then
- USE_PII=$val
- else
- USE_PII=1
+ CPUTYPE=$val
fi
;;
--ignore-myself)
@@ -305,7 +326,7 @@
echo "CTR1_COUNT $CTR1_COUNT"
echo "CTR0_OSUSR $CTR0_OSUSR"
echo "CTR1_OSUSR $CTR1_OSUSR"
- echo "USE_PII $USE_PII"
+ echo "CPUTYPE $CPUTYPE"
echo "IGNORE_MYSELF $IGNORE_MYSELF"
echo "DIR $DIR"
echo "LOG_FILE $LOG_FILE"
@@ -421,7 +442,7 @@
$SYSCTL -w dev.oprofile.1.event=$CTR1_EVENT_VAL >/dev/null
fi
-OPD_ARGS="--buffer-size=$BUF_SIZE --use-cpu=$USE_PII --ignore-myself=$IGNORE_MYSELF \
+OPD_ARGS="--buffer-size=$BUF_SIZE --use-cpu=$CPUTYPE --ignore-myself=$IGNORE_MYSELF \
--log-file=$LOG_FILE --base-dir=$DIR --samples-dir=$SAMPLES_DIR \
--device-file=$DEVICE_FILE --kernel-only=$KERNEL_ONLY \
--hash-map-device-file=$HASH_MAP_DEVICE_FILE --vmlinux=$VMLINUX"
Index: oprofiled.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/dae/oprofiled.c,v
retrieving revision 1.38
retrieving revision 1.39
diff -u -d -r1.38 -r1.39
--- oprofiled.c 2001/07/25 02:22:44 1.38
+++ oprofiled.c 2001/08/19 20:09:17 1.39
@@ -206,6 +206,8 @@
if (ret & OP_CTR1_PII_EVENT) fprintf(stderr, "oprofiled: ctr1: %d: event only available on PII\n", footer.ctr1_type_val);
if (ret & OP_CTR0_PIII_EVENT) fprintf(stderr, "oprofiled: ctr0: %d: event only available on PIII\n", footer.ctr0_type_val);
if (ret & OP_CTR1_PIII_EVENT) fprintf(stderr, "oprofiled: ctr1: %d: event only available on PIII\n", footer.ctr1_type_val);
+ if (ret & OP_CTR0_ATHLON_EVENT) fprintf(stderr, "oprofiled: ctr0: %d: event only available on AMD Athlon\n", footer.ctr0_type_val);
+ if (ret & OP_CTR1_ATHLON_EVENT) fprintf(stderr, "oprofiled: ctr1: %d: event only available on AMD Athlon\n", footer.ctr1_type_val);
if (ret != OP_EVENTS_OK) {
poptPrintHelp(optcon, stderr, 0);
Index: oprofiled.h
===================================================================
RCS file: /cvsroot/oprofile/oprofile/dae/oprofiled.h,v
retrieving revision 1.29
retrieving revision 1.30
diff -u -d -r1.29 -r1.30
--- oprofiled.h 2001/08/10 13:36:37 1.29
+++ oprofiled.h 2001/08/19 20:09:17 1.30
@@ -96,6 +96,8 @@
#define OP_CTR1_PII_EVENT 0x80
#define OP_CTR0_PIII_EVENT 0x100
#define OP_CTR1_PIII_EVENT 0x200
+#define OP_CTR0_ATHLON_EVENT 0x400
+#define OP_CTR1_ATHLON_EVENT 0x800
/* FIXME : Carefull these are also present in pp/oprofpp.h */
#define OPD_MAGIC 0xdeb6
|
|
From: John L. <mov...@us...> - 2001-08-19 20:09:20
|
Update of /cvsroot/oprofile/oprofile
In directory usw-pr-cvs1:/tmp/cvs-serv3584
Modified Files:
ChangeLog README TODO op_events.c op_init.c oprofile.c
oprofile.h
Log Message:
the athlon patch. more work needed, but it works. Thanks Dave !!
Index: ChangeLog
===================================================================
RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v
retrieving revision 1.85
retrieving revision 1.86
diff -u -d -r1.85 -r1.86
--- ChangeLog 2001/08/19 18:50:54 1.85
+++ ChangeLog 2001/08/19 20:09:17 1.86
@@ -1,3 +1,19 @@
+2001-08-19 John Levon <mo...@co...>
+
+ * op_init.c:
+ * oprofile.h:
+ * oprofile.c: small tidies of the Athlon support
+
+2001-08-19 Dave Jones <da...@su...>
+
+ * op_events.c:
+ * op_init.c:
+ * oprofile.h:
+ * oprofile.c:
+ * dae/op_start:
+ * dae/oprofiled.h:
+ * dae/oprofiled.c: initial Athlon support
+
2001-08-19 Philippe Elie <ph...@cl...>
* oprofile.c: use symbolic constant for all apic setup, no generated
Index: README
===================================================================
RCS file: /cvsroot/oprofile/oprofile/README,v
retrieving revision 1.11
retrieving revision 1.12
diff -u -d -r1.11 -r1.12
--- README 2001/08/11 01:38:29 1.11
+++ README 2001/08/19 20:09:17 1.12
@@ -11,5 +11,6 @@
wrote the tcl/tk oprofile interface and the source annotation facility,
and more.
-Dave Jones <da...@su...> and Bob Montgomery <bo...@fc...> provided
-bug fixes and useful testing.
+Dave Jones <da...@su...> provided bug fixes and the Athlon support.
+
+Bob Montgomery <bo...@fc...> provided bug fixes and useful testing.
