diff options
Diffstat (limited to 'drivers/gpu/drm/i915/gt/intel_ringbuffer.c')
-rw-r--r-- | drivers/gpu/drm/i915/gt/intel_ringbuffer.c | 2337 |
1 files changed, 2337 insertions, 0 deletions
diff --git a/drivers/gpu/drm/i915/gt/intel_ringbuffer.c b/drivers/gpu/drm/i915/gt/intel_ringbuffer.c new file mode 100644 index 000000000000..f0d60affdba3 --- /dev/null +++ b/drivers/gpu/drm/i915/gt/intel_ringbuffer.c @@ -0,0 +1,2337 @@ +/* + * Copyright © 2008-2010 Intel Corporation + * + * 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. + * + * Authors: + * Eric Anholt <eric@anholt.net> + * Zou Nan hai <nanhai.zou@intel.com> + * Xiang Hai hao<haihao.xiang@intel.com> + * + */ + +#include <linux/log2.h> + +#include <drm/i915_drm.h> + +#include "i915_drv.h" +#include "i915_gem_render_state.h" +#include "i915_trace.h" +#include "intel_reset.h" +#include "intel_workarounds.h" + +/* Rough estimate of the typical request size, performing a flush, + * set-context and then emitting the batch. + */ +#define LEGACY_REQUEST_SIZE 200 + +unsigned int intel_ring_update_space(struct intel_ring *ring) +{ + unsigned int space; + + space = __intel_ring_space(ring->head, ring->emit, ring->size); + + ring->space = space; + return space; +} + +static int +gen2_render_ring_flush(struct i915_request *rq, u32 mode) +{ + unsigned int num_store_dw; + u32 cmd, *cs; + + cmd = MI_FLUSH; + num_store_dw = 0; + if (mode & EMIT_INVALIDATE) + cmd |= MI_READ_FLUSH; + if (mode & EMIT_FLUSH) + num_store_dw = 4; + + cs = intel_ring_begin(rq, 2 + 3 * num_store_dw); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + *cs++ = cmd; + while (num_store_dw--) { + *cs++ = MI_STORE_DWORD_IMM | MI_MEM_VIRTUAL; + *cs++ = i915_scratch_offset(rq->i915); + *cs++ = 0; + } + *cs++ = MI_FLUSH | MI_NO_WRITE_FLUSH; + + intel_ring_advance(rq, cs); + + return 0; +} + +static int +gen4_render_ring_flush(struct i915_request *rq, u32 mode) +{ + u32 cmd, *cs; + int i; + + /* + * read/write caches: + * + * I915_GEM_DOMAIN_RENDER is always invalidated, but is + * only flushed if MI_NO_WRITE_FLUSH is unset. On 965, it is + * also flushed at 2d versus 3d pipeline switches. + * + * read-only caches: + * + * I915_GEM_DOMAIN_SAMPLER is flushed on pre-965 if + * MI_READ_FLUSH is set, and is always flushed on 965. + * + * I915_GEM_DOMAIN_COMMAND may not exist? + * + * I915_GEM_DOMAIN_INSTRUCTION, which exists on 965, is + * invalidated when MI_EXE_FLUSH is set. + * + * I915_GEM_DOMAIN_VERTEX, which exists on 965, is + * invalidated with every MI_FLUSH. + * + * TLBs: + * + * On 965, TLBs associated with I915_GEM_DOMAIN_COMMAND + * and I915_GEM_DOMAIN_CPU in are invalidated at PTE write and + * I915_GEM_DOMAIN_RENDER and I915_GEM_DOMAIN_SAMPLER + * are flushed at any MI_FLUSH. + */ + + cmd = MI_FLUSH; + if (mode & EMIT_INVALIDATE) { + cmd |= MI_EXE_FLUSH; + if (IS_G4X(rq->i915) || IS_GEN(rq->i915, 5)) + cmd |= MI_INVALIDATE_ISP; + } + + i = 2; + if (mode & EMIT_INVALIDATE) + i += 20; + + cs = intel_ring_begin(rq, i); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + *cs++ = cmd; + + /* + * A random delay to let the CS invalidate take effect? Without this + * delay, the GPU relocation path fails as the CS does not see + * the updated contents. Just as important, if we apply the flushes + * to the EMIT_FLUSH branch (i.e. immediately after the relocation + * write and before the invalidate on the next batch), the relocations + * still fail. This implies that is a delay following invalidation + * that is required to reset the caches as opposed to a delay to + * ensure the memory is written. + */ + if (mode & EMIT_INVALIDATE) { + *cs++ = GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE; + *cs++ = i915_scratch_offset(rq->i915) | PIPE_CONTROL_GLOBAL_GTT; + *cs++ = 0; + *cs++ = 0; + + for (i = 0; i < 12; i++) + *cs++ = MI_FLUSH; + + *cs++ = GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE; + *cs++ = i915_scratch_offset(rq->i915) | PIPE_CONTROL_GLOBAL_GTT; + *cs++ = 0; + *cs++ = 0; + } + + *cs++ = cmd; + + intel_ring_advance(rq, cs); + + return 0; +} + +/* + * Emits a PIPE_CONTROL with a non-zero post-sync operation, for + * implementing two workarounds on gen6. From section 1.4.7.1 + * "PIPE_CONTROL" of the Sandy Bridge PRM volume 2 part 1: + * + * [DevSNB-C+{W/A}] Before any depth stall flush (including those + * produced by non-pipelined state commands), software needs to first + * send a PIPE_CONTROL with no bits set except Post-Sync Operation != + * 0. + * + * [Dev-SNB{W/A}]: Before a PIPE_CONTROL with Write Cache Flush Enable + * =1, a PIPE_CONTROL with any non-zero post-sync-op is required. + * + * And the workaround for these two requires this workaround first: + * + * [Dev-SNB{W/A}]: Pipe-control with CS-stall bit set must be sent + * BEFORE the pipe-control with a post-sync op and no write-cache + * flushes. + * + * And this last workaround is tricky because of the requirements on + * that bit. From section 1.4.7.2.3 "Stall" of the Sandy Bridge PRM + * volume 2 part 1: + * + * "1 of the following must also be set: + * - Render Target Cache Flush Enable ([12] of DW1) + * - Depth Cache Flush Enable ([0] of DW1) + * - Stall at Pixel Scoreboard ([1] of DW1) + * - Depth Stall ([13] of DW1) + * - Post-Sync Operation ([13] of DW1) + * - Notify Enable ([8] of DW1)" + * + * The cache flushes require the workaround flush that triggered this + * one, so we can't use it. Depth stall would trigger the same. + * Post-sync nonzero is what triggered this second workaround, so we + * can't use that one either. Notify enable is IRQs, which aren't + * really our business. That leaves only stall at scoreboard. + */ +static int +gen6_emit_post_sync_nonzero_flush(struct i915_request *rq) +{ + u32 scratch_addr = i915_scratch_offset(rq->i915) + 2 * CACHELINE_BYTES; + u32 *cs; + + cs = intel_ring_begin(rq, 6); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + *cs++ = GFX_OP_PIPE_CONTROL(5); + *cs++ = PIPE_CONTROL_CS_STALL | PIPE_CONTROL_STALL_AT_SCOREBOARD; + *cs++ = scratch_addr | PIPE_CONTROL_GLOBAL_GTT; + *cs++ = 0; /* low dword */ + *cs++ = 0; /* high dword */ + *cs++ = MI_NOOP; + intel_ring_advance(rq, cs); + + cs = intel_ring_begin(rq, 6); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + *cs++ = GFX_OP_PIPE_CONTROL(5); + *cs++ = PIPE_CONTROL_QW_WRITE; + *cs++ = scratch_addr | PIPE_CONTROL_GLOBAL_GTT; + *cs++ = 0; + *cs++ = 0; + *cs++ = MI_NOOP; + intel_ring_advance(rq, cs); + + return 0; +} + +static int +gen6_render_ring_flush(struct i915_request *rq, u32 mode) +{ + u32 scratch_addr = i915_scratch_offset(rq->i915) + 2 * CACHELINE_BYTES; + u32 *cs, flags = 0; + int ret; + + /* Force SNB workarounds for PIPE_CONTROL flushes */ + ret = gen6_emit_post_sync_nonzero_flush(rq); + if (ret) + return ret; + + /* Just flush everything. Experiments have shown that reducing the + * number of bits based on the write domains has little performance + * impact. + */ + if (mode & EMIT_FLUSH) { + flags |= PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH; + flags |= PIPE_CONTROL_DEPTH_CACHE_FLUSH; + /* + * Ensure that any following seqno writes only happen + * when the render cache is indeed flushed. + */ + flags |= PIPE_CONTROL_CS_STALL; + } + if (mode & EMIT_INVALIDATE) { + flags |= PIPE_CONTROL_TLB_INVALIDATE; + flags |= PIPE_CONTROL_INSTRUCTION_CACHE_INVALIDATE; + flags |= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE; + flags |= PIPE_CONTROL_VF_CACHE_INVALIDATE; + flags |= PIPE_CONTROL_CONST_CACHE_INVALIDATE; + flags |= PIPE_CONTROL_STATE_CACHE_INVALIDATE; + /* + * TLB invalidate requires a post-sync write. + */ + flags |= PIPE_CONTROL_QW_WRITE | PIPE_CONTROL_CS_STALL; + } + + cs = intel_ring_begin(rq, 4); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + *cs++ = GFX_OP_PIPE_CONTROL(4); + *cs++ = flags; + *cs++ = scratch_addr | PIPE_CONTROL_GLOBAL_GTT; + *cs++ = 0; + intel_ring_advance(rq, cs); + + return 0; +} + +static u32 *gen6_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs) +{ + /* First we do the gen6_emit_post_sync_nonzero_flush w/a */ + *cs++ = GFX_OP_PIPE_CONTROL(4); + *cs++ = PIPE_CONTROL_CS_STALL | PIPE_CONTROL_STALL_AT_SCOREBOARD; + *cs++ = 0; + *cs++ = 0; + + *cs++ = GFX_OP_PIPE_CONTROL(4); + *cs++ = PIPE_CONTROL_QW_WRITE; + *cs++ = i915_scratch_offset(rq->i915) | PIPE_CONTROL_GLOBAL_GTT; + *cs++ = 0; + + /* Finally we can flush and with it emit the breadcrumb */ + *cs++ = GFX_OP_PIPE_CONTROL(4); + *cs++ = (PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH | + PIPE_CONTROL_DEPTH_CACHE_FLUSH | + PIPE_CONTROL_DC_FLUSH_ENABLE | + PIPE_CONTROL_QW_WRITE | + PIPE_CONTROL_CS_STALL); + *cs++ = rq->timeline->hwsp_offset | PIPE_CONTROL_GLOBAL_GTT; + *cs++ = rq->fence.seqno; + + *cs++ = MI_USER_INTERRUPT; + *cs++ = MI_NOOP; + + rq->tail = intel_ring_offset(rq, cs); + assert_ring_tail_valid(rq->ring, rq->tail); + + return cs; +} + +static int +gen7_render_ring_cs_stall_wa(struct i915_request *rq) +{ + u32 *cs; + + cs = intel_ring_begin(rq, 4); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + *cs++ = GFX_OP_PIPE_CONTROL(4); + *cs++ = PIPE_CONTROL_CS_STALL | PIPE_CONTROL_STALL_AT_SCOREBOARD; + *cs++ = 0; + *cs++ = 0; + intel_ring_advance(rq, cs); + + return 0; +} + +static int +gen7_render_ring_flush(struct i915_request *rq, u32 mode) +{ + u32 scratch_addr = i915_scratch_offset(rq->i915) + 2 * CACHELINE_BYTES; + u32 *cs, flags = 0; + + /* + * Ensure that any following seqno writes only happen when the render + * cache is indeed flushed. + * + * Workaround: 4th PIPE_CONTROL command (except the ones with only + * read-cache invalidate bits set) must have the CS_STALL bit set. We + * don't try to be clever and just set it unconditionally. + */ + flags |= PIPE_CONTROL_CS_STALL; + + /* Just flush everything. Experiments have shown that reducing the + * number of bits based on the write domains has little performance + * impact. + */ + if (mode & EMIT_FLUSH) { + flags |= PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH; + flags |= PIPE_CONTROL_DEPTH_CACHE_FLUSH; + flags |= PIPE_CONTROL_DC_FLUSH_ENABLE; + flags |= PIPE_CONTROL_FLUSH_ENABLE; + } + if (mode & EMIT_INVALIDATE) { + flags |= PIPE_CONTROL_TLB_INVALIDATE; + flags |= PIPE_CONTROL_INSTRUCTION_CACHE_INVALIDATE; + flags |= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE; + flags |= PIPE_CONTROL_VF_CACHE_INVALIDATE; + flags |= PIPE_CONTROL_CONST_CACHE_INVALIDATE; + flags |= PIPE_CONTROL_STATE_CACHE_INVALIDATE; + flags |= PIPE_CONTROL_MEDIA_STATE_CLEAR; + /* + * TLB invalidate requires a post-sync write. + */ + flags |= PIPE_CONTROL_QW_WRITE; + flags |= PIPE_CONTROL_GLOBAL_GTT_IVB; + + flags |= PIPE_CONTROL_STALL_AT_SCOREBOARD; + + /* Workaround: we must issue a pipe_control with CS-stall bit + * set before a pipe_control command that has the state cache + * invalidate bit set. */ + gen7_render_ring_cs_stall_wa(rq); + } + + cs = intel_ring_begin(rq, 4); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + *cs++ = GFX_OP_PIPE_CONTROL(4); + *cs++ = flags; + *cs++ = scratch_addr; + *cs++ = 0; + intel_ring_advance(rq, cs); + + return 0; +} + +static u32 *gen7_rcs_emit_breadcrumb(struct i915_request *rq, u32 *cs) +{ + *cs++ = GFX_OP_PIPE_CONTROL(4); + *cs++ = (PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH | + PIPE_CONTROL_DEPTH_CACHE_FLUSH | + PIPE_CONTROL_DC_FLUSH_ENABLE | + PIPE_CONTROL_FLUSH_ENABLE | + PIPE_CONTROL_QW_WRITE | + PIPE_CONTROL_GLOBAL_GTT_IVB | + PIPE_CONTROL_CS_STALL); + *cs++ = rq->timeline->hwsp_offset; + *cs++ = rq->fence.seqno; + + *cs++ = MI_USER_INTERRUPT; + *cs++ = MI_NOOP; + + rq->tail = intel_ring_offset(rq, cs); + assert_ring_tail_valid(rq->ring, rq->tail); + + return cs; +} + +static u32 *gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs) +{ + GEM_BUG_ON(rq->timeline->hwsp_ggtt != rq->engine->status_page.vma); + GEM_BUG_ON(offset_in_page(rq->timeline->hwsp_offset) != I915_GEM_HWS_SEQNO_ADDR); + + *cs++ = MI_FLUSH_DW | MI_FLUSH_DW_OP_STOREDW | MI_FLUSH_DW_STORE_INDEX; + *cs++ = I915_GEM_HWS_SEQNO_ADDR | MI_FLUSH_DW_USE_GTT; + *cs++ = rq->fence.seqno; + + *cs++ = MI_USER_INTERRUPT; + + rq->tail = intel_ring_offset(rq, cs); + assert_ring_tail_valid(rq->ring, rq->tail); + + return cs; +} + +#define GEN7_XCS_WA 32 +static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs) +{ + int i; + + GEM_BUG_ON(rq->timeline->hwsp_ggtt != rq->engine->status_page.vma); + GEM_BUG_ON(offset_in_page(rq->timeline->hwsp_offset) != I915_GEM_HWS_SEQNO_ADDR); + + *cs++ = MI_FLUSH_DW | MI_FLUSH_DW_OP_STOREDW | MI_FLUSH_DW_STORE_INDEX; + *cs++ = I915_GEM_HWS_SEQNO_ADDR | MI_FLUSH_DW_USE_GTT; + *cs++ = rq->fence.seqno; + + for (i = 0; i < GEN7_XCS_WA; i++) { + *cs++ = MI_STORE_DWORD_INDEX; + *cs++ = I915_GEM_HWS_SEQNO_ADDR; + *cs++ = rq->fence.seqno; + } + + *cs++ = MI_FLUSH_DW; + *cs++ = 0; + *cs++ = 0; + + *cs++ = MI_USER_INTERRUPT; + *cs++ = MI_NOOP; + + rq->tail = intel_ring_offset(rq, cs); + assert_ring_tail_valid(rq->ring, rq->tail); + + return cs; +} +#undef GEN7_XCS_WA + +static void set_hwstam(struct intel_engine_cs *engine, u32 mask) +{ + /* + * Keep the render interrupt unmasked as this papers over + * lost interrupts following a reset. + */ + if (engine->class == RENDER_CLASS) { + if (INTEL_GEN(engine->i915) >= 6) + mask &= ~BIT(0); + else + mask &= ~I915_USER_INTERRUPT; + } + + intel_engine_set_hwsp_writemask(engine, mask); +} + +static void set_hws_pga(struct intel_engine_cs *engine, phys_addr_t phys) +{ + struct drm_i915_private *dev_priv = engine->i915; + u32 addr; + + addr = lower_32_bits(phys); + if (INTEL_GEN(dev_priv) >= 4) + addr |= (phys >> 28) & 0xf0; + + I915_WRITE(HWS_PGA, addr); +} + +static struct page *status_page(struct intel_engine_cs *engine) +{ + struct drm_i915_gem_object *obj = engine->status_page.vma->obj; + + GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj)); + return sg_page(obj->mm.pages->sgl); +} + +static void ring_setup_phys_status_page(struct intel_engine_cs *engine) +{ + set_hws_pga(engine, PFN_PHYS(page_to_pfn(status_page(engine)))); + set_hwstam(engine, ~0u); +} + +static void set_hwsp(struct intel_engine_cs *engine, u32 offset) +{ + struct drm_i915_private *dev_priv = engine->i915; + i915_reg_t hwsp; + + /* + * The ring status page addresses are no longer next to the rest of + * the ring registers as of gen7. + */ + if (IS_GEN(dev_priv, 7)) { + switch (engine->id) { + /* + * No more rings exist on Gen7. Default case is only to shut up + * gcc switch check warning. + */ + default: + GEM_BUG_ON(engine->id); + /* fallthrough */ + case RCS0: + hwsp = RENDER_HWS_PGA_GEN7; + break; + case BCS0: + hwsp = BLT_HWS_PGA_GEN7; + break; + case VCS0: + hwsp = BSD_HWS_PGA_GEN7; + break; + case VECS0: + hwsp = VEBOX_HWS_PGA_GEN7; + break; + } + } else if (IS_GEN(dev_priv, 6)) { + hwsp = RING_HWS_PGA_GEN6(engine->mmio_base); + } else { + hwsp = RING_HWS_PGA(engine->mmio_base); + } + + I915_WRITE(hwsp, offset); + POSTING_READ(hwsp); +} + +static void flush_cs_tlb(struct intel_engine_cs *engine) +{ + struct drm_i915_private *dev_priv = engine->i915; + + if (!IS_GEN_RANGE(dev_priv, 6, 7)) + return; + + /* ring should be idle before issuing a sync flush*/ + WARN_ON((ENGINE_READ(engine, RING_MI_MODE) & MODE_IDLE) == 0); + + ENGINE_WRITE(engine, RING_INSTPM, + _MASKED_BIT_ENABLE(INSTPM_TLB_INVALIDATE | + INSTPM_SYNC_FLUSH)); + if (intel_wait_for_register(engine->uncore, + RING_INSTPM(engine->mmio_base), + INSTPM_SYNC_FLUSH, 0, + 1000)) + DRM_ERROR("%s: wait for SyncFlush to complete for TLB invalidation timed out\n", + engine->name); +} + +static void ring_setup_status_page(struct intel_engine_cs *engine) +{ + set_hwsp(engine, i915_ggtt_offset(engine->status_page.vma)); + set_hwstam(engine, ~0u); + + flush_cs_tlb(engine); +} + +static bool stop_ring(struct intel_engine_cs *engine) +{ + struct drm_i915_private *dev_priv = engine->i915; + + if (INTEL_GEN(dev_priv) > 2) { + ENGINE_WRITE(engine, + RING_MI_MODE, _MASKED_BIT_ENABLE(STOP_RING)); + if (intel_wait_for_register(engine->uncore, + RING_MI_MODE(engine->mmio_base), + MODE_IDLE, + MODE_IDLE, + 1000)) { + DRM_ERROR("%s : timed out trying to stop ring\n", + engine->name); + + /* + * Sometimes we observe that the idle flag is not + * set even though the ring is empty. So double + * check before giving up. + */ + if (ENGINE_READ(engine, RING_HEAD) != + ENGINE_READ(engine, RING_TAIL)) + return false; + } + } + + ENGINE_WRITE(engine, RING_HEAD, ENGINE_READ(engine, RING_TAIL)); + + ENGINE_WRITE(engine, RING_HEAD, 0); + ENGINE_WRITE(engine, RING_TAIL, 0); + + /* The ring must be empty before it is disabled */ + ENGINE_WRITE(engine, RING_CTL, 0); + + return (ENGINE_READ(engine, RING_HEAD) & HEAD_ADDR) == 0; +} + +static int xcs_resume(struct intel_engine_cs *engine) +{ + struct drm_i915_private *dev_priv = engine->i915; + struct intel_ring *ring = engine->buffer; + int ret = 0; + + GEM_TRACE("%s: ring:{HEAD:%04x, TAIL:%04x}\n", + engine->name, ring->head, ring->tail); + + intel_uncore_forcewake_get(engine->uncore, FORCEWAKE_ALL); + + if (!stop_ring(engine)) { + /* G45 ring initialization often fails to reset head to zero */ + DRM_DEBUG_DRIVER("%s head not reset to zero " + "ctl %08x head %08x tail %08x start %08x\n", + engine->name, + ENGINE_READ(engine, RING_CTL), + ENGINE_READ(engine, RING_HEAD), + ENGINE_READ(engine, RING_TAIL), + ENGINE_READ(engine, RING_START)); + + if (!stop_ring(engine)) { + DRM_ERROR("failed to set %s head to zero " + "ctl %08x head %08x tail %08x start %08x\n", + engine->name, + ENGINE_READ(engine, RING_CTL), + ENGINE_READ(engine, RING_HEAD), + ENGINE_READ(engine, RING_TAIL), + ENGINE_READ(engine, RING_START)); + ret = -EIO; + goto out; + } + } + + if (HWS_NEEDS_PHYSICAL(dev_priv)) + ring_setup_phys_status_page(engine); + else + ring_setup_status_page(engine); + + intel_engine_reset_breadcrumbs(engine); + + /* Enforce ordering by reading HEAD register back */ + ENGINE_READ(engine, RING_HEAD); + + /* Initialize the ring. This must happen _after_ we've cleared the ring + * registers with the above sequence (the readback of the HEAD registers + * also enforces ordering), otherwise the hw might lose the new ring + * register values. */ + ENGINE_WRITE(engine, RING_START, i915_ggtt_offset(ring->vma)); + + /* WaClearRingBufHeadRegAtInit:ctg,elk */ + if (ENGINE_READ(engine, RING_HEAD)) + DRM_DEBUG_DRIVER("%s initialization failed [head=%08x], fudging\n", + engine->name, ENGINE_READ(engine, RING_HEAD)); + + /* Check that the ring offsets point within the ring! */ + GEM_BUG_ON(!intel_ring_offset_valid(ring, ring->head)); + GEM_BUG_ON(!intel_ring_offset_valid(ring, ring->tail)); + intel_ring_update_space(ring); + + /* First wake the ring up to an empty/idle ring */ + ENGINE_WRITE(engine, RING_HEAD, ring->head); + ENGINE_WRITE(engine, RING_TAIL, ring->head); + ENGINE_POSTING_READ(engine, RING_TAIL); + + ENGINE_WRITE(engine, RING_CTL, RING_CTL_SIZE(ring->size) | RING_VALID); + + /* If the head is still not zero, the ring is dead */ + if (intel_wait_for_register(engine->uncore, + RING_CTL(engine->mmio_base), + RING_VALID, RING_VALID, + 50)) { + DRM_ERROR("%s initialization failed " + "ctl %08x (valid? %d) head %08x [%08x] tail %08x [%08x] start %08x [expected %08x]\n", + engine->name, + ENGINE_READ(engine, RING_CTL), + ENGINE_READ(engine, RING_CTL) & RING_VALID, + ENGINE_READ(engine, RING_HEAD), ring->head, + ENGINE_READ(engine, RING_TAIL), ring->tail, + ENGINE_READ(engine, RING_START), + i915_ggtt_offset(ring->vma)); + ret = -EIO; + goto out; + } + + if (INTEL_GEN(dev_priv) > 2) + ENGINE_WRITE(engine, + RING_MI_MODE, _MASKED_BIT_DISABLE(STOP_RING)); + + /* Now awake, let it get started */ + if (ring->tail != ring->head) { + ENGINE_WRITE(engine, RING_TAIL, ring->tail); + ENGINE_POSTING_READ(engine, RING_TAIL); + } + + /* Papering over lost _interrupts_ immediately following the restart */ + intel_engine_queue_breadcrumbs(engine); +out: + intel_uncore_forcewake_put(engine->uncore, FORCEWAKE_ALL); + + return ret; +} + +static void reset_prepare(struct intel_engine_cs *engine) +{ + intel_engine_stop_cs(engine); +} + +static void reset_ring(struct intel_engine_cs *engine, bool stalled) +{ + struct i915_timeline *tl = &engine->timeline; + struct i915_request *pos, *rq; + unsigned long flags; + u32 head; + + rq = NULL; + spin_lock_irqsave(&tl->lock, flags); + list_for_each_entry(pos, &tl->requests, link) { + if (!i915_request_completed(pos)) { + rq = pos; + break; + } + } + + /* + * The guilty request will get skipped on a hung engine. + * + * Users of client default contexts do not rely on logical + * state preserved between batches so it is safe to execute + * queued requests following the hang. Non default contexts + * rely on preserved state, so skipping a batch loses the + * evolution of the state and it needs to be considered corrupted. + * Executing more queued batches on top of corrupted state is + * risky. But we take the risk by trying to advance through + * the queued requests in order to make the client behaviour + * more predictable around resets, by not throwing away random + * amount of batches it has prepared for execution. Sophisticated + * clients can use gem_reset_stats_ioctl and dma fence status + * (exported via sync_file info ioctl on explicit fences) to observe + * when it loses the context state and should rebuild accordingly. + * + * The context ban, and ultimately the client ban, mechanism are safety + * valves if client submission ends up resulting in nothing more than + * subsequent hangs. + */ + + if (rq) { + /* + * Try to restore the logical GPU state to match the + * continuation of the request queue. If we skip the + * context/PD restore, then the next request may try to execute + * assuming that its context is valid and loaded on the GPU and + * so may try to access invalid memory, prompting repeated GPU + * hangs. + * + * If the request was guilty, we still restore the logical + * state in case the next request requires it (e.g. the + * aliasing ppgtt), but skip over the hung batch. + * + * If the request was innocent, we try to replay the request + * with the restored context. + */ + i915_reset_request(rq, stalled); + + GEM_BUG_ON(rq->ring != engine->buffer); + head = rq->head; + } else { + head = engine->buffer->tail; + } + engine->buffer->head = intel_ring_wrap(engine->buffer, head); + + spin_unlock_irqrestore(&tl->lock, flags); +} + +static void reset_finish(struct intel_engine_cs *engine) +{ +} + +static int intel_rcs_ctx_init(struct i915_request *rq) +{ + int ret; + + ret = intel_engine_emit_ctx_wa(rq); + if (ret != 0) + return ret; + + ret = i915_gem_render_state_emit(rq); + if (ret) + return ret; + + return 0; +} + +static int rcs_resume(struct intel_engine_cs *engine) +{ + struct drm_i915_private *dev_priv = engine->i915; + + /* + * Disable CONSTANT_BUFFER before it is loaded from the context + * image. For as it is loaded, it is executed and the stored + * address may no longer be valid, leading to a GPU hang. + * + * This imposes the requirement that userspace reload their + * CONSTANT_BUFFER on every batch, fortunately a requirement + * they are already accustomed to from before contexts were + * enabled. + */ + if (IS_GEN(dev_priv, 4)) + I915_WRITE(ECOSKPD, + _MASKED_BIT_ENABLE(ECO_CONSTANT_BUFFER_SR_DISABLE)); + + /* WaTimedSingleVertexDispatch:cl,bw,ctg,elk,ilk,snb */ + if (IS_GEN_RANGE(dev_priv, 4, 6)) + I915_WRITE(MI_MODE, _MASKED_BIT_ENABLE(VS_TIMER_DISPATCH)); + + /* We need to disable the AsyncFlip performance optimisations in order + * to use MI_WAIT_FOR_EVENT within the CS. It should already be + * programmed to '1' on all products. + * + * WaDisableAsyncFlipPerfMode:snb,ivb,hsw,vlv + */ + if (IS_GEN_RANGE(dev_priv, 6, 7)) + I915_WRITE(MI_MODE, _MASKED_BIT_ENABLE(ASYNC_FLIP_PERF_DISABLE)); + + /* Required for the hardware to program scanline values for waiting */ + /* WaEnableFlushTlbInvalidationMode:snb */ + if (IS_GEN(dev_priv, 6)) + I915_WRITE(GFX_MODE, + _MASKED_BIT_ENABLE(GFX_TLB_INVALIDATE_EXPLICIT)); + + /* WaBCSVCSTlbInvalidationMode:ivb,vlv,hsw */ + if (IS_GEN(dev_priv, 7)) + I915_WRITE(GFX_MODE_GEN7, + _MASKED_BIT_ENABLE(GFX_TLB_INVALIDATE_EXPLICIT) | + _MASKED_BIT_ENABLE(GFX_REPLAY_MODE)); + + if (IS_GEN(dev_priv, 6)) { + /* From the Sandybridge PRM, volume 1 part 3, page 24: + * "If this bit is set, STCunit will have LRA as replacement + * policy. [...] This bit must be reset. LRA replacement + * policy is not supported." + */ + I915_WRITE(CACHE_MODE_0, + _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB)); + } + + if (IS_GEN_RANGE(dev_priv, 6, 7)) + I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_FORCE_ORDERING)); + + return xcs_resume(engine); +} + +static void cancel_requests(struct intel_engine_cs *engine) +{ + struct i915_request *request; + unsigned long flags; + + spin_lock_irqsave(&engine->timeline.lock, flags); + + /* Mark all submitted requests as skipped. */ + list_for_each_entry(request, &engine->timeline.requests, link) { + if (!i915_request_signaled(request)) + dma_fence_set_error(&request->fence, -EIO); + + i915_request_mark_complete(request); + } + + /* Remaining _unready_ requests will be nop'ed when submitted */ + + spin_unlock_irqrestore(&engine->timeline.lock, flags); +} + +static void i9xx_submit_request(struct i915_request *request) +{ + i915_request_submit(request); + + ENGINE_WRITE(request->engine, RING_TAIL, + intel_ring_set_tail(request->ring, request->tail)); +} + +static u32 *i9xx_emit_breadcrumb(struct i915_request *rq, u32 *cs) +{ + GEM_BUG_ON(rq->timeline->hwsp_ggtt != rq->engine->status_page.vma); + GEM_BUG_ON(offset_in_page(rq->timeline->hwsp_offset) != I915_GEM_HWS_SEQNO_ADDR); + + *cs++ = MI_FLUSH; + + *cs++ = MI_STORE_DWORD_INDEX; + *cs++ = I915_GEM_HWS_SEQNO_ADDR; + *cs++ = rq->fence.seqno; + + *cs++ = MI_USER_INTERRUPT; + *cs++ = MI_NOOP; + + rq->tail = intel_ring_offset(rq, cs); + assert_ring_tail_valid(rq->ring, rq->tail); + + return cs; +} + +#define GEN5_WA_STORES 8 /* must be at least 1! */ +static u32 *gen5_emit_breadcrumb(struct i915_request *rq, u32 *cs) +{ + int i; + + GEM_BUG_ON(rq->timeline->hwsp_ggtt != rq->engine->status_page.vma); + GEM_BUG_ON(offset_in_page(rq->timeline->hwsp_offset) != I915_GEM_HWS_SEQNO_ADDR); + + *cs++ = MI_FLUSH; + + BUILD_BUG_ON(GEN5_WA_STORES < 1); + for (i = 0; i < GEN5_WA_STORES; i++) { + *cs++ = MI_STORE_DWORD_INDEX; + *cs++ = I915_GEM_HWS_SEQNO_ADDR; + *cs++ = rq->fence.seqno; + } + + *cs++ = MI_USER_INTERRUPT; + + rq->tail = intel_ring_offset(rq, cs); + assert_ring_tail_valid(rq->ring, rq->tail); + + return cs; +} +#undef GEN5_WA_STORES + +static void +gen5_irq_enable(struct intel_engine_cs *engine) +{ + gen5_enable_gt_irq(engine->i915, engine->irq_enable_mask); +} + +static void +gen5_irq_disable(struct intel_engine_cs *engine) +{ + gen5_disable_gt_irq(engine->i915, engine->irq_enable_mask); +} + +static void +i9xx_irq_enable(struct intel_engine_cs *engine) +{ + engine->i915->irq_mask &= ~engine->irq_enable_mask; + intel_uncore_write(engine->uncore, GEN2_IMR, engine->i915->irq_mask); + intel_uncore_posting_read_fw(engine->uncore, GEN2_IMR); +} + +static void +i9xx_irq_disable(struct intel_engine_cs *engine) +{ + engine->i915->irq_mask |= engine->irq_enable_mask; + intel_uncore_write(engine->uncore, GEN2_IMR, engine->i915->irq_mask); +} + +static void +i8xx_irq_enable(struct intel_engine_cs *engine) +{ + struct drm_i915_private *dev_priv = engine->i915; + + dev_priv->irq_mask &= ~engine->irq_enable_mask; + I915_WRITE16(GEN2_IMR, dev_priv->irq_mask); + POSTING_READ16(RING_IMR(engine->mmio_base)); +} + +static void +i8xx_irq_disable(struct intel_engine_cs *engine) +{ + struct drm_i915_private *dev_priv = engine->i915; + + dev_priv->irq_mask |= engine->irq_enable_mask; + I915_WRITE16(GEN2_IMR, dev_priv->irq_mask); +} + +static int +bsd_ring_flush(struct i915_request *rq, u32 mode) +{ + u32 *cs; + + cs = intel_ring_begin(rq, 2); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + *cs++ = MI_FLUSH; + *cs++ = MI_NOOP; + intel_ring_advance(rq, cs); + return 0; +} + +static void +gen6_irq_enable(struct intel_engine_cs *engine) +{ + ENGINE_WRITE(engine, RING_IMR, + ~(engine->irq_enable_mask | engine->irq_keep_mask)); + + /* Flush/delay to ensure the RING_IMR is active before the GT IMR */ + ENGINE_POSTING_READ(engine, RING_IMR); + + gen5_enable_gt_irq(engine->i915, engine->irq_enable_mask); +} + +static void +gen6_irq_disable(struct intel_engine_cs *engine) +{ + ENGINE_WRITE(engine, RING_IMR, ~engine->irq_keep_mask); + gen5_disable_gt_irq(engine->i915, engine->irq_enable_mask); +} + +static void +hsw_vebox_irq_enable(struct intel_engine_cs *engine) +{ + ENGINE_WRITE(engine, RING_IMR, ~engine->irq_enable_mask); + + /* Flush/delay to ensure the RING_IMR is active before the GT IMR */ + ENGINE_POSTING_READ(engine, RING_IMR); + + gen6_unmask_pm_irq(engine->i915, engine->irq_enable_mask); +} + +static void +hsw_vebox_irq_disable(struct intel_engine_cs *engine) +{ + ENGINE_WRITE(engine, RING_IMR, ~0); + gen6_mask_pm_irq(engine->i915, engine->irq_enable_mask); +} + +static int +i965_emit_bb_start(struct i915_request *rq, + u64 offset, u32 length, + unsigned int dispatch_flags) +{ + u32 *cs; + + cs = intel_ring_begin(rq, 2); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + *cs++ = MI_BATCH_BUFFER_START | MI_BATCH_GTT | (dispatch_flags & + I915_DISPATCH_SECURE ? 0 : MI_BATCH_NON_SECURE_I965); + *cs++ = offset; + intel_ring_advance(rq, cs); + + return 0; +} + +/* Just userspace ABI convention to limit the wa batch bo to a resonable size */ +#define I830_BATCH_LIMIT SZ_256K +#define I830_TLB_ENTRIES (2) +#define I830_WA_SIZE max(I830_TLB_ENTRIES*4096, I830_BATCH_LIMIT) +static int +i830_emit_bb_start(struct i915_request *rq, + u64 offset, u32 len, + unsigned int dispatch_flags) +{ + u32 *cs, cs_offset = i915_scratch_offset(rq->i915); + + GEM_BUG_ON(rq->i915->gt.scratch->size < I830_WA_SIZE); + + cs = intel_ring_begin(rq, 6); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + /* Evict the invalid PTE TLBs */ + *cs++ = COLOR_BLT_CMD | BLT_WRITE_RGBA; + *cs++ = BLT_DEPTH_32 | BLT_ROP_COLOR_COPY | 4096; + *cs++ = I830_TLB_ENTRIES << 16 | 4; /* load each page */ + *cs++ = cs_offset; + *cs++ = 0xdeadbeef; + *cs++ = MI_NOOP; + intel_ring_advance(rq, cs); + + if ((dispatch_flags & I915_DISPATCH_PINNED) == 0) { + if (len > I830_BATCH_LIMIT) + return -ENOSPC; + + cs = intel_ring_begin(rq, 6 + 2); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + /* Blit the batch (which has now all relocs applied) to the + * stable batch scratch bo area (so that the CS never + * stumbles over its tlb invalidation bug) ... + */ + *cs++ = SRC_COPY_BLT_CMD | BLT_WRITE_RGBA; + *cs++ = BLT_DEPTH_32 | BLT_ROP_SRC_COPY | 4096; + *cs++ = DIV_ROUND_UP(len, 4096) << 16 | 4096; + *cs++ = cs_offset; + *cs++ = 4096; + *cs++ = offset; + + *cs++ = MI_FLUSH; + *cs++ = MI_NOOP; + intel_ring_advance(rq, cs); + + /* ... and execute it. */ + offset = cs_offset; + } + + cs = intel_ring_begin(rq, 2); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + *cs++ = MI_BATCH_BUFFER_START | MI_BATCH_GTT; + *cs++ = offset | (dispatch_flags & I915_DISPATCH_SECURE ? 0 : + MI_BATCH_NON_SECURE); + intel_ring_advance(rq, cs); + + return 0; +} + +static int +i915_emit_bb_start(struct i915_request *rq, + u64 offset, u32 len, + unsigned int dispatch_flags) +{ + u32 *cs; + + cs = intel_ring_begin(rq, 2); + if (IS_ERR(cs)) + return PTR_ERR(cs); + + *cs++ = MI_BATCH_BUFFER_START | MI_BATCH_GTT; + *cs++ = offset | (dispatch_flags & I915_DISPATCH_SECURE ? 0 : + MI_BATCH_NON_SECURE); + intel_ring_advance(rq, cs); + + return 0; +} + +int intel_ring_pin(struct intel_ring *ring) +{ + struct i915_vma *vma = ring->vma; + enum i915_map_type map = i915_coherent_map_type(vma->vm->i915); + unsigned int flags; + void *addr; + int ret; + + GEM_BUG_ON(ring->vaddr); + + ret = i915_timeline_pin(ring->timeline); + if (ret) + return ret; + + flags = PIN_GLOBAL; + + /* Ring wraparound at offset 0 sometimes hangs. No idea why. */ + flags |= PIN_OFFSET_BIAS | i915_ggtt_pin_bias(vma); |