| Message ID | 20180727085350.321-3-jakub.bartminski@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v6,1/6] drm/i915/guc: Avoid wasting memory on incorrect GuC pin bias | expand |
Quoting Jakub Bartmiński (2018-07-27 09:53:47) > diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c > index d0acef299b9c..8ac5214b3648 100644 > --- a/drivers/gpu/drm/i915/i915_gem_gtt.c > +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c > @@ -2917,6 +2917,14 @@ int i915_gem_init_ggtt(struct drm_i915_private *dev_priv) > struct drm_mm_node *entry; > int ret; > > + /* > + * GuC requires the ring to be placed in Non-WOPCM memory. If GuC is not > + * present or not in use we still need a small bias as ring wraparound > + * at offset 0 sometimes hangs. No idea why. > + */ > + ggtt->pin_bias = max_t(u32, I915_GTT_PAGE_SIZE, > + intel_guc_wopcm_region_size(&dev_priv->guc)); Hmm, it feels like this should fit in with ggtt_init_hw, but the current setup makes that impossible. intel_guc_wopcm_region_size() is a peculiar path to access dev_priv->wopcm. Do we not want something like intel_wopcm_lower_reserved_size()? > +/** > + * intel_guc_wopcm_region_size() - Get the size of the GuC region of WOPCM. > + * @guc: intel_guc structure. > + * > + * Returns: > + * 0 if GuC is not present or not in use. > + * Otherwise, the GuC WOPCM size. > + */ > +u32 intel_guc_wopcm_region_size(struct intel_guc *guc) > +{ > + struct intel_wopcm *wopcm = &guc_to_i915(guc)->wopcm; > + > + /* WOPCM must have already been partitioned */ > + GEM_BUG_ON(!wopcm->size); Unrelated bug-on, since what you want is a statement that intel_wopcm_init() was called and that doesn't provide it. -Chris
On Fri, 27 Jul 2018 12:29:15 +0200, Chris Wilson <chris@chris-wilson.co.uk> wrote: > Quoting Jakub Bartmiński (2018-07-27 09:53:47) >> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c >> b/drivers/gpu/drm/i915/i915_gem_gtt.c >> index d0acef299b9c..8ac5214b3648 100644 >> --- a/drivers/gpu/drm/i915/i915_gem_gtt.c >> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c >> @@ -2917,6 +2917,14 @@ int i915_gem_init_ggtt(struct drm_i915_private >> *dev_priv) >> struct drm_mm_node *entry; >> int ret; >> >> + /* >> + * GuC requires the ring to be placed in Non-WOPCM memory. If >> GuC is not >> + * present or not in use we still need a small bias as ring >> wraparound >> + * at offset 0 sometimes hangs. No idea why. >> + */ >> + ggtt->pin_bias = max_t(u32, I915_GTT_PAGE_SIZE, >> + >> intel_guc_wopcm_region_size(&dev_priv->guc)); > > Hmm, it feels like this should fit in with ggtt_init_hw, but the current > setup makes that impossible. > > intel_guc_wopcm_region_size() is a peculiar path to access > dev_priv->wopcm. > > Do we not want something like intel_wopcm_lower_reserved_size()? Earlier I was suggesting intel_guc_get_ggtt_pin_bias() as we should not care how bias this is related to wopcm - only GuC should know that. > >> +/** >> + * intel_guc_wopcm_region_size() - Get the size of the GuC region of >> WOPCM. >> + * @guc: intel_guc structure. >> + * >> + * Returns: >> + * 0 if GuC is not present or not in use. >> + * Otherwise, the GuC WOPCM size. >> + */ >> +u32 intel_guc_wopcm_region_size(struct intel_guc *guc) >> +{ >> + struct intel_wopcm *wopcm = &guc_to_i915(guc)->wopcm; >> + >> + /* WOPCM must have already been partitioned */ >> + GEM_BUG_ON(!wopcm->size); > > Unrelated bug-on, since what you want is a statement that > intel_wopcm_init() was called and that doesn't provide it. I guess GEM_BUG_ON(!wopcm->guc.size) should be better here. Michal
