Random memory corruption may occur due to incorrent tlb flushes

Random memory corruption may occur due to incorrent tlb flushes

[Santosh Sivaraj]

Hi Greg/Sasha,

The commit a46cc7a90fd (powerpc/mm/radix: Improve
TLB/PWC flushes) picked up in 4.14 release has the potential to cause random
memory corruption. This was fixed in 5.5 by the following patches.

12e4d53f3f powerpc/mmu_gather: enable RCU_TABLE_FREE even for !SMP case
0ed1325967 mm/mmu_gather: invalidate TLB correctly on batch allocation failure and flush
0758cd8304 asm-generic/tlb: avoid potential double flush

It's a bit tricky to backport to 4.14 stable (though I have a backport to 4.19
stable, which I will post shortly). If you think it's important to fix this in
4.14, it would easier to revert the above mentioned commit (a46cc7a90fd). 

Please let me know your thoughts.

Thanks,
Santosh

Re: Random memory corruption may occur due to incorrent tlb flushes

[Greg KH]

On Tue, Feb 18, 2020 at 01:41:07PM +0530, Santosh Sivaraj wrote:
> Hi Greg/Sasha,
> 
> The commit a46cc7a90fd (powerpc/mm/radix: Improve
> TLB/PWC flushes) picked up in 4.14 release has the potential to cause random
> memory corruption. This was fixed in 5.5 by the following patches.
> 
> 12e4d53f3f powerpc/mmu_gather: enable RCU_TABLE_FREE even for !SMP case
> 0ed1325967 mm/mmu_gather: invalidate TLB correctly on batch allocation failure and flush
> 0758cd8304 asm-generic/tlb: avoid potential double flush
> 
> It's a bit tricky to backport to 4.14 stable (though I have a backport to 4.19
> stable, which I will post shortly). If you think it's important to fix this in
> 4.14, it would easier to revert the above mentioned commit (a46cc7a90fd). 
> 
> Please let me know your thoughts.

A revert is probably best, can you send it?

thanks,

greg k-h

Re: Random memory corruption may occur due to incorrent tlb flushes

[Sasha Levin]

On Tue, Feb 18, 2020 at 09:49:01AM +0100, Greg KH wrote:
>On Tue, Feb 18, 2020 at 01:41:07PM +0530, Santosh Sivaraj wrote:
>> Hi Greg/Sasha,
>>
>> The commit a46cc7a90fd (powerpc/mm/radix: Improve
>> TLB/PWC flushes) picked up in 4.14 release has the potential to cause random
>> memory corruption. This was fixed in 5.5 by the following patches.
>>
>> 12e4d53f3f powerpc/mmu_gather: enable RCU_TABLE_FREE even for !SMP case
>> 0ed1325967 mm/mmu_gather: invalidate TLB correctly on batch allocation failure and flush
>> 0758cd8304 asm-generic/tlb: avoid potential double flush
>>
>> It's a bit tricky to backport to 4.14 stable (though I have a backport to 4.19
>> stable, which I will post shortly). If you think it's important to fix this in
>> 4.14, it would easier to revert the above mentioned commit (a46cc7a90fd).
>>
>> Please let me know your thoughts.
>
>A revert is probably best, can you send it?

This is a bit tricky because a46cc7a90fd wasn't picked up by as, but
rather is part of the 4.14 kernel. Please make sure you Cc the revert to
the PowerPC mm folks so they could give it a careful review, as it's
much more complex than just reverting a stable commit we queued up.

-- 
Thanks,
Sasha