[PATCH] drm/i915: Support to enable TRTT on GEN9

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



From: Akash Goel <akash.goel@xxxxxxxxx>

Gen9 has an additional address translation hardware support in form of
Tiled Resource Translation Table (TR-TT) which provides an extra level
of abstraction over PPGTT.
This is useful for mapping Sparse/Tiled texture resources.
Sparse resources are created as virtual-only allocations. Regions of the
resource that the application intends to use is bound to the physical memory
on the fly and can be re-bound to different memory allocations over the
lifetime of the resource.

TR-TT is tightly coupled with PPGTT, a new instance of TR-TT will be required
for a new PPGTT instance, but TR-TT may not enabled for every context.
1/16th of the 48bit PPGTT space is earmarked for the translation by TR-TT,
which such chunk to use is conveyed to HW through a register.
Any GFX address, which lies in that reserved 44 bit range will be translated
through TR-TT first and then through PPGTT to get the actual physical address,
so the output of translation from TR-TT will be a PPGTT offset.

TRTT is constructed as a 3 level tile Table. Each tile is 64KB is size which
leaves behind 44-16=28 address bits. 28bits are partitioned as 9+9+10, and
each level is contained within a 4KB page hence L3 and L2 is composed of
512 64b entries and L1 is composed of 1024 32b entries.

There is a provision to keep TR-TT Tables in virtual space, where the pages of
TRTT tables will be mapped to PPGTT.
Currently this is the supported mode, in this mode UMD will have a full control
on TR-TT management, with bare minimum support from KMD.
So the entries of L3 table will contain the PPGTT offset of L2 Table pages,
similarly entries of L2 table will contain the PPGTT offset of L1 Table pages.
The entries of L1 table will contain the PPGTT offset of BOs actually backing
the Sparse resources.
The assumption here is that UMD only will do the complete PPGTT address space
management and use the Soft Pin API for all the buffer objects associated with
a given Context. So UMD will also have to allocate the L3/L2/L1 table pages
as a regular GEM BO only & assign them a PPGTT address through the Soft Pin API.
UMD would have to emit the MI_STORE_DATA_IMM commands in the batch buffer to
program the relevant entries of L3/L2/L1 tables.

Any space in TR-TT segment not bound to any Sparse texture, will be handled
through Invalid tile, User is expected to initialize the entries of a new
L3/L2/L1 table page with the Invalid tile pattern. The entries corresponding to
the holes in the Sparse texture resource will be set with the Null tile pattern
The improper programming of TRTT should only lead to a recoverable GPU hang,
eventually leading to banning of the culprit context without victimizing others.

The association of any Sparse resource with the BOs will be known only to UMD,
and only the Sparse resources shall be assigned an offset from the TR-TT segment
by UMD. The use of TR-TT segment or mapping of Sparse resources will be
abstracted from the KMD, UMD can do the address assignment from TR-TT segment
autonomously and KMD will be oblivious of it.
The BOs must not be assigned an address from TR-TT segment, they will be mapped
to PPGTT in a regular way by KMD, using the Soft Pin offset provided by UMD.

This patch provides an interface through which UMD can convey KMD to enable
TR-TT for a given context. A new I915_CONTEXT_PARAM_ENABLE_TRTT param has been
added to I915_GEM_CONTEXT_SETPARAM ioctl for that purpose.
UMD will have to pass the GFX address of L3 table page, pattern value for the
Null & invalid Tile registers.

