[User-mode-linux-devel] CVS checkins

[Jeff Dike <jdike@ka...>]

I've checked in Andrea's LFS patch, the putative fix for the 
module-creating-kernel-thread bug, and some more exported symbols.

				Jeff

Re: [User-mode-linux-devel] panic initializing module and thread

[Jeff Dike <jdike@ka...>]

jhh@... said:
> Creating a server thread in init_module causes a kernel panic:
> Kernel panic: Kernel mode fault at addr 0x518000a0, ip 0x518000a0 

Try this:

--- arch/um/kernel/process_kern.c~      Sun Aug 27 10:49:03 2000
+++ arch/um/kernel/process_kern.c       Wed Aug 30 20:09:01 2000
@@ -120,6 +120,7 @@
        task->thread.extern_pid = pid;
        stop_pid(pid);
        set_cmdline("(kernel thread)");
+       init_flush_vm();
        if(current->thread.request.u.cswitch.from != NULL)
                schedule_tail(current->thread.request.u.cswitch.from);
        return((*fn)(arg));


> Also, it would be nice if ksyms.c exported block_signals and
> unblock_signals. I lifted much of my code from nfsd, and it calls
> these. 

That will be in test8 and in CVS shortly.

				Jeff

[User-mode-linux-devel] panic initializing module and thread

[John H. Hartman <jhh@pr...>]

I've been working on a module that implements a simple SUNRPC server.
Creating a server thread in init_module causes a kernel panic:

Kernel panic: Kernel mode fault at addr 0x518000a0, ip 0x518000a0

The address 0x518000a0 is the starting address of the thread's function,
which is inside the module. Could there be a race between installing a
module and setting up its memory, and starting a thread that uses that
memory? This all works fine on a real machine. I'm using
linux-2.4.0-test7. 

Also, it would be nice if ksyms.c exported block_signals and
unblock_signals. I lifted much of my code from nfsd, and it calls these.

-- 
John


[User-mode-linux-devel] test7 checkins

[Jeff Dike <jdike@ka...>]

test7 has been checked in.  There were a couple of minor header file changes 
needed.  I also fixed the build of arch/um/kernel so that symbols can be 
exported from it.

				Jeff

[User-mode-linux-devel] cvs checkins

[Jeff Dike <jdike@ka...>]

These ones make the debugging options configurable, add some comparibility 
with 2.2 and /proc/ide to try to make the Debian install work, clean up some 
old unused code, and fix the compile of arch/um/kernel/unmap_fin.o.

				Jeff

[User-mode-linux-devel] UML only serves up one packet

[William Stearns <wstearns@po...>]

Good day, Rusty and James,
	Jeff and I have been working on the UML networking code.  
(Translation: "_He's_ been doing all the work, I just break things and
whine loudly." :-)
	If I run apache in UML, it serves up files to the _host_ quite
nicely.  If I try to see those same files from another machine on the real
ethernet segment (my laptop), a bunch of bytes get stuck in the SendQ
according to "netstat -an" in the UML.  Just for reference, the host is
proxyarping the uml.  My laptop can successfully ping the uml, and the uml
can successfully ping my laptop.  For the moment, I'm using the umn-sl0
networking.
	The easiest way to reproduce it is to whip up a script called
/usr/sbin/in.fingerd that is:

#!/bin/bash
cat /home/httpd/html/index.html

	or some other file that's larger than 512 bytes.  When I do

finger @the_uml_ip_address

	i get back on the first part of the file.  

finger @the_uml_ip_address | wc -c

	returns 530 bytes; when I strip out the finger header from the
output ("[the_uml_ip_address]"):

finger @the_uml_ip_address | grep -v '\[the_uml_ip_address\]' | wc -c

	I get exactly 512 bytes.  And nothing more.  Every time,
repeatably.
	I get the impression that when UML has to feed a tcp stream back
to a client application running on some machine other than the uml or the
host, that it simply fails to schedule more than one 512 byte
packet.  Jeff gave me the impression he's really not sure about the
networking code in UML; is there any chance that once of you networking
gods could take a look at the code and see if we can coax a second or
third packet out of it?
	Please let me know if I ran short on details; I'll be glad to try
again or provide more specifics.
	Cheers,
	- Bill

---------------------------------------------------------------------------
	"SCSI is *NOT* magic.  There are *fundamental technical reasons*
why it is necessary to sacrifice a young goat to your SCSI chain now and
then." 
	-- jfw@...
--------------------------------------------------------------------------
William Stearns (wstearns@...).  Mason, Buildkernel, named2hosts, 
and ipfwadm2ipchains are at:                http://www.pobox.com/~wstearns
LinuxMonth; articles for Linux Enthusiasts! http://www.linuxmonth.com
--------------------------------------------------------------------------

[User-mode-linux-devel] A fix to that last fix

[Jeff Dike <jdike@ka...>]

I just noticed, fixed, and checked in a moronic bug in my last fix:

--- arch/um/drivers/umn_kern.c~ Wed Aug 16 14:27:33 2000
+++ arch/um/drivers/umn_kern.c  Wed Aug 16 22:39:25 2000
@@ -40,7 +40,7 @@
        spin_lock_irqsave(&umn.lock, flags);
        ret = umn_send_packet(umn.tty_fd, skb->data, skb->len);
        dev_kfree_skb(skb);
-       spin_unlock_irqrestore(&lp->lock, flags);
+       spin_unlock_irqrestore(&umn.lock, flags);
        return(ret);
 }

What's even better is that this appears to fix the occasional process 
segfaults I've been seeing.  I had a way to reproduce it intermittently 
yesterday, but I can't reproduce it today with this fix.

So, if this turns out to be true, and it also fixes the ping flood crash, I'd 
have to say that test6 has gone from cr*p to the best release yet.

				Jeff

[User-mode-linux-devel] that umn_kern.c fix is now checked in

[Jeff Dike <jdike@ka...>]

[User-mode-linux-devel] Re: Could you try the ping flood of death on test6?

[Jeff Dike <jdike@ka...>]

wstearns@... said:
> I've got one more for you.  I'm using your test6 kernel from last
> night with an rh6.2 root filesystem with apache.  Running apachebench
> on the host (ab, in the apache package) against apache in the uml
> gives:
> bash# Kernel panic: Kernel mode fault at addr 0x0, ip 0x1007479c 

Nice one.  Here's the fix.  Could you see if it fixes the ping flood crash, 
too?

				Jeff

--- arch/um/drivers/umn_kern.c~ Sat Jun 17 19:29:56 2000
+++ arch/um/drivers/umn_kern.c  Wed Aug 16 14:27:33 2000
@@ -8,6 +8,7 @@
 #include "linux/skbuff.h"
 #include "linux/if_arp.h"
 #include "linux/init.h"
+#include <linux/spinlock.h>
 #include "umn.h"
 #include "user_util.h"
 #include "kern.h"
@@ -25,16 +26,21 @@
        int buffsize;
        struct net_device dev;
        char *ptp_addr;
+       spinlock_t lock;
 };
 
-struct umn umn = { 0, -1, -1, -1, -1, -1, 0, { }, 0, { }, "192.168.0.254" };
+struct umn umn = { 0, -1, -1, -1, -1, -1, 0, { }, 0, { }, "192.168.0.254",
+                   SPIN_LOCK_UNLOCKED };
 
 static int xmit(struct sk_buff *skb, struct net_device *dev)
 {
+       unsigned long flags;
        int ret;
 
+       spin_lock_irqsave(&umn.lock, flags);
        ret = umn_send_packet(umn.tty_fd, skb->data, skb->len);
        dev_kfree_skb(skb);
+       spin_unlock_irqrestore(&lp->lock, flags);
        return(ret);
 }
 

Re: [User-mode-linux-devel] a final test6 checkin

[Jeff Dike <jdike@ka...>]

> I've been running small network simulations with the virtual ethernet
> driver and your last update and it seems stable to me.

Cool.

Do you have any neato network things that you've been testing with?  I'd 
wouldn't mind getting some of them documented and up on the web site and in 
the HOWTO.

				Jeff

Re: [User-mode-linux-devel] a final test6 checkin

[James R. Leu <jleu@mi...>]

I've been running small network simulations with the virtual ethernet driver
and your last update and it seems stable to me.  Seems like the interrupt
system is solid.  Good work!

Laters,
Jim

On Tue, Aug 15, 2000 at 01:29:06PM -0500, Jeff Dike wrote:
> This one is just a bit of cleanup.  This kernel passed my stress tests, so 
> it's now being tagged v_2_4_0_test6, and it's going to be what I release to 
> the masses.
> 
> 				Jeff
> 
> 
> 
> _______________________________________________
> User-mode-linux-devel mailing list
> User-mode-linux-devel@...
> http://lists.sourceforge.net/mailman/listinfo/user-mode-linux-devel

-- 
James R. Leu

[User-mode-linux-devel] a final test6 checkin

[Jeff Dike <jdike@ka...>]

This one is just a bit of cleanup.  This kernel passed my stress tests, so 
it's now being tagged v_2_4_0_test6, and it's going to be what I release to 
the masses.

				Jeff

[User-mode-linux-devel] cvs checkins

[Jeff Dike <jdike@ka...>]

test6 is better now.  I found a stupid casting bug which was causing pointer 
arithmetic to happen instead of integer arithetic.  With this patch, uml now 
at least boots.

The patch is below.

				Jeff

--- cvs/linux/include/asm-um/page.h	Thu Aug 10 14:33:23 2000
+++ um/include/asm-um/page.h	Tue Aug 15 00:46:29 2000
@@ -32,7 +32,7 @@
 
 #define __va_space (8*1024*1024)
 
-#define __pa(x)	((unsigned long) ((x) - (physmem)))
+#define __pa(x)	((unsigned long) (x) - (physmem))
 #define __va(x)	((void *) ((x) + (physmem)))
 
 #define virt_to_page(kaddr)	(mem_map + (__pa(kaddr) >> PAGE_SHIFT))
--- cvs/linux/include/asm-um/pgtable.h	Thu Aug 10 14:33:23 2000
+++ um/include/asm-um/pgtable.h	Tue Aug 15 00:46:42 2000
@@ -151,7 +151,7 @@
 
 #define BAD_PAGETABLE __bad_pagetable()
 #define BAD_PAGE __bad_page()
-#define ZERO_PAGE(vaddr) (virt_to_page(((unsigned long) empty_zero_page)))
+#define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
 
 /* number of bits that fit into a memory pointer */
 #define BITS_PER_PTR			(8*sizeof(unsigned long))

[User-mode-linux-devel] test6 changes are checked in

[Jeff Dike <jdike@ka...>]

There were some minor softirq and page table changes.

However, this is not stable.  Grab this only if you to help debug it.  The 
debugging interface I just put in might help :-)

				Jeff

[User-mode-linux-devel] Re: [PATCH] test6

[Jeff Dike <jdike@ka...>]

> It also turns um.a into um.o so ksyms doesn't get dropped by the
> linker (which means no export of those symbols, hence modules break).

That's something I've been meaning to do...

> Result is running now (so `it works for me!').

I can't get it to boot my Slackware fs.  It boots tomsrtbt, but I got a crash 
in procfs when I shut it down.

>  I suspect this may be the case on an unchecked malloc).  I *don't*
> think this is a UML issue...

Well, I think I have some problems here.  And I think you're hitting them.

				Jeff

[User-mode-linux-devel] [PATCH] test6

[Rusty Russell <rusty@li...>]

These are the changes I needed to make for compilation with 6-9 (and
probably for test6 proper).

It also turns um.a into um.o so ksyms doesn't get dropped by the
linker (which means no export of those symbols, hence modules break).
Adds some more exports.  Result is running now (so `it works for
me!').

I get apt-get to seg fault sometimes (and non-repeatedly); I'm running
without swap in um kernel, so it may well be running out of memory
(given that uml reports dying with NULL access, I suspect this may be
the case on an unchecked malloc).  I *don't* think this is a UML
issue...

I do get crashes on halt frequently (after `Sending KILL Signal...'),
but no biggie (fsck is fast on a 500MB root file).

