2007-09-06 11:25:20

by Michael Stiller

Subject: WARNING native_smp_call_function mask 2.6.22.6 SMP on Single CPU System

Hi,

i get the following warning if i call vfree some memory allocated by
vmalloc on a single cpu machine running 2.6.22.6 SMP:

WARNING: at arch/i386/kernel/smp.c:559 native_smp_call_function_mask()
[<c010e77c>] native_smp_call_function_mask+0x145/0x14a
[<c01cb32d>] idr_remove+0x102/0x163
[<c010e43f>] do_flush_tlb_all+0x0/0x59
[<c010e43f>] do_flush_tlb_all+0x0/0x59
[<c010ea33>] smp_call_function+0x1c/0x29
[<c0120922>] on_each_cpu+0x18/0x29
[<c010e2e9>] flush_tlb_all+0x1b/0x1f
[<c0155227>] remove_vm_area+0x4c/0x59
[<c0155252>] __vunmap+0x1e/0xe5
[<c011ccb7>] printk+0x1b/0x1f
[<e0b15f21>] ring_release+0x1f7/0x28a [ring]
[<c0247305>] sock_fasync+0x99/0x146
[<c024826b>] sock_release+0x14/0x66
[<c02486a8>] sock_close+0x1e/0x38
[<c01600ee>] __fput+0x99/0x166
[<c015db9f>] filp_close+0x3e/0x62
[<c015ebf0>] sys_close+0x5f/0x9e
[<c0103dba>] syscall_call+0x7/0xb
=======================

I use the rvmalloc / rvfree functions copied from usbvideo.c:

static void rvfree(void *mem, unsigned long size)
{
unsigned long adr;
unsigned long pages = 0;

#if defined(RING_DEBUG)
printk("RING: rvfree: %lu bytes\n", size);
#endif

if (!mem)
return;

adr = (unsigned long) mem;
while ((long) size > 0) {
ClearPageReserved(vmalloc_to_page((void *)adr));
pages++;
adr += PAGE_SIZE;
size -= PAGE_SIZE;
}
#if defined(RING_DEBUG)
printk("RING: rvfree: %lu pages\n", pages);
printk("RING: rvfree: calling vfree....\n");
#endif
vfree(mem); // WARNING occurs here
printk("RING: rvfree: after vfree....\n");
}


Any clues?

Please cc me for answers. Thanks in advance.

-Michael

2007-09-06 12:22:23

by Satyam Sharma

Subject: Re: WARNING native_smp_call_function mask 2.6.22.6 SMP on Single CPU System

On Thu, 6 Sep 2007, Michael Stiller wrote:

> Hi,
>
> i get the following warning if i call vfree some memory allocated by
> vmalloc on a single cpu machine running 2.6.22.6 SMP:
>
> WARNING: at arch/i386/kernel/smp.c:559 native_smp_call_function_mask()
> [<c010e77c>] native_smp_call_function_mask+0x145/0x14a
> [<c01cb32d>] idr_remove+0x102/0x163
> [<c010e43f>] do_flush_tlb_all+0x0/0x59
> [<c010e43f>] do_flush_tlb_all+0x0/0x59
> [<c010ea33>] smp_call_function+0x1c/0x29
> [<c0120922>] on_each_cpu+0x18/0x29
> [<c010e2e9>] flush_tlb_all+0x1b/0x1f
> [<c0155227>] remove_vm_area+0x4c/0x59
> [<c0155252>] __vunmap+0x1e/0xe5
> [<c011ccb7>] printk+0x1b/0x1f
> [<e0b15f21>] ring_release+0x1f7/0x28a [ring]

There's no ring_release() in 2.6.22 -- at least doesn't show up in a grep.

> [<c0247305>] sock_fasync+0x99/0x146
> [<c024826b>] sock_release+0x14/0x66
> [<c02486a8>] sock_close+0x1e/0x38
> [<c01600ee>] __fput+0x99/0x166
> [<c015db9f>] filp_close+0x3e/0x62
> [<c015ebf0>] sys_close+0x5f/0x9e
> [<c0103dba>] syscall_call+0x7/0xb
> =======================
>
> I use the rvmalloc / rvfree functions copied from usbvideo.c:
>
> static void rvfree(void *mem, unsigned long size)
> {
> unsigned long adr;
> unsigned long pages = 0;
>
> #if defined(RING_DEBUG)
> printk("RING: rvfree: %lu bytes\n", size);
> #endif
>
> if (!mem)
> return;
>
> adr = (unsigned long) mem;
> while ((long) size > 0) {
> ClearPageReserved(vmalloc_to_page((void *)adr));
> pages++;
> adr += PAGE_SIZE;
> size -= PAGE_SIZE;
> }
> #if defined(RING_DEBUG)
> printk("RING: rvfree: %lu pages\n", pages);
> printk("RING: rvfree: calling vfree....\n");
> #endif
> vfree(mem); // WARNING occurs here
> printk("RING: rvfree: after vfree....\n");
> }
>
>
> Any clues?

You can't call vfree() with irqs disabled. Looks like you called
ring_release() / rvfree() that way ...

2007-09-06 12:22:50

by Michael Stiller

Subject: Re: WARNING native_smp_call_function mask 2.6.22.6 SMP on Single CPU System

On Thu, 2007-09-06 at 13:00 +0200, Michael Stiller wrote:
> Hi,
>
> i get the following warning if i call vfree some memory allocated by
> vmalloc on a single cpu machine running 2.6.22.6 SMP:
>
> WARNING: at arch/i386/kernel/smp.c:559 native_smp_call_function_mask()

Resolved. I called vfree with irqs disabled. Me stupid ;)

-Michael