Re: [PATCH] x86/mm/pti: in pti_clone_pgtable() don't increase addr by PUD_SIZE

[Song Liu <songliubraving@fb.com>]

> On Aug 20, 2019, at 3:00 AM, Peter Zijlstra <peterz@infradead.org> wrote:
> 
> On Tue, Aug 20, 2019 at 12:51:28AM -0700, Song Liu wrote:
>> pti_clone_pgtable() increases addr by PUD_SIZE for pud_none(*pud) case.
>> This is not accurate because addr may not be PUD_SIZE aligned.
>> 
>> In our x86_64 kernel, pti_clone_pgtable() fails to clone 7 PMDs because
>> of this issuse, including PMD for the irq entry table. For a memcache
>> like workload, this introduces about 4.5x more iTLB-load and about 2.5x
>> more iTLB-load-misses on a Skylake CPU.
>> 
>> This patch fixes this issue by adding PMD_SIZE to addr for pud_none()
>> case.
> 
>> diff --git a/arch/x86/mm/pti.c b/arch/x86/mm/pti.c
>> index b196524759ec..5a67c3015f59 100644
>> --- a/arch/x86/mm/pti.c
>> +++ b/arch/x86/mm/pti.c
>> @@ -330,7 +330,7 @@ pti_clone_pgtable(unsigned long start, unsigned long end,
>> 
>> 		pud = pud_offset(p4d, addr);
>> 		if (pud_none(*pud)) {
>> -			addr += PUD_SIZE;
>> +			addr += PMD_SIZE;
>> 			continue;
>> 		}
> 
> I'm thinking you're right in that there's a bug here, but I'm also
> thinking your patch is both incomplete and broken.
> 
> What that code wants to do is skip to the end of the pud, a pmd_size
> increase will not do that. And right below this, there's a second
> instance of this exact pattern.
> 
> Did I get the below right?

Yes, your are right. 

Thanks,
Song

> 
> ---
> diff --git a/arch/x86/mm/pti.c b/arch/x86/mm/pti.c
> index b196524759ec..32b20b3cb227 100644
> --- a/arch/x86/mm/pti.c
> +++ b/arch/x86/mm/pti.c
> @@ -330,12 +330,14 @@ pti_clone_pgtable(unsigned long start, unsigned long end,
> 
> 		pud = pud_offset(p4d, addr);
> 		if (pud_none(*pud)) {
> +			addr &= PUD_MASK;
> 			addr += PUD_SIZE;
> 			continue;
> 		}
> 
> 		pmd = pmd_offset(pud, addr);
> 		if (pmd_none(*pmd)) {
> +			addr &= PMD_MASK;
> 			addr += PMD_SIZE;
> 			continue;
> 		}