Testcase: igt/gem_trtt

Signed-off-by: Akash Goel <akash.goel@xxxxxxxxx>
---
 drivers/gpu/drm/i915/i915_dma.c         |  3 ++
 drivers/gpu/drm/i915/i915_drv.h         | 12 +++++++
 drivers/gpu/drm/i915/i915_gem_context.c | 45 ++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_gem_gtt.c     | 57 +++++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_gem_gtt.h     |  6 ++++
 drivers/gpu/drm/i915/i915_reg.h         | 19 +++++++++++
 drivers/gpu/drm/i915/intel_lrc.c        | 41 ++++++++++++++++++++++++
 include/uapi/drm/i915_drm.h             |  8 +++++
 8 files changed, 191 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c
index 988a380..c247c25 100644
--- a/drivers/gpu/drm/i915/i915_dma.c
+++ b/drivers/gpu/drm/i915/i915_dma.c
@@ -172,6 +172,9 @@ static int i915_getparam(struct drm_device *dev, void *data,
 	case I915_PARAM_HAS_EXEC_SOFTPIN:
 		value = 1;
 		break;
+	case I915_PARAM_HAS_TRTT:
+		value = HAS_TRTT(dev);
+		break;
 	default:
 		DRM_DEBUG("Unknown parameter %d\n", param->param);
 		return -EINVAL;
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index c6dd4db..12c612e 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -839,6 +839,7 @@ struct i915_ctx_hang_stats {
 #define DEFAULT_CONTEXT_HANDLE 0
 
 #define CONTEXT_NO_ZEROMAP (1<<0)
+#define CONTEXT_USE_TRTT   (1<<1)
 /**
  * struct intel_context - as the name implies, represents a context.
  * @ref: reference count.
@@ -881,6 +882,15 @@ struct intel_context {
 		int pin_count;
 	} engine[I915_NUM_RINGS];
 
+	/* TRTT info */
+	struct {
+		uint32_t invd_tile_val;
+		uint32_t null_tile_val;
+		uint64_t l3_table_address;
+		struct i915_vma *vma;
+		bool update_trtt_params;
+	} trtt_info;
+
 	struct list_head link;
 };
 
@@ -2626,6 +2636,8 @@ struct drm_i915_cmd_table {
 				 !IS_VALLEYVIEW(dev) && !IS_CHERRYVIEW(dev) && \
 				 !IS_BROXTON(dev))
 
+#define HAS_TRTT(dev)		(IS_GEN9(dev))
+
 #define INTEL_PCH_DEVICE_ID_MASK		0xff00
 #define INTEL_PCH_IBX_DEVICE_ID_TYPE		0x3b00
 #define INTEL_PCH_CPT_DEVICE_ID_TYPE		0x1c00
diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index 900ffd0..ae9fc34 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -146,6 +146,9 @@ static void i915_gem_context_clean(struct intel_context *ctx)
 		if (WARN_ON(__i915_vma_unbind_no_wait(vma)))
 			break;
 	}
+
+	if (ctx->flags & CONTEXT_USE_TRTT)
+		i915_gem_destroy_trtt_vma(ctx->trtt_info.vma);
 }
 
 void i915_gem_context_free(struct kref *ctx_ref)
@@ -512,6 +515,35 @@ i915_gem_context_get(struct drm_i915_file_private *file_priv, u32 id)
 	return ctx;
 }
 
+static int
+i915_setup_trtt_ctx(struct intel_context *ctx,
+		    struct drm_i915_gem_context_trtt_param *trtt_params)
+{
+	if (ctx->flags & CONTEXT_USE_TRTT)
+		return -EEXIST;
+
+	/* basic sanity checks for the l3 table pointer */
+	if ((ctx->trtt_info.l3_table_address >= GEN9_TRTT_SEGMENT_START) &&
+	    (ctx->trtt_info.l3_table_address <
+			(GEN9_TRTT_SEGMENT_START + GEN9_TRTT_SEGMENT_SIZE)))
+		return -EINVAL;
+
+	if (ctx->trtt_info.l3_table_address & ~GEN9_TRTT_L3_GFXADDR_MASK)
+		return -EINVAL;
+
+	ctx->trtt_info.vma = i915_gem_setup_trtt_vma(&ctx->ppgtt->base);
+	if (IS_ERR(ctx->trtt_info.vma))
+		return PTR_ERR(ctx->trtt_info.vma);
+
+	ctx->trtt_info.null_tile_val = trtt_params->null_tile_val;
+	ctx->trtt_info.invd_tile_val = trtt_params->invd_tile_val;
+	ctx->trtt_info.l3_table_address = trtt_params->l3_table_address;
+	ctx->trtt_info.update_trtt_params = 1;
+
+	ctx->flags |= CONTEXT_USE_TRTT;
+	return 0;
+}
+
 static inline int
 mi_set_context(struct drm_i915_gem_request *req, u32 hw_flags)
 {
@@ -952,6 +984,7 @@ int i915_gem_context_setparam_ioctl(struct drm_device *dev, void *data,
 {
 	struct drm_i915_file_private *file_priv = file->driver_priv;
 	struct drm_i915_gem_context_param *args = data;
+	struct drm_i915_gem_context_trtt_param trtt_params;
 	struct intel_context *ctx;
 	int ret;
 
@@ -983,6 +1016,18 @@ int i915_gem_context_setparam_ioctl(struct drm_device *dev, void *data,
 			ctx->flags |= args->value ? CONTEXT_NO_ZEROMAP : 0;
 		}
 		break;
+	case I915_CONTEXT_PARAM_ENABLE_TRTT:
+		if (args->size < sizeof(trtt_params))
+			ret = -EINVAL;
+		else if (!HAS_TRTT(dev) || !USES_FULL_48BIT_PPGTT(dev))
+			ret = -ENODEV;
+		else if (copy_from_user(&trtt_params,
+					to_user_ptr(args->value),
+					sizeof(trtt_params)))
+			ret = -EFAULT;
+		else
+			ret = i915_setup_trtt_ctx(ctx, &trtt_params);
+		break;
 	default:
 		ret = -EINVAL;
 		break;
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 56f4f2e..28fc1ea 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -2146,6 +2146,13 @@ int i915_ppgtt_init(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt)
 
 int i915_ppgtt_init_hw(struct drm_device *dev)
 {
+	if (HAS_TRTT(dev) && USES_FULL_48BIT_PPGTT(dev)) {
+		struct drm_i915_private *dev_priv = dev->dev_private;
+
+		I915_WRITE(GEN9_TR_CHICKEN_BIT_VECTOR,
+			   GEN9_TRTT_BYPASS_DISABLE);
+	}
+
 	/* In the case of execlists, PPGTT is enabled by the context descriptor
 	 * and the PDPs are contained within the context itself.  We don't
 	 * need to do anything here. */
@@ -3328,6 +3335,56 @@ i915_gem_obj_lookup_or_create_ggtt_vma(struct drm_i915_gem_object *obj,
 
 }
 
+void i915_gem_destroy_trtt_vma(struct i915_vma *vma)
+{
+	struct i915_address_space *vm = vma->vm;
+
+	WARN_ON(!list_empty(&vma->vma_link));
+	WARN_ON(!list_empty(&vma->mm_list));
+	WARN_ON(!list_empty(&vma->exec_list));
+
+	drm_mm_remove_node(&vma->node);
+	i915_ppgtt_put(i915_vm_to_ppgtt(vm));
+	kmem_cache_free(to_i915(vm->dev)->vmas, vma);
+}
+
+struct i915_vma *
+i915_gem_setup_trtt_vma(struct i915_address_space *vm)
+{
+	struct i915_vma *vma;
+	int ret;
+
+	vma = kmem_cache_zalloc(to_i915(vm->dev)->vmas, GFP_KERNEL);
+	if (vma == NULL)
+		return ERR_PTR(-ENOMEM);
+
+	INIT_LIST_HEAD(&vma->vma_link);
+	INIT_LIST_HEAD(&vma->mm_list);
+	INIT_LIST_HEAD(&vma->exec_list);
+	vma->vm = vm;
+	i915_ppgtt_get(i915_vm_to_ppgtt(vm));
+
+	/* Mark the vma as perennially pinned */
+	vma->pin_count = 1;
+
+	/* Reserve from the 48 bit PPGTT space */
+	vma->node.start = GEN9_TRTT_SEGMENT_START;
+	vma->node.size = GEN9_TRTT_SEGMENT_SIZE;
+	ret = drm_mm_reserve_node(&vm->mm, &vma->node);
+	if (ret) {
+		ret = i915_gem_evict_for_vma(vma);
+		if (ret == 0)
+			ret = drm_mm_reserve_node(&vm->mm, &vma->node);
+	}
+	if (ret) {
+		DRM_ERROR("Reservation for TRTT segment failed: %i\n", ret);
+		i915_gem_destroy_trtt_vma(vma);
+		return ERR_PTR(ret);
+	}
+
+	return vma;
+}
+
 static struct scatterlist *
 rotate_pages(dma_addr_t *in, unsigned int offset,
 	     unsigned int width, unsigned int height,
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index b448ad8..acb942d 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -129,6 +129,10 @@ typedef uint64_t gen8_ppgtt_pml4e_t;
 #define GEN8_PPAT_ELLC_OVERRIDE		(0<<2)
 #define GEN8_PPAT(i, x)			((uint64_t) (x) << ((i) * 8))
 
