drm/i915/tgl: Register state context definition for Gen12
authorMichel Thierry <michel.thierry@intel.com>
Fri, 6 Sep 2019 12:23:14 +0000 (15:23 +0300)
committerChris Wilson <chris@chris-wilson.co.uk>
Fri, 6 Sep 2019 17:12:58 +0000 (18:12 +0100)
Gen12 has subtle changes in the reg state context offsets (some fields
are gone, some are in a different location), compared to previous Gens.

The simplest approach seems to be keeping Gen12 (and future platform)
changes apart from the previous gens, while keeping the registers that
are contiguous in functions we can reuse.

v2: alias, virtual engine, rpcs, prune unused regs
v3: use engine base (Daniele), take ctx_bb for all

Bspec: 46255
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Tested-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
[ickle: Tweaked the GEM_WARN_ON after settling on a compromise with
Daniele]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190906122314.2146-2-mika.kuoppala@linux.intel.com
drivers/gpu/drm/i915/gt/intel_lrc.c
drivers/gpu/drm/i915/gt/intel_lrc_reg.h

index de86ca6add09e333b680b0d5fc3a2c8461ee333e..0ddfbebbcbbc2de3aedd40b1dc51455ef2f4178f 100644 (file)
@@ -808,8 +808,12 @@ static void virtual_update_register_offsets(u32 *regs,
 {
        u32 base = engine->mmio_base;
 
+       /* Refactor so that we only have one place that knows all the offsets! */
+       GEM_WARN_ON(INTEL_GEN(engine->i915) >= 12);
+
        /* Must match execlists_init_reg_state()! */
 
+       /* Common part */
        regs[CTX_CONTEXT_CONTROL] =
                i915_mmio_reg_offset(RING_CONTEXT_CONTROL(base));
        regs[CTX_RING_HEAD] = i915_mmio_reg_offset(RING_HEAD(base));
@@ -820,13 +824,16 @@ static void virtual_update_register_offsets(u32 *regs,
        regs[CTX_BB_HEAD_U] = i915_mmio_reg_offset(RING_BBADDR_UDW(base));
        regs[CTX_BB_HEAD_L] = i915_mmio_reg_offset(RING_BBADDR(base));
        regs[CTX_BB_STATE] = i915_mmio_reg_offset(RING_BBSTATE(base));
+
        regs[CTX_SECOND_BB_HEAD_U] =
                i915_mmio_reg_offset(RING_SBBADDR_UDW(base));
        regs[CTX_SECOND_BB_HEAD_L] = i915_mmio_reg_offset(RING_SBBADDR(base));
        regs[CTX_SECOND_BB_STATE] = i915_mmio_reg_offset(RING_SBBSTATE(base));
 
+       /* PPGTT part */
        regs[CTX_CTX_TIMESTAMP] =
                i915_mmio_reg_offset(RING_CTX_TIMESTAMP(base));
+
        regs[CTX_PDP3_UDW] = i915_mmio_reg_offset(GEN8_RING_PDP_UDW(base, 3));
        regs[CTX_PDP3_LDW] = i915_mmio_reg_offset(GEN8_RING_PDP_LDW(base, 3));
        regs[CTX_PDP2_UDW] = i915_mmio_reg_offset(GEN8_RING_PDP_UDW(base, 2));
@@ -3122,39 +3129,13 @@ static u32 intel_lr_indirect_ctx_offset(struct intel_engine_cs *engine)
        return indirect_ctx_offset;
 }
 
-static struct i915_ppgtt *vm_alias(struct i915_address_space *vm)
-{
-       if (i915_is_ggtt(vm))
-               return i915_vm_to_ggtt(vm)->alias;
-       else
-               return i915_vm_to_ppgtt(vm);
-}
 
