From: Antoine M. <an...@na...> - 2005-08-25 09:05:04
|
On Fri, 2005-08-12 at 18:14 +0200, Blaisorblade wrote: > On Friday 12 August 2005 01:22, Jeff Dike wrote: > > On Thu, Aug 11, 2005 at 06:17:03PM +0200, eri...@we... wrote: > > > - Opteron Support > > > I shall setup several services on a Dual Opteron 250. I came across some > > > issues with Opterons and I wonder if Opterons are ok for UML? Are they > > > stable? Or ist this a problem with 64 Bit Mode only and UML works fine on > > > Opterons in 32 Bit Mode? 32-bit mode works fine 64-bit is broken atm, you'll run out of memory very quickly. > > > UML/x86_64 is new, so it hasn't had as much exercise as UML/i386. There's > > also an as-yet-unfixed page leak which causes it to run out of memory after > > a while. > What is the /proc/meminfo and/or /proc/slabinfo content? Which is the culprit > item? Files attached. Hope this helps. before=on startup, during=after 1 emerge, after=after 2 emerge Antoine PS: I also got this error, not sure it has anything to do with the above (probably not): x86_64-pc-linux-gnu-gcc -O2 -fno-unit-at-a-time -DHAVE_CONFIG_H -D_U_="__attribute__((unused))" -I. -I./missing -I/usr/include -c ./smbutil.c In file included from ./tcpdump-stdinc.h:96, from ./smbutil.c:18: /usr/include/unistd.h:533: internal compiler error: Segmentation fault Please submit a full bug report, with preprocessed source if appropriate. See <URL:http://bugs.gentoo.org/> for instructions. The bug is not reproducible, so it is likely a hardware or OS problem. make: *** [smbutil.o] Error 1 !!! ERROR: net-analyzer/tcpdump-3.8.3-r4 failed. !!! Function src_compile, Line 47, Exitcode 2 !!! (no error message) !!! If you need support, post the topmost build error, NOT this status message. |
From: Jeff D. <jd...@ad...> - 2005-08-31 17:45:12
|
On Wed, Aug 31, 2005 at 02:21:29PM +0100, Antoine Martin wrote: > And I've had the case (now 4 times and I ended up loosing some work - > but that's my own fault!) where other processes on the same box get > affected (they segfault). The guest is doing nothing (idle - no cron) > I stopped the guest and now the problems have gone away... > Something is corrupting my processes. What's the host, exactly? Jeff |
From: Antoine M. <an...@na...> - 2005-08-31 18:05:29
|
On Wed, 2005-08-31 at 11:41 -0400, Jeff Dike wrote: > On Wed, Aug 31, 2005 at 02:21:29PM +0100, Antoine Martin wrote: > > And I've had the case (now 4 times and I ended up loosing some work - > > but that's my own fault!) where other processes on the same box get > > affected (they segfault). The guest is doing nothing (idle - no cron) > > I stopped the guest and now the problems have gone away... > > Something is corrupting my processes. > > What's the host, exactly? It's an Acer 1500 laptop with amd64-3000 and 1.2GB ram. It's been running fine since I stopped the guest, but then again that does not mean that was the problem - it could just be putting more pressure on I/O or ram consumption and triggering something else. But my gut feeling is that it is doing something.. I'm getting back my amd64 test rig (not a laptop) in the next few days so I'll be able to run some proper tests then. Antoine |
From: Jeff D. <jd...@ad...> - 2005-08-28 13:55:19
|
On Thu, Aug 25, 2005 at 10:04:38AM +0100, Antoine Martin wrote: > 64-bit is broken atm, you'll run out of memory very quickly. I found the leak, and have been making a UML doing a kernel build loop all weekend, with no memory loss. The patch is below. > PS: I also got this error, not sure it has anything to do with the above > (probably not): > x86_64-pc-linux-gnu-gcc -O2 -fno-unit-at-a-time -DHAVE_CONFIG_H > -D_U_="__attribute__((unused))" -I. -I./missing -I/usr/include > -c ./smbutil.c > In file included from ./tcpdump-stdinc.h:96, > from ./smbutil.c:18: > /usr/include/unistd.h:533: internal compiler error: Segmentation fault Is this reproducable? Jeff |
From: Jeff D. <jd...@ad...> - 2005-08-28 14:29:54
Attachments:
page-leak
|
On Thu, Aug 25, 2005 at 10:04:38AM +0100, Antoine Martin wrote: > 64-bit is broken atm, you'll run out of memory very quickly. Um, and the patch is really included this time. Jeff |
From: Antoine M. <an...@na...> - 2005-08-30 23:28:04
|
