Linus,
Attached patch implements a /proc/#/wchan which provides a pre-decoded
wchan via kallsyms. I.e.,
[15:05:06]rml@phantasy:~$ cat /proc/1228/wchan
wait4
Which is damn cool to me and will let ps(1) grab wchan information
without having to parse System.map. It also means procps will not need
System.map.
If CONFIG_KALLSYMS is enabled, /proc/#/wchan exists and exports the
pre-decoded symbol name. The old wchan value in /proc/#/stat is
hard-coded to zero.
If CONFIG_KALLSYMS is not enabled, /proc/#/wchan does not exist to
conserve memory. In that case, the old wchan field in /proc/#/stat will
export the usual wchan address.
This will not break procps, however the wchan field will be zero without
an updated version if CONFIG_KALLSYMS is set. That is fine as 2.5
requires an updated procps anyhow. If CONFIG_KALLSYMS is not set,
things are unchanged.
The procps patch is available at:
http://tech9.net/rml/procps/patches/procps-wchan-rml-2.0.10.20021011-1.patch
and ready for merging into procps CVS.
Patch is against current BK. Please, apply.
Robert Love
Implement /proc/#/wchan, a pre-decoded wchan
Documentation/filesystems/proc.txt | 1 +
fs/proc/array.c | 10 ++++++----
fs/proc/base.c | 33 +++++++++++++++++++++++++++++++++
3 files changed, 40 insertions(+), 4 deletions(-)
diff -urN linux-2.5.43/fs/proc/array.c linux/fs/proc/array.c
--- linux-2.5.43/fs/proc/array.c 2002-10-16 22:55:13.000000000 -0400
+++ linux/fs/proc/array.c 2002-10-16 23:26:19.000000000 -0400
@@ -298,7 +298,7 @@
int proc_pid_stat(struct task_struct *task, char * buffer)
{
- unsigned long vsize, eip, esp, wchan;
+ unsigned long vsize, eip, esp;
long priority, nice;
int tty_pgrp = -1, tty_nr = 0;
sigset_t sigign, sigcatch;
@@ -331,8 +331,6 @@
up_read(&mm->mmap_sem);
}
- wchan = get_wchan(task);
-
collect_sigign_sigcatch(task, &sigign, &sigcatch);
/* scale priority and nice values from timeslices to -20..20 */
@@ -384,7 +382,11 @@
task->blocked.sig[0] & 0x7fffffffUL,
sigign .sig[0] & 0x7fffffffUL,
sigcatch .sig[0] & 0x7fffffffUL,
- wchan,
+#ifndef CONFIG_KALLSYMS
+ get_wchan(task),
+#else
+ 0UL,
+#endif
task->nswap,
task->cnswap,
task->exit_signal,
diff -urN linux-2.5.43/fs/proc/base.c linux/fs/proc/base.c
--- linux-2.5.43/fs/proc/base.c 2002-10-15 23:28:25.000000000 -0400
+++ linux/fs/proc/base.c 2002-10-16 23:24:17.000000000 -0400
@@ -28,6 +28,7 @@
#include <linux/namespace.h>
#include <linux/mm.h>
#include <linux/smp_lock.h>
+#include <linux/kallsyms.h>
/*
* For hysterical raisins we keep the same inumbers as in the old procfs.
@@ -54,6 +55,7 @@
PROC_PID_MAPS,
PROC_PID_CPU,
PROC_PID_MOUNTS,
+ PROC_PID_WCHAN,
PROC_PID_FD_DIR = 0x8000, /* 0x8000-0xffff */
};
@@ -81,6 +83,9 @@
E(PROC_PID_ROOT, "root", S_IFLNK|S_IRWXUGO),
E(PROC_PID_EXE, "exe", S_IFLNK|S_IRWXUGO),
E(PROC_PID_MOUNTS, "mounts", S_IFREG|S_IRUGO),
+#ifdef CONFIG_KALLSYMS
+ E(PROC_PID_WCHAN, "wchan", S_IFREG|S_IRUGO),
+#endif
{0,0,NULL,0}
};
#undef E
@@ -245,6 +250,28 @@
return res;
}
+#ifdef CONFIG_KALLSYMS
+/*
+ * Provides a wchan file via kallsyms in a proper one-value-per-file format.
+ * Returns the resolved symbol. If that fails, simply return the address.