-static void execlists_init_reg_state(u32 *regs,
-                                    struct intel_context *ce,
-                                    struct intel_engine_cs *engine,
-                                    struct intel_ring *ring)
+static void init_common_reg_state(u32 * const regs,
+                                 struct i915_ppgtt * const ppgtt,
+                                 struct intel_engine_cs *engine,
+                                 struct intel_ring *ring)
 {
-       struct i915_ppgtt *ppgtt = vm_alias(ce->vm);
-       const bool rcs = engine->class == RENDER_CLASS;
        const u32 base = engine->mmio_base;
-       const u32 lri_base =
-               intel_engine_has_relative_mmio(engine) ? MI_LRI_CS_MMIO : 0;
-
-       /*
-        * A context is actually a big batch buffer with several
-        * MI_LOAD_REGISTER_IMM commands followed by (reg, value) pairs. The
-        * values we are setting here are only for the first context restore:
-        * on a subsequent save, the GPU will recreate this batchbuffer with new
-        * values (including all the missing MI_LOAD_REGISTER_IMM commands that
-        * we are not initializing here).
-        *
-        * Must keep consistent with virtual_update_register_offsets().
-        */
-       regs[CTX_LRI_HEADER_0] =
-               MI_LOAD_REGISTER_IMM(rcs ? 14 : 11) |
-               MI_LRI_FORCE_POSTED |
-               lri_base;
 
        CTX_REG(regs, CTX_CONTEXT_CONTROL, RING_CONTEXT_CONTROL(base),
                _MASKED_BIT_DISABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT) |
@@ -3172,41 +3153,43 @@ static void execlists_init_reg_state(u32 *regs,
        CTX_REG(regs, CTX_BB_HEAD_U, RING_BBADDR_UDW(base), 0);
        CTX_REG(regs, CTX_BB_HEAD_L, RING_BBADDR(base), 0);
        CTX_REG(regs, CTX_BB_STATE, RING_BBSTATE(base), RING_BB_PPGTT);
-       CTX_REG(regs, CTX_SECOND_BB_HEAD_U, RING_SBBADDR_UDW(base), 0);
-       CTX_REG(regs, CTX_SECOND_BB_HEAD_L, RING_SBBADDR(base), 0);
-       CTX_REG(regs, CTX_SECOND_BB_STATE, RING_SBBSTATE(base), 0);
-       if (rcs) {
-               struct i915_ctx_workarounds *wa_ctx = &engine->wa_ctx;
-
-               CTX_REG(regs, CTX_RCS_INDIRECT_CTX, RING_INDIRECT_CTX(base), 0);
-               CTX_REG(regs, CTX_RCS_INDIRECT_CTX_OFFSET,
-                       RING_INDIRECT_CTX_OFFSET(base), 0);
-               if (wa_ctx->indirect_ctx.size) {
-                       u32 ggtt_offset = i915_ggtt_offset(wa_ctx->vma);
+}
 
-                       regs[CTX_RCS_INDIRECT_CTX + 1] =
-                               (ggtt_offset + wa_ctx->indirect_ctx.offset) |
-                               (wa_ctx->indirect_ctx.size / CACHELINE_BYTES);
+static void init_wa_bb_reg_state(u32 * const regs,
+                                struct intel_engine_cs *engine,
+                                u32 pos_bb_per_ctx)
+{
+       struct i915_ctx_workarounds * const wa_ctx = &engine->wa_ctx;
+       const u32 base = engine->mmio_base;
+       const u32 pos_indirect_ctx = pos_bb_per_ctx + 2;
+       const u32 pos_indirect_ctx_offset = pos_indirect_ctx + 2;
 
-                       regs[CTX_RCS_INDIRECT_CTX_OFFSET + 1] =
-                               intel_lr_indirect_ctx_offset(engine) << 6;
-               }
+       CTX_REG(regs, pos_indirect_ctx, RING_INDIRECT_CTX(base), 0);
+       CTX_REG(regs, pos_indirect_ctx_offset,
+               RING_INDIRECT_CTX_OFFSET(base), 0);
+       if (wa_ctx->indirect_ctx.size) {
+               const u32 ggtt_offset = i915_ggtt_offset(wa_ctx->vma);
 
-               CTX_REG(regs, CTX_BB_PER_CTX_PTR, RING_BB_PER_CTX_PTR(base), 0);
-               if (wa_ctx->per_ctx.size) {
-                       u32 ggtt_offset = i915_ggtt_offset(wa_ctx->vma);
+               regs[pos_indirect_ctx + 1] =
+                       (ggtt_offset + wa_ctx->indirect_ctx.offset) |
+                       (wa_ctx->indirect_ctx.size / CACHELINE_BYTES);
 
-                       regs[CTX_BB_PER_CTX_PTR + 1] =
-                               (ggtt_offset + wa_ctx->per_ctx.offset) | 0x01;
-               }
+               regs[pos_indirect_ctx_offset + 1] =
+                       intel_lr_indirect_ctx_offset(engine) << 6;
        }
 
-       regs[CTX_LRI_HEADER_1] =
-               MI_LOAD_REGISTER_IMM(9) |
-               MI_LRI_FORCE_POSTED |
-               lri_base;
+       CTX_REG(regs, pos_bb_per_ctx, RING_BB_PER_CTX_PTR(base), 0);
+       if (wa_ctx->per_ctx.size) {
+               const u32 ggtt_offset = i915_ggtt_offset(wa_ctx->vma);
 
-       CTX_REG(regs, CTX_CTX_TIMESTAMP, RING_CTX_TIMESTAMP(base), 0);
+               regs[pos_bb_per_ctx + 1] =
+                       (ggtt_offset + wa_ctx->per_ctx.offset) | 0x01;
+       }
+}
+
+static void init_ppgtt_reg_state(u32 *regs, u32 base,
+                                struct i915_ppgtt *ppgtt)
+{
        /* PDP values well be assigned later if needed */
        CTX_REG(regs, CTX_PDP3_UDW, GEN8_RING_PDP_UDW(base, 3), 0);
        CTX_REG(regs, CTX_PDP3_LDW, GEN8_RING_PDP_LDW(base, 3), 0);
@@ -3229,6 +3212,47 @@ static void execlists_init_reg_state(u32 *regs,
                ASSIGN_CTX_PDP(ppgtt, regs, 1);
                ASSIGN_CTX_PDP(ppgtt, regs, 0);
        }
+}
+
+static struct i915_ppgtt *vm_alias(struct i915_address_space *vm)
+{
+       if (i915_is_ggtt(vm))
+               return i915_vm_to_ggtt(vm)->alias;
+       else
+               return i915_vm_to_ppgtt(vm);
+}
+
+static void gen8_init_reg_state(u32 * const regs,
+                               struct intel_context *ce,
+                               struct intel_engine_cs *engine,
+                               struct intel_ring *ring)
+{
+       struct i915_ppgtt * const ppgtt = vm_alias(ce->vm);
+       const bool rcs = engine->class == RENDER_CLASS;
+       const u32 base = engine->mmio_base;
+       const u32 lri_base =
+               intel_engine_has_relative_mmio(engine) ? MI_LRI_CS_MMIO : 0;
+
+       regs[CTX_LRI_HEADER_0] =
+               MI_LOAD_REGISTER_IMM(rcs ? 14 : 11) |
+               MI_LRI_FORCE_POSTED |
+               lri_base;
+
+       init_common_reg_state(regs, ppgtt, engine, ring);
+       CTX_REG(regs, CTX_SECOND_BB_HEAD_U, RING_SBBADDR_UDW(base), 0);
+       CTX_REG(regs, CTX_SECOND_BB_HEAD_L, RING_SBBADDR(base), 0);
+       CTX_REG(regs, CTX_SECOND_BB_STATE, RING_SBBSTATE(base), 0);
+       if (rcs)
+               init_wa_bb_reg_state(regs, engine, CTX_BB_PER_CTX_PTR);
+
+       regs[CTX_LRI_HEADER_1] =
+               MI_LOAD_REGISTER_IMM(9) |
+               MI_LRI_FORCE_POSTED |
+               lri_base;
+
+       CTX_REG(regs, CTX_CTX_TIMESTAMP, RING_CTX_TIMESTAMP(base), 0);
+
+       init_ppgtt_reg_state(regs, base, ppgtt);
 
        if (rcs) {
                regs[CTX_LRI_HEADER_2] = MI_LOAD_REGISTER_IMM(1) | lri_base;
@@ -3240,6 +3264,66 @@ static void execlists_init_reg_state(u32 *regs,
                regs[CTX_END] |= BIT(0);
 }
 
+static void gen12_init_reg_state(u32 * const regs,
+                                struct intel_context *ce,
+                                struct intel_engine_cs *engine,
+                                struct intel_ring *ring)
+{
+       struct i915_ppgtt * const ppgtt = i915_vm_to_ppgtt(ce->vm);
+       const bool rcs = engine->class == RENDER_CLASS;
+       const u32 base = engine->mmio_base;
+       const u32 lri_base =
+               intel_engine_has_relative_mmio(engine) ? MI_LRI_CS_MMIO : 0;
+
+       regs[CTX_LRI_HEADER_0] =
+               MI_LOAD_REGISTER_IMM(rcs ? 11 : 9) |
+               MI_LRI_FORCE_POSTED |
+               lri_base;
+
+       init_common_reg_state(regs, ppgtt, engine, ring);
+
+       /* We want ctx_ptr for all engines to be set */
+       init_wa_bb_reg_state(regs, engine, GEN12_CTX_BB_PER_CTX_PTR);
+
+       regs[CTX_LRI_HEADER_1] =
+               MI_LOAD_REGISTER_IMM(9) |
+               MI_LRI_FORCE_POSTED |
+               lri_base;
+
+       CTX_REG(regs, CTX_CTX_TIMESTAMP, RING_CTX_TIMESTAMP(base), 0);
+
+       init_ppgtt_reg_state(regs, base, ppgtt);
+
+       if (rcs) {
+               regs[GEN12_CTX_LRI_HEADER_3] =
+                       MI_LOAD_REGISTER_IMM(1) | lri_base;
+               CTX_REG(regs, CTX_R_PWR_CLK_STATE, GEN8_R_PWR_CLK_STATE, 0);
+
+               /* TODO: oa_init_reg_state ? */
+       }
+}
+
+static void execlists_init_reg_state(u32 *regs,
+                                    struct intel_context *ce,
+                                    struct intel_engine_cs *engine,
+                                    struct intel_ring *ring)
+{
+       /*
+        * A context is actually a big batch buffer with several
+        * MI_LOAD_REGISTER_IMM commands followed by (reg, value) pairs. The
+        * values we are setting here are only for the first context restore:
+        * on a subsequent save, the GPU will recreate this batchbuffer with new
+        * values (including all the missing MI_LOAD_REGISTER_IMM commands that
+        * we are not initializing here).
+        *
+        * Must keep consistent with virtual_update_register_offsets().
+        */
+       if (INTEL_GEN(engine->i915) >= 12)
+               gen12_init_reg_state(regs, ce, engine, ring);
+       else
+               gen8_init_reg_state(regs, ce, engine, ring);
+}
+
 static int
 populate_lr_context(struct intel_context *ce,
                    struct drm_i915_gem_object *ctx_obj,
index b8f20ad711692f7843d15f1e27e92763d7eed75f..68caf85418660ccbaa88384a7c4594bb54483b86 100644 (file)
@@ -9,7 +9,7 @@
 
 #include <linux/types.h>
 
-/* GEN8+ Reg State Context */
+/* GEN8 to GEN11 Reg State Context */
 #define CTX_LRI_HEADER_0               0x01
 #define CTX_CONTEXT_CONTROL            0x02
 #define CTX_RING_HEAD                  0x04
 #define CTX_R_PWR_CLK_STATE            0x42
 #define CTX_END                                0x44
 
+/* GEN12+ Reg State Context */
+#define GEN12_CTX_BB_PER_CTX_PTR               0x12
+#define GEN12_CTX_LRI_HEADER_3                 0x41
+
 #define CTX_REG(reg_state, pos, reg, val) do { \
        u32 *reg_state__ = (reg_state); \
        const u32 pos__ = (pos); \