Re: [PATCH] hpwd watchdog mark page executable

[Maxim Uvarov <maxim.uvarov@oracle.com>]

On 12/20/2011 04:50 PM, Andrew Morton wrote:
> On Mon,  5 Dec 2011 16:25:39 -0800
> Maxim Uvarov<maxim.uvarov@oracle.com>  wrote:
>
>> Mark hpwdt watchdog pages executable to prevent failing:
>> BUG: unable to handle kernel paging request at c00f0000
>>   IP: [<c00f0000>] 0xc00effff
>>   *pdpt = 0000000000b7c001 *pde = 0000000000cf5067 *pte = 80000000000f0163
>>   Oops: 0011 [#1] SMP
>>
>> ...
>>
>> --- a/drivers/watchdog/hpwdt.c
>> +++ b/drivers/watchdog/hpwdt.c
>> @@ -335,6 +335,8 @@ static int __devinit detect_cru_service(void)
>>   	if (p == NULL)
>>   		return -ENOMEM;
>>
>> +	set_memory_x((unsigned long)p&  PAGE_MASK, ROM_SIZE>>  PAGE_SHIFT);
>> +
>>   	for (q = p; q<  p + ROM_SIZE; q += 16) {
>>   		rc = bios32_present(q);
>>   		if (!rc)
>
> Odd.  Either nobody else is using this driver, or there's something
> special about your setup.  Are you able to explain this?
>

Code drivers/watchdog/hpwdt.c is ifdefed on 2 branches for 32 and 64 bit 
accordingly. For 64 bit address for watchdog handler is used from 
smbios, for 32 bit - from ioremap(0x000F0000, ). Accordingly 64 kernel 
unsets  _PAGE_NX flag to pages. I.e. calls set_memory_x(). But 32 bit 
did not do so. It worked on 2.6.32 and since 3.0 looks like additional 
page protection code was added. I can't find the right commit which 
changed it right now.

> Also, do you believe that this fix should be backported into earlier
> stable kernels and if so, why?
>
Andrew, I found this problem in 3.0 than checked that it exist in 
linux-next when it was 3.1+. Because it's panic on module loading,  yes 
this is critical bug and it's needs to be back-ported. (Actually this 
driver was not changed so simple cherry-pick one line patch will work.).

Maxim.
  > Thanks.