On Tue, Aug 22, 2023 at 11:28:57AM -0400, Oak Zeng wrote:
> From: Nirmoy Das <[email protected]>
>
> Create a separate blitter context if a platform requires
> GGTT updates using MI_UPDATE_GTT blitter command.
>
> Subsequent patch will introduce methods to update
> GGTT using this blitter context and MI_UPDATE_GTT blitter
> command.
>
> v2: Fix a typo in comment. (Oak)
>
> Signed-off-by: Nirmoy Das <[email protected]>
> Signed-off-by: Oak Zeng <[email protected]>
> ---
> drivers/gpu/drm/i915/gt/intel_engine.h | 4 ++
> drivers/gpu/drm/i915/gt/intel_engine_cs.c | 44 +++++++++++++++++++-
> drivers/gpu/drm/i915/gt/intel_engine_types.h | 3 ++
> drivers/gpu/drm/i915/gt/intel_gtt.c | 4 ++
> drivers/gpu/drm/i915/gt/intel_gtt.h | 2 +
> 5 files changed, 56 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/i915/gt/intel_engine.h
> b/drivers/gpu/drm/i915/gt/intel_engine.h
> index b58c30ac8ef0..ee36db2fdaa7 100644
> --- a/drivers/gpu/drm/i915/gt/intel_engine.h
> +++ b/drivers/gpu/drm/i915/gt/intel_engine.h
> @@ -170,6 +170,8 @@ intel_write_status_page(struct intel_engine_cs *engine,
> int reg, u32 value)
> #define I915_GEM_HWS_SEQNO 0x40
> #define I915_GEM_HWS_SEQNO_ADDR (I915_GEM_HWS_SEQNO *
> sizeof(u32))
> #define I915_GEM_HWS_MIGRATE (0x42 * sizeof(u32))
> +#define I915_GEM_HWS_GGTT_BLIT 0x46
> +#define I915_GEM_HWS_GGTT_BLIT_ADDR (I915_GEM_HWS_GGTT_BLIT * sizeof(u32))
> #define I915_GEM_HWS_PXP 0x60
> #define I915_GEM_HWS_PXP_ADDR (I915_GEM_HWS_PXP * sizeof(u32))
> #define I915_GEM_HWS_GSC 0x62
> @@ -356,4 +358,6 @@ u64 intel_clamp_preempt_timeout_ms(struct intel_engine_cs
> *engine, u64 value);
> u64 intel_clamp_stop_timeout_ms(struct intel_engine_cs *engine, u64 value);
> u64 intel_clamp_timeslice_duration_ms(struct intel_engine_cs *engine, u64
> value);
>
> +void intel_engine_blitter_context_set_ready(struct intel_gt *gt, bool ready);
> +bool intel_engine_blitter_context_ready(struct intel_gt *gt);
> #endif /* _INTEL_RINGBUFFER_H_ */
> diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> index dfb69fc977a0..d8c492a507a4 100644
> --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> @@ -27,6 +27,7 @@
> #include "intel_gt_mcr.h"
> #include "intel_gt_pm.h"
> #include "intel_gt_requests.h"
> +#include "intel_gtt.h"
> #include "intel_lrc.h"
> #include "intel_lrc_reg.h"
> #include "intel_reset.h"
> @@ -1419,6 +1420,34 @@ void intel_engine_destroy_pinned_context(struct
> intel_context *ce)
> intel_context_put(ce);
> }
>
> +void intel_engine_blitter_context_set_ready(struct intel_gt *gt, bool ready)
> +{
> + struct intel_engine_cs *engine = gt->engine[BCS0];
> +
> + if (engine && engine->blitter_context)
> + atomic_set(&engine->blitter_context_ready, ready ? 1 : 0);
> +}
> +
> +bool intel_engine_blitter_context_ready(struct intel_gt *gt)
> +{
> + struct intel_engine_cs *engine = gt->engine[BCS0];
> +
> + if (engine)
> + return atomic_read(&engine->blitter_context_ready) == 1;
> +
> + return false;
> +}
> +
> +static struct intel_context *
> +create_ggtt_blitter_context(struct intel_engine_cs *engine)
> +{
> + static struct lock_class_key kernel;
> +
> + /* MI_UPDATE_GTT can insert up to 512 PTE entries so get a bigger ring
> */
> + return intel_engine_create_pinned_context(engine, engine->gt->vm,
> SZ_512K,
> + I915_GEM_HWS_GGTT_BLIT_ADDR,
> + &kernel,
> "ggtt_blitter_context");
> +}
> static struct intel_context *
> create_kernel_context(struct intel_engine_cs *engine)
> {
> @@ -1442,7 +1471,7 @@ create_kernel_context(struct intel_engine_cs *engine)
> */
> static int engine_init_common(struct intel_engine_cs *engine)
> {
> - struct intel_context *ce;
> + struct intel_context *ce, *bce = NULL;
> int ret;
>
> engine->set_default_submission(engine);
> @@ -1458,6 +1487,15 @@ static int engine_init_common(struct intel_engine_cs
> *engine)
> ce = create_kernel_context(engine);
> if (IS_ERR(ce))
> return PTR_ERR(ce);
> + /*
> + * Create a separate pinned context for GGTT update using blitter
> + * if a platform require such service.
> + */
> + if (i915_ggtt_require_blitter(engine->i915) && engine->id == BCS0) {
> + bce = create_ggtt_blitter_context(engine);
> + if (IS_ERR(bce))
> + return PTR_ERR(bce);
> + }
>
> ret = measure_breadcrumb_dw(ce);
> if (ret < 0)
> @@ -1465,6 +1503,7 @@ static int engine_init_common(struct intel_engine_cs
> *engine)
>
> engine->emit_fini_breadcrumb_dw = ret;
> engine->kernel_context = ce;
> + engine->blitter_context = bce;
>
> return 0;
>
> @@ -1537,6 +1576,9 @@ void intel_engine_cleanup_common(struct intel_engine_cs
> *engine)
>
> if (engine->kernel_context)
> intel_engine_destroy_pinned_context(engine->kernel_context);
> + if (engine->blitter_context)
> + intel_engine_destroy_pinned_context(engine->blitter_context);
> +
>
> GEM_BUG_ON(!llist_empty(&engine->barrier_tasks));
> cleanup_status_page(engine);
> diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h
> b/drivers/gpu/drm/i915/gt/intel_engine_types.h
> index e99a6fa03d45..62095c0d8783 100644
> --- a/drivers/gpu/drm/i915/gt/intel_engine_types.h
> +++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h
> @@ -415,6 +415,9 @@ struct intel_engine_cs {
> struct llist_head barrier_tasks;
>
> struct intel_context *kernel_context; /* pinned */
> + struct intel_context *blitter_context; /* pinned, only for BCS0 */
> + /* mark the blitter engine's availability status */
> + atomic_t blitter_context_ready;
>
> /**
> * pinned_contexts_list: List of pinned contexts. This list is only
> diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c
> b/drivers/gpu/drm/i915/gt/intel_gtt.c
> index 13944a14ea2d..9c77c97670fe 100644
> --- a/drivers/gpu/drm/i915/gt/intel_gtt.c
> +++ b/drivers/gpu/drm/i915/gt/intel_gtt.c
> @@ -21,6 +21,10 @@
> #include "intel_gt_regs.h"
> #include "intel_gtt.h"
>
> +bool i915_ggtt_require_blitter(struct drm_i915_private *i915)
> +{
> + return IS_METEORLAKE(i915);
Drive-by comment: this workaround is tied to the Xe_LPM+ media IP, not
to the MTL platform. Other platforms that re-use Xe_LPM+ IP will also
be affected, whereas MTL platforms that lack media, or integrate a
different media chiplet will not be affected. So the condition here
should be:
/* Wa_13010847436 */
return MEDIA_VER_FULL(i915) == IP_VER(13, 0);
But does this even belong in this patch? It sounds like patch #3 of the
series is where you intended to hook up this programming to the specific
workaround.
Matt
> +}
>
> static bool intel_ggtt_update_needs_vtd_wa(struct drm_i915_private *i915)
> {
> diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.h
> b/drivers/gpu/drm/i915/gt/intel_gtt.h
> index 4d6296cdbcfd..9710eb031fb2 100644
> --- a/drivers/gpu/drm/i915/gt/intel_gtt.h
> +++ b/drivers/gpu/drm/i915/gt/intel_gtt.h
> @@ -688,4 +688,6 @@ static inline struct sgt_dma {
> return (struct sgt_dma){ sg, addr, addr + sg_dma_len(sg) };
> }
>
> +bool i915_ggtt_require_blitter(struct drm_i915_private *i915);
> +
> #endif
> --
> 2.26.3
>
--
Matt Roper
Graphics Software Engineer
Linux GPU Platform Enablement
Intel Corporation
