[Intel-gfx] [PATCH v3] drm/i915: Add Guc/HuC firmware details to error state

[Michal Wajdeczko]

On Fri, 20 Oct 2017 14:21:16 +0200, Chris Wilson  
<chris at chris-wilson.co.uk> wrote:

> Quoting Michal Wajdeczko (2017-10-20 13:14:45)
>> On Fri, 20 Oct 2017 13:44:04 +0200, Chris Wilson
>> <chris at chris-wilson.co.uk> wrote:
>>
>> > Quoting Joonas Lahtinen (2017-10-20 10:41:57)
>> >> On Thu, 2017-10-19 at 20:23 +0000, Michal Wajdeczko wrote:
>> >> > @@ -774,6 +793,11 @@ int i915_error_state_to_str(struct
>> >> drm_i915_error_state_buf *m,
>> >> >       err_print_capabilities(m, &error->device_info);
>> >> >       err_print_params(m, &error->params);
>> >> >
>> >> > +     if (error->device_info.has_guc) {
>> >> > +             intel_uc_fw_dump(&error->guc_fw, &p);
>> >> > +             intel_uc_fw_dump(&error->huc_fw, &p);
>> >> > +     }
>> >>
>> >> I might use "error->{g,h}uc_fw.path" as the condition, for both
>> >> individually. We will have DMC here in the future, too.
>> >
>> > That's the type of predicate I was looking for. If we can tell when  
>> the
>> > fw has been loaded just by looking at the uc_fw struct, please do so.
>>
>> If you want to include uc fw info only when fw was loaded at capture
>> moment,
>> then we can look directly at its "load_status" field:
>>
>>         if (error->guc_fw.load_status == INTEL_UC_FIRMWARE_SUCCESS)
>>
>> But what if fw was loaded earlier, before error capture is started?
>> Don't we want to capture whole driver configuration?
>
> The goal is to know what the driver/hw was doing at the time of the
> capture (which we presume is still identical to the hang).
>
> At this point, we are just asking ourselves what is the most unambiguous
> way of printing valid data. We already show has_guc in the output, so it
> boils down to what was the uc_fw state. It doesn't need to be
> load_status == SUCCESS, as we should show a failed attempt to upload the
> fw as well.
>
> So intel_uc_fw.size?

But then in case of zero size fw you will not get info about load failure  
;)

What about leaving here "if (error->device_info.has_guc)" as main guard
for both GuC and HuC fw dumps (note that upcoming series make it explicit
that HuC is always in pair with GuC) but then also add "if (uc_fw->path)"
condition to the uc fw pretty printer to stop printing fw details if no
path was specified (all fields will be void anyway). Then we may get:

has_guc: yes
...
GuC firmware: (null)
HuC firmware: (null)


Michal