drm/amdgpu:implement CONTEXT_CONTROL (v5)

v1:
for gfx8, use CONTEXT_CONTROL package to dynamically
skip preamble CEIB and other load_xxx command in sequence.

v2:
support GFX7 as well.
remove cntxcntl in compute ring funcs because CPC doesn't
support this packet.

v3: fix reduntant judgement in cntxcntl.
v4: some cleanups, don't change cs_submit()
v5: keep old MESA supported & bump up KMS version.
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Ack-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -320,6 +320,7 @@ struct amdgpu_ring_funcs {
 	void (*begin_use)(struct amdgpu_ring *ring);
 	void (*end_use)(struct amdgpu_ring *ring);
 	void (*emit_switch_buffer) (struct amdgpu_ring *ring);
+	void (*emit_cntxcntl) (struct amdgpu_ring *ring, uint32_t flags);
 };
 
 /*
@@ -966,6 +967,7 @@ struct amdgpu_ctx {
 	spinlock_t		ring_lock;
 	struct fence            **fences;
 	struct amdgpu_ctx_ring	rings[AMDGPU_MAX_RINGS];
+	bool preamble_presented;
 };
 
 struct amdgpu_ctx_mgr {
@@ -1231,6 +1233,10 @@ struct amdgpu_cs_parser {
 	struct amdgpu_bo_list_entry	uf_entry;
 };
 
+#define AMDGPU_PREAMBLE_IB_PRESENT          (1 << 0) /* bit set means command submit involves a preamble IB */
+#define AMDGPU_PREAMBLE_IB_PRESENT_FIRST    (1 << 1) /* bit set means preamble IB is first presented in belonging context */
+#define AMDGPU_HAVE_CTX_SWITCH              (1 << 2) /* bit set means context switch occured */
+
 struct amdgpu_job {
 	struct amd_sched_job    base;
 	struct amdgpu_device	*adev;
@@ -1239,6 +1245,7 @@ struct amdgpu_job {
 	struct amdgpu_sync	sync;
 	struct amdgpu_ib	*ibs;
 	struct fence		*fence; /* the hw fence */
+	uint32_t		preamble_status;
 	uint32_t		num_ibs;
 	void			*owner;
 	uint64_t		fence_ctx; /* the fence_context this job uses */
@@ -2276,6 +2283,7 @@ amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
 #define amdgpu_ring_emit_hdp_flush(r) (r)->funcs->emit_hdp_flush((r))
 #define amdgpu_ring_emit_hdp_invalidate(r) (r)->funcs->emit_hdp_invalidate((r))
 #define amdgpu_ring_emit_switch_buffer(r) (r)->funcs->emit_switch_buffer((r))
+#define amdgpu_ring_emit_cntxcntl(r, d) (r)->funcs->emit_cntxcntl((r), (d))
 #define amdgpu_ring_pad_ib(r, ib) ((r)->funcs->pad_ib((r), (ib)))
 #define amdgpu_ring_init_cond_exec(r) (r)->funcs->init_cond_exec((r))
 #define amdgpu_ring_patch_cond_exec(r,o) (r)->funcs->patch_cond_exec((r),(o))

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c

@@ -850,6 +850,14 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 		if (r)
 			return r;
 
+		if (ib->flags & AMDGPU_IB_FLAG_PREAMBLE) {
+			parser->job->preamble_status |= AMDGPU_PREAMBLE_IB_PRESENT;
+			if (!parser->ctx->preamble_presented) {
+				parser->job->preamble_status |= AMDGPU_PREAMBLE_IB_PRESENT_FIRST;
+				parser->ctx->preamble_presented = true;
+			}
+		}
+
 		if (parser->job->ring && parser->job->ring != ring)
 			return -EINVAL;
 

