/* * Copyright (c) 2018 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. */ /** * The aux map provides a multi-level lookup of the main surface address which * ends up providing information about the auxiliary surface data, including * the address where the auxiliary data resides. * * The 48-bit VMA (GPU) address of the main surface is split to do the address * lookup: * * 48 bit address of main surface * +--------+--------+--------+------+ * | 47:36 | 35:24 | 23:16 | 15:0 | * | L3-idx | L2-idx | L1-idx | ... | * +--------+--------+--------+------+ * * The GFX_AUX_TABLE_BASE_ADDR points to a buffer. The L3 Table Entry is * located by indexing into this buffer as a uint64_t array using the L3-idx * value. The 64-bit L3 entry is defined as: * * +-------+-------------+------+---+ * | 63:48 | 47:15 | 14:1 | 0 | * | ... | L2-tbl-addr | ... | V | * +-------+-------------+------+---+ * * If the `V` (valid) bit is set, then the L2-tbl-addr gives the address for * the level-2 table entries, with the lower address bits filled with zero. * The L2 Table Entry is located by indexing into this buffer as a uint64_t * array using the L2-idx value. The 64-bit L2 entry is similar to the L3 * entry, except with 2 additional address bits: * * +-------+-------------+------+---+ * | 63:48 | 47:13 | 12:1 | 0 | * | ... | L1-tbl-addr | ... | V | * +-------+-------------+------+---+ * * If the `V` bit is set, then the L1-tbl-addr gives the address for the * level-1 table entries, with the lower address bits filled with zero. The L1 * Table Entry is located by indexing into this buffer as a uint64_t array * using the L1-idx value. The 64-bit L1 entry is defined as: * * +--------+------+-------+-------+-------+---------------+-----+---+ * | 63:58 | 57 | 56:54 | 53:52 | 51:48 | 47:8 | 7:1 | 0 | * | Format | Y/Cr | Depth | TM | ... | aux-data-addr | ... | V | * +--------+------+-------+-------+-------+---------------+-----+---+ * * Where: * - Format: See `get_format_encoding` * - Y/Cr: 0=Y(Luma), 1=Cr(Chroma) * - (bit) Depth: See `get_bpp_encoding` * - TM (Tile-mode): 0=Ys, 1=Y, 2=rsvd, 3=rsvd * - aux-data-addr: VMA/GPU address for the aux-data * - V: entry is valid */ #include "intel_aux_map.h" #include "intel_gem.h" #include "dev/intel_device_info.h" #include "isl/isl.h" #include "drm-uapi/i915_drm.h" #include "util/list.h" #include "util/ralloc.h" #include "util/u_atomic.h" #include "util/u_math.h" #include #include #include #include static const bool aux_map_debug = false; struct aux_map_buffer { struct list_head link; struct intel_buffer *buffer; }; struct intel_aux_map_context { void *driver_ctx; pthread_mutex_t mutex; struct intel_mapped_pinned_buffer_alloc *buffer_alloc; uint32_t num_buffers; struct list_head buffers; uint64_t level3_base_addr; uint64_t *level3_map; uint32_t tail_offset, tail_remaining; uint32_t state_num; }; static bool add_buffer(struct intel_aux_map_context *ctx) { struct aux_map_buffer *buf = ralloc(ctx, struct aux_map_buffer); if (!buf) return false; const uint32_t size = 0x100000; buf->buffer = ctx->buffer_alloc->alloc(ctx->driver_ctx, size); if (!buf->buffer) { ralloc_free(buf); return false; } assert(buf->buffer->map != NULL); list_addtail(&buf->link, &ctx->buffers); ctx->tail_offset = 0; ctx->tail_remaining = size; p_atomic_inc(&ctx->num_buffers); return true; } static void advance_current_pos(struct intel_aux_map_context *ctx, uint32_t size) { assert(ctx->tail_remaining >= size); ctx->tail_remaining -= size; ctx->tail_offset += size; } static bool align_and_verify_space(struct intel_aux_map_context *ctx, uint32_t size, uint32_t align) { if (ctx->tail_remaining < size) return false; struct aux_map_buffer *tail = list_last_entry(&ctx->buffers, struct aux_map_buffer, link); uint64_t gpu = tail->buffer->gpu + ctx->tail_offset; uint64_t aligned = align64(gpu, align); if ((aligned - gpu) + size > ctx->tail_remaining) { return false; } else { if (aligned - gpu > 0) advance_current_pos(ctx, aligned - gpu); return true; } } static void get_current_pos(struct intel_aux_map_context *ctx, uint64_t *gpu, uint64_t **map) { assert(!list_is_empty(&ctx->buffers)); struct aux_map_buffer *tail = list_last_entry(&ctx->buffers, struct aux_map_buffer, link); if (gpu) *gpu = tail->buffer->gpu + ctx->tail_offset; if (map) *map = (uint64_t*)((uint8_t*)tail->buffer->map + ctx->tail_offset); } static bool add_sub_table(struct intel_aux_map_context *ctx, uint32_t size, uint32_t align, uint64_t *gpu, uint64_t **map) { if (!align_and_verify_space(ctx, size, align)) { if (!add_buffer(ctx)) return false; UNUSED bool aligned = align_and_verify_space(ctx, size, align); assert(aligned); } get_current_pos(ctx, gpu, map); memset(*map, 0, size); advance_current_pos(ctx, size); return true; } uint32_t intel_aux_map_get_state_num(struct intel_aux_map_context *ctx) { return p_atomic_read(&ctx->state_num); } struct intel_aux_map_context * intel_aux_map_init(void *driver_ctx, struct intel_mapped_pinned_buffer_alloc *buffer_alloc, const struct intel_device_info *devinfo) { struct intel_aux_map_context *ctx; if (devinfo->ver < 12) return NULL; ctx = ralloc(NULL, struct intel_aux_map_context); if (!ctx) return NULL; if (pthread_mutex_init(&ctx->mutex, NULL)) return NULL; ctx->driver_ctx = driver_ctx; ctx->buffer_alloc = buffer_alloc; ctx->num_buffers = 0; list_inithead(&ctx->buffers); ctx->tail_offset = 0; ctx->tail_remaining = 0; ctx->state_num = 0; if (add_sub_table(ctx, 32 * 1024, 32 * 1024, &ctx->level3_base_addr, &ctx->level3_map)) { if (aux_map_debug) fprintf(stderr, "AUX-MAP L3: 0x%"PRIx64", map=%p\n", ctx->level3_base_addr, ctx->level3_map); p_atomic_inc(&ctx->state_num); return ctx; } else { ralloc_free(ctx); return NULL; } } void intel_aux_map_finish(struct intel_aux_map_context *ctx) { if (!ctx) return; pthread_mutex_destroy(&ctx->mutex); list_for_each_entry_safe(struct aux_map_buffer, buf, &ctx->buffers, link) { ctx->buffer_alloc->free(ctx->driver_ctx, buf->buffer); list_del(&buf->link); p_atomic_dec(&ctx->num_buffers); ralloc_free(buf); } ralloc_free(ctx); } uint64_t intel_aux_map_get_base(struct intel_aux_map_context *ctx) { /** * This get initialized in intel_aux_map_init, and never changes, so there is * no need to lock the mutex. */ return ctx->level3_base_addr; } static struct aux_map_buffer * find_buffer(struct intel_aux_map_context *ctx, uint64_t addr) { list_for_each_entry(struct aux_map_buffer, buf, &ctx->buffers, link) { if (buf->buffer->gpu <= addr && buf->buffer->gpu_end > addr) { return buf; } } return NULL; } static uint64_t * get_u64_entry_ptr(struct intel_aux_map_context *ctx, uint64_t addr) { struct aux_map_buffer *buf = find_buffer(ctx, addr); assert(buf); uintptr_t map_offset = addr - buf->buffer->gpu; return (uint64_t*)((uint8_t*)buf->buffer->map + map_offset); } static uint8_t get_bpp_encoding(enum isl_format format) { if (isl_format_is_yuv(format)) { switch (format) { case ISL_FORMAT_YCRCB_NORMAL: case ISL_FORMAT_YCRCB_SWAPY: case ISL_FORMAT_PLANAR_420_8: return 3; case ISL_FORMAT_PLANAR_420_12: return 2; case ISL_FORMAT_PLANAR_420_10: return 1; case ISL_FORMAT_PLANAR_420_16: return 0; default: unreachable("Unsupported format!"); return 0; } } else { switch (isl_format_get_layout(format)->bpb) { case 16: return 0; case 8: return 