Re: [PATCH v2 3/5] drm/xe/xe_ras: Add support to query error counter for CRI

[Raag Jadav <raag.jadav@intel.com>]

On Mon, Apr 20, 2026 at 12:03:06PM +0530, Tauro, Riana wrote:
> On 4/13/2026 2:49 PM, Raag Jadav wrote:
> > On Mon, Apr 06, 2026 at 08:24:42PM +0530, Riana Tauro wrote:
> > > Add support to get error counter value for CRI.
> > > 
> > > When userspace queries a drm_ras error counter, fetch the
> > > latest counter value from system controller.

...

> > > +static int get_error_counter(struct xe_device *xe, struct xe_ras_error_class *error_class,
> > > +			     u32 *value)
> > > +{
> > > +	struct xe_ras_get_counter_response response = {0};
> > > +	struct xe_ras_get_counter_request request = {0};
> > > +	struct xe_sysctrl_mailbox_command command = {0};
> > > +	size_t rlen;
> > > +	int ret;
> > > +
> > > +	request.error_class = *error_class;
> > > +
> > > +	prepare_sysctrl_command(&command, XE_SYSCTRL_CMD_GET_COUNTER, &request, sizeof(request),
> > > +				&response, sizeof(response));
> > > +
> > > +	ret = xe_sysctrl_send_command(&xe->sc, &command, &rlen);
> > > +	if (ret) {
> > > +		xe_err(xe, "[RAS]: Sysctrl error ret %d\n", ret);
> > This gives the impression of RAS error, but is it really?
> 
> This command can be used from different components. This will help
> differentiate
> the file.

File is not much useful for debugging, what's probably more useful is
distinguishable error message.

	xe_err(xe, "sysctrl: failed to get counter %d\n", ret);

or perhaps

	xe_err(xe, "sysctrl: command %x failed with %d\n", cmd, ret);

Also, I think [RAS] tag should be exclusive to hardware errors but I'll
leave it to you all.

> > > +		return ret;
> > > +	}
> > > +
> > > +	if (rlen != sizeof(response)) {
> > > +		xe_err(xe, "[RAS]: Sysctrl response size mismatch. Expected %zu, got %zu\n",
> > Ditto.
> > 
> > > +		       sizeof(response), rlen);
> > > +		return -EINVAL;
> > Is this propagated back to the user? If yes, is this the correct error
> > code for the scenario?
> 
> Yes the error code will be propagated back to the user.  Any suggestions?
> EIO for system controller errors?

Yep, we tried to query system controller but it had different (or perhaps
no) idea what it was.

PS: Probably another reason to not have [RAS] in error message ;)

Raag

> > > +	}
> > > +
> > > +	*value = response.counter_value;
> > > +
> > > +	return 0;
> > > +}