+/* Lies at the top of 48 bit PPGTT space */
+#define GEN9_TRTT_SEGMENT_START		((1ULL << 48) - (1ULL << 44))
+#define GEN9_TRTT_SEGMENT_SIZE		(1ULL << 44)
+
 enum i915_ggtt_view_type {
 	I915_GGTT_VIEW_NORMAL = 0,
 	I915_GGTT_VIEW_ROTATED,
@@ -559,4 +563,6 @@ size_t
 i915_ggtt_view_size(struct drm_i915_gem_object *obj,
 		    const struct i915_ggtt_view *view);
 
+struct i915_vma *i915_gem_setup_trtt_vma(struct i915_address_space *vm);
+void i915_gem_destroy_trtt_vma(struct i915_vma *vma);
 #endif
diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
index 007ae83..5859be6 100644
--- a/drivers/gpu/drm/i915/i915_reg.h
+++ b/drivers/gpu/drm/i915/i915_reg.h
@@ -186,6 +186,25 @@ static inline bool i915_mmio_reg_valid(i915_reg_t reg)
 #define   GEN8_RPCS_EU_MIN_SHIFT	0
 #define   GEN8_RPCS_EU_MIN_MASK		(0xf << GEN8_RPCS_EU_MIN_SHIFT)
 
+#define GEN9_TR_CHICKEN_BIT_VECTOR	_MMIO(0x4DFC)
+#define   GEN9_TRTT_BYPASS_DISABLE	(1<<0)
+
+/* TRTT registers in the H/W Context */
+#define GEN9_TRTT_L3_POINTER_DW0	_MMIO(0x4DE0)
+#define GEN9_TRTT_L3_POINTER_DW1	_MMIO(0x4DE4)
+#define   GEN9_TRTT_L3_GFXADDR_MASK	0xFFFFFFFF0000
+
+#define GEN9_TRTT_NULL_TILE_REG		_MMIO(0x4DE8)
+#define GEN9_TRTT_INVD_TILE_REG		_MMIO(0x4DEC)
+
+#define GEN9_TRTT_VA_MASKDATA		_MMIO(0x4DF0)
+#define   GEN9_TRVA_MASK_VALUE		0xF0
+#define   GEN9_TRVA_DATA_VALUE		0xF
+
+#define GEN9_TRTT_TABLE_CONTROL		_MMIO(0x4DF4)
+#define   GEN9_TRTT_IN_GFX_VA_SPACE	(1<<1)
+#define   GEN9_TRTT_ENABLE		(1<<0)
+
 #define GAM_ECOCHK			_MMIO(0x4090)
 #define   BDW_DISABLE_HDC_INVALIDATION	(1<<25)
 #define   ECOCHK_SNB_BIT		(1<<10)
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 8096c6a..a8b795d 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -183,6 +183,12 @@
 #define CTX_LRI_HEADER_2		0x41
 #define CTX_R_PWR_CLK_STATE		0x42
 #define CTX_GPGPU_CSR_BASE_ADDRESS	0x44
+#define CTX_TRTT_L3_PTR_DW0		0x202
+#define CTX_TRTT_L3_PTR_DW1		0x204
+#define CTX_TRTT_NULL_TILE		0x206
+#define CTX_TRTT_INVD_TILE		0x208
+#define CTX_TRTT_VA_MASKDATA		0x20A
+#define CTX_TRTT_TBL_CTL		0x20C
 
 #define GEN8_CTX_VALID (1<<0)
 #define GEN8_CTX_FORCE_PD_RESTORE (1<<1)
@@ -228,6 +234,8 @@ enum {
 static int intel_lr_context_pin(struct drm_i915_gem_request *rq);
 static void lrc_setup_hardware_status_page(struct intel_engine_cs *ring,
 		struct drm_i915_gem_object *default_ctx_obj);
+static void populate_lr_context_trtt(struct intel_context *ctx,
+		uint32_t *reg_state);
 
