Message ID | 1461177750-20187-11-git-send-email-chris@chris-wilson.co.uk (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On ke, 2016-04-20 at 19:42 +0100, Chris Wilson wrote: > The code to switch_mm() is already handled by i915_switch_context(), the > only difference required to setup the aliasing ppgtt is that we need to > emit te switch_mm() on the first context, i.e. when transitioning from > engine->last_context == NULL. This allows us to defer the > initialisation of the GPU from early device initialisation to first use, > which should marginally speed up both. The caveat is that we then defer > the context initialisation until first use - i.e. we cannot assume that > the GPU engines are initialised. For example, this means that power > contexts for rc6 (Ironlake) need to explicitly loaded, as they are. > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Mika Kuoppala <mika.kuoppala@intel.com> Some comments below. Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c > index aafcb4942acf..14c9b29294c5 100644 > --- a/drivers/gpu/drm/i915/i915_gem.c > +++ b/drivers/gpu/drm/i915/i915_gem.c > @@ -4910,34 +4910,6 @@ i915_gem_init_hw(struct drm_device *dev) > if (ret) > goto out; This whole if (ret) goto out can be removed as out is right after the removed code block. > > - /* Now it is safe to go back round and do everything else: */ > - for_each_engine(engine, dev_priv) { > - struct drm_i915_gem_request *req; > - > - req = i915_gem_request_alloc(engine, NULL); > - if (IS_ERR(req)) { > - ret = PTR_ERR(req); > - break; > - } > - > - ret = i915_ppgtt_init_ring(req); > - if (ret) > - goto err_request; > - > - ret = i915_gem_context_enable(req); > - if (ret) > - goto err_request; > - > -err_request: > - i915_add_request_no_flush(req); > - if (ret) { > - DRM_ERROR("Failed to enable %s, error=%d\n", > - engine->name, ret); > - i915_gem_cleanup_engines(dev); > - break; > - } > - } > - > out: > intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); > return ret; > diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c > index e5b3d74f8222..4d376f984a8c 100644 > --- a/drivers/gpu/drm/i915/i915_gem_context.c > +++ b/drivers/gpu/drm/i915/i915_gem_context.c > @@ -431,27 +431,6 @@ void i915_gem_context_fini(struct drm_device *dev) > dev_priv->kernel_context = NULL; > } > <SNIP> > > static bool > -needs_pd_load_pre(struct intel_engine_cs *engine, struct intel_context *to) > +needs_pd_load_pre(struct i915_hw_ppgtt *ppgtt, > + struct intel_engine_cs *engine, > + struct intel_context *to) > { > - if (!to->ppgtt) > + if (ppgtt == NULL) Code checker will scream and ask for !ppgtt. And I'm pretty sure we should not keep flip-flopping between two styles. Also, you're not doing != NULL either to check if pointer is not NULL, but just if (foo), so I do not see why to avoid if (!foo). Regards, Joonas
On Thu, Apr 21, 2016 at 09:48:36AM +0300, Joonas Lahtinen wrote: > On ke, 2016-04-20 at 19:42 +0100, Chris Wilson wrote: > > static bool > > -needs_pd_load_pre(struct intel_engine_cs *engine, struct intel_context *to) > > +needs_pd_load_pre(struct i915_hw_ppgtt *ppgtt, > > + struct intel_engine_cs *engine, > > + struct intel_context *to) > > { > > - if (!to->ppgtt) > > + if (ppgtt == NULL) > > Code checker will scream and ask for !ppgtt. And I'm pretty sure we > should not keep flip-flopping between two styles. Also, you're not > doing != NULL either to check if pointer is not NULL, but just if > (foo), so I do not see why to avoid if (!foo). I normally do do != NULL :( Oh, well. I give in. Can I also stop writing NULL everywhere and just use 0 in pointer contexts? -Chris
On Thu, Apr 21, 2016 at 09:48:36AM +0300, Joonas Lahtinen wrote: > On ke, 2016-04-20 at 19:42 +0100, Chris Wilson wrote: > > The code to switch_mm() is already handled by i915_switch_context(), the > > only difference required to setup the aliasing ppgtt is that we need to > > emit te switch_mm() on the first context, i.e. when transitioning from > > engine->last_context == NULL. This allows us to defer the > > initialisation of the GPU from early device initialisation to first use, > > which should marginally speed up both. The caveat is that we then defer > > the context initialisation until first use - i.e. we cannot assume that > > the GPU engines are initialised. For example, this means that power > > contexts for rc6 (Ironlake) need to explicitly loaded, as they are. > > > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > Cc: Mika Kuoppala <mika.kuoppala@intel.com> > > Some comments below. > > Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c > > index aafcb4942acf..14c9b29294c5 100644 > > --- a/drivers/gpu/drm/i915/i915_gem.c > > +++ b/drivers/gpu/drm/i915/i915_gem.c > > @@ -4910,34 +4910,6 @@ i915_gem_init_hw(struct drm_device *dev) > > if (ret) > > goto out; > > This whole if (ret) goto out can be removed as out is right after the > removed code block. Oh, this piece of code is earmarked to be removed and now I even have a bugzilla for it! https://bugs.freedesktop.org/show_bug.cgi?id=95023 is an example where we are doing the wrong thing with the current i915_gem_init_seqno. -Chris
On Thu, Apr 21, 2016 at 08:01:37AM +0100, Chris Wilson wrote: > Oh, this piece of code is earmarked to be removed and now I even have a > bugzilla for it! > > https://bugs.freedesktop.org/show_bug.cgi?id=95023 > is an example where we are doing the wrong thing with the current > i915_gem_init_seqno. Too excited this morning. Thought I could explain some of the discrepancy in the error state by convincing myself we currently didn't reset all register state for seqno on init. -Chris
On to, 2016-04-21 at 08:01 +0100, Chris Wilson wrote: > On Thu, Apr 21, 2016 at 09:48:36AM +0300, Joonas Lahtinen wrote: > > > > On ke, 2016-04-20 at 19:42 +0100, Chris Wilson wrote: > > > > > > The code to switch_mm() is already handled by i915_switch_context(), the > > > only difference required to setup the aliasing ppgtt is that we need to > > > emit te switch_mm() on the first context, i.e. when transitioning from > > > engine->last_context == NULL. This allows us to defer the > > > initialisation of the GPU from early device initialisation to first use, > > > which should marginally speed up both. The caveat is that we then defer > > > the context initialisation until first use - i.e. we cannot assume that > > > the GPU engines are initialised. For example, this means that power > > > contexts for rc6 (Ironlake) need to explicitly loaded, as they are. > > > > > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > > Cc: Mika Kuoppala <mika.kuoppala@intel.com> > > Some comments below. > > > > Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > > > > > > > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c > > > index aafcb4942acf..14c9b29294c5 100644 > > > --- a/drivers/gpu/drm/i915/i915_gem.c > > > +++ b/drivers/gpu/drm/i915/i915_gem.c > > > @@ -4910,34 +4910,6 @@ i915_gem_init_hw(struct drm_device *dev) > > > if (ret) > > > goto out; > > This whole if (ret) goto out can be removed as out is right after the > > removed code block. > Oh, this piece of code is earmarked to be removed and now I even have a > bugzilla for it! > Huh? I was commeting that you just literally removed everything between "goto out;" and "out:" with this commit. So you can just drop the whole if (ret) goto out; > https://bugs.freedesktop.org/show_bug.cgi?id=95023 > is an example where we are doing the wrong thing with the current > i915_gem_init_seqno. > -Chris >
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 7dd79c9a8ce3..7adc7c83a116 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -3304,7 +3304,6 @@ void i915_gem_context_lost(struct drm_i915_private *dev_priv); void i915_gem_context_fini(struct drm_device *dev); void i915_gem_context_reset(struct drm_device *dev); int i915_gem_context_open(struct drm_device *dev, struct drm_file *file); -int i915_gem_context_enable(struct drm_i915_gem_request *req); void i915_gem_context_close(struct drm_device *dev, struct drm_file *file); int i915_switch_context(struct drm_i915_gem_request *req); struct intel_context * diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index aafcb4942acf..14c9b29294c5 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -4910,34 +4910,6 @@ i915_gem_init_hw(struct drm_device *dev) if (ret) goto out; - /* Now it is safe to go back round and do everything else: */ - for_each_engine(engine, dev_priv) { - struct drm_i915_gem_request *req; - - req = i915_gem_request_alloc(engine, NULL); - if (IS_ERR(req)) { - ret = PTR_ERR(req); - break; - } - - ret = i915_ppgtt_init_ring(req); - if (ret) - goto err_request; - - ret = i915_gem_context_enable(req); - if (ret) - goto err_request; - -err_request: - i915_add_request_no_flush(req); - if (ret) { - DRM_ERROR("Failed to enable %s, error=%d\n", - engine->name, ret); - i915_gem_cleanup_engines(dev); - break; - } - } - out: intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); return ret; diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c index e5b3d74f8222..4d376f984a8c 100644 --- a/drivers/gpu/drm/i915/i915_gem_context.c +++ b/drivers/gpu/drm/i915/i915_gem_context.c @@ -431,27 +431,6 @@ void i915_gem_context_fini(struct drm_device *dev) dev_priv->kernel_context = NULL; } -int i915_gem_context_enable(struct drm_i915_gem_request *req) -{ - struct intel_engine_cs *engine = req->engine; - int ret; - - if (i915.enable_execlists) { - if (engine->init_context == NULL) - return 0; - - ret = engine->init_context(req); - } else - ret = i915_switch_context(req); - - if (ret) { - DRM_ERROR("ring init context: %d\n", ret); - return ret; - } - - return 0; -} - static int context_idr_cleanup(int id, void *p, void *data) { struct intel_context *ctx = p; @@ -630,7 +609,8 @@ static int remap_l3(struct drm_i915_gem_request *req, int slice) return 0; } -static inline bool skip_rcs_switch(struct intel_engine_cs *engine, +static inline bool skip_rcs_switch(struct i915_hw_ppgtt *ppgtt, + struct intel_engine_cs *engine, struct intel_context *to) { if (to->remap_slice) @@ -639,21 +619,27 @@ static inline bool skip_rcs_switch(struct intel_engine_cs *engine, if (!to->legacy_hw_ctx.initialized) return false; - if (to->ppgtt && - !(intel_engine_flag(engine) & to->ppgtt->pd_dirty_rings)) + if (ppgtt && (intel_engine_flag(engine) & ppgtt->pd_dirty_rings)) return false; return to == engine->last_context; } static bool -needs_pd_load_pre(struct intel_engine_cs *engine, struct intel_context *to) +needs_pd_load_pre(struct i915_hw_ppgtt *ppgtt, + struct intel_engine_cs *engine, + struct intel_context *to) { - if (!to->ppgtt) + if (ppgtt == NULL) return false; + /* Always load the ppgtt on first use */ + if (engine->last_context == NULL) + return true; + + /* Same context without new entries, skip */ if (engine->last_context == to && - !(intel_engine_flag(engine) & to->ppgtt->pd_dirty_rings)) + !(intel_engine_flag(engine) & ppgtt->pd_dirty_rings)) return false; if (engine->id != RCS) @@ -666,9 +652,11 @@ needs_pd_load_pre(struct intel_engine_cs *engine, struct intel_context *to) } static bool -needs_pd_load_post(struct intel_context *to, u32 hw_flags) +needs_pd_load_post(struct i915_hw_ppgtt *ppgtt, + struct intel_context *to, + u32 hw_flags) { - if (!to->ppgtt) + if (ppgtt == NULL) return false; if (!IS_GEN8(to->i915)) @@ -684,11 +672,12 @@ static int do_rcs_switch(struct drm_i915_gem_request *req) { struct intel_context *to = req->ctx; struct intel_engine_cs *engine = req->engine; + struct i915_hw_ppgtt *ppgtt = to->ppgtt ?: req->i915->mm.aliasing_ppgtt; struct intel_context *from; u32 hw_flags; int ret, i; - if (skip_rcs_switch(engine, to)) + if (skip_rcs_switch(ppgtt, engine, to)) return 0; /* Trying to pin first makes error handling easier. */ @@ -719,13 +708,13 @@ static int do_rcs_switch(struct drm_i915_gem_request *req) if (ret) goto unpin_out; - if (needs_pd_load_pre(engine, to)) { + if (needs_pd_load_pre(ppgtt, engine, to)) { /* Older GENs and non render rings still want the load first, * "PP_DCLV followed by PP_DIR_BASE register through Load * Register Immediate commands in Ring Buffer before submitting * a context."