Index: TODO
===================================================================
RCS file: /cvsroot/oprofile/oprofile/TODO,v
retrieving revision 1.11
retrieving revision 1.12
diff -u -d -r1.11 -r1.12
--- TODO 2001/08/19 18:50:54 1.11
+++ TODO 2001/08/19 20:09:17 1.12
@@ -2,8 +2,7 @@
-----------
o Save counter state and restore on a finish
-o Athlon: values in oprofile.c not op_init.c, bit 22 is per-counter, determine working event types,
- fix ctr_overflowed(), fix set_perfctr to set high to -1
+o Athlon: bit 22 is per-counter, determine working event types, support 4 counters, documentation
o SMP mappings - place in all buffers perhaps ?
o check chroot() processes and the path hash stuff
o There is no need for the daemon to run as root as long as every binary image
Index: op_events.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/op_events.c,v
retrieving revision 1.19
retrieving revision 1.20
diff -u -d -r1.19 -r1.20
--- op_events.c 2001/07/21 22:53:38 1.19
+++ op_events.c 2001/08/19 20:09:17 1.20
@@ -26,6 +26,7 @@
#define OP_PII_PIII 3
#define OP_PII_ONLY 4
#define OP_PIII_ONLY 5
+#define OP_ATHLON_ONLY 6
#ifdef __KERNEL__
#include <linux/string.h>
@@ -87,13 +88,39 @@
{ 2, utm_bitmask, { 0x0, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0 }, },
};
+/* Allowed, Event #, unit mask, name, minimum event value */
static struct op_event op_events[] = {
/* Data Cache Unit (DCU) */
+ {OP_ATHLON_ONLY,0x40,0,"DATA_CACHE_ACCESSES", 500,},
+ {OP_ATHLON_ONLY,0x41,0,"DATA_CACHE_MISSES", 500,},
+ {OP_ATHLON_ONLY,0x42,0,"DATA_CACHE_REFILLS_FROM_L2", 500,},
+ {OP_ATHLON_ONLY,0x43,0,"DATA_CACHE_REFILLS_FROM_SYSTEM", 500,},
+ {OP_ATHLON_ONLY,0x44,0,"DATA_CACHE_WRITEBACKS", 500,},
+ {OP_ATHLON_ONLY,0x45,0,"L1_DTLB_MISSES_L2_DTLD_HITS", 500,},
+ {OP_ATHLON_ONLY,0x46,0,"L1_AND_L2_DTLB_MISSES", 500,},
+ {OP_ATHLON_ONLY,0x47,0,"MISALIGNED_DATA_REFS", 500,},
+ {OP_ATHLON_ONLY,0x80,0,"ICACHE_FETCHES", 500,},
+ {OP_ATHLON_ONLY,0x81,0,"ICACHE_MISSES", 500,},
+ {OP_ATHLON_ONLY,0x84,0,"L1_ITLB_MISSES_L2_ITLB_HITS", 500,},
+ {OP_ATHLON_ONLY,0x85,0,"L1_AND_L2_ITLB_MISSES", 500,},
+ {OP_ATHLON_ONLY,0xc0,0,"RETIRED_INSNS", 500,},
+ {OP_ATHLON_ONLY,0xc1,0,"RETIRED_OPS", 500,},
+ {OP_ATHLON_ONLY,0xc2,0,"RETIRED_BRANCHES", 500,},
+ {OP_ATHLON_ONLY,0xc3,0,"RETIRED_BRANCHES_MISPREDICTED", 500,},
+ {OP_ATHLON_ONLY,0xc4,0,"RETIRED_TAKEN_BRANCHES", 500,},
+ {OP_ATHLON_ONLY,0xc5,0,"RETIRED_TAKEN_BRANCHES_MISPREDICTED", 500,},
+ {OP_ATHLON_ONLY,0xc6,0,"RETIRED_FAR_CONTROL_TRANSFERS", 500,},
+ {OP_ATHLON_ONLY,0xc7,0,"RETIRED_RESYNC_BRANCHES", 500,},
+ {OP_ATHLON_ONLY,0xcd,0,"INTERRUPTS_MASKED", 500,},
+ {OP_ATHLON_ONLY,0xce,0,"INTERRUPTS_MASKED_PENDING", 500,},
+ {OP_ATHLON_ONLY,0xcf,0,"HARDWARE_INTERRUPTS", 500,},
+
{OP_ANY,0x43,0,"DATA_MEM_REFS", 500 },
{OP_ANY,0x45,0,"DCU_LINES_IN", 500 },
{OP_ANY,0x46,0,"DCU_M_LINES_IN", 500 },
{OP_ANY,0x47,0,"DCU_M_LINES_OUT", 500},
{OP_ANY,0x48,0,"DCU_MISS_OUTSTANDING", 500 },
+
/* Intruction Fetch Unit (IFU) */
{OP_ANY,0x80,0,"IFU_IFETCH", 500 },
{OP_ANY,0x81,0,"IFU_IFETCH_MISS", 500 },
@@ -201,6 +228,8 @@
#define OP_CTR1_PII_EVENT 0x80
#define OP_CTR0_PIII_EVENT 0x100
#define OP_CTR1_PIII_EVENT 0x200
+#define OP_CTR0_ATHLON_EVENT 0x400
+#define OP_CTR1_ATHLON_EVENT 0x800
/**
* op_check_unit_mask - sanity check unit mask value
@@ -291,6 +320,8 @@
*
* 2 Pentium III
*
+ * 3 AMD Athlon
+ *
* Use 0 values for @ctr0_type and @ctr1_type if the
* counter is not used.
*
@@ -325,6 +356,11 @@
if (!proc)
ret |= OP_CTR0_PII_EVENT;
break;
+
+ case OP_ATHLON_ONLY:
+ if (proc != 3)
+ ret |= OP_CTR0_ATHLON_EVENT;
+ break;
default:
break;
}
@@ -359,6 +395,12 @@
if (!proc)
ret |= OP_CTR1_PII_EVENT;
break;
+
+ case OP_ATHLON_ONLY:
+ if (proc != 3)
+ ret |= OP_CTR1_ATHLON_EVENT;
+ break;
+
default:
break;
}
@@ -397,6 +439,8 @@
*
* 2 Pentium III
*
+ * 3 AMD Athlon
+ *
* Use "" strings for @ctr0_type and @ctr1_type if the
* counter is not used.
*
@@ -702,6 +746,7 @@
case 3: printf("- Pentium II/III only\n"); break;
case 4: printf("- Pentium II only\n"); break;
case 5: printf("- Pentium III only\n"); break;
+ case 6: printf("- AMD Athlon only\n"); break;
default: printf("\n"); break;
}
if (op_events[i].unit) {
Index: op_init.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/op_init.c,v
retrieving revision 1.5
retrieving revision 1.6
diff -u -d -r1.5 -r1.6
--- op_init.c 2001/06/22 00:19:31 1.5
+++ op_init.c 2001/08/19 20:09:17 1.6
@@ -26,21 +26,26 @@
/* we want to include all P6 processors (i.e. > Pentium Classic,
* < Pentium IV
*/
- if (current_cpu_data.x86_vendor != X86_VENDOR_INTEL ||
- current_cpu_data.x86 != 6) {
- printk(KERN_ERR "oprofile: not an Intel P6 processor. Sorry.\n");
- return 0;
+ if ((current_cpu_data.x86_vendor != X86_VENDOR_INTEL &&
+ current_cpu_data.x86 != 6) ||
+ (current_cpu_data.x86_vendor != X86_VENDOR_AMD &&
+ current_cpu_data.x86 != 6)) {
+ printk(KERN_ERR "oprofile: not an Intel P6 or AMD Athlon processor. Sorry.\n");
+ return CPU_NO_GOOD;
}
- /* 0 if PPro, 1 if PII, 2 if PIII */
- cpu_type = (current_cpu_data.x86_model > 5) ? 2 :
- (current_cpu_data.x86_model > 2);
- return 1;
+ /* 0 if PPro, 1 if PII, 2 if PIII, 3 if Athlon */
+ if (current_cpu_data.x86_vendor == X86_VENDOR_AMD)
+ cpu_type = CPU_ATHLON;
+ else
+ cpu_type = (current_cpu_data.x86_model > 5) ? CPU_PIII :
+ (current_cpu_data.x86_model > 2);
+ return cpu_type;
}
int __init stub_init(void)
{
- if (!hw_ok())
+ if (hw_ok() == CPU_NO_GOOD)
return -EINVAL;
return oprof_init();
Index: oprofile.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/oprofile.c,v
retrieving revision 1.72
retrieving revision 1.73
diff -u -d -r1.72 -r1.73
--- oprofile.c 2001/08/19 18:50:54 1.72
+++ oprofile.c 2001/08/19 20:09:17 1.73
@@ -34,9 +34,20 @@
static int op_ctr_val[OP_MAX_COUNTERS];
static int op_ctr_kernel[OP_MAX_COUNTERS];
static int op_ctr_user[OP_MAX_COUNTERS];
+
pid_t pid_filter;
pid_t pgrp_filter;
+/* the MSRs we need */
+uint ctrlreg0 = MSR_IA32_EVNTSEL0;
+uint ctrlreg1 = MSR_IA32_EVNTSEL1;
+uint ctrlreg2 = MSR_K7_PERFCTL2;
+uint ctrlreg3 = MSR_K7_PERFCTL3;
+uint ctrreg0 = MSR_IA32_PERFCTR0;
+uint ctrreg1 = MSR_IA32_PERFCTR1;
+uint ctrreg2 = MSR_K7_PERFCTR2;
+uint ctrreg3 = MSR_K7_PERFCTR3;
+
u32 prof_on __cacheline_aligned;
static int op_major;
@@ -50,6 +61,15 @@
extern spinlock_t map_lock;
+extern unsigned int ctrlreg0;
+extern unsigned int ctrlreg1;
+extern unsigned int ctrlreg2;
+extern unsigned int ctrlreg3;
+extern unsigned int ctrreg0;
+extern unsigned int ctrreg1;
+extern unsigned int ctrreg2;
+extern unsigned int ctrreg3;
+
/* ---------------- NMI handler ------------------ */
/* FIXME: this whole handler would probably be better in straight asm */
@@ -313,6 +333,7 @@
return 0;
}
+
/* ugly hack */
/* PHE : the memory must be uncachable! this is perhaps more a
* problem for Athlon than for PII.
@@ -409,7 +430,7 @@
return 0;
not_local_p6_apic:
- printk(KERN_ERR "oprofile: no local P6 APIC\n");
+ printk(KERN_ERR "oprofile: no local P6 APIC. Your laptop doesn't have one !\n");
/* IA32 V3, 7.4.2 */
rdmsr(MSR_IA32_APICBASE, msr_low, msr_high);
wrmsr(MSR_IA32_APICBASE, msr_low & ~(1<<11), msr_high);
@@ -438,12 +459,24 @@
{
uint low, high;
- rdmsr(MSR_IA32_EVNTSEL0, low, high);
- wrmsr(MSR_IA32_EVNTSEL0, low & ~(1<<22), high);
+ // first, let's use the right MSRs
+ switch (cpu_type) {
+ case CPU_ATHLON:
+ ctrlreg0 = MSR_K7_PERFCTL0;
+ ctrlreg1 = MSR_K7_PERFCTL1;
+ ctrreg0 = MSR_K7_PERFCTR0;
+ ctrreg1 = MSR_K7_PERFCTR1;
+ break;
+ default:;
+ }
+
+ rdmsr(ctrlreg0, low, high);
+ // FIXME: enable bit is per-counter on athlon
+ wrmsr(ctrlreg0, low & ~(1<<22), high);
/* IA Vol. 3 Figure 15-3 */
- rdmsr(MSR_IA32_EVNTSEL0, low, high);
+ rdmsr(ctrlreg0, low, high);
/* clear */
low &= (1<<21);
@@ -452,24 +485,26 @@
pmc_fill_in(&low, op_ctr_kernel[0], op_ctr_user[0], op_ctr_val[0], op_ctr_um[0]);
}
- wrmsr(MSR_IA32_EVNTSEL0, low, 0);
+ wrmsr(ctrlreg0, low, 0);
- rdmsr(MSR_IA32_EVNTSEL1, low, high);
+ rdmsr(ctrlreg1, low, high);
/* clear */
low &= (3<<21);
if (op_ctr_val[1]) {
set_perfctr(op_ctr_count[1], 1);
pmc_fill_in(&low, op_ctr_kernel[1], op_ctr_user[1], op_ctr_val[1], op_ctr_um[1]);
- wrmsr(MSR_IA32_EVNTSEL1, low, high);
+ wrmsr(ctrlreg1, low, high);
}
/* disable ctr1 if the UP oopser might be on, but we can't do anything
* interesting with the NMIs
*/
/* PHE FIXME Have always a meaning ? */
+ /* this is pretty bogus really. especially as we don't re-enable it.
+ * Instead, save state set up, and restore with pmc_unsetup or similar */
#if !defined(CONFIG_X86_UP_APIC) || !defined(OP_EXPORTED_DO_NMI)
- wrmsr(MSR_IA32_EVNTSEL1, low, high);
+ wrmsr(ctrlreg1, low, high);
#endif
}
@@ -481,8 +516,9 @@
return;
/* enable counters */
- rdmsr(MSR_IA32_EVNTSEL0, low, high);
- wrmsr(MSR_IA32_EVNTSEL0, low | (1<<22), high);
+ rdmsr(ctrlreg0, low, high);
+ // FIXME: bit 22 per-counter on athlon
+ wrmsr(ctrlreg0, low | (1<<22), high);
}
static void pmc_stop(void *info)
@@ -493,8 +529,9 @@
return;
/* disable counters */
- rdmsr(MSR_IA32_EVNTSEL0, low, high);
- wrmsr(MSR_IA32_EVNTSEL0, low & ~(1<<22), high);
+ rdmsr(ctrlreg0, low, high);
+ // FIXME: bit 22 per-counter on athlon
+ wrmsr(ctrlreg0, low & ~(1<<22), high);
}
inline static void pmc_select_start(uint cpu)
@@ -847,6 +884,8 @@
if (ret & OP_CTR1_PII_EVENT) printk(KERN_ERR "oprofile: ctr1: event only available on PII\n");
if (ret & OP_CTR0_PIII_EVENT) printk(KERN_ERR "oprofile: ctr0: event only available on PIII\n");
if (ret & OP_CTR1_PIII_EVENT) printk(KERN_ERR "oprofile: ctr1: event only available on PIII\n");
+ if (ret & OP_CTR0_ATHLON_EVENT) printk(KERN_ERR "oprofile: ctr1: event only available on Athlon\n");
+ if (ret & OP_CTR1_ATHLON_EVENT) printk(KERN_ERR "oprofile: ctr1: event only available on Athlon\n");
if (ret)
return 0;
@@ -874,7 +913,7 @@
static int oprof_start(void)
{
int err = 0;
-
+
down(&sysctlsem);
if ((err = oprof_init_data()))
@@ -885,7 +924,7 @@
err = -EINVAL;
goto out;
}
-
+
if ((smp_call_function(pmc_setup, NULL, 0, 1))) {
oprof_free_mem(smp_num_cpus);
err = -EINVAL;
@@ -893,7 +932,7 @@
}
pmc_setup(NULL);
-
+
install_nmi();
if (!kernel_only)
@@ -905,7 +944,7 @@
pmc_start(NULL);
prof_on = 1;
-
+
out:
up(&sysctlsem);
return err;
Index: oprofile.h
===================================================================
RCS file: /cvsroot/oprofile/oprofile/oprofile.h,v
retrieving revision 1.45
retrieving revision 1.46
diff -u -d -r1.45 -r1.46
--- oprofile.h 2001/08/11 12:37:56 1.45
+++ oprofile.h 2001/08/19 20:09:17 1.46
@@ -53,6 +53,12 @@
#define OP_CTR_0 0x1
#define OP_CTR_1 0x2
+#define CPU_NO_GOOD -1
+#define CPU_PPRO 0
+#define CPU_PII 1
+#define CPU_PIII 2
+#define CPU_ATHLON 3
+
/* MSRs */
#ifndef MSR_IA32_PERFCTR0
#define MSR_IA32_PERFCTR0 0xc1
@@ -69,6 +75,30 @@
#ifndef MSR_IA32_APICBASE
#define MSR_IA32_APICBASE 0x1B
#endif
+#ifndef MSR_K7_PERFCTL0
+#define MSR_K7_PERFCTL0 0xc0010000
+#endif
+#ifndef MSR_K7_PERFCTL1
+#define MSR_K7_PERFCTL1 0xc0010001
+#endif
+#ifndef MSR_K7_PERFCTL2
+#define MSR_K7_PERFCTL2 0xc0010002
+#endif
+#ifndef MSR_K7_PERFCTL3
+#define MSR_K7_PERFCTL3 0xc0010003
+#endif
+#ifndef MSR_K7_PERFCTR0
+#define MSR_K7_PERFCTR0 0xc0010004
+#endif
+#ifndef MSR_K7_PERFCTR1
+#define MSR_K7_PERFCTR1 0xc0010005
+#endif
+#ifndef MSR_K7_PERFCTR2
+#define MSR_K7_PERFCTR2 0xc0010006
+#endif
+#ifndef MSR_K7_PERFCTR3
+#define MSR_K7_PERFCTR3 0xc0010007
+#endif
#ifndef APIC_SPIV_APIC_ENABLED
#define APIC_SPIV_APIC_ENABLED (1<<8)
@@ -114,8 +144,8 @@
^ (ctr<<8)) & (data->hash_size - 1)
/* relying on MSR numbers being neighbours */
-#define get_perfctr(l,h,c) do { rdmsr(MSR_IA32_PERFCTR0 + c, (l), (h)); } while (0)
-#define set_perfctr(l,c) do { wrmsr(MSR_IA32_PERFCTR0 + c, -(u32)(l), 0); } while (0)
+#define get_perfctr(l,h,c) do { rdmsr(ctrreg0 + c, (l), (h)); } while (0)
+#define set_perfctr(l,c) do { wrmsr(ctrreg0 + c, -(u32)(l), -1); } while (0)
#define ctr_overflowed(n) (!((n) & (1U<<31)))
#define OP_EVENTS_OK 0x0
@@ -129,6 +159,8 @@
#define OP_CTR1_PII_EVENT 0x80
#define OP_CTR0_PIII_EVENT 0x100
#define OP_CTR1_PIII_EVENT 0x200
+#define OP_CTR0_ATHLON_EVENT 0x400
+#define OP_CTR1_ATHLON_EVENT 0x800
#define op_check_range(val,l,h,str) do { \
if ((val) < (l) || (val) > (h)) { \
@@ -182,6 +214,16 @@
#else
#include <linux/completion.h>
#endif
+
+// FIXME: these names are too easy to mis-type !
+extern uint ctrlreg0;
+extern uint ctrlreg1;
+extern uint ctrlreg2;
+extern uint ctrlreg3;
+extern uint ctrreg0;
+extern uint ctrreg1;
+extern uint ctrreg2;
+extern uint ctrreg3;
int oprof_init(void);
void oprof_exit(void);
|
|
From: John L. <mov...@us...> - 2001-08-19 18:50:59
|
Update of /cvsroot/oprofile/oprofile
In directory usw-pr-cvs1:/tmp/cvs-serv21972
Modified Files:
ChangeLog TODO oprofile.c
Log Message:
symbolic changes from philippe. lots of stuff TODO ;)
Index: ChangeLog
===================================================================
RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v
retrieving revision 1.84
retrieving revision 1.85
diff -u -d -r1.84 -r1.85
--- ChangeLog 2001/08/18 16:11:17 1.84
+++ ChangeLog 2001/08/19 18:50:54 1.85
@@ -1,3 +1,8 @@
+2001-08-19 Philippe Elie <ph...@cl...>
+
+ * oprofile.c: use symbolic constant for all apic setup, no generated
+ code change
+
2001-08-18 Philippe Elie <ph...@cl...>
* Makefile.in:
Index: TODO
===================================================================
RCS file: /cvsroot/oprofile/oprofile/TODO,v
retrieving revision 1.10
retrieving revision 1.11
diff -u -d -r1.10 -r1.11
--- TODO 2001/08/17 16:47:10 1.10
+++ TODO 2001/08/19 18:50:54 1.11
@@ -1,6 +1,9 @@
Correctness
-----------
+o Save counter state and restore on a finish
+o Athlon: values in oprofile.c not op_init.c, bit 22 is per-counter, determine working event types,
+ fix ctr_overflowed(), fix set_perfctr to set high to -1
o SMP mappings - place in all buffers perhaps ?
o check chroot() processes and the path hash stuff
o There is no need for the daemon to run as root as long as every binary image
@@ -22,6 +25,7 @@
WBNI
----
+o good support for determining changed profiles
o compile-time interrupt accounting (sysctl-read/writeable) - then re-do results/rates/
o we don't really need a vmlinux file during profiling ...
o Other x86 processors
Index: oprofile.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/oprofile.c,v
retrieving revision 1.71
retrieving revision 1.72
diff -u -d -r1.71 -r1.72
--- oprofile.c 2001/08/14 02:38:29 1.71
+++ oprofile.c 2001/08/19 18:50:54 1.72
@@ -176,6 +176,9 @@
/* ---------------- APIC setup ------------------ */
+/* PHE : this would be probably an unconditionnaly restore state from a saved
+ *state
+ */
static void disable_local_P6_APIC(void *dummy)
{
#ifndef CONFIG_X86_UP_APIC
@@ -184,6 +187,11 @@
uint h;
/* FIXME: maybe this should go at end of function ? */
+ /* PHE I think when the doc says : -if you disable the apic the bits
+ * of LVT cannot be reset- it talk about the SW disable through bit 8
+ * of SPIV see 7.4.14 (7.5.14) not the hardware disable, so it is ok
+ * but I perhaps we need a software disable of the APIC at the end
+ */
/* first disable via MSR */
/* IA32 V3, 7.4.2 */
rdmsr(MSR_IA32_APICBASE, l, h);
@@ -231,11 +239,9 @@
/* IA32 V3, Figure 7.8 */
old_lvtpc[smp_processor_id()] = val = apic_read(APIC_LVTPC);
/* allow PC overflow interrupts */
- val &= ~(1<<16);
+ val &= ~APIC_LVT_MASKED;
/* set delivery to NMI */
- val |= (1<<10);
- val &= ~(1<<9);
- val &= ~(1<<8);
+ val = SET_APIC_DELIVERY_MODE(val, APIC_MODE_NMI);
apic_write(APIC_LVTPC, val);
}
@@ -244,11 +250,44 @@
apic_write(APIC_LVTPC, old_lvtpc[smp_processor_id()]);
}
+/* PHE bit 8 of APIC_BASE_MSR is the BSP flags, it is set by harware at boot
+ * up sequence and would be keeped ok by software to allow #INIT or #ERR, else
+ * all the processor on the next MP sequence fall in shutdown.
+ *
+ * On UP architecture it should always be zero (ok for my PII) 7.4.8 (7.5.8),
+ * I am unsure but I have presumption than kernel cannot boot properly if bsp
+ * flags is set on UP machine.
+ *
+ * I think it is ok in all case, perhaps corner case on SMP with only one
+ * processor plugged ?
+ */
+inline static int is_bsp_proc(void)
+{
+ uint l;
+ uint h;
+ rdmsr(MSR_IA32_APICBASE, l, h);
+ wrmsr(MSR_IA32_APICBASE, l & ~(1<<11), h);
+
+ return (h & (1 << 8)) ? 1 : 0;
+}
+
static int __init apic_needs_setup(void)
{
/* FIXME: we need to detect UP kernel, SMP hardware, and
* return 0 here, to avoid screwing up the APIC
*/
+ /* PHE: for UP kernel SMP machine this processor should have the bsp
+ * flags set, something like:
+
+#ifndef X86_UP_APIC
+ if (smp_num_cpus == 1) {
+ return !is_bsp_proc();
+ }
+#endif
+ return 0;
+
+ would work
+ */
return
/* if enabled, the kernel has already set it up */
#ifdef CONFIG_X86_UP_APIC
@@ -257,6 +296,13 @@
smp_num_cpus == 1;
}
+/* PHE, would : save all state change for later restoration, minimalise the
+ * amount of change. Basically we need to test if apic is HW disable, SW
+ * disable, and change only the LVTPC setup, all other must be keep as it
+ * (except the LINT1 for bsp ?) We must rely on the kernel to setup all things.
+ * also : do we need write_apic --> write_apic_around to support
+ * GOOD_APIC/BAD_APIC ?
+ */
static int __init apic_setup(void)
{
uint msr_low, msr_high;
@@ -268,6 +314,12 @@
}
/* ugly hack */
+ /* PHE : the memory must be uncachable! this is perhaps more a
+ * problem for Athlon than for PII.
+ * Also this stuff probably mishandled SMP ?
+ * The _PAGE_GLOBAL attribute should be probably avoided but make
+ * this later
+ */
my_set_fixmap();
/* FIXME: NMI delivery for SMP ? */
@@ -288,6 +340,7 @@
if (GET_APIC_MAXLVT(apic_read(APIC_LVR)) != 4)
goto not_local_p6_apic;
+ /* __global_cli(); ? an IPI can occur inside this stuff ? */
__cli();
/* enable APIC locally */
@@ -296,36 +349,59 @@
apic_write(APIC_SPIV, val | APIC_SPIV_APIC_ENABLED);
/* FIXME: examine this stuff */
- val = 0x00008700;
+ val = APIC_LVT_LEVEL_TRIGGER;
+ val = SET_APIC_DELIVERY_MODE(val, APIC_MODE_EXINT);
apic_write(APIC_LVT0, val);
- val = 0x00000400;
+ /* edge triggered, IA 7.4.11 */
+ /* PHE SMP: only the BSP must see the LINT1 IRQ (from kernel source)
+ * recheck this later
+ */
+ val = SET_APIC_DELIVERY_MODE(0, APIC_MODE_NMI);
apic_write(APIC_LVT1, val);
/* clear error register */
/* IA32 V3, 7.4.17 */
+ /* PHE must be cleared after unmasking by a back-to-back write,
+ * but it is probably ok because we mask only, the ESR is not updated
+ * is this a real problem ?
+ */
apic_write(APIC_ESR, 0);
/* mask error interrupt */
/* IA32 V3, Figure 7.8 */
val = apic_read(APIC_LVTERR);
- val |= 0x00010000;
+ val |= APIC_LVT_MASKED;
apic_write(APIC_LVTERR, val);
/* setup timer vector */
/* IA32 V3, 7.4.8 */
- apic_write(APIC_LVTT, 0x0001031);
+ /* PHE actually it is ok but kernel change can hang up the machine
+ * after this point.
+ */
+ apic_write(APIC_LVTT, APIC_SEND_PENDING | 0x31);
/* Divide configuration register */
+ /* PHE the apic clock is based on the FSB. This should only changed
+ * with a calibration method.
+ */
val = APIC_TDR_DIV_1;
apic_write(APIC_TDCR, val);
+ /* PHE __global_sti() probably */
__sti();
/* If the local APIC NMI watchdog has been disabled, we'll need
* to set up NMI delivery anyway ...
*/
+ /* PHE : ? why not use a FixedMode rather than NMI mode for profiling,
+ * this avoid all problem with parity error nmi and watchdog, but this
+ * implies than the actual nmi handler must be hardware interruptible,
+ * basically this is better at my eyes but there is a few visible
+ * (and hidden) problem with this issue. This can probably wait unless
+ * hidden problem occur on SMP, recheck doc about NMI
+ */
lvtpc_apic_setup(NULL);
printk(KERN_INFO "oprofile: enabled local APIC\n");
@@ -391,6 +467,7 @@
/* disable ctr1 if the UP oopser might be on, but we can't do anything
* interesting with the NMIs
*/
+ /* PHE FIXME Have always a meaning ? */
#if !defined(CONFIG_X86_UP_APIC) || !defined(OP_EXPORTED_DO_NMI)
wrmsr(MSR_IA32_EVNTSEL1, low, high);
#endif
|
|
From: John L. <mov...@us...> - 2001-08-18 16:11:21
|
Update of /cvsroot/oprofile/oprofile/doc In directory usw-pr-cvs1:/tmp/cvs-serv29460/doc Modified Files: Makefile.in oprofile.sgml Log Message: uninstall patch from philippe Index: Makefile.in =================================================================== RCS file: /cvsroot/oprofile/oprofile/doc/Makefile.in,v retrieving revision 1.5 retrieving revision 1.6 diff -u -d -r1.5 -r1.6 --- Makefile.in 2001/08/16 01:57:58 1.5 +++ Makefile.in 2001/08/18 16:11:17 1.6 @@ -1,15 +1,17 @@ # $Id$ prefix=@prefix@ -MANDIR=@mandir@ +MANDIR=@mandir@/man1 +CATDIR=@mandir@/cat1 LN_SF=ln -sf RM=rm MKDIR=mkdir MKDIR_P=$(MKDIR) -p DB2HTML=db2html +LINK_LIST=oprofiled.1 op_start.1 op_stop.1 op_help.1 oprofpp.1 -.PHONY: all clean install - +.PHONY: all clean install uninstall + all: -$(RM) -rf oprofile.junk $(DB2HTML) oprofile.sgml @@ -20,12 +22,18 @@ -$(RM) -rf oprofile install: - -$(MKDIR_P) $(MANDIR)/man1 - cp oprofile.1 $(MANDIR)/man1/oprofile.1 - -$(LN_SF) $(MANDIR)/man1/oprofile.1 $(MANDIR)/man1/oprofiled.1 - -$(LN_SF) $(MANDIR)/man1/oprofile.1 $(MANDIR)/man1/op_start.1 - -$(LN_SF) $(MANDIR)/man1/oprofile.1 $(MANDIR)/man1/op_stop.1 - -$(LN_SF) $(MANDIR)/man1/oprofile.1 $(MANDIR)/man1/op_help.1 - -$(LN_SF) $(MANDIR)/man1/oprofile.1 $(MANDIR)/man1/oprofpp.1 - + -$(MKDIR_P) $(MANDIR) + cp oprofile.1 $(MANDIR)/oprofile.1 + @for f in $(LINK_LIST); do \ + $(LN_SF) $(MANDIR)/oprofile.1 $(MANDIR)/$$f; \ + done + +uninstall: + rm -f $(MANDIR)/oprofile.1 + rm -f @mandir@/cat1/oprofile.1.gz + @for f in $(LINK_LIST); do \ + rm -f $(MANDIR)/$$f; \ + rm -f $(CATDIR)/cat1/$$f.gz; \ + done + clean: Index: oprofile.sgml =================================================================== RCS file: /cvsroot/oprofile/oprofile/doc/oprofile.sgml,v retrieving revision 1.19 retrieving revision 1.20 diff -u -d -r1.19 -r1.20 --- oprofile.sgml 2001/08/11 01:45:54 1.19 +++ oprofile.sgml 2001/08/18 16:11:17 1.20 @@ -181,6 +181,14 @@ <para>Be warned than you cannot convert from new version to old version. If you convert to new version you cannot reverse the conversion. </sect1> + +<sect1 id="uninstall"> +<title>Uninstalling oprofile</title> +<para> +You must have the source tree installed to uninstall oprofile, then a <command>make uninstall</command> will +remove all installed file except your configuration file in the directory <filename>~/.oprofile</filename> +</para> +</sect1> </chapter> |
|
From: John L. <mov...@us...> - 2001-08-18 16:11:21
|
Update of /cvsroot/oprofile/oprofile/gui In directory usw-pr-cvs1:/tmp/cvs-serv29460/gui Modified Files: Makefile.in Log Message: uninstall patch from philippe Index: Makefile.in =================================================================== RCS file: /cvsroot/oprofile/oprofile/gui/Makefile.in,v retrieving revision 1.2 retrieving revision 1.3 diff -u -d -r1.2 -r1.3 --- Makefile.in 2001/08/16 01:57:58 1.2 +++ Makefile.in 2001/08/18 16:11:17 1.3 @@ -1,9 +1,9 @@ all: - -.PHONY: all clean - + +.PHONY: all clean install uninstall + clean: - + prefix=@prefix@ exec_prefix=@exec_prefix@ BINDIR=@bindir@ @@ -13,3 +13,10 @@ -$(MKDIR_P) $(BINDIR) cp oprofile $(BINDIR)/oprofile cd $(BINDIR) && chmod 755 oprofile + +uninstall: + rm -f $(BINDIR)/oprofile + @echo "---------------------------------------------------------------" + @echo "The uninstall target does not remove your ~/.oprofile directory" + @echo "---------------------------------------------------------------" + |
|
From: John L. <mov...@us...> - 2001-08-18 16:11:21
|
Update of /cvsroot/oprofile/oprofile/dae In directory usw-pr-cvs1:/tmp/cvs-serv29460/dae Modified Files: Makefile.in Log Message: uninstall patch from philippe Index: Makefile.in =================================================================== RCS file: /cvsroot/oprofile/oprofile/dae/Makefile.in,v retrieving revision 1.13 retrieving revision 1.14 diff -u -d -r1.13 -r1.14 --- Makefile.in 2001/08/16 01:57:58 1.13 +++ Makefile.in 2001/08/18 16:11:17 1.14 @@ -1,7 +1,7 @@ # $Id$ -.PHONY: all clean srcdoc install - +.PHONY: all clean srcdoc install uninstall + all: oprofiled op_help prefix=@prefix@ @@ -10,19 +10,22 @@ MANDIR=@mandir@ CC=@CC@ MKDIR_P=mkdir -p +INSTALL_LIST=oprofiled op_start op_dump op_stop op_help -install: oprofiled op_start op_dump op_stop op_help +install: $(INSTALL_LIST) -$(MKDIR_P) $(BINDIR) - cp oprofiled $(BINDIR)/oprofiled - cp op_start $(BINDIR)/op_start - cp op_dump $(BINDIR)/op_dump - cp op_stop $(BINDIR)/op_stop - cp op_help $(BINDIR)/op_help - cd $(BINDIR) && chmod 755 oprofiled op_start op_dump op_stop op_help - + @for f in $(INSTALL_LIST); do \ + cp $$f $(BINDIR)/$$f && chmod 755 $(BINDIR)/$$f; \ + done + +uninstall: + @for f in $(INSTALL_LIST); do \ + rm -f $(BINDIR)/$$f; \ + done + clean: rm -f oprofiled op_help *.o oprofiled-src.html - + CFLAGS=-Wall -Wstrict-prototypes -O2 -pipe -Wstrict-prototypes -Wunused -W -Wshadow -Wmissing-prototypes -Winline @CFLAGS@ LIBS=@extra_libraries@ -lpopt INCLUDES=-I. @extra_includes@ |
|
From: John L. <mov...@us...> - 2001-08-18 16:11:21
|
Update of /cvsroot/oprofile/oprofile In directory usw-pr-cvs1:/tmp/cvs-serv29460 Modified Files: ChangeLog Makefile.in Log Message: uninstall patch from philippe Index: ChangeLog =================================================================== RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v retrieving revision 1.83 retrieving revision 1.84 diff -u -d -r1.83 -r1.84 --- ChangeLog 2001/08/18 15:58:31 1.83 +++ ChangeLog 2001/08/18 16:11:17 1.84 @@ -1,3 +1,11 @@ +2001-08-18 Philippe Elie <ph...@cl...> + + * Makefile.in: + * dae/Makefile.in: + * pp/Makefile.in: + * doc/Makefile.in: + * gui/Makefile.in: add uninstall target + 2001-08-18 John Levon <mo...@co...> * dae/opd_proc.c: tiny cleanup Index: Makefile.in =================================================================== RCS file: /cvsroot/oprofile/oprofile/Makefile.in,v retrieving revision 1.22 retrieving revision 1.23 diff -u -d -r1.22 -r1.23 --- Makefile.in 2001/08/16 01:57:58 1.22 +++ Makefile.in 2001/08/18 16:11:17 1.23 @@ -6,18 +6,20 @@ CC=@CC@ LD=@LD@ -.PHONY: all install clean distclean tags +.PHONY: all install clean distclean maintainerclean uninstall tags all: oprofile.o $(MAKE) -C dae $(MAKE) -C pp $(MAKE) -C gui +# doc/Makefile is deleted here but should be deleted in distclean. +# When the dist target works move it as action of distclean. maintainerclean: distclean - rm -f configure + rm -f configure doc/Makefile distclean: clean - rm -f Makefile dae/Makefile pp/Makefile tags + rm -f Makefile dae/Makefile pp/Makefile gui/Makefile tags # broken dist: @@ -30,7 +32,7 @@ ./configure ; \ cp doc/Makefile doc/Makefile.tmp ; \ make distclean ; \ - cd doc && make && rm -f Makefile && cd .. ; \ + cd doc && make -f Makefile.tmp && rm -f Makefile.tmp && cd .. ; \ cd .. ; \ mv oprofile oprofile-@VERSION@ ; \ tar cvfz ../oprofile-@VERSION@.tar.gz oprofile-@VERSION@ ; \ @@ -49,6 +51,16 @@ $(MAKE) -C pp install $(MAKE) -C doc install $(MAKE) -C gui install + +uninstall: + -rm -f $(MODINSTALLDIR)/oprofile.o + -rmdir --ignore-fail-on-non-empty $(MODINSTALLDIR) + depmod -a + $(MAKE) -C dae uninstall + $(MAKE) -C pp uninstall + $(MAKE) -C doc uninstall + $(MAKE) -C gui uninstall + @echo "Goodbye." clean: rm -f *.o config.status config.log config.cache |