Thanks Jeff, Works for me (tm) I reckon that this patch is small enough to go into 2.6.13.1 It fixes a bug, so it should be included, right? There are still problems that I can't track down. On occasion I ended up with a system hosed up, where I couldn't even do a 'ps -ef' without hanging the terminal. Not sure this is relevant, but I caught this on the host (it could just be that my latop caught some MCE errors after overheating, but it happened more than once so I thought I'd post it): [13783.971150] ----------- [cut here ] --------- [please bite here ] --------- [13783.971159] Kernel BUG at "mm/prio_tree.c":125 [13783.971162] invalid operand: 0000 [1] PREEMPT [13783.971165] CPU 0 [13783.971167] Modules linked in: ipt_MASQUERADE iptable_nat tun parport_pc lp parport eeprom i2c_sensor i2c_viapro i2c_dev i2c_core rfcomm l2cap bluetooth af_packet ipt_REJECT ipt_state ip_conntrack iptable_filter ip_tables dm_mod container ipv6 ohci1394 ieee1394 uhci_hcd ehci_hcd shpchp via_ircc irda crc_ccitt tg3 usbcore joydev tsdev unix [13783.971185] Pid: 24646, comm: vmlinux-2.6.13- Tainted: G M 2.6.13 [13783.971188] RIP: 0010:[<ffffffff8016ea98>] <ffffffff8016ea98>{vma_prio_tree_remove+88} [13783.971199] RSP: 0018:ffff81004a225df8 EFLAGS: 00010283 [13783.971203] RAX: 00000000000000fd RBX: ffff8100497aa528 RCX: fffffffffffffffc [13783.971208] RDX: ffff810036ec4e78 RSI: ffff810036ec4e98 RDI: ffff81004a1e5c48 [13783.971213] RBP: ffff81004a1e5c48 R08: ffff810036ec4e98 R09: ffff81004a225df0 [13783.971218] R10: ffff8100494cd6f8 R11: ffff8100494cd6f8 R12: ffff81003747b380 [13783.971222] R13: ffff81004a1e5c48 R14: ffff81004a1e5cf8 R15: ffff81004a1e5d10 [13783.971227] FS: 000000006069e860(0000) GS:ffffffff806ec800(0000) knlGS:000000006471aff8 [13783.971230] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b [13783.971233] CR2: 0000007fbffff030 CR3: 000000000459c000 CR4: 00000000000006e0 [13783.971238] Process vmlinux-2.6.13- (pid: 24646, threadinfo ffff81004a224000, task ffff81000f991560) [13783.971240] Stack: ffffffff8065d080 ffff810036ec4e78 ffff81004a1e5c48 ffffffff80174d51 [13783.971248] 0000000000000000 ffff81000df04180 ffff81004a1e5c48 ffffffff80175c8e [13783.971253] 0000000000000000 ffff81000df04188 [13783.971257] Call Trace:<ffffffff80174d51>{remove_vm_struct+65} <ffffffff80175c8e>{do_munmap+670} [13783.971270] <ffffffff8010e2dd>{do_signal+157} <ffffffff80176a43>{do_mmap_pgoff+931} [13783.971281] <ffffffff8015f122>{audit_syscall_entry+338} <ffffffff8047e525>{__down_write+53} [13783.971294] <ffffffff80115425>{sys_mmap+165} <ffffffff8010ed52>{tracesys+209} [13783.971306] [13783.971311] [13783.971312] Code: 0f 0b a3 e0 33 4a 80 ff ff ff ff c2 7d 00 48 83 7f 60 00 74 [13783.971322] RIP <ffffffff8016ea98>{vma_prio_tree_remove+88} RSP <ffff81004a225df8> [13783.971328] <6>note: vmlinux-2.6.13-[24646] exited with preempt_count 1 [13783.971337] scheduling while atomic: vmlinux-2.6.13-/0x00000001/24646 [13783.971339] [13783.971340] Call Trace:<ffffffff8047ce5a>{schedule+122} <ffffffff8047e4c5>{__down_read+213} [13783.971351] <ffffffff80138a59>{exit_mm+57} <ffffffff80139874>{do_exit+484} [13783.971361] <ffffffff8038bd0e>{do_unblank_screen+110} <ffffffff80110535>{die+69} [13783.971372] <ffffffff80111073>{do_invalid_op+163} <ffffffff8016ea98>{vma_prio_tree_remove+88} [13783.971381] <ffffffff80313e4a>{avc_has_perm+90} <ffffffff8010f4e1>{error_exit+0} [13783.971393] <ffffffff8016ea98>{vma_prio_tree_remove+88} <ffffffff80174d51>{remove_vm_struct+65} [13783.971406] <ffffffff80175c8e>{do_munmap+670} <ffffffff8010e2dd>{do_signal+157} [13783.971414] <ffffffff80176a43>{do_mmap_pgoff+931} <ffffffff8015f122>{audit_syscall_entry+338} [13783.971425] <ffffffff8047e525>{__down_write+53} <ffffffff80115425>{sys_mmap+165} [13783.971435] <ffffffff8010ed52>{tracesys+209} On Sun, 2005-08-28 at 10:03 -0400, Jeff Dike wrote: > On Thu, Aug 25, 2005 at 10:04:38AM +0100, Antoine Martin wrote: > > 64-bit is broken atm, you'll run out of memory very quickly. > > Um, and the patch is really included this time. > > Jeff |
From: Antoine M. <an...@na...> - 2005-08-31 13:21:51
|
On Wed, 2005-08-31 at 01:27 +0100, Antoine Martin wrote: > Thanks Jeff, > > Works for me (tm) > I reckon that this patch is small enough to go into 2.6.13.1 > It fixes a bug, so it should be included, right? > > There are still problems that I can't track down. On occasion I ended up > with a system hosed up, where I couldn't even do a 'ps -ef' without > hanging the terminal. And I've had the case (now 4 times and I ended up loosing some work - but that's my own fault!) where other processes on the same box get affected (they segfault). The guest is doing nothing (idle - no cron) I stopped the guest and now the problems have gone away... Something is corrupting my processes. Antoine |