[PATCH] OProfile: Fix oops on undetected CPU type

[PATCH] OProfile: Fix oops on undetected CPU type

[Zwane Mwaikambo]

Running opcontrol --reset on an em64t P4 yields the following (clearly 
there must be a reason why it's disabled but i'll address that  
seperately);

Unable to handle kernel NULL pointer dereference at virtual address 
00000000
 printing eip:
c0562cf4
*pde = 00000000
Oops: 0000 [#1]
PREEMPT SMP DEBUG_PAGEALLOC
Modules linked in:
CPU:    1
EIP:    0060:[<c0562cf4>]    Not tainted VLI
EFLAGS: 00010246   (2.6.11-rc2)
EIP is at oprofilefs_str_to_user+0x24/0x50
eax: 00000000   ebx: 00000000   ecx: ffffffff   edx: ffffffff
esi: 00001000   edi: 00000000   ebp: ec477f78   esp: ec477f60
ds: 007b   es: 007b   ss: 0068
Process cat (pid: 2810, threadinfo=ec476000 task=f739ead0)
Stack: 00000000 c016806a 0804d858 c072aa00 ec9eff58 00001000 ec477f9c c015e1e7
       ec477fa8 00000000 00000000 0804d858 ec9eff58 fffffff7 0804d858 ec477fbc
       c015e50d ec477fa8 00000000 00000000 00000000 00000003 00001000 ec476000
Call Trace:
 [<c0103fea>] show_stack+0x7a/0x90
 [<c0104176>] show_registers+0x156/0x1d0
 [<c01043a0>] die+0x100/0x190
 [<c01170be>] do_page_fault+0x45e/0x644
 [<c0103c77>] error_code+0x2b/0x30
 [<c015e1e7>] vfs_read+0xc7/0x160
 [<c015e50d>] sys_read+0x3d/0x70
 [<c0103105>] sysenter_past_esp+0x52/0x75
Code: 90 90 90 90 90 90 90 55 89 e5 83 ec 18 89 5d f4 31 db 89 55 f0 ba ff 
ff ff ff 89 75 f8 89 ce 89 d1 89 7d fc 89 c7 89 04 24 89 d8 <f2> ae f7 d1 
49 89 4c 24 04 8b 45 f0 89 f2 8b 4d 08 e8 66 e4 c1

->cpu_type is NULL since p4_init skipped this specific processor.

Signed-off-by: Zwane Mwaikambo <zwane@fsmlabs.com>

===== drivers/oprofile/oprofile_files.c 1.7 vs edited =====
--- 1.7/drivers/oprofile/oprofile_files.c	2005-01-04 19:48:23 -07:00
+++ edited/drivers/oprofile/oprofile_files.c	2005-01-28 11:36:25 -07:00
@@ -63,7 +63,9 @@ static struct file_operations pointer_si
 
 static ssize_t cpu_type_read(struct file * file, char __user * buf, size_t count, loff_t * offset)
 {
-	return oprofilefs_str_to_user(oprofile_ops.cpu_type, buf, count, offset);
+	if (oprofile_ops.cpu_type)
+		return oprofilefs_str_to_user(oprofile_ops.cpu_type, buf, count, offset);
+	return -EIO;
 }
  
  

Re: [PATCH] OProfile: Fix oops on undetected CPU type

[John Levon]

On Fri, Jan 28, 2005 at 12:06:19PM -0700, Zwane Mwaikambo wrote:

> ===== drivers/oprofile/oprofile_files.c 1.7 vs edited =====
> --- 1.7/drivers/oprofile/oprofile_files.c	2005-01-04 19:48:23 -07:00
> +++ edited/drivers/oprofile/oprofile_files.c	2005-01-28 11:36:25 -07:00
> @@ -63,7 +63,9 @@ static struct file_operations pointer_si
>  
>  static ssize_t cpu_type_read(struct file * file, char __user * buf, size_t count, loff_t * offset)
>  {
> -	return oprofilefs_str_to_user(oprofile_ops.cpu_type, buf, count, offset);
> +	if (oprofile_ops.cpu_type)
> +		return oprofilefs_str_to_user(oprofile_ops.cpu_type, buf, count, offset);
> +	return -EIO;

This is wrong: you need to investigate why .cpu_type isn't set: in
particular, it should have fallen back to timer mode.
oprofile_arch_init() should have returned -ENODEV, and that should have
set timer mode.

Unfortunately bkcvs seems out of date so I can't even look at this
myself.

john

Re: [PATCH] OProfile: Fix oops on undetected CPU type

[Zwane Mwaikambo]

On Sat, 29 Jan 2005, John Levon wrote:

> On Fri, Jan 28, 2005 at 12:06:19PM -0700, Zwane Mwaikambo wrote:
> 
> > ===== drivers/oprofile/oprofile_files.c 1.7 vs edited =====
> > --- 1.7/drivers/oprofile/oprofile_files.c	2005-01-04 19:48:23 -07:00
> > +++ edited/drivers/oprofile/oprofile_files.c	2005-01-28 11:36:25 -07:00
> > @@ -63,7 +63,9 @@ static struct file_operations pointer_si
> >  
> >  static ssize_t cpu_type_read(struct file * file, char __user * buf, size_t count, loff_t * offset)
> >  {
> > -	return oprofilefs_str_to_user(oprofile_ops.cpu_type, buf, count, offset);
> > +	if (oprofile_ops.cpu_type)
> > +		return oprofilefs_str_to_user(oprofile_ops.cpu_type, buf, count, offset);
> > +	return -EIO;
> 
> This is wrong: you need to investigate why .cpu_type isn't set: in
> particular, it should have fallen back to timer mode.
> oprofile_arch_init() should have returned -ENODEV, and that should have
> set timer mode.
> 
> Unfortunately bkcvs seems out of date so I can't even look at this
> myself.

Yes you are right, i checked bk and there was a lot of shuffling about due 
to the timer override. But it looks like we're depending on the timer 
variable being set. We could always just run timer_init if cpu_type is not 
set.

Signed-off-by: Zwane Mwaikambo <zwane@arm.linux.org.uk>

===== drivers/oprofile/oprof.c 1.11 vs edited =====
--- 1.11/drivers/oprofile/oprof.c	2005-01-04 19:48:23 -07:00
+++ edited/drivers/oprofile/oprof.c	2005-01-29 09:38:24 -07:00
@@ -157,7 +157,7 @@ static int __init oprofile_init(void)
 
 	oprofile_arch_init(&oprofile_ops);
 
-	if (timer) {
+	if (timer || !oprofile_ops.cpu_type) {
 		printk(KERN_INFO "oprofile: using timer interrupt.\n");
 		oprofile_timer_init(&oprofile_ops);
 	}

Re: [PATCH] OProfile: Fix oops on undetected CPU type

[John Levon]

On Sat, Jan 29, 2005 at 09:47:42AM -0700, Zwane Mwaikambo wrote:

> > Unfortunately bkcvs seems out of date so I can't even look at this
> > myself.
> 
> Yes you are right, i checked bk and there was a lot of shuffling about due 
> to the timer override. But it looks like we're depending on the timer 
> variable being set. We could always just run timer_init if cpu_type is not 
> set.
> 
> Signed-off-by: Zwane Mwaikambo <zwane@arm.linux.org.uk>
> 
> ===== drivers/oprofile/oprof.c 1.11 vs edited =====
> --- 1.11/drivers/oprofile/oprof.c	2005-01-04 19:48:23 -07:00
> +++ edited/drivers/oprofile/oprof.c	2005-01-29 09:38:24 -07:00
> @@ -157,7 +157,7 @@ static int __init oprofile_init(void)
>  
>  	oprofile_arch_init(&oprofile_ops);
>  
> -	if (timer) {
> +	if (timer || !oprofile_ops.cpu_type) {

Looks like you're still on the broken bkcvs, which is missing this
patch:

http://linus.bkbits.net:8080/linux-2.5/cset@1.1966.1.72?nav=index.html|ChangeSet@-2w

which AFAICS is the correct solution to the problem.

regards,
john

Re: [PATCH] OProfile: Fix oops on undetected CPU type

[Zwane Mwaikambo]

On Sat, 29 Jan 2005, John Levon wrote:

> Looks like you're still on the broken bkcvs, which is missing this
> patch:
> 
> http://linus.bkbits.net:8080/linux-2.5/cset@1.1966.1.72?nav=index.html|ChangeSet@-2w
> 
> which AFAICS is the correct solution to the problem.

Hmm i was actually using BK and not BKCVS and had pulled after the 25th. 
Anyway thanks, that should do it then.