Re: [LKP] Re: [mm/vunmap] e47110e905: WARNING:at_mm/vmalloc.c:#__vunmap

[Xing Zhengjun <zhengjun.xing@linux.intel.com>]

Hi Linus,

On 4/24/2021 1:18 AM, Linus Torvalds wrote:
> On Thu, Apr 22, 2021 at 11:15 PM kernel test robot
> <oliver.sang@intel.com> wrote:
>>
>> commit: e47110e90584a22e9980510b00d0dfad3a83354e ("mm/vunmap: add cond_resched() in vunmap_pmd_range")
> 
> Funky. That commit doesn't seem to have anything to do with the oops.
> 
> The oops is odd too:
> 
>> [  198.731223] WARNING: CPU: 0 PID: 1948 at mm/vmalloc.c:2247 __vunmap (kbuild/src/consumer/mm/vmalloc.c:2247 (discriminator 1))
> 
> That's the warning for an unaligned vunmap():
> 
>    2247          if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad
> address (%p)\n",
>    2248                          addr))
>    2249                  return;
> 
>> [  198.744933] Call Trace:
>> [  198.745229] free_module (kbuild/src/consumer/kernel/module.c:2251)
> 
>    2248          /* This may be empty, but that's OK */
>    2249          module_arch_freeing_init(mod);
>    2250          module_memfree(mod->init_layout.base);

We add debug code to print logs when mod->init_layout.base is NULL, 
after more than 100 times test, we find that when mod->init_layout.base 
is NULL, no align warning happened. From the descriptions of vfree, if 
@addr is NULL, no operation is performed. So when the warning happened, 
the mod->init_layout.base is not a NULL.

void vfree(const void *addr)
{
	BUG_ON(in_nmi());

	kmemleak_free(addr);

	might_sleep_if(!in_interrupt());

	if (!addr)
		return;

	__vfree(addr);
}

static void __vfree(const void *addr)
{
	if (unlikely(in_interrupt()))
		__vfree_deferred(addr);
	else
		__vunmap(addr, 1);
}


>    2251          kfree(mod->args);
> 
> That's the "module_memfree()" - the return address points to the
> return point, which is the next line.
> 
> And as far as I can tell, the only thing that assigns anything but
> NULL to that init_layout.base is
> 
>                  ptr = module_alloc(mod->init_layout.size);
> 
> which uses __vmalloc_node_range() for the allocation.
> 
> So absolutely nothing in this report makes sense to me. I suspect it's
> some odd memory corruption.
> 
> Oliver - how reliable is that bisection?
> 
> Does anybody else see what might be up?
> 
>              Linus
> _______________________________________________
> LKP mailing list -- lkp@lists.01.org
> To unsubscribe send an email to lkp-leave@lists.01.org
> 

-- 
Zhengjun Xing