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c

@@ -55,9 +55,10 @@
  * - 3.3.0 - Add VM support for UVD on supported hardware.
  * - 3.4.0 - Add AMDGPU_INFO_NUM_EVICTIONS.
  * - 3.5.0 - Add support for new UVD_NO_OP register.
+ * - 3.6.0 - kmd involves use CONTEXT_CONTROL in ring buffer.
  */
 #define KMS_DRIVER_MAJOR	3
-#define KMS_DRIVER_MINOR	5
+#define KMS_DRIVER_MINOR	6
 #define KMS_DRIVER_PATCHLEVEL	0
 
 int amdgpu_vram_limit = 0;

drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c

@@ -125,6 +125,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 	unsigned patch_offset = ~0;
 	struct amdgpu_vm *vm;
 	uint64_t fence_ctx;
+	uint32_t status = 0;
 
 	unsigned i;
 	int r = 0;
@@ -176,11 +177,20 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 
 	skip_preamble = ring->current_ctx == fence_ctx;
 	need_ctx_switch = ring->current_ctx != fence_ctx;
+	if (job && ring->funcs->emit_cntxcntl) {
+		if (need_ctx_switch)
+			status |= AMDGPU_HAVE_CTX_SWITCH;
+		status |= job->preamble_status;
+		amdgpu_ring_emit_cntxcntl(ring, status);
+	}
+
 	for (i = 0; i < num_ibs; ++i) {
 		ib = &ibs[i];
 
 		/* drop preamble IBs if we don't have a context switch */
-		if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) && skip_preamble)
+		if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) &&
+			skip_preamble &&
+			!(status & AMDGPU_PREAMBLE_IB_PRESENT_FIRST))
 			continue;
 
 		amdgpu_ring_emit_ib(ring, ib, job ? job->vm_id : 0,

drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c

@@ -2096,6 +2096,25 @@ static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, control);
 }
 
+static void gfx_v7_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
+{
+	uint32_t dw2 = 0;
+
+	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
+	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
+		/* set load_global_config & load_global_uconfig */
+		dw2 |= 0x8001;
+		/* set load_cs_sh_regs */
+		dw2 |= 0x01000000;
+		/* set load_per_context_state & load_gfx_sh_regs */
+		dw2 |= 0x10002;
+	}
+
+	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
+	amdgpu_ring_write(ring, dw2);
+	amdgpu_ring_write(ring, 0);
+}
+
 /**
  * gfx_v7_0_ring_test_ib - basic ring IB test
  *
@@ -4938,6 +4957,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
 	.test_ib = gfx_v7_0_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
 	.pad_ib = amdgpu_ring_generic_pad_ib,
+	.emit_cntxcntl = gfx_v7_ring_emit_cntxcntl,
 };
 
 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {

drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c

@@ -6076,6 +6076,35 @@ static void gfx_v8_ring_emit_sb(struct amdgpu_ring *ring)
 	amdgpu_ring_write(ring, 0);
 }
 
+static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
+{
+	uint32_t dw2 = 0;
+
+	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
+	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
+		/* set load_global_config & load_global_uconfig */
+		dw2 |= 0x8001;
+		/* set load_cs_sh_regs */
+		dw2 |= 0x01000000;
+		/* set load_per_context_state & load_gfx_sh_regs for GFX */
+		dw2 |= 0x10002;
+
+		/* set load_ce_ram if preamble presented */
+		if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
+			dw2 |= 0x10000000;
+	} else {
+		/* still load_ce_ram if this is the first time preamble presented
+		 * although there is no context switch happens.
+		 */
+		if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
+			dw2 |= 0x10000000;
+	}
+
+	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
+	amdgpu_ring_write(ring, dw2);
+	amdgpu_ring_write(ring, 0);
+}
+
 static void gfx_v8_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
 						 enum amdgpu_interrupt_state state)
 {
@@ -6258,6 +6287,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
 	.insert_nop = amdgpu_ring_insert_nop,
 	.pad_ib = amdgpu_ring_generic_pad_ib,
 	.emit_switch_buffer = gfx_v8_ring_emit_sb,
+	.emit_cntxcntl = gfx_v8_ring_emit_cntxcntl,
 };
 
 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {