Re: [PATCH 2/2] platform/x86/intel/ifs: Initialize AMX state for the scan test

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi,

On 4/30/24 11:25 PM, Chang S. Bae wrote:
> The scan test does not start when the AMX state remains active and is not
> re-initialized. With the extension of kernel_fpu_begin_mask(), the driver
> code can now initialize the state properly.
> 
> Introduce custom FPU handling wrappers to ensure compliant with the
> established FPU API semantics, as kernel_fpu_begin() exclusively sets
> legacy states. This follows the EFI case from commit b0dc553cfc9d
> ("x86/fpu: Make the EFI FPU calling convention explicit").
> 
> Then, use these wrappers to surround the MSR_ACTIVATE_SCAN write to
> minimize the critical section. To prevent unnecessary delays, invoke
> ifs_fpu_begin() before entering the rendezvous loop.
> 
> Signed-off-by: Chang S. Bae <chang.seok.bae@xxxxxxxxx>
> Reviewed-by: Jithu Joseph <jithu.joseph@xxxxxxxxx>
> Tested-by: Jithu Joseph <jithu.joseph@xxxxxxxxx>

Thanks, the patch looks good to me.

I believe this is best merged through the tip/x86 tree together
with patch 1/2 :

Acked-by: Hans de Goede <hdegoede@xxxxxxxxxx>

I have checked and this should not cause any conflicts with
the ifs changes current in platform-drivers-x86/for-next.

Regards,

Hans




> ---
>  drivers/platform/x86/intel/ifs/ifs.h     | 14 ++++++++++++++
>  drivers/platform/x86/intel/ifs/runtest.c |  6 ++++++
>  2 files changed, 20 insertions(+)
> 
> diff --git a/drivers/platform/x86/intel/ifs/ifs.h b/drivers/platform/x86/intel/ifs/ifs.h
> index 56b9f3e3cf76..71d8b50854b2 100644
> --- a/drivers/platform/x86/intel/ifs/ifs.h
> +++ b/drivers/platform/x86/intel/ifs/ifs.h
> @@ -325,4 +325,18 @@ int do_core_test(int cpu, struct device *dev);
>  extern struct attribute *plat_ifs_attrs[];
>  extern struct attribute *plat_ifs_array_attrs[];
>  
> +static inline void ifs_fpu_begin(void)
> +{
> +	/*
> +	 * The AMX state must be initialized prior to executing In-Field
> +	 * Scan tests, according to Intel SDM.
> +	 */
> +	kernel_fpu_begin_mask(KFPU_AMX);
> +}
> +
> +static inline void ifs_fpu_end(void)
> +{
> +	kernel_fpu_end();
> +}
> +
>  #endif
> diff --git a/drivers/platform/x86/intel/ifs/runtest.c b/drivers/platform/x86/intel/ifs/runtest.c
> index 95b4b71fab53..a35eac7c0b44 100644
> --- a/drivers/platform/x86/intel/ifs/runtest.c
> +++ b/drivers/platform/x86/intel/ifs/runtest.c
> @@ -188,6 +188,9 @@ static int doscan(void *data)
>  	/* Only the first logical CPU on a core reports result */
>  	first = cpumask_first(cpu_smt_mask(cpu));
>  
> +	/* Prepare FPU state before entering the rendezvous loop*/
> +	ifs_fpu_begin();
> +
>  	wait_for_sibling_cpu(&scan_cpus_in, NSEC_PER_SEC);
>  
>  	/*
> @@ -199,6 +202,9 @@ static int doscan(void *data)
>  	 * are processed in a single pass) before it retires.
>  	 */
>  	wrmsrl(MSR_ACTIVATE_SCAN, params->activate->data);
> +
> +	ifs_fpu_end();
> +
>  	rdmsrl(MSR_SCAN_STATUS, status.data);
>  
>  	trace_ifs_status(ifsd->cur_batch, start, stop, status.data);





[Index of Archives]     [Linux Kernel Development]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux