Re: [Intel-gfx] [PATCH 08/26] drm/i915/guc: Add multi-lrc context registration

[Matthew Brost <matthew.brost@intel.com>]

On Fri, Oct 08, 2021 at 10:20:16AM -0700, John Harrison wrote:
> On 10/7/2021 12:50, John Harrison wrote:
> > On 10/4/2021 15:06, Matthew Brost wrote:
> > > Add multi-lrc context registration H2G. In addition a workqueue and
> > > process descriptor are setup during multi-lrc context registration as
> > > these data structures are needed for multi-lrc submission.
> > > 
> > > v2:
> > >   (John Harrison)
> > >    - Move GuC specific fields into sub-struct
> > >    - Clean up WQ defines
> > >    - Add comment explaining math to derive WQ / PD address
> > > 
> > > Signed-off-by: Matthew Brost <matthew.brost@intel.com>
> > > ---
> > >   drivers/gpu/drm/i915/gt/intel_context_types.h |  12 ++
> > >   drivers/gpu/drm/i915/gt/intel_lrc.c           |   5 +
> > >   .../gpu/drm/i915/gt/uc/abi/guc_actions_abi.h  |   1 +
> > >   drivers/gpu/drm/i915/gt/uc/intel_guc_fwif.h   |   2 -
> > >   .../gpu/drm/i915/gt/uc/intel_guc_submission.c | 114 +++++++++++++++++-
> > >   5 files changed, 131 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/gt/intel_context_types.h
> > > b/drivers/gpu/drm/i915/gt/intel_context_types.h
> > > index 76dfca57cb45..48decb5ee954 100644
> > > --- a/drivers/gpu/drm/i915/gt/intel_context_types.h
> > > +++ b/drivers/gpu/drm/i915/gt/intel_context_types.h
> > > @@ -239,6 +239,18 @@ struct intel_context {
> > >           struct intel_context *parent;
> > >           /** @number_children: number of children if parent */
> > >           u8 number_children;
> > > +        /** @guc: GuC specific members for parallel submission */
> > > +        struct {
> > > +            /** @wqi_head: head pointer in work queue */
> > > +            u16 wqi_head;
> > > +            /** @wqi_tail: tail pointer in work queue */
> > > +            u16 wqi_tail;
> PS: As per comments on previous rev, something somewhere needs to explicitly
> state what WQI means. One suggestion was to do that here, ideally with maybe
> a brief description of what the queue is, how it is used, etc. Although
> probably it would be better kept in a GuC specific file. E.g. added to
> guc_fwif.h in patch #12.
> 

I think this should just be in the main GuC kernel doc. I can include an
update to the kernel DoC in a patch at the end of the next rev of the
series. That patch doesn't necessarily have to included in the initial
merge of parallel submission if it takes a bit more time to review.