Quoting Michal Wajdeczko (2018-07-27 13:03:14) > On Fri, 27 Jul 2018 12:29:15 +0200, Chris Wilson > <chris@chris-wilson.co.uk> wrote: > > > Quoting Jakub Bartmiński (2018-07-27 09:53:47) > >> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c > >> b/drivers/gpu/drm/i915/i915_gem_gtt.c > >> index d0acef299b9c..8ac5214b3648 100644 > >> --- a/drivers/gpu/drm/i915/i915_gem_gtt.c > >> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c > >> @@ -2917,6 +2917,14 @@ int i915_gem_init_ggtt(struct drm_i915_private > >> *dev_priv) > >> struct drm_mm_node *entry; > >> int ret; > >> > >> + /* > >> + * GuC requires the ring to be placed in Non-WOPCM memory. If > >> GuC is not > >> + * present or not in use we still need a small bias as ring > >> wraparound > >> + * at offset 0 sometimes hangs. No idea why. > >> + */ > >> + ggtt->pin_bias = max_t(u32, I915_GTT_PAGE_SIZE, > >> + > >> intel_guc_wopcm_region_size(&dev_priv->guc)); > > > > Hmm, it feels like this should fit in with ggtt_init_hw, but the current > > setup makes that impossible. > > > > intel_guc_wopcm_region_size() is a peculiar path to access > > dev_priv->wopcm. > > > > Do we not want something like intel_wopcm_lower_reserved_size()? > > Earlier I was suggesting intel_guc_get_ggtt_pin_bias() as we should not > care how bias this is related to wopcm - only GuC should know that. :) Here I only care that the reason we exclude anything from the GGTT is self-evident. intel_guc_reserved_gtt_size(); intel_wopcm_reserved_gtt_size(); make more sense to me as to why we would even have a ggtt->pin_bias = intel_X_reserved_gtt_size() here. -Chris
diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c index b10770cfccd2..32f96b8cd9c4 100644 --- a/drivers/gpu/drm/i915/i915_gem_context.c +++ b/drivers/gpu/drm/i915/i915_gem_context.c @@ -329,15 +329,7 @@ __create_hw_context(struct drm_i915_private *dev_priv, ctx->desc_template = default_desc_template(dev_priv, dev_priv->mm.aliasing_ppgtt); - /* - * GuC requires the ring to be placed in Non-WOPCM memory. If GuC is not - * present or not in use we still need a small bias as ring wraparound - * at offset 0 sometimes hangs. No idea why. - */ - if (USES_GUC(dev_priv)) - ctx->ggtt_offset_bias = dev_priv->guc.ggtt_pin_bias; - else - ctx->ggtt_offset_bias = I915_GTT_PAGE_SIZE; + ctx->ggtt_offset_bias = dev_priv->ggtt.pin_bias; return ctx; diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c index d0acef299b9c..8ac5214b3648 100644 --- a/drivers/gpu/drm/i915/i915_gem_gtt.c +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c @@ -2917,6 +2917,14 @@ int i915_gem_init_ggtt(struct drm_i915_private *dev_priv) struct drm_mm_node *entry; int ret; + /* + * GuC requires the ring to be placed in Non-WOPCM memory. If GuC is not + * present or not in use we still need a small bias as ring wraparound + * at offset 0 sometimes hangs. No idea why. + */ + ggtt->pin_bias = max_t(u32, I915_GTT_PAGE_SIZE, + intel_guc_wopcm_region_size(&dev_priv->guc)); + ret = intel_vgt_balloon(dev_priv); if (ret) return ret; diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h index 14e62651010b..71e8cf24c800 100644 --- a/drivers/gpu/drm/i915/i915_gem_gtt.h +++ b/drivers/gpu/drm/i915/i915_gem_gtt.h @@ -396,6 +396,8 @@ struct i915_ggtt { int mtrr; + u32 pin_bias; + struct drm_mm_node error_capture; }; diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h index f06d66377107..abf6144e3296 100644 --- a/drivers/gpu/drm/i915/i915_vma.h +++ b/drivers/gpu/drm/i915/i915_vma.h @@ -207,6 +207,11 @@ static inline u32 i915_ggtt_offset(const struct i915_vma *vma) return lower_32_bits(vma->node.start); } +static inline u32 i915_ggtt_pin_bias(struct i915_vma *vma) +{ + return i915_vm_to_ggtt(vma->vm)->pin_bias; +} + static inline struct i915_vma *i915_vma_get(struct i915_vma *vma) { i915_gem_object_get(vma->obj); diff --git a/drivers/gpu/drm/i915/intel_guc.c b/drivers/gpu/drm/i915/intel_guc.c index aa28368f8ba7..8ab4653e89af 100644 --- a/drivers/gpu/drm/i915/intel_guc.c +++ b/drivers/gpu/drm/i915/intel_guc.c @@ -27,8 +27,6 @@ #include "intel_guc_submission.h" #include "i915_drv.h" -static void guc_init_ggtt_pin_bias(struct intel_guc *guc); - static void gen8_guc_raise_irq(struct intel_guc *guc) { struct drm_i915_private *dev_priv = guc_to_i915(guc); @@ -142,8 +140,6 @@ int intel_guc_init_misc(struct intel_guc *guc) struct drm_i915_private *i915 = guc_to_i915(guc); int ret; - guc_init_ggtt_pin_bias(guc); - ret = guc_init_wq(guc); if (ret) return ret; @@ -603,22 +599,6 @@ int intel_guc_resume(struct intel_guc *guc) * to DRAM. The value of the GuC ggtt_pin_bias is the GuC WOPCM size. */ -/** - * guc_init_ggtt_pin_bias() - Initialize the GuC ggtt_pin_bias value. - * @guc: intel_guc structure. - * - * This function will calculate and initialize the ggtt_pin_bias value - * based on the GuC WOPCM size. - */ -static void guc_init_ggtt_pin_bias(struct intel_guc *guc) -{ - struct drm_i915_private *i915 = guc_to_i915(guc); - - GEM_BUG_ON(!i915->wopcm.size); - - guc->ggtt_pin_bias = i915->wopcm.guc.size; -} - /** * intel_guc_allocate_vma() - Allocate a GGTT VMA for GuC usage * @guc: the guc @@ -637,6 +617,7 @@ struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size) struct drm_i915_private *dev_priv = guc_to_i915(guc); struct drm_i915_gem_object *obj; struct i915_vma *vma; + u64 flags; int ret; obj = i915_gem_object_create(dev_priv, size); @@ -647,8 +628,8 @@ struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size) if (IS_ERR(vma)) goto err; - ret = i915_vma_pin(vma, 0, PAGE_SIZE, - PIN_GLOBAL | PIN_OFFSET_BIAS | guc->ggtt_pin_bias); + flags = PIN_GLOBAL | PIN_OFFSET_BIAS | i915_ggtt_pin_bias(vma); + ret = i915_vma_pin(vma, 0, PAGE_SIZE, flags); if (ret) { vma = ERR_PTR(ret); goto err; @@ -660,3 +641,21 @@ struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size) i915_gem_object_put(obj); return vma; } + +/** + * intel_guc_wopcm_region_size() - Get the size of the GuC region of WOPCM. + * @guc: intel_guc structure. + * + * Returns: + * 0 if GuC is not present or not in use. + * Otherwise, the GuC WOPCM size. + */ +u32 intel_guc_wopcm_region_size(struct intel_guc *guc) +{ + struct intel_wopcm *wopcm = &guc_to_i915(guc)->wopcm; + + /* WOPCM must have already been partitioned */ + GEM_BUG_ON(!wopcm->size); + + return wopcm->guc.size; +} diff --git a/drivers/gpu/drm/i915/intel_guc.h b/drivers/gpu/drm/i915/intel_guc.h index 4121928a495e..61fe3034cbe2 100644 --- a/drivers/gpu/drm/i915/intel_guc.h +++ b/drivers/gpu/drm/i915/intel_guc.h @@ -49,9 +49,6 @@ struct intel_guc { struct intel_guc_log log; struct intel_guc_ct ct; - /* Offset where Non-WOPCM memory starts. */ - u32 ggtt_pin_bias; - /* Log snapshot if GuC errors during load */ struct drm_i915_gem_object *load_err_log; @@ -130,10 +127,10 @@ static inline void intel_guc_to_host_event_handler(struct intel_guc *guc) * @vma: i915 graphics virtual memory area. * * GuC does not allow any gfx GGTT address that falls into range - * [0, GuC ggtt_pin_bias), which is reserved for Boot ROM, SRAM and WOPCM. - * Currently, in order to exclude [0, GuC ggtt_pin_bias) address space from + * [0, ggtt.pin_bias), which is reserved for Boot ROM, SRAM and WOPCM. + * Currently, in order to exclude [0, ggtt.pin_bias) address space from * GGTT, all gfx objects used by GuC are allocated with intel_guc_allocate_vma() - * and pinned with PIN_OFFSET_BIAS along with the value of GuC ggtt_pin_bias. + * and pinned with PIN_OFFSET_BIAS along with the value of ggtt.pin_bias. * * Return: GGTT offset of the @vma. */ @@ -142,7 +139,7 @@ static inline u32 intel_guc_ggtt_offset(struct intel_guc *guc, { u32 offset = i915_ggtt_offset(vma); - GEM_BUG_ON(offset < guc->ggtt_pin_bias); + GEM_BUG_ON(offset < i915_ggtt_pin_bias(vma)); GEM_BUG_ON(range_overflows_t(u64, offset, vma->size, GUC_GGTT_TOP)); return offset; @@ -168,6 +165,7 @@ int intel_guc_auth_huc(struct intel_guc *guc, u32 rsa_offset); int intel_guc_suspend(struct intel_guc *guc); int intel_guc_resume(struct intel_guc *guc); struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size); +u32 intel_guc_wopcm_region_size(struct intel_guc *guc); static inline int intel_guc_sanitize(struct intel_guc *guc) { diff --git a/drivers/gpu/drm/i915/intel_huc.c b/drivers/gpu/drm/i915/intel_huc.c index ffcad5fad6a7..37ef540dd280 100644 --- a/drivers/gpu/drm/i915/intel_huc.c +++ b/drivers/gpu/drm/i915/intel_huc.c @@ -63,7 +63,7 @@ int intel_huc_auth(struct intel_huc *huc) return -ENOEXEC; vma = i915_gem_object_ggtt_pin(huc->fw.obj, NULL, 0, 0, - PIN_OFFSET_BIAS | guc->ggtt_pin_bias); + PIN_OFFSET_BIAS | i915->ggtt.pin_bias); if (IS_ERR(vma)) { ret = PTR_ERR(vma); DRM_ERROR("HuC: Failed to pin huc fw object %d\n", ret); diff --git a/drivers/gpu/drm/i915/intel_uc_fw.c b/drivers/gpu/drm/i915/intel_uc_fw.c index 6e8e0b546743..fd496416087c 100644 --- a/drivers/gpu/drm/i915/intel_uc_fw.c +++ b/drivers/gpu/drm/i915/intel_uc_fw.c @@ -222,7 +222,7 @@ int intel_uc_fw_upload(struct intel_uc_fw *uc_fw, goto fail; } - ggtt_pin_bias = to_i915(uc_fw->obj->base.dev)->guc.ggtt_pin_bias; + ggtt_pin_bias = to_i915(uc_fw->obj->base.dev)->ggtt.pin_bias; vma = i915_gem_object_ggtt_pin(uc_fw->obj, NULL, 0, 0, PIN_OFFSET_BIAS | ggtt_pin_bias); if (IS_ERR(vma)) {
Removing the pin bias from GuC allows us to not check for GuC every time we pin a context, which fixes the assertion error on unresolved GuC platform default in mock contexts selftest. It also seems that we were using uninitialized WOPCM variables when setting the GuC pin bias. The pin bias has to be set after the WOPCM, but before the call to i915_gem_contexts_init where the first contexts are pinned. v2: This also makes it so that there's no need to set GuC variables from within the WOPCM init function or to move the WOPCM init, while keeping the correct initialization order. Also for mock tests the pin bias is left at 0 and we make sure that the pin bias with GuC will not be smaller than without GuC. v3: Avoid unused i915 in intel_guc_ggtt_offset if debug is disabled. v4: Squash with WOPCM init reordering. Moved the i915_ggtt_pin_bias helper to this patch, and made some functions use it instead of directly dereferencing i915->ggtt. v5: Since we now don't use wopcm.guc.base for the pin bias there's no need to validate it. It also has already been verified in WOPCM init. v6: Moved and renamed the function which now returns the wopcm.guc.size to intel_guc.c:intel_guc_wopcm_region_size to avoid any possible confusion with the pin_bias in ggtt, which should be used for pinning. Deleted the now unnecessarily introduced includes from previous versions. Dropped naming changes from dev_priv to i915 for better patch readability. Fixes: f7dc0157e4b5 ("drm/i915/uc: Fetch GuC/HuC firmwares from guc/huc specific init") Testcase: igt/drv_selftest/mock_contexts #GuC Signed-off-by: Jakub Bartmiński <jakub.bartminski@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Michał Winiarski <michal.winiarski@intel.com> Cc: Michal Wajdeczko <michal.wajdeczko@intel.com> --- drivers/gpu/drm/i915/i915_gem_context.c | 10 +----- drivers/gpu/drm/i915/i915_gem_gtt.c | 8 +++++ drivers/gpu/drm/i915/i915_gem_gtt.h | 2 ++ drivers/gpu/drm/i915/i915_vma.h | 5 +++ drivers/gpu/drm/i915/intel_guc.c | 43 ++++++++++++------------- drivers/gpu/drm/i915/intel_guc.h | 12 +++---- drivers/gpu/drm/i915/intel_huc.c | 2 +- drivers/gpu/drm/i915/intel_uc_fw.c | 2 +- 8 files changed, 44 insertions(+), 40 deletions(-)