|
From: John L. <mov...@us...> - 2001-08-18 16:11:21
|
Update of /cvsroot/oprofile/oprofile/pp In directory usw-pr-cvs1:/tmp/cvs-serv29460/pp Modified Files: Makefile.in Log Message: uninstall patch from philippe Index: Makefile.in =================================================================== RCS file: /cvsroot/oprofile/oprofile/pp/Makefile.in,v retrieving revision 1.13 retrieving revision 1.14 diff -u -d -r1.13 -r1.14 --- Makefile.in 2001/08/16 01:57:58 1.13 +++ Makefile.in 2001/08/18 16:11:17 1.14 @@ -1,6 +1,6 @@ all: oprofpp oprof_convert opf_filter -.PHONY: all clean srcdoc +.PHONY: all clean srcdoc install uninstall clean: rm -f oprofpp *.o oprofpp-src.html oprof_convert opf_filter @@ -11,17 +11,18 @@ BINDIR=@bindir@ MANDIR=@mandir@ MKDIR_P=mkdir -p +INSTALL_LIST=oprofpp oprof_convert opf_filter op_to_source -install: oprofpp oprof_convert opf_filter op_to_source +install: $(INSTALL_LIST) -$(MKDIR_P) $(BINDIR) - cp oprofpp $(BINDIR)/oprofpp - chmod 755 $(BINDIR)/oprofpp - cp oprof_convert $(BINDIR)/oprof_convert - chmod 755 $(BINDIR)/oprof_convert - cp opf_filter $(BINDIR)/opf_filter - chmod 755 $(BINDIR)/opf_filter - cp op_to_source $(BINDIR)/op_to_source - chmod 755 $(BINDIR)/op_to_source + @for f in $(INSTALL_LIST); do \ + cp $$f $(BINDIR)/$$f && chmod 755 $(BINDIR)/$$f; \ + done + +uninstall: + @for f in $(INSTALL_LIST); do \ + rm -f $(BINDIR)/$$f; \ + done CFLAGS=-Wall -Wstrict-prototypes -W -pipe -DOP_EVENTS_DESC -L/usr/lib @CFLAGS@ LIBS=-lpopt -lbfd -liberty -ldl |
|
From: John L. <mov...@us...> - 2001-08-18 15:58:35
|
Update of /cvsroot/oprofile/oprofile In directory usw-pr-cvs1:/tmp/cvs-serv26988 Modified Files: ChangeLog Log Message: small fix Index: ChangeLog =================================================================== RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v retrieving revision 1.82 retrieving revision 1.83 diff -u -d -r1.82 -r1.83 --- ChangeLog 2001/08/17 16:47:10 1.82 +++ ChangeLog 2001/08/18 15:58:31 1.83 @@ -1,3 +1,7 @@ +2001-08-18 John Levon <mo...@co...> + + * dae/opd_proc.c: tiny cleanup + 2001-08-17 Philippe Elie <ph...@cl...> * dae/opd_proc.c: |
|
From: John L. <mov...@us...> - 2001-08-18 15:58:35
|
Update of /cvsroot/oprofile/oprofile/dae
In directory usw-pr-cvs1:/tmp/cvs-serv26988/dae
Modified Files:
opd_proc.c
Log Message:
small fix
Index: opd_proc.c
===================================================================
RCS file: /cvsroot/oprofile/oprofile/dae/opd_proc.c,v
retrieving revision 1.63
retrieving revision 1.64
diff -u -d -r1.63 -r1.64
--- opd_proc.c 2001/08/17 16:47:10 1.63
+++ opd_proc.c 2001/08/18 15:58:31 1.64
@@ -1040,6 +1040,21 @@
}
/**
+ * verb_show_sample - print the sample out to the log
+ * @offset: the offset value
+ * @map: map to print
+ */
+inline static void verb_show_sample(u32 offset, struct opd_map *map)
+{
+ if (!verbose)
+ return;
+
+ printf("DO_PUT_SAMPLE (LAST_MAP): calc offset 0x%.8x, map start 0x%.8x,"
+ " end 0x%.8x, offset 0x%.8x, name $%s$\n",
+ offset, map->start, map->end, map->offset, opd_images[map->image].name);
+}
+
+/**
* opd_map_offset - return offset of sample against map
* @map: map to use
* @eip: EIP value to use
@@ -1090,8 +1105,7 @@
if (opd_is_in_map(&proc->maps[proc->last_map], sample->eip)) {
i = proc->last_map;
if (proc->maps[i].image != -1) {
- verbprintf("DO_PUT_SAMPLE (LAST_MAP): calc offset 0x%.8x, map start 0x%.8x, end 0x%.8x, offset 0x%.8x, name $%s$\n",
- opd_map_offset(&proc->maps[i], sample->eip), proc->maps[i].start, proc->maps[i].end, proc->maps[i].offset, opd_images[proc->maps[i].image].name);
+ verb_show_sample(opd_map_offset(&proc->maps[i], sample->eip), &proc->maps[i]);
opd_put_image_sample(&opd_images[proc->maps[i].image], opd_map_offset(&proc->maps[i], sample->eip), sample->count);
}
opd_stats[OPD_PROCESS]++;
@@ -1103,8 +1117,7 @@
if (opd_is_in_map(&proc->maps[i], sample->eip)) {
u32 offset = opd_map_offset(&proc->maps[i], sample->eip);
if (proc->maps[i].image != -1) {
- verbprintf("DO_PUT_SAMPLE: calc offset 0x%.8x, map start 0x%.8x, end 0x%.8x, offset 0x%.8x, name $%s$\n",
- offset, proc->maps[i].start, proc->maps[i].end, proc->maps[i].offset, opd_images[proc->maps[i].image].name);
+ verb_show_sample(offset, &proc->maps[i]);
opd_put_image_sample(&opd_images[proc->maps[i].image], offset, sample->count);
}
proc->last_map = i;
|
40 messages has been excluded from this view by a project administrator.