Cheers!
Rusty.

diff -urN -X /tmp/filele7AXi --minimal linux-2.4.0-test6-9-uml/arch/um/Makefile working-2.4.0-test6-9-uml/arch/um/Makefile
--- linux-2.4.0-test6-9-uml/arch/um/Makefile	Thu Aug 10 03:07:55 2000
+++ working-2.4.0-test6-9-uml/arch/um/Makefile	Thu Aug 10 23:58:42 2000
@@ -27,7 +27,7 @@
 SUBDIRS += $(ARCH_DIR)/drivers $(ARCH_DIR)/kernel $(ARCH_DIR)/sys-$(SUBARCH) \
 	$(ARCH_DIR)/ptproxy
 
-LIBS += $(ARCH_DIR)/kernel/um.a $(ARCH_DIR)/drivers/um_drivers.o \
+LIBS += $(ARCH_DIR)/kernel/um.o $(ARCH_DIR)/drivers/um_drivers.o \
 	$(ARCH_DIR)/sys-$(SUBARCH)/sys.o $(ARCH_DIR)/ptproxy/ptproxy.a
 
 $(ARCH_DIR)/link.ld:
diff -urN -X /tmp/filele7AXi --minimal linux-2.4.0-test6-9-uml/arch/um/kernel/Makefile working-2.4.0-test6-9-uml/arch/um/kernel/Makefile
--- linux-2.4.0-test6-9-uml/arch/um/kernel/Makefile	Thu Aug 10 03:07:56 2000
+++ working-2.4.0-test6-9-uml/arch/um/kernel/Makefile	Fri Aug 11 01:07:08 2000
@@ -1,4 +1,4 @@
-LIB = um.a
+OBJ := um.o
 
 OBJS = process.o current.o exec_kern.o exec_user.o init_task.o irq.o \
 	irq_user.o mem.o process_kern.o ptrace.o reboot.o resource.o \
@@ -13,7 +13,7 @@
 USER_CFLAGS = $(patsubst -I%,,$(CFLAGS))
 USER_CFLAGS += -I../include
 
-all: $(LIB) unmap_fin.o
+all: $(OBJ) unmap_fin.o
 
 exec_user.o: exec_user.c
 	gcc -D__KERNEL__ $(USER_CFLAGS) $(EXTRA_CFLAGS) -c -o $@ $<
@@ -43,12 +43,12 @@
 unmap_fin.o : unmap.o
 	ld -r -o $@ $< -lc
 
-$(LIB): $(OBJS) $(OX_OBJS)
+$(OBJ): $(OBJS) $(OX_OBJS)
 	rm -f $@
-	ar cr $@ $^
+	$(LD) $(LINKFLAGS) $^ -o $@
 
 clean:
-	rm -f $(OBJS) $(OX_OBJS)
+	rm -f $(OBJ) $(OBJS) $(OX_OBJS)
 
 modules:
 
diff -urN -X /tmp/filele7AXi --minimal linux-2.4.0-test6-9-uml/arch/um/kernel/irq.c working-2.4.0-test6-9-uml/arch/um/kernel/irq.c
--- linux-2.4.0-test6-9-uml/arch/um/kernel/irq.c	Thu Aug 10 03:07:56 2000
+++ working-2.4.0-test6-9-uml/arch/um/kernel/irq.c	Thu Aug 10 03:25:47 2000
@@ -328,7 +328,7 @@
 	desc->handler->end(irq);
 	spin_unlock(&desc->lock);
 
-	if (softirq_state[cpu].active & softirq_state[cpu].mask)
+	if (softirq_active(cpu) & softirq_mask(cpu))
 		do_softirq();
 	return 1;
 }
diff -urN -X /tmp/filele7AXi --minimal linux-2.4.0-test6-9-uml/arch/um/kernel/ksyms.c working-2.4.0-test6-9-uml/arch/um/kernel/ksyms.c
--- linux-2.4.0-test6-9-uml/arch/um/kernel/ksyms.c	Thu Aug 10 03:07:56 2000
+++ working-2.4.0-test6-9-uml/arch/um/kernel/ksyms.c	Fri Aug 11 01:16:44 2000
@@ -1,6 +1,7 @@
 #include "linux/module.h"
 #include "linux/string.h"
 #include "asm/current.h"
+#include "asm/uaccess.h"
 #include "kern_util.h"
 #include "user_util.h"
 
@@ -9,3 +10,5 @@
 EXPORT_SYMBOL(physmem);
 EXPORT_SYMBOL(current_task);
 EXPORT_SYMBOL(set_signals);
+EXPORT_SYMBOL(set_fault_addr);
+EXPORT_SYMBOL(block_signals);
diff -urN -X /tmp/filele7AXi --minimal linux-2.4.0-test6-9-uml/arch/um/kernel/process_kern.c working-2.4.0-test6-9-uml/arch/um/kernel/process_kern.c
--- linux-2.4.0-test6-9-uml/arch/um/kernel/process_kern.c	Thu Aug 10 03:07:56 2000
+++ working-2.4.0-test6-9-uml/arch/um/kernel/process_kern.c	Thu Aug 10 03:31:00 2000
@@ -218,7 +218,7 @@
 void do_bh(void)
 {
 #ifndef __SMP__
-	if (softirq_state[0].active&softirq_state[0].mask)
+	if (softirq_active(0) & softirq_mask(0))
 		do_softirq();
 #else
 #error Need to update do_bh
--
Hacking time.

[User-mode-linux-devel] Re: ssh user key

[Jeff Dike <jdike@ka...>]

> I get a "Seg fault in signals" with a set_signals(0) after 10 minutes
> or so of flood pinging a machine across an ethernet cable from the
> host.

Well, I've been running ping floods in both directions for the last 20 minutes 
with no problems.  So, I'm gonna need a better way of reproducing the problem.

				Jeff

[User-mode-linux-devel] Re: Making incoming connections to a uml

[Jeff Dike <jdike@ka...>]

> I tried to get yesterdays (8/8) cvs kernel to compile, but commenting
> out a lot of references to init_ptproxy and a few other _ptproxy
> functions gave me a compiled executable, but it didn't run.

How are you pulling it and building it?

cvs says there aren't any differences between the CVS pool and my pool except 
a few trivialities that you wouldn't notice.  And my pool builds and works 
fine as far as I can tell.

I'll grab the pool from sourceforge and build it just to sanity-check things.

				Jeff

[User-mode-linux-devel] Re: Making incoming connections to a uml

[Jeff Dike <jdike@ka...>]

> For tips like that, do you have any preference about whether they show
> up on the web site proper or in the howto?  I'd be glad to add it to
> the howto if that's useful.

Both, I think.  Right now, the only place to get the HOWTO is off the web 
site, which makes it a bit redundant, but ultimately, it should get 
distributed from LDP or something.

> It wasn't clear whether "set_signals(0)" or "set_signals(-1)" was the
> correct value in system.h:local_irq_save(flags), so I'm trying them
> both.

The quickie fix I gave you was -1, which is sort of wrong, so in the real 
code, I changed it back to 0, and fixed set_signals to correspond to that.

I've been running ping floods against uml and I can't get anything bad to 
happen.

> I tried to get yesterdays (8/8) cvs kernel to compile, but commenting
> out a lot of references to init_ptproxy and a few other _ptproxy
> functions gave me a compiled executable, but it didn't run.

Rats.  I'll figure out what's wrong with CVS and fix it.

> I'm sure you knew the problem existed and are working on it.  

No I didn't.

> I was
> mostly doing it to see Rusty's new menuconfig stuff.

It is pretty nice.

				Jeff

[User-mode-linux-devel] Re: Making incoming connections to a uml

[William Stearns <wstearns@po...>]

Good afternoon, Jeff,

On Tue, 8 Aug 2000, Jeff Dike wrote:

> I'm revamping the web site, and I'd like to describe how to allow incoming 
> connections.  That's the port forwarding thing you showed me a while back.  
> Can you repeat it so I can toss it in?

	I need to look that up again; I'll send it to you when I have it.
	For tips like that, do you have any preference about whether they
show up on the web site proper or in the howto?  I'd be glad to add it to
the howto if that's useful.