*/ trace_switch_mm(engine, to); - ret = to->ppgtt->switch_mm(to->ppgtt, req); + ret = ppgtt->switch_mm(ppgtt, req); if (ret) goto unpin_out; } @@ -736,15 +725,14 @@ static int do_rcs_switch(struct drm_i915_gem_request *req) * space. This means we must enforce that a page table load * occur when this occurs. */ hw_flags = MI_RESTORE_INHIBIT; - else if (to->ppgtt && - intel_engine_flag(engine) & to->ppgtt->pd_dirty_rings) + else if (ppgtt && intel_engine_flag(engine) & ppgtt->pd_dirty_rings) hw_flags = MI_FORCE_RESTORE; else hw_flags = 0; /* We should never emit switch_mm more than once */ - WARN_ON(needs_pd_load_pre(engine, to) && - needs_pd_load_post(to, hw_flags)); + WARN_ON(needs_pd_load_pre(ppgtt, engine, to) && + needs_pd_load_post(ppgtt, to, hw_flags)); if (to != from || (hw_flags & MI_FORCE_RESTORE)) { ret = mi_set_context(req, hw_flags); @@ -780,9 +768,9 @@ static int do_rcs_switch(struct drm_i915_gem_request *req) /* GEN8 does *not* require an explicit reload if the PDPs have been * setup, and we do not wish to move them. */ - if (needs_pd_load_post(to, hw_flags)) { + if (needs_pd_load_post(ppgtt, to, hw_flags)) { trace_switch_mm(engine, to); - ret = to->ppgtt->switch_mm(to->ppgtt, req); + ret = ppgtt->switch_mm(ppgtt, req); /* The hardware context switch is emitted, but we haven't * actually changed the state - so it's probably safe to bail * here. Still, let the user know something dangerous has @@ -792,8 +780,8 @@ static int do_rcs_switch(struct drm_i915_gem_request *req) return ret; } - if (to->ppgtt) - to->ppgtt->pd_dirty_rings &= ~intel_engine_flag(engine); + if (ppgtt) + ppgtt->pd_dirty_rings &= ~intel_engine_flag(engine); for (i = 0; i < MAX_L3_SLICES; i++) { if (!(to->remap_slice & (1<<i))) @@ -846,17 +834,18 @@ int i915_switch_context(struct drm_i915_gem_request *req) if (engine->id != RCS || req->ctx->legacy_hw_ctx.rcs_state == NULL) { struct intel_context *to = req->ctx; + struct i915_hw_ppgtt *ppgtt = + to->ppgtt ?: req->i915->mm.aliasing_ppgtt; - if (needs_pd_load_pre(engine, to)) { + if (needs_pd_load_pre(ppgtt, engine, to)) { int ret; trace_switch_mm(engine, to); - ret = to->ppgtt->switch_mm(to->ppgtt, req); + ret = ppgtt->switch_mm(ppgtt, req); if (ret) return ret; - /* Doing a PD load always reloads the page dirs */ - to->ppgtt->pd_dirty_rings &= ~intel_engine_flag(engine); + ppgtt->pd_dirty_rings &= ~intel_engine_flag(engine); } if (to != engine->last_context) { diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c index 69191ae2cd6a..45a17a9bf3ce 100644 --- a/drivers/gpu/drm/i915/i915_gem_gtt.c +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c @@ -2187,20 +2187,6 @@ int i915_ppgtt_init_hw(struct drm_device *dev) return 0; } -int i915_ppgtt_init_ring(struct drm_i915_gem_request *req) -{ - struct drm_i915_private *dev_priv = req->i915; - struct i915_hw_ppgtt *ppgtt = dev_priv->mm.aliasing_ppgtt; - - if (i915.enable_execlists) - return 0; - - if (!ppgtt) - return 0; - - return ppgtt->switch_mm(ppgtt, req); -} - struct i915_hw_ppgtt * i915_ppgtt_create(struct drm_device *dev, struct drm_i915_file_private *fpriv) { diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h index b0ae6632c01a..114850368bca 100644 --- a/drivers/gpu/drm/i915/i915_gem_gtt.h +++ b/drivers/gpu/drm/i915/i915_gem_gtt.h @@ -522,7 +522,6 @@ void i915_ggtt_cleanup_hw(struct drm_device *dev); int i915_ppgtt_init(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt); int i915_ppgtt_init_hw(struct drm_device *dev); -int i915_ppgtt_init_ring(struct drm_i915_gem_request *req); void i915_ppgtt_release(struct kref *kref); struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_device *dev, struct drm_i915_file_private *fpriv);
The code to switch_mm() is already handled by i915_switch_context(), the only difference required to setup the aliasing ppgtt is that we need to emit te switch_mm() on the first context, i.e. when transitioning from engine->last_context == NULL. This allows us to defer the initialisation of the GPU from early device initialisation to first use, which should marginally speed up both. The caveat is that we then defer the context initialisation until first use - i.e. we cannot assume that the GPU engines are initialised. For example, this means that power contexts for rc6 (Ironlake) need to explicitly loaded, as they are. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Mika Kuoppala <mika.kuoppala@intel.com> --- drivers/gpu/drm/i915/i915_drv.h | 1 - drivers/gpu/drm/i915/i915_gem.c | 28 ------------ drivers/gpu/drm/i915/i915_gem_context.c | 77 ++++++++++++++------------------- drivers/gpu/drm/i915/i915_gem_gtt.c | 14 ------ drivers/gpu/drm/i915/i915_gem_gtt.h | 1 - 5 files changed, 33 insertions(+), 88 deletions(-)