 
 /**
@@ -390,6 +398,14 @@ static int execlists_update_context(struct drm_i915_gem_request *rq)
 		ASSIGN_CTX_PDP(ppgtt, reg_state, 0);
 	}
 
+	if (ring->id == RCS && rq->ctx->trtt_info.update_trtt_params) {
+		/* The same page of the context object also contain fields
+		 * related for TRTT setup.
+		 */
+		populate_lr_context_trtt(rq->ctx, reg_state);
+		rq->ctx->trtt_info.update_trtt_params = 0;
+	}
+
 	kunmap_atomic(reg_state);
 
 	return 0;
@@ -2247,6 +2263,31 @@ make_rpcs(struct drm_device *dev)
 	return rpcs;
 }
 
+static void
+populate_lr_context_trtt(struct intel_context *ctx, uint32_t *reg_state)
+{
+	unsigned long masked_l3_gfx_address =
+		ctx->trtt_info.l3_table_address & GEN9_TRTT_L3_GFXADDR_MASK;
+
+	ASSIGN_CTX_REG(reg_state, CTX_TRTT_L3_PTR_DW0, GEN9_TRTT_L3_POINTER_DW0,
+		       lower_32_bits(masked_l3_gfx_address));
+
+	ASSIGN_CTX_REG(reg_state, CTX_TRTT_L3_PTR_DW1, GEN9_TRTT_L3_POINTER_DW1,
+		       upper_32_bits(masked_l3_gfx_address));
+
+	ASSIGN_CTX_REG(reg_state, CTX_TRTT_NULL_TILE, GEN9_TRTT_NULL_TILE_REG,
+		       ctx->trtt_info.null_tile_val);
+
+	ASSIGN_CTX_REG(reg_state, CTX_TRTT_INVD_TILE, GEN9_TRTT_INVD_TILE_REG,
+		       ctx->trtt_info.invd_tile_val);
+
+	ASSIGN_CTX_REG(reg_state, CTX_TRTT_VA_MASKDATA, GEN9_TRTT_VA_MASKDATA,
+		       GEN9_TRVA_MASK_VALUE | GEN9_TRVA_DATA_VALUE);
+
+	ASSIGN_CTX_REG(reg_state, CTX_TRTT_TBL_CTL, GEN9_TRTT_TABLE_CONTROL,
+		       GEN9_TRTT_IN_GFX_VA_SPACE | GEN9_TRTT_ENABLE);
+}
+
 static int
 populate_lr_context(struct intel_context *ctx, struct drm_i915_gem_object *ctx_obj,
 		    struct intel_engine_cs *ring, struct intel_ringbuffer *ringbuf)
diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
index acf2102..6d6f448 100644
--- a/include/uapi/drm/i915_drm.h
+++ b/include/uapi/drm/i915_drm.h
@@ -357,6 +357,7 @@ typedef struct drm_i915_irq_wait {
 #define I915_PARAM_HAS_GPU_RESET	 35
 #define I915_PARAM_HAS_RESOURCE_STREAMER 36
 #define I915_PARAM_HAS_EXEC_SOFTPIN	 37
+#define I915_PARAM_HAS_TRTT		 38
 
 typedef struct drm_i915_getparam {
 	__s32 param;
@@ -1140,7 +1141,14 @@ struct drm_i915_gem_context_param {
 #define I915_CONTEXT_PARAM_BAN_PERIOD	0x1
 #define I915_CONTEXT_PARAM_NO_ZEROMAP	0x2
 #define I915_CONTEXT_PARAM_GTT_SIZE	0x3
+#define I915_CONTEXT_PARAM_ENABLE_TRTT	0x4
 	__u64 value;
 };
 
+struct drm_i915_gem_context_trtt_param {
+	__u64 l3_table_address;
+	__u32 invd_tile_val;
+	__u32 null_tile_val;
+};
+
 #endif /* _UAPI_I915_DRM_H_ */
-- 
1.9.2

_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/intel-gfx




[Index of Archives]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]
  Powered by Linux