[Intel-gfx] drm/i915/slpc: Update freq min/max softlimits
Kamble, Sagar A
sagar.a.kamble at intel.com
Sun Aug 21 16:09:17 UTC 2016
On 8/21/2016 2:09 PM, Chris Wilson wrote:
> On Sun, Aug 21, 2016 at 11:39:22AM +0530, Kamble, Sagar A wrote:
>>
>> On 8/20/2016 1:32 PM, Chris Wilson wrote:
>>> On Sat, Aug 20, 2016 at 10:39:25AM +0530, Sagar Arun Kamble wrote:
>>>> + obj = dev_priv->guc.slpc.vma->obj;
>>>> + if (obj) {
>>> OOPS.
>> Fixed in next series.
>>>> + intel_slpc_query_task_state(dev_priv);
>>>> +
>>>> + page = i915_gem_object_get_page(obj, 0);
>>>> + if (page)
>>>> + pv = kmap_atomic(page);
>>>> + }
>>>> +
>>>> + if (pv) {
>>>> + data = *(struct slpc_shared_data *) pv;
>>>> + kunmap_atomic(pv);
>>> Can kmap_atomic return zero?
>> Fixed in next series.
>>>> +
>>>> + /*
>>>> + * TODO: Define separate variables for slice and unslice
>>>> + * frequencies for driver state variable.
>>>> + */
>>>> + dev_priv->rps.max_freq_softlimit =
>>>> + data.task_state_data.freq_unslice_max;
>>>> + dev_priv->rps.min_freq_softlimit =
>>>> + data.task_state_data.freq_unslice_min;
>>> These are user values, you do not get to arbitrarily rewrite them.
>>>
>>> You control dev_priv->rps.[min|max]_freq.
>> With SLPC, GuC firmware SLPC S/W requested frequency be operated in
>> the softlimits analogous to
>> Host softlimits. Limits might be different with SLPC and can be
>> controlled through regular interfaces.
>> dev_priv->rps.[min|max]_freq are HW Min/Max.
> Exactly. The soft limits are *only* set by the user. They are not to
> modified by the driver. (The caveat would be a dynamic update of the hw
> range, but that too should never be required.)
> -Chris
This initialization is similar to following from intel_init_gt_powersave
dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
I assume min_freq is hw min(RPn). With SLPC, min_freq(RPn) will not be
requested.
SLPC operating range today is (>Rpe, Rp0) so I wanted user to know
the min_softlimit being initialized by SLPC by default.
>
More information about the Intel-gfx
mailing list