Matt 

> John.
> 
> > > +            /**
> > > +             * @parent_page: page in context state (ce->state) used
> > > +             * by parent for work queue, process descriptor
> > > +             */
> > > +            u8 parent_page;
> > > +        } guc;
> > >       } parallel;
> > >     #ifdef CONFIG_DRM_I915_SELFTEST
> > > diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c
> > > b/drivers/gpu/drm/i915/gt/intel_lrc.c
> > > index 3ef9eaf8c50e..57339d5c1fc8 100644
> > > --- a/drivers/gpu/drm/i915/gt/intel_lrc.c
> > > +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c
> > > @@ -942,6 +942,11 @@ __lrc_alloc_state(struct intel_context *ce,
> > > struct intel_engine_cs *engine)
> > >           context_size += PAGE_SIZE;
> > >       }
> > >   +    if (intel_context_is_parent(ce) &&
> > > intel_engine_uses_guc(engine)) {
> > > +        ce->parallel.guc.parent_page = context_size / PAGE_SIZE;
> > > +        context_size += PAGE_SIZE;
> > > +    }
> > > +
> > >       obj = i915_gem_object_create_lmem(engine->i915, context_size,
> > >                         I915_BO_ALLOC_PM_VOLATILE);
> > >       if (IS_ERR(obj))
> > > diff --git a/drivers/gpu/drm/i915/gt/uc/abi/guc_actions_abi.h
> > > b/drivers/gpu/drm/i915/gt/uc/abi/guc_actions_abi.h
> > > index 8ff582222aff..ba10bd374cee 100644
> > > --- a/drivers/gpu/drm/i915/gt/uc/abi/guc_actions_abi.h
> > > +++ b/drivers/gpu/drm/i915/gt/uc/abi/guc_actions_abi.h
> > > @@ -142,6 +142,7 @@ enum intel_guc_action {
> > >       INTEL_GUC_ACTION_REGISTER_COMMAND_TRANSPORT_BUFFER = 0x4505,
> > >       INTEL_GUC_ACTION_DEREGISTER_COMMAND_TRANSPORT_BUFFER = 0x4506,
> > >       INTEL_GUC_ACTION_DEREGISTER_CONTEXT_DONE = 0x4600,
> > > +    INTEL_GUC_ACTION_REGISTER_CONTEXT_MULTI_LRC = 0x4601,
> > >       INTEL_GUC_ACTION_RESET_CLIENT = 0x5507,
> > >       INTEL_GUC_ACTION_LIMIT
> > >   };
> > > diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_fwif.h
> > > b/drivers/gpu/drm/i915/gt/uc/intel_guc_fwif.h
> > > index fa4be13c8854..0eeb2a9feeed 100644
> > > --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_fwif.h
> > > +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_fwif.h
> > > @@ -52,8 +52,6 @@
> > >     #define GUC_DOORBELL_INVALID        256
> > >   -#define GUC_WQ_SIZE            (PAGE_SIZE * 2)
> > > -
> > >   /* Work queue item header definitions */
> > >   #define WQ_STATUS_ACTIVE        1
> > >   #define WQ_STATUS_SUSPENDED        2
> > > diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> > > b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> > > index 451d9ae861a6..ab6d7fc1b0b1 100644
> > > --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> > > +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> > > @@ -344,6 +344,45 @@ static inline struct i915_priolist
> > > *to_priolist(struct rb_node *rb)
> > >       return rb_entry(rb, struct i915_priolist, node);
> > >   }
> > >   +/*
> > > + * When using multi-lrc submission an extra page in the context
> > > state is
> > > + * reserved for the process descriptor and work queue.
> > > + *
> > > + * The layout of this page is below:
> > > + * 0                        guc_process_desc
> > > + * ...                        unused
> > > + * PAGE_SIZE / 2                work queue start
> > > + * ...                        work queue
> > > + * PAGE_SIZE - 1                work queue end
> > > + */
> > > +#define WQ_SIZE            (PAGE_SIZE / 2)
> > > +#define WQ_OFFSET        (PAGE_SIZE - WQ_SIZE)
> > I thought you were going with '#define PARENT_SCRATCH SIZE PAGE_SIZE'
> > and then using that everywhere else? Unless there is a fundamental
> > reason why the above must be exactly a page in size then I think the
> > size should be defined once and re-used rather than assumed in multiple
> > places (including in the description comment).
> > 
> > > +static u32 __get_process_desc_offset(struct intel_context *ce)
> > > +{
> > > +    GEM_BUG_ON(!ce->parallel.guc.parent_page);
> > > +
> > > +    return ce->parallel.guc.parent_page * PAGE_SIZE;
> > > +}
> > > +
> > > +static u32 __get_wq_offset(struct intel_context *ce)
> > > +{
> > > +    return __get_process_desc_offset(ce) + WQ_OFFSET;
> > > +}
> > > +
> > > +static struct guc_process_desc *
> > > +__get_process_desc(struct intel_context *ce)
> > > +{
> > > +    /*
> > > +     * Need to subtract LRC_STATE_OFFSET here as the
> > > +     * parallel.guc.parent_page is the offset into ce->state while
> > > +     * ce->lrc_reg_reg is ce->state + LRC_STATE_OFFSET.
> > > +     */
> > > +    return (struct guc_process_desc *)
> > > +        (ce->lrc_reg_state +
> > > +         ((__get_process_desc_offset(ce) -
> > > +           LRC_STATE_OFFSET) / sizeof(u32)));
> > > +}
> > > +
> > >   static struct guc_lrc_desc *__get_lrc_desc(struct intel_guc *guc,
> > > u32 index)
> > >   {
> > >       struct guc_lrc_desc *base = guc->lrc_desc_pool_vaddr;
> > > @@ -1365,6 +1404,30 @@ static void unpin_guc_id(struct intel_guc
> > > *guc, struct intel_context *ce)
> > >       spin_unlock_irqrestore(&guc->submission_state.lock, flags);
> > >   }
> > >   +static int __guc_action_register_multi_lrc(struct intel_guc *guc,
> > > +                       struct intel_context *ce,
> > > +                       u32 guc_id,
> > > +                       u32 offset,
> > > +                       bool loop)
> > > +{
> > > +    struct intel_context *child;
> > > +    u32 action[4 + MAX_ENGINE_INSTANCE];
> > > +    int len = 0;
> > > +
> > > +    GEM_BUG_ON(ce->parallel.number_children > MAX_ENGINE_INSTANCE);
> > > +
> > > +    action[len++] = INTEL_GUC_ACTION_REGISTER_CONTEXT_MULTI_LRC;
> > > +    action[len++] = guc_id;
> > > +    action[len++] = ce->parallel.number_children + 1;
> > > +    action[len++] = offset;
> > > +    for_each_child(ce, child) {
> > > +        offset += sizeof(struct guc_lrc_desc);
> > > +        action[len++] = offset;
> > > +    }
> > > +
> > > +    return guc_submission_send_busy_loop(guc, action, len, 0, loop);
> > > +}
> > > +
> > >   static int __guc_action_register_context(struct intel_guc *guc,
> > >                        u32 guc_id,
> > >                        u32 offset,
> > > @@ -1387,9 +1450,15 @@ static int register_context(struct
> > > intel_context *ce, bool loop)
> > >           ce->guc_id.id * sizeof(struct guc_lrc_desc);
> > >       int ret;
> > >   +    GEM_BUG_ON(intel_context_is_child(ce));
> > >       trace_intel_context_register(ce);
> > >   -    ret = __guc_action_register_context(guc, ce->guc_id.id,
> > > offset, loop);
> > > +    if (intel_context_is_parent(ce))
> > > +        ret = __guc_action_register_multi_lrc(guc, ce, ce->guc_id.id,
> > > +                              offset, loop);
> > > +    else
> > > +        ret = __guc_action_register_context(guc, ce->guc_id.id, offset,
> > > +                            loop);
> > >       if (likely(!ret)) {
> > >           unsigned long flags;
> > >   @@ -1418,6 +1487,7 @@ static int deregister_context(struct
> > > intel_context *ce, u32 guc_id)
> > >   {
> > >       struct intel_guc *guc = ce_to_guc(ce);
> > >   +    GEM_BUG_ON(intel_context_is_child(ce));
> > >       trace_intel_context_deregister(ce);
> > >         return __guc_action_deregister_context(guc, guc_id);
> > > @@ -1445,6 +1515,7 @@ static int guc_lrc_desc_pin(struct
> > > intel_context *ce, bool loop)
> > >       struct guc_lrc_desc *desc;
> > >       bool context_registered;
> > >       intel_wakeref_t wakeref;
> > > +    struct intel_context *child;
> > >       int ret = 0;
> > >         GEM_BUG_ON(!engine->mask);
> > > @@ -1470,6 +1541,41 @@ static int guc_lrc_desc_pin(struct
> > > intel_context *ce, bool loop)
> > >       desc->context_flags = CONTEXT_REGISTRATION_FLAG_KMD;
> > >       guc_context_policy_init(engine, desc);
> > >   +    /*
> > > +     * Context is a parent, we need to register a process descriptor
> > > +     * describing a work queue and register all child contexts.
> > > +     */
> > This was now meant to say 'If the context is a parent...'?
> > 
> > John.
> > 
> > > +    if (intel_context_is_parent(ce)) {
> > > +        struct guc_process_desc *pdesc;
> > > +
> > > +        ce->parallel.guc.wqi_tail = 0;
> > > +        ce->parallel.guc.wqi_head = 0;
> > > +
> > > +        desc->process_desc = i915_ggtt_offset(ce->state) +
> > > +            __get_process_desc_offset(ce);
> > > +        desc->wq_addr = i915_ggtt_offset(ce->state) +
> > > +            __get_wq_offset(ce);
> > > +        desc->wq_size = WQ_SIZE;
> > > +
> > > +        pdesc = __get_process_desc(ce);
> > > +        memset(pdesc, 0, sizeof(*(pdesc)));
> > > +        pdesc->stage_id = ce->guc_id.id;
> > > +        pdesc->wq_base_addr = desc->wq_addr;
> > > +        pdesc->wq_size_bytes = desc->wq_size;
> > > +        pdesc->wq_status = WQ_STATUS_ACTIVE;
> > > +
> > > +        for_each_child(ce, child) {
> > > +            desc = __get_lrc_desc(guc, child->guc_id.id);
> > > +
> > > +            desc->engine_class =
> > > +                engine_class_to_guc_class(engine->class);
> > > +            desc->hw_context_desc = child->lrc.lrca;
> > > +            desc->priority = ce->guc_state.prio;
> > > +            desc->context_flags = CONTEXT_REGISTRATION_FLAG_KMD;
> > > +            guc_context_policy_init(engine, desc);
> > > +        }
> > > +    }
> > > +
> > >       /*
> > >        * The context_lookup xarray is used to determine if the hardware
> > >        * context is currently registered. There are two cases in
> > > which it
> > > @@ -2804,6 +2910,12 @@ g2h_context_lookup(struct intel_guc *guc, u32
> > > desc_idx)
> > >           return NULL;
> > >       }
> > >   +    if (unlikely(intel_context_is_child(ce))) {
> > > +        drm_err(&guc_to_gt(guc)->i915->drm,
> > > +            "Context is child, desc_idx %u", desc_idx);
> > > +        return NULL;
> > > +    }
> > > +
> > >       return ce;
> > >   }
> > 
>