Re: [PATCH] x86/PAT: Report PAT on CPUs that support PAT without MTRR

[Juergen Gross]

On 12.07.22 20:20, Chuck Zmudzinski wrote:
> The commit 99c13b8c8896d7bcb92753bf
> ("x86/mm/pat: Don't report PAT on CPUs that don't support it")
> incorrectly failed to account for the case in init_cache_modes() when
> CPUs do support PAT and falsely reported PAT to be disabled when in
> fact PAT is enabled. In some environments, notably in Xen PV domains,
> MTRR is disabled but PAT is still enabled, and that is the case
> that the aforementioned commit failed to account for.
>
> As an unfortunate consequnce, the pat_enabled() function currently does
> not correctly report that PAT is enabled in such environments. The fix
> is implemented in init_cache_modes() by setting pat_bp_enabled to true
> in init_cache_modes() for the case that commit 99c13b8c8896d7bcb92753bf
> ("x86/mm/pat: Don't report PAT on CPUs that don't support it") failed
> to account for.
>
> This patch fixes a regression that some users are experiencing with
> Linux as a Xen Dom0 driving particular Intel graphics devices by
> correctly reporting to the Intel i915 driver that PAT is enabled where
> previously it was falsely reporting that PAT is disabled.
>
> Fixes: 99c13b8c8896d7bcb92753bf ("x86/mm/pat: Don't report PAT on CPUs that don't support it")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Chuck Zmudzinski <brchuckz@xxxxxxx>
> ---
> Reminder: This patch is a regression fix that is needed on stable
> versions 5.17 and later.
>
>   arch/x86/mm/pat/memtype.c | 12 ++++++++++++
>   1 file changed, 12 insertions(+)
>
> diff --git a/arch/x86/mm/pat/memtype.c b/arch/x86/mm/pat/memtype.c
> index d5ef64ddd35e..0f2417bd1b40 100644
> --- a/arch/x86/mm/pat/memtype.c
> +++ b/arch/x86/mm/pat/memtype.c
> @@ -315,6 +315,18 @@ void init_cache_modes(void)
>   		      PAT(4, WB) | PAT(5, WT) | PAT(6, UC_MINUS) | PAT(7, UC);
>   	}
>   
> +	else if (!pat_bp_enabled) {

Please put the "else if {" into the same line as the "}" above.

> +	/*
> +	 * In some environments, specifically Xen PV, PAT
> +	 * initialization is skipped because MTRRs are disabled even
> +	 * though PAT is available. In such environments, set PAT to
> +	 * enabled to correctly indicate to callers of pat_enabled()
> +	 * that CPU support for PAT is available.
> +	 */
> +	pat_bp_enabled = true;
> +	pr_info("x86/PAT: PAT enabled by init_cache_modes\n");

Wrong indentation.

> +	}
> +
>   	__init_cache_modes(pat);
>   }
>   

Any reason you didn't fix the "nopat" issue Jan mentioned?

I asked you twice to add this fix.


Juergen

Attachment: OpenPGP_0xB0DE9DD628BF132F.asc
Description: OpenPGP public key

Attachment: OpenPGP_signature
Description: OpenPGP digital signature