diff --git a/drivers/gpu/drm/i915/gvt/execlist.c b/drivers/gpu/drm/i915/gvt/execlist.c index 3c38e2a2dba343ea6740b87e3daf4470fee2649e..f149847cbd17bcd2a649ace7440352ce70c0b2c9 100644 --- a/drivers/gpu/drm/i915/gvt/execlist.c +++ b/drivers/gpu/drm/i915/gvt/execlist.c @@ -167,7 +167,7 @@ static void emulate_csb_update(struct intel_vgpu_execlist *execlist, ring_id_to_context_switch_event(execlist->ring_id)); } -int emulate_execlist_ctx_schedule_out( +static int emulate_execlist_ctx_schedule_out( struct intel_vgpu_execlist *execlist, struct execlist_ctx_descriptor_format *ctx) { @@ -260,7 +260,7 @@ static struct intel_vgpu_execlist_slot *get_next_execlist_slot( return &execlist->slot[status.execlist_write_pointer]; } -int emulate_execlist_schedule_in(struct intel_vgpu_execlist *execlist, +static int emulate_execlist_schedule_in(struct intel_vgpu_execlist *execlist, struct execlist_ctx_descriptor_format ctx[2]) { struct intel_vgpu_execlist_slot *running = execlist->running_slot; @@ -353,6 +353,279 @@ int emulate_execlist_schedule_in(struct intel_vgpu_execlist *execlist, return 0; } +static void free_workload(struct intel_vgpu_workload *workload) +{ + intel_vgpu_unpin_mm(workload->shadow_mm); + intel_gvt_mm_unreference(workload->shadow_mm); + kmem_cache_free(workload->vgpu->workloads, workload); +} + +#define get_desc_from_elsp_dwords(ed, i) \ + ((struct execlist_ctx_descriptor_format *)&((ed)->data[i * 2])) + +static int prepare_execlist_workload(struct intel_vgpu_workload *workload) +{ + struct intel_vgpu *vgpu = workload->vgpu; + struct execlist_ctx_descriptor_format ctx[2]; + int ring_id = workload->ring_id; + + intel_vgpu_pin_mm(workload->shadow_mm); + intel_vgpu_sync_oos_pages(workload->vgpu); + intel_vgpu_flush_post_shadow(workload->vgpu); + if (!workload->emulate_schedule_in) + return 0; + + ctx[0] = *get_desc_from_elsp_dwords(&workload->elsp_dwords, 1); + ctx[1] = *get_desc_from_elsp_dwords(&workload->elsp_dwords, 0); + + return emulate_execlist_schedule_in(&vgpu->execlist[ring_id], ctx); +} + +static int complete_execlist_workload(struct intel_vgpu_workload *workload) +{ + struct intel_vgpu *vgpu = workload->vgpu; + struct intel_vgpu_execlist *execlist = + &vgpu->execlist[workload->ring_id]; + struct intel_vgpu_workload *next_workload; + struct list_head *next = workload_q_head(vgpu, workload->ring_id)->next; + bool lite_restore = false; + int ret; + + gvt_dbg_el("complete workload %p status %d\n", workload, + workload->status); + + if (workload->status) + goto out; + + if (!list_empty(workload_q_head(vgpu, workload->ring_id))) { + struct execlist_ctx_descriptor_format *this_desc, *next_desc; + + next_workload = container_of(next, + struct intel_vgpu_workload, list); + this_desc = &workload->ctx_desc; + next_desc = &next_workload->ctx_desc; + + lite_restore = same_context(this_desc, next_desc); + } + + if (lite_restore) { + gvt_dbg_el("next context == current - no schedule-out\n"); + free_workload(workload); + return 0; + } + + ret = emulate_execlist_ctx_schedule_out(execlist, &workload->ctx_desc); + if (ret) + goto err; +out: + free_workload(workload); + return 0; +err: + free_workload(workload); + return ret; +} + +#define RING_CTX_OFF(x) \ + offsetof(struct execlist_ring_context, x) + +static void read_guest_pdps(struct intel_vgpu *vgpu, + u64 ring_context_gpa, u32 pdp[8]) +{ + u64 gpa; + int i; + + gpa = ring_context_gpa + RING_CTX_OFF(pdp3_UDW.val); + + for (i = 0; i < 8; i++) + intel_gvt_hypervisor_read_gpa(vgpu, + gpa + i * 8, &pdp[7 - i], 4); +} + +static int prepare_mm(struct intel_vgpu_workload *workload) +{ + struct execlist_ctx_descriptor_format *desc = &workload->ctx_desc; + struct intel_vgpu_mm *mm; + int page_table_level; + u32 pdp[8]; + + if (desc->addressing_mode == 1) { /* legacy 32-bit */ + page_table_level = 3; + } else if (desc->addressing_mode == 3) { /* legacy 64 bit */ + page_table_level = 4; + } else { + gvt_err("Advanced Context mode(SVM) is not supported!\n"); + return -EINVAL; + } + + read_guest_pdps(workload->vgpu, workload->ring_context_gpa, pdp); + + mm = intel_vgpu_find_ppgtt_mm(workload->vgpu, page_table_level, pdp); + if (mm) { + intel_gvt_mm_reference(mm); + } else { + + mm = intel_vgpu_create_mm(workload->vgpu, INTEL_GVT_MM_PPGTT, + pdp, page_table_level, 0); + if (IS_ERR(mm)) { + gvt_err("fail to create mm object.\n"); + return PTR_ERR(mm); + } + } + workload->shadow_mm = mm; + return 0; +} + +#define get_last_workload(q) \ + (list_empty(q) ? NULL : container_of(q->prev, \ + struct intel_vgpu_workload, list)) + +bool submit_context(struct intel_vgpu *vgpu, int ring_id, + struct execlist_ctx_descriptor_format *desc, + bool emulate_schedule_in) +{ + struct list_head *q = workload_q_head(vgpu, ring_id); + struct intel_vgpu_workload *last_workload = get_last_workload(q); + struct intel_vgpu_workload *workload = NULL; + u64 ring_context_gpa; + u32 head, tail, start, ctl, ctx_ctl; + int ret; + + ring_context_gpa = intel_vgpu_gma_to_gpa(vgpu->gtt.ggtt_mm, + (u32)((desc->lrca + 1) << GTT_PAGE_SHIFT)); + if (ring_context_gpa == INTEL_GVT_INVALID_ADDR) { + gvt_err("invalid guest context LRCA: %x\n", desc->lrca); + return -EINVAL; + } + + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(ring_header.val), &head, 4); + + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(ring_tail.val), &tail, 4); + + head &= RB_HEAD_OFF_MASK; + tail &= RB_TAIL_OFF_MASK; + + if (last_workload && same_context(&last_workload->ctx_desc, desc)) { + gvt_dbg_el("ring id %d cur workload == last\n", ring_id); + gvt_dbg_el("ctx head %x real head %lx\n", head, + last_workload->rb_tail); + /* + * cannot use guest context head pointer here, + * as it might not be updated at this time + */ + head = last_workload->rb_tail; + } + + gvt_dbg_el("ring id %d begin a new workload\n", ring_id); + + workload = kmem_cache_zalloc(vgpu->workloads, GFP_KERNEL); + if (!workload) + return -ENOMEM; + + /* record some ring buffer register values for scan and shadow */ + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(rb_start.val), &start, 4); + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(rb_ctrl.val), &ctl, 4); + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(ctx_ctrl.val), &ctx_ctl, 4); + + INIT_LIST_HEAD(&workload->list); + + init_waitqueue_head(&workload->shadow_ctx_status_wq); + atomic_set(&workload->shadow_ctx_active, 0); + + workload->vgpu = vgpu; + workload->ring_id = ring_id; + workload->ctx_desc = *desc; + workload->ring_context_gpa = ring_context_gpa; + workload->rb_head = head; + workload->rb_tail = tail; + workload->rb_start = start; + workload->rb_ctl = ctl; + workload->prepare = prepare_execlist_workload; + workload->complete = complete_execlist_workload; + workload->status = -EINPROGRESS; + workload->emulate_schedule_in = emulate_schedule_in; + + if (emulate_schedule_in) + memcpy(&workload->elsp_dwords, + &vgpu->execlist[ring_id].elsp_dwords, + sizeof(workload->elsp_dwords)); + + gvt_dbg_el("workload %p ring id %d head %x tail %x start %x ctl %x\n", + workload, ring_id, head, tail, start, ctl); + + gvt_dbg_el("workload %p emulate schedule_in %d\n", workload, + emulate_schedule_in); + + ret = prepare_mm(workload); + if (ret) { + kmem_cache_free(vgpu->workloads, workload); + return ret; + } + + queue_workload(workload); + return 0; +} + +int intel_vgpu_submit_execlist(struct intel_vgpu *vgpu, int ring_id) +{ + struct intel_vgpu_execlist *execlist = &vgpu->execlist[ring_id]; + struct execlist_ctx_descriptor_format *desc[2], valid_desc[2]; + unsigned long valid_desc_bitmap = 0; + bool emulate_schedule_in = true; + int ret; + int i; + + memset(valid_desc, 0, sizeof(valid_desc)); + + desc[0] = get_desc_from_elsp_dwords(&execlist->elsp_dwords, 1); + desc[1] = get_desc_from_elsp_dwords(&execlist->elsp_dwords, 0); + + for (i = 0; i < 2; i++) { + if (!desc[i]->valid) + continue; + + if (!desc[i]->privilege_access) { + gvt_err("vgpu%d: unexpected GGTT elsp submission\n", + vgpu->id); + return -EINVAL; + } + + /* TODO: add another guest context checks here. */ + set_bit(i, &valid_desc_bitmap); + valid_desc[i] = *desc[i]; + } + + if (!valid_desc_bitmap) { + gvt_err("vgpu%d: no valid desc in a elsp submission\n", + vgpu->id); + return -EINVAL; + } + + if (!test_bit(0, (void *)&valid_desc_bitmap) && + test_bit(1, (void *)&valid_desc_bitmap)) { + gvt_err("vgpu%d: weird elsp submission, desc 0 is not valid\n", + vgpu->id); + return -EINVAL; + } + + /* submit workload */ + for_each_set_bit(i, (void *)&valid_desc_bitmap, 2) { + ret = submit_context(vgpu, ring_id, &valid_desc[i], + emulate_schedule_in); + if (ret) { + gvt_err("vgpu%d: fail to schedule workload\n", + vgpu->id); + return ret; + } + emulate_schedule_in = false; + } + return 0; +} + static void init_vgpu_execlist(struct intel_vgpu *vgpu, int ring_id) { struct intel_vgpu_execlist *execlist = &vgpu->execlist[ring_id]; @@ -374,13 +647,28 @@ static void init_vgpu_execlist(struct intel_vgpu *vgpu, int ring_id) vgpu_vreg(vgpu, ctx_status_ptr_reg) = ctx_status_ptr.dw; } +void intel_vgpu_clean_execlist(struct intel_vgpu *vgpu) +{ + kmem_cache_destroy(vgpu->workloads); +} + int intel_vgpu_init_execlist(struct intel_vgpu *vgpu) { int i; /* each ring has a virtual execlist engine */ - for (i = 0; i < I915_NUM_ENGINES; i++) + for (i = 0; i < I915_NUM_ENGINES; i++) { init_vgpu_execlist(vgpu, i); + INIT_LIST_HEAD(&vgpu->workload_q_head[i]); + } + + vgpu->workloads = kmem_cache_create("gvt-g vgpu workload", + sizeof(struct intel_vgpu_workload), 0, + SLAB_HWCACHE_ALIGN, + NULL); + + if (!vgpu->workloads) + return -ENOMEM; return 0; } diff --git a/drivers/gpu/drm/i915/gvt/execlist.h b/drivers/gpu/drm/i915/gvt/execlist.h index 428f54358ae85bc07439ed0466986d5c3313e87e..13614fb4e5724f0d45f4556da6813285a172f3bc 100644 --- a/drivers/gpu/drm/i915/gvt/execlist.h +++ b/drivers/gpu/drm/i915/gvt/execlist.h @@ -118,6 +118,49 @@ struct execlist_context_status_format { }; }; +struct execlist_mmio_pair { + u32 addr; + u32 val; +}; + +/* The first 52 dwords in register state context */ +struct execlist_ring_context { + u32 nop1; + u32 lri_cmd_1; + struct execlist_mmio_pair ctx_ctrl; + struct execlist_mmio_pair ring_header; + struct execlist_mmio_pair ring_tail; + struct execlist_mmio_pair rb_start; + struct execlist_mmio_pair rb_ctrl; + struct execlist_mmio_pair bb_cur_head_UDW; + struct execlist_mmio_pair bb_cur_head_LDW; + struct execlist_mmio_pair bb_state; + struct execlist_mmio_pair second_bb_addr_UDW; + struct execlist_mmio_pair second_bb_addr_LDW; + struct execlist_mmio_pair second_bb_state; + struct execlist_mmio_pair bb_per_ctx_ptr; + struct execlist_mmio_pair rcs_indirect_ctx; + struct execlist_mmio_pair rcs_indirect_ctx_offset; + u32 nop2; + u32 nop3; + u32 nop4; + u32 lri_cmd_2; + struct execlist_mmio_pair ctx_timestamp; + struct execlist_mmio_pair pdp3_UDW; + struct execlist_mmio_pair pdp3_LDW; + struct execlist_mmio_pair pdp2_UDW; + struct execlist_mmio_pair pdp2_LDW; + struct execlist_mmio_pair pdp1_UDW; + struct execlist_mmio_pair pdp1_LDW; + struct execlist_mmio_pair pdp0_UDW; + struct execlist_mmio_pair pdp0_LDW; +}; + +struct intel_vgpu_elsp_dwords { + u32 data[4]; + u32 index; +}; + struct intel_vgpu_execlist_slot { struct execlist_ctx_descriptor_format ctx[2]; u32 index; @@ -130,8 +173,13 @@ struct intel_vgpu_execlist { struct execlist_ctx_descriptor_format *running_context; int ring_id; struct intel_vgpu *vgpu; + struct intel_vgpu_elsp_dwords elsp_dwords; }; +void intel_vgpu_clean_execlist(struct intel_vgpu *vgpu); + int intel_vgpu_init_execlist(struct intel_vgpu *vgpu); +int intel_vgpu_submit_execlist(struct intel_vgpu *vgpu, int ring_id); + #endif /*_GVT_EXECLIST_H_*/ diff --git a/drivers/gpu/drm/i915/gvt/gvt.h b/drivers/gpu/drm/i915/gvt/gvt.h index 82f932a24e7d41a2f3924ad65800813757cd0819..414163695e07addf4d6bc4e8a2e60532182b54bd 100644 --- a/drivers/gpu/drm/i915/gvt/gvt.h +++ b/drivers/gpu/drm/i915/gvt/gvt.h @@ -42,6 +42,7 @@ #include "display.h" #include "edid.h" #include "execlist.h" +#include "scheduler.h" #define GVT_MAX_VGPU 8 @@ -147,8 +148,9 @@ struct intel_vgpu { struct intel_vgpu_gtt gtt; struct intel_vgpu_opregion opregion; struct intel_vgpu_display display; - /* TODO: move the declaration of intel_gvt.h to a proper place. */ struct intel_vgpu_execlist execlist[I915_NUM_ENGINES]; + struct list_head workload_q_head[I915_NUM_ENGINES]; + struct kmem_cache *workloads; }; struct intel_gvt_gm { @@ -193,6 +195,7 @@ struct intel_gvt { struct intel_gvt_irq irq; struct intel_gvt_gtt gtt; struct intel_gvt_opregion opregion; + struct intel_gvt_workload_scheduler scheduler; struct task_struct *service_thread; wait_queue_head_t service_thread_wq; diff --git a/drivers/gpu/drm/i915/gvt/handlers.c b/drivers/gpu/drm/i915/gvt/handlers.c index 194778b374fff789b1ab1df64bec615848296c82..970804aed381d597d22ebb8b5cc934bedd509991 100644 --- a/drivers/gpu/drm/i915/gvt/handlers.c +++ b/drivers/gpu/drm/i915/gvt/handlers.c @@ -128,6 +128,18 @@ static int new_mmio_info(struct intel_gvt *gvt, return 0; } +static int render_mmio_to_ring_id(struct intel_gvt *gvt, unsigned int reg) +{ + int i; + + reg &= ~GENMASK(11, 0); + for (i = 0; i < I915_NUM_ENGINES; i++) { + if (gvt->dev_priv->engine[i].mmio_base == reg) + return i; + } + return -1; +} + #define offset_to_fence_num(offset) \ ((offset - i915_mmio_reg_offset(FENCE_REG_GEN6_LO(0))) >> 3) @@ -1262,6 +1274,28 @@ static int ring_timestamp_mmio_read(struct intel_vgpu *vgpu, return intel_vgpu_default_mmio_read(vgpu, offset, p_data, bytes); } +static int elsp_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + int ring_id = render_mmio_to_ring_id(vgpu->gvt, offset); + struct intel_vgpu_execlist *execlist; + u32 data = *(u32 *)p_data; + int ret; + + if (WARN_ON(ring_id < 0)) + return -EINVAL; + + execlist = &vgpu->execlist[ring_id]; + + execlist->elsp_dwords.data[execlist->elsp_dwords.index] = data; + if (execlist->elsp_dwords.index == 3) + ret = intel_vgpu_submit_execlist(vgpu, ring_id); + + ++execlist->elsp_dwords.index; + execlist->elsp_dwords.index &= 0x3; + return 0; +} + #define MMIO_F(reg, s, f, am, rm, d, r, w) do { \ ret = new_mmio_info(gvt, INTEL_GVT_MMIO_OFFSET(reg), \ f, s, am, rm, d, r, w); \ @@ -2169,8 +2203,8 @@ static int init_broadwell_mmio_info(struct intel_gvt *gvt) MMIO_RING_D(RING_ACTHD_UDW, D_BDW_PLUS); #define RING_REG(base) (base + 0x230) - MMIO_RING_DFH(RING_REG, D_BDW_PLUS, 0, NULL, NULL); - MMIO_DH(RING_REG(GEN8_BSD2_RING_BASE), D_BDW_PLUS, NULL, NULL); + MMIO_RING_DFH(RING_REG, D_BDW_PLUS, 0, NULL, elsp_mmio_write); + MMIO_DH(RING_REG(GEN8_BSD2_RING_BASE), D_BDW_PLUS, NULL, elsp_mmio_write); #undef RING_REG #define RING_REG(base) (base + 0x234) diff --git a/drivers/gpu/drm/i915/gvt/reg.h b/drivers/gpu/drm/i915/gvt/reg.h index 4842cb97c892a575407f4824984d7189cd4daf32..0dfe789d8f02b64ade88381b2a69fbca09f2862a 100644 --- a/drivers/gpu/drm/i915/gvt/reg.h +++ b/drivers/gpu/drm/i915/gvt/reg.h @@ -72,4 +72,9 @@ #define FORCEWAKE_ACK_MEDIA_GEN9_REG 0x0D88 #define FORCEWAKE_ACK_HSW_REG 0x130044 +#define RB_HEAD_OFF_MASK ((1U << 21) - (1U << 2)) +#define RB_TAIL_OFF_MASK ((1U << 21) - (1U << 3)) +#define RB_TAIL_SIZE_MASK ((1U << 21) - (1U << 12)) +#define _RING_CTL_BUF_SIZE(ctl) (((ctl) & RB_TAIL_SIZE_MASK) + GTT_PAGE_SIZE) + #endif diff --git a/drivers/gpu/drm/i915/gvt/scheduler.h b/drivers/gpu/drm/i915/gvt/scheduler.h new file mode 100644 index 0000000000000000000000000000000000000000..8884749f0bd4d113f2d0fbfdec383345a41cadae --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/scheduler.h @@ -0,0 +1,64 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#ifndef _GVT_SCHEDULER_H_ +#define _GVT_SCHEDULER_H_ + +struct intel_gvt_workload_scheduler { + struct list_head workload_q_head[I915_NUM_ENGINES]; +}; + +struct intel_vgpu_workload { + struct intel_vgpu *vgpu; + int ring_id; + struct drm_i915_gem_request *req; + /* if this workload has been dispatched to i915? */ + bool dispatched; + int status; + + struct intel_vgpu_mm *shadow_mm; + + /* different submission model may need different handler */ + int (*prepare)(struct intel_vgpu_workload *); + int (*complete)(struct intel_vgpu_workload *); + struct list_head list; + + /* execlist context information */ + struct execlist_ctx_descriptor_format ctx_desc; + struct execlist_ring_context *ring_context; + unsigned long rb_head, rb_tail, rb_ctl, rb_start; + struct intel_vgpu_elsp_dwords elsp_dwords; + bool emulate_schedule_in; + atomic_t shadow_ctx_active; + wait_queue_head_t shadow_ctx_status_wq; + u64 ring_context_gpa; +}; + +#define workload_q_head(vgpu, ring_id) \ + (&(vgpu->workload_q_head[ring_id])) + +#define queue_workload(workload) \ + list_add_tail(&workload->list, \ + workload_q_head(workload->vgpu, workload->ring_id)) + +#endif diff --git a/drivers/gpu/drm/i915/gvt/vgpu.c b/drivers/gpu/drm/i915/gvt/vgpu.c index e806b0e4035a5f02e1565cdebf7417ee5dc1729e..705a23c1ed857f28bdf1ab25176a3e8939fdf5b0 100644 --- a/drivers/gpu/drm/i915/gvt/vgpu.c +++ b/drivers/gpu/drm/i915/gvt/vgpu.c @@ -146,6 +146,7 @@ void intel_gvt_destroy_vgpu(struct intel_vgpu *vgpu) vgpu->active = false; idr_remove(&gvt->vgpu_idr, vgpu->id); + intel_vgpu_clean_execlist(vgpu); intel_vgpu_clean_display(vgpu); intel_vgpu_clean_opregion(vgpu); intel_vgpu_clean_gtt(vgpu);