+ */
+static int proc_pid_wchan(struct task_struct *task, char *buffer)
+{
+ const char *sym_name, *ignore;
+ unsigned long wchan, dummy;
+
+ wchan = get_wchan(task);
+
+ if (!kallsyms_address_to_symbol(wchan, &ignore, &dummy, &dummy,
+ &ignore, &dummy, &dummy, &sym_name,
+ &dummy, &dummy)) {
+ return sprintf(buffer, "%lu", wchan);
+ }
+
+ return sprintf(buffer, "%s", sym_name);
+}
+#endif
+
/************************************************************************/
/* Here the fs part begins */
/************************************************************************/
@@ -1016,6 +1043,12 @@
case PROC_PID_MOUNTS:
inode->i_fop = &proc_mounts_operations;
break;
+#ifdef CONFIG_KALLSYMS
+ case PROC_PID_WCHAN:
+ inode->i_fop = &proc_info_file_operations;
+ ei->op.proc_read = proc_pid_wchan;
+ break;
+#endif
default:
printk("procfs: impossible type (%d)",p->type);
iput(inode);
diff -urN linux-2.5.43/Documentation/filesystems/proc.txt linux/Documentation/filesystems/proc.txt
--- linux-2.5.43/Documentation/filesystems/proc.txt 2002-10-15 23:27:48.000000000 -0400
+++ linux/Documentation/filesystems/proc.txt 2002-10-17 00:14:36.000000000 -0400
@@ -130,6 +130,7 @@
stat Process status
statm Process memory status information
status Process status in human readable form
+ wchan If CONFIG_KALLSYMS is set, a pre-decoded wchan
..............................................................................
For example, to get the status information of a process, all you have to do is
On Thu, Oct 17, 2002 at 03:14:02PM -0400, Robert Love wrote:
> Linus,
>
> Attached patch implements a /proc/#/wchan which provides a pre-decoded
> wchan via kallsyms. I.e.,
>
> [15:05:06]rml@phantasy:~$ cat /proc/1228/wchan
> wait4
>
> Which is damn cool to me and will let ps(1) grab wchan information
> without having to parse System.map. It also means procps will not need
> System.map.
>
> If CONFIG_KALLSYMS is enabled, /proc/#/wchan exists and exports the
> pre-decoded symbol name. The old wchan value in /proc/#/stat is
> hard-coded to zero.
>
> If CONFIG_KALLSYMS is not enabled, /proc/#/wchan does not exist to
> conserve memory. In that case, the old wchan field in /proc/#/stat will
> export the usual wchan address.
>
> This will not break procps, however the wchan field will be zero without
> an updated version if CONFIG_KALLSYMS is set. That is fine as 2.5
> requires an updated procps anyhow. If CONFIG_KALLSYMS is not set,
> things are unchanged.
Can't you just left the old, nuerical one in even if CONFIG_KALLSYMS
ise set? One ifdef less and far less surprises..
On Thu, 2002-10-17 at 15:58, Christoph Hellwig wrote:
> Can't you just left the old, nuerical one in even if CONFIG_KALLSYMS
> ise set? One ifdef less and far less surprises..
But why? Save the call to get_wchan()...
Ideally I would like to remove the field altogether but thats too much
breakage.
A value of zero in the field position is an indication the wchan file
exists, too.
The ifdefs there do not bother me. I think its pretty clear why its
being done, although I could add a comment to the "0UL" value saying
"now in /proc/#/wchan" or whatever.
But, no, I do not agree. No human can parse /proc/#/stats anyhow.
Since procps needs to be updated for 3.0 anyhow, it will be fine.
Robert Love
On Oct 17 at 03:14:02, Robert Love wrote:
> Which is damn cool to me and will let ps(1) grab wchan information
> without having to parse System.map. It also means procps will not need
> System.map.
Hope this patch gets included in kernel 2.5 ;). I think is very nice to make
ps not to depend on System.map for resolving wchan. I'll give a try to your
patch right now.
--
David G?mez
"The question of whether computers can think is just like the question of
whether submarines can swim." -- Edsger W. Dijkstra
Hi!
> > Can't you just left the old, nuerical one in even if CONFIG_KALLSYMS
> > ise set? One ifdef less and far less surprises..
>
> But why? Save the call to get_wchan()...
Yes, and force users to update procps for no good reason. And "new"
procps will still need code to deal with get_wchan themselves... Plus
you loose information by killing get_wchan() -- two different wait
points in one function seems very possible to me.
Pavel
--
Worst form of spam? Adding advertisment signatures ala sourceforge.net.
What goes next? Inserting advertisment *into* email?
On Mon, 2002-10-21 at 19:08, Pavel Machek wrote:
> Yes, and force users to update procps for no good reason. And "new"
> procps will still need code to deal with get_wchan themselves... Plus
> you loose information by killing get_wchan() -- two different wait
> points in one function seems very possible to me.
- users of 2.5 need a new procps but only wchan is "broken" and
only if CONFIG_KALLSYMS is set anyhow
- I did not kill get_wchan, just do not use it in stat - in fact
I use it in the new wchan approach, too
- its not an issue now because the updated patch (posted 5 days
after this email you are replying to) keeps wchan.
So you got your way. Look at the patch posted 22 Oct. It is now in
2.5-mm.
Robert Love