> How are things going?  I haven't heard anything from you in a while, so I 
> assume that uml is behaving itself.

	Kinda-sorta.  It wasn't clear whether "set_signals(0)" or
"set_signals(-1)" was the correct value in system.h:local_irq_save(flags),
so I'm trying them both.  I get a "Seg fault in signals" with a
set_signals(0) after 10 minutes or so of flood pinging a machine across
an ethernet cable from the host.  I also get the exact same error with
set_signals(-1)
	I hope you understand that I'm not fixated on pingflooding people
on the local lan!  It just seems like an easy way to reproduce a
networking issue that might take days under normal circumstances to show
up.

	I tried to get yesterdays (8/8) cvs kernel to compile, but
commenting out a lot of references to init_ptproxy and a few other
_ptproxy functions gave me a compiled executable, but it didn't run.  No
answer necessary; I'm sure you knew the problem existed and are working on
it.  I was mostly doing it to see Rusty's new menuconfig stuff.
	Cheers,
	- Bill

---------------------------------------------------------------------------
	People standing in the middle of the road look like 
roadkill to me.
	-- Linus Torvalds, keynote speech at LWCE.
--------------------------------------------------------------------------
William Stearns (wstearns@...).  Mason, Buildkernel, named2hosts, 
and ipfwadm2ipchains are at:                http://www.pobox.com/~wstearns
LinuxMonth; articles for Linux Enthusiasts! http://www.linuxmonth.com
--------------------------------------------------------------------------

[User-mode-linux-devel] Last batch of the day :-)

[Jeff Dike <jdike@ka...>]

I merged in Rusty's patches, which give uml a complete config process for the 
first time and which export enough symbols to allow loading the netfilter 
stuff as modules, which I'm sure all of you have been waiting impatiently for.

I also fixed a couple of irq bugs.  As a result, the virtual ethernet driver 
seems better behaved.

				Jeff

[User-mode-linux-devel] more debugging checkins

[Jeff Dike <jdike@ka...>]

This batch of changes cleans things up a bit, and adds a 'debug' command line 
switch.  This causes uml to start a gdb in an xterm, and leaves it stopped at 
the beginning of start_kernel for your debugging pleasure.

The proxy causes occasional strange things to happen, like illegal 
instructions and process segfaults.

				Jeff

[User-mode-linux-devel] more ptrace proxy checkins

[Jeff Dike <jdike@ka...>]

I checked in some more fixes.  Now you can set a breakpoint and gdb will catch 
whatever thread happens to hit it.  'next' and 'step' seem to work ok.

You still can't step across a context switch.  I'm working on that.

				Jeff

[User-mode-linux-devel] ptrace proxy checkins

[Jeff Dike <jdike@ka...>]

I've integrated Lars Brinkhoff's ptrace proxy into uml and checked it in to 
cvs.  This is a clever bit of code that gets around the problem of gdb not 
being able to attach to threads which are under syscall trace by putting gdb 
under syscall trace and faking it into believing that it is really attached to 
a thread when it's really not.

The way this works in uml is that when you want to start up gdb, send a 
SIGUSR1 to the tracing thread.  It will fire up an xterm with a gdb running in 
it.  Attach to whatever thread is currently running.  If the idle thread is 
'running' ps will show it as 'S', otherwise (as with a panic), the running 
thread will show up as 'R' like you'd expect.

Limitations:  
You can't follow the kernel from one thread to another.  
Breakpoints set in one thread disappear in another.

I'm working on these.

				Jeff

Re: UML-CC Re: [User-mode-linux-devel] An irq fix check-in

[Jeff Dike <jdike@ka...>]

> What, if anything, should be under /usr/src/linux while I'm building
> uml in a different directory?  Where should /usr/include/
> {asm|linux|scsi} point?

That's probably the problem.  /usr/src/linux should contain the sources for 
the native kernel.  /usr/include/asm probably points to /usr/src/linux/asm, 
which probably now points at /usr/src/linux/asm-um.  Point it back to 
/usr/src/linux/asm-i386, and things should build again.

				Jeff