RE: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS

[Joshi, Mukul]

[AMD Official Use Only]



> -----Original Message-----
> From: Ghannam, Yazen <Yazen.Ghannam@xxxxxxx>
> Sent: Thursday, September 23, 2021 10:29 AM
> To: Joshi, Mukul <Mukul.Joshi@xxxxxxx>
> Cc: linux-edac@xxxxxxxxxxxxxxx; x86@xxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx;
> bp@xxxxxxxxx; mingo@xxxxxxxxxx; mchehab@xxxxxxxxxx; amd-
> gfx@xxxxxxxxxxxxxxxxxxxxx
> Subject: Re: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran
> RAS
> 
> On Wed, Sep 22, 2021 at 03:36:20PM -0400, Mukul Joshi wrote:
> > On Aldebaran, GPU driver will handle bad page retirement even though
> > UMC is host managed. As a result, register a bad page retirement
> > handler on the mce notifier chain to retire bad pages on Aldebaran.
> >
> 
> I think this should state that the driver will do page retirement for GPU-managed
> memory. As written, it implies that the driver do page retirement in general for
> the system.
> 
ACK. I will update the description.
> ...
> 
> > +
> > +static int amdgpu_bad_page_notifier(struct notifier_block *nb,
> > +				    unsigned long val, void *data) {
> > +	struct mce *m = (struct mce *)data;
> > +	struct amdgpu_device *adev = NULL;
> > +	uint32_t gpu_id = 0;
> > +	uint32_t umc_inst = 0;
> > +	uint32_t ch_inst, channel_index = 0;
> > +	struct ras_err_data err_data = {0, 0, 0, NULL};
> > +	struct eeprom_table_record err_rec;
> > +	uint64_t retired_page;
> > +
> > +	/*
> > +	 * If the error was generated in UMC_V2, which belongs to GPU UMCs,
> > +	 * and error occurred in DramECC (Extended error code = 0) then only
> > +	 * process the error, else bail out.
> > +	 */
> > +	if (!m || !((smca_get_bank_type(m->bank) == SMCA_UMC_V2) &&
> > +		    (XEC(m->status, 0x1f) == 0x0)))
> 
> The MCA_STATUS[ErrorCodeExt] field is bits [21:16], so the mask should be
> 0x3f.

Ack. Thanks for catching this.
> 
> > +		return NOTIFY_DONE;
> > +
> > +	/*
> > +	 * If it is correctable error, return.
> > +	 */
> > +	if (mce_is_correctable(m))
> > +		return NOTIFY_OK;
> 
> Shouldn't this be "NOTIFY_DONE" if "don't care" about this error?

The thinking is we want to stop calling further consumers since it's a correctable error in GPU UMC and we are not taking any action about the correctable errors.

Thanks,
Mukul

> 
> Thanks,
> Yazen