4; case 32: return 5; case 64: return 6; case 128: return 7; default: unreachable("Unsupported bpp!"); return 0; } } } #define INTEL_AUX_MAP_ENTRY_Y_TILED_BIT (0x1ull << 52) uint64_t intel_aux_map_format_bits(enum isl_tiling tiling, enum isl_format format, uint8_t plane) { if (aux_map_debug) fprintf(stderr, "AUX-MAP entry %s, bpp_enc=%d\n", isl_format_get_name(format), isl_format_get_aux_map_encoding(format)); assert(isl_tiling_is_any_y(tiling)); uint64_t format_bits = ((uint64_t)isl_format_get_aux_map_encoding(format) << 58) | ((uint64_t)(plane > 0) << 57) | ((uint64_t)get_bpp_encoding(format) << 54) | INTEL_AUX_MAP_ENTRY_Y_TILED_BIT; assert((format_bits & INTEL_AUX_MAP_FORMAT_BITS_MASK) == format_bits); return format_bits; } uint64_t intel_aux_map_format_bits_for_isl_surf(const struct isl_surf *isl_surf) { assert(!isl_format_is_planar(isl_surf->format)); return intel_aux_map_format_bits(isl_surf->tiling, isl_surf->format, 0); } static void get_aux_entry(struct intel_aux_map_context *ctx, uint64_t address, uint32_t *l1_index_out, uint64_t *l1_entry_addr_out, uint64_t **l1_entry_map_out) { uint32_t l3_index = (address >> 36) & 0xfff; uint64_t *l3_entry = &ctx->level3_map[l3_index]; uint64_t *l2_map; if ((*l3_entry & INTEL_AUX_MAP_ENTRY_VALID_BIT) == 0) { uint64_t l2_gpu; if (add_sub_table(ctx, 32 * 1024, 32 * 1024, &l2_gpu, &l2_map)) { if (aux_map_debug) fprintf(stderr, "AUX-MAP L3[0x%x]: 0x%"PRIx64", map=%p\n", l3_index, l2_gpu, l2_map); } else { unreachable("Failed to add L2 Aux-Map Page Table!"); } *l3_entry = (l2_gpu & 0xffffffff8000ULL) | 1; } else { uint64_t l2_addr = intel_canonical_address(*l3_entry & ~0x7fffULL); l2_map = get_u64_entry_ptr(ctx, l2_addr); } uint32_t l2_index = (address >> 24) & 0xfff; uint64_t *l2_entry = &l2_map[l2_index]; uint64_t l1_addr, *l1_map; if ((*l2_entry & INTEL_AUX_MAP_ENTRY_VALID_BIT) == 0) { if (add_sub_table(ctx, 8 * 1024, 8 * 1024, &l1_addr, &l1_map)) { if (aux_map_debug) fprintf(stderr, "AUX-MAP L2[0x%x]: 0x%"PRIx64", map=%p\n", l2_index, l1_addr, l1_map); } else { unreachable("Failed to add L1 Aux-Map Page Table!"); } *l2_entry = (l1_addr & 0xffffffffe000ULL) | 1; } else { l1_addr = intel_canonical_address(*l2_entry & ~0x1fffULL); l1_map = get_u64_entry_ptr(ctx, l1_addr); } uint32_t l1_index = (address >> 16) & 0xff; if (l1_index_out) *l1_index_out = l1_index; if (l1_entry_addr_out) *l1_entry_addr_out = l1_addr + l1_index * sizeof(*l1_map); if (l1_entry_map_out) *l1_entry_map_out = &l1_map[l1_index]; } static void add_mapping(struct intel_aux_map_context *ctx, uint64_t address, uint64_t aux_address, uint64_t format_bits, bool *state_changed) { if (aux_map_debug) fprintf(stderr, "AUX-MAP 0x%"PRIx64" => 0x%"PRIx64"\n", address, aux_address); uint32_t l1_index; uint64_t *l1_entry; get_aux_entry(ctx, address, &l1_index, NULL, &l1_entry); const uint64_t l1_data = (aux_address & INTEL_AUX_MAP_ADDRESS_MASK) | format_bits | INTEL_AUX_MAP_ENTRY_VALID_BIT; const uint64_t current_l1_data = *l1_entry; if ((current_l1_data & INTEL_AUX_MAP_ENTRY_VALID_BIT) == 0) { assert((aux_address & 0xffULL) == 0); if (aux_map_debug) fprintf(stderr, "AUX-MAP L1[0x%x] 0x%"PRIx64" -> 0x%"PRIx64"\n", l1_index, current_l1_data, l1_data); /** * We use non-zero bits in 63:1 to indicate the entry had been filled * previously. If these bits are non-zero and they don't exactly match * what we want to program into the entry, then we must force the * aux-map tables to be flushed. */ if (current_l1_data != 0 && \ (current_l1_data | INTEL_AUX_MAP_ENTRY_VALID_BIT) != l1_data) *state_changed = true; *l1_entry = l1_data; } else { if (aux_map_debug) fprintf(stderr, "AUX-MAP L1[0x%x] is already marked valid!\n", l1_index); assert(*l1_entry == l1_data); } } uint64_t * intel_aux_map_get_entry(struct intel_aux_map_context *ctx, uint64_t address, uint64_t *entry_address) { pthread_mutex_lock(&ctx->mutex); uint64_t *l1_entry_map; get_aux_entry(ctx, address, NULL, entry_address, &l1_entry_map); pthread_mutex_unlock(&ctx->mutex); return l1_entry_map; } void intel_aux_map_add_mapping(struct intel_aux_map_context *ctx, uint64_t address, uint64_t aux_address, uint64_t main_size_B, uint64_t format_bits) { bool state_changed = false; pthread_mutex_lock(&ctx->mutex); uint64_t map_addr = address; uint64_t dest_aux_addr = aux_address; assert(align64(address, INTEL_AUX_MAP_MAIN_PAGE_SIZE) == address); assert(align64(aux_address, INTEL_AUX_MAP_AUX_PAGE_SIZE) == aux_address); while (map_addr - address < main_size_B) { add_mapping(ctx, map_addr, dest_aux_addr, format_bits, &state_changed); map_addr += INTEL_AUX_MAP_MAIN_PAGE_SIZE; dest_aux_addr += INTEL_AUX_MAP_AUX_PAGE_SIZE; } pthread_mutex_unlock(&ctx->mutex); if (state_changed) p_atomic_inc(&ctx->state_num); } /** * We mark the leaf entry as invalid, but we don't attempt to cleanup the * other levels of translation mappings. Since we attempt to re-use VMA * ranges, hopefully this will not lead to unbounded growth of the translation * tables. */ static void remove_mapping(struct intel_aux_map_context *ctx, uint64_t address, bool *state_changed) { uint32_t l3_index = (address >> 36) & 0xfff; uint64_t *l3_entry = &ctx->level3_map[l3_index]; uint64_t *l2_map; if ((*l3_entry & INTEL_AUX_MAP_ENTRY_VALID_BIT) == 0) { return; } else { uint64_t l2_addr = intel_canonical_address(*l3_entry & ~0x7fffULL); l2_map = get_u64_entry_ptr(ctx, l2_addr); } uint32_t l2_index = (address >> 24) & 0xfff; uint64_t *l2_entry = &l2_map[l2_index]; uint64_t *l1_map; if ((*l2_entry & INTEL_AUX_MAP_ENTRY_VALID_BIT) == 0) { return; } else { uint64_t l1_addr = intel_canonical_address(*l2_entry & ~0x1fffULL); l1_map = get_u64_entry_ptr(ctx, l1_addr); } uint32_t l1_index = (address >> 16) & 0xff; uint64_t *l1_entry = &l1_map[l1_index]; const uint64_t current_l1_data = *l1_entry; const uint64_t l1_data = current_l1_data & ~1ull; if ((current_l1_data & INTEL_AUX_MAP_ENTRY_VALID_BIT) == 0) { return; } else { if (aux_map_debug) fprintf(stderr, "AUX-MAP [0x%x][0x%x][0x%x] L1 entry removed!\n", l3_index, l2_index, l1_index); /** * We use non-zero bits in 63:1 to indicate the entry had been filled * previously. In the unlikely event that these are all zero, we force a * flush of the aux-map tables. */ if (unlikely(l1_data == 0)) *state_changed = true; *l1_entry = l1_data; } } void intel_aux_map_unmap_range(struct intel_aux_map_context *ctx, uint64_t address, uint64_t size) { bool state_changed = false; pthread_mutex_lock(&ctx->mutex); if (aux_map_debug) fprintf(stderr, "AUX-MAP remove 0x%"PRIx64"-0x%"PRIx64"\n", address, address + size); uint64_t map_addr = address; assert(align64(address, INTEL_AUX_MAP_MAIN_PAGE_SIZE) == address); while (map_addr - address < size) { remove_mapping(ctx, map_addr, &state_changed); map_addr += 64 * 1024; } pthread_mutex_unlock(&ctx->mutex); if (state_changed) p_atomic_inc(&ctx->state_num); } uint32_t intel_aux_map_get_num_buffers(struct intel_aux_map_context *ctx) { return p_atomic_read(&ctx->num_buffers); } void intel_aux_map_fill_bos(struct intel_aux_map_context *ctx, void **driver_bos, uint32_t max_bos) { assert(p_atomic_read(&ctx->num_buffers) >= max_bos); uint32_t i = 0; list_for_each_entry(struct aux_map_buffer, buf, &ctx->buffers, link) { if (i >= max_bos) return; driver_bos[i++] = buf->buffer->driver_bo; } }