diff options
Diffstat (limited to 'drivers/gpu/drm/i915/i915_gem_gtt.c')
| -rw-r--r-- | drivers/gpu/drm/i915/i915_gem_gtt.c | 2043 |
1 files changed, 839 insertions, 1204 deletions
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c index 7015a97b1097..c3028722d4e3 100644 --- a/drivers/gpu/drm/i915/i915_gem_gtt.c +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c @@ -36,6 +36,7 @@ #include <drm/i915_drm.h> #include "display/intel_frontbuffer.h" +#include "gt/intel_gt.h" #include "i915_drv.h" #include "i915_scatterlist.h" @@ -45,6 +46,12 @@ #define I915_GFP_ALLOW_FAIL (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN) +#if IS_ENABLED(CONFIG_DRM_I915_TRACE_GTT) +#define DBG(...) trace_printk(__VA_ARGS__) +#else +#define DBG(...) +#endif + /** * DOC: Global GTT views * @@ -106,12 +113,14 @@ * */ +#define as_pd(x) container_of((x), typeof(struct i915_page_directory), pt) + static int i915_get_ggtt_vma_pages(struct i915_vma *vma); -static void gen6_ggtt_invalidate(struct drm_i915_private *i915) +static void gen6_ggtt_invalidate(struct i915_ggtt *ggtt) { - struct intel_uncore *uncore = &i915->uncore; + struct intel_uncore *uncore = &ggtt->vm.i915->uncore; /* * Note that as an uncached mmio write, this will flush the @@ -120,24 +129,19 @@ static void gen6_ggtt_invalidate(struct drm_i915_private *i915) intel_uncore_write_fw(uncore, GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN); } -static void guc_ggtt_invalidate(struct drm_i915_private *i915) +static void guc_ggtt_invalidate(struct i915_ggtt *ggtt) { - struct intel_uncore *uncore = &i915->uncore; + struct intel_uncore *uncore = &ggtt->vm.i915->uncore; - gen6_ggtt_invalidate(i915); + gen6_ggtt_invalidate(ggtt); intel_uncore_write_fw(uncore, GEN8_GTCR, GEN8_GTCR_INVALIDATE); } -static void gmch_ggtt_invalidate(struct drm_i915_private *i915) +static void gmch_ggtt_invalidate(struct i915_ggtt *ggtt) { intel_gtt_chipset_flush(); } -static inline void i915_ggtt_invalidate(struct drm_i915_private *i915) -{ - i915->ggtt.invalidate(i915); -} - static int ppgtt_bind_vma(struct i915_vma *vma, enum i915_cache_level cache_level, u32 unused) @@ -215,10 +219,10 @@ static u64 gen8_pte_encode(dma_addr_t addr, return pte; } -static gen8_pde_t gen8_pde_encode(const dma_addr_t addr, - const enum i915_cache_level level) +static u64 gen8_pde_encode(const dma_addr_t addr, + const enum i915_cache_level level) { - gen8_pde_t pde = _PAGE_PRESENT | _PAGE_RW; + u64 pde = _PAGE_PRESENT | _PAGE_RW; pde |= addr; if (level != I915_CACHE_NONE) pde |= PPAT_CACHED_PDE; @@ -227,9 +231,6 @@ static gen8_pde_t gen8_pde_encode(const dma_addr_t addr, return pde; } -#define gen8_pdpe_encode gen8_pde_encode -#define gen8_pml4e_encode gen8_pde_encode - static u64 snb_pte_encode(dma_addr_t addr, enum i915_cache_level level, u32 flags) @@ -482,9 +483,69 @@ static void vm_free_page(struct i915_address_space *vm, struct page *page) spin_unlock(&vm->free_pages.lock); } +static void i915_address_space_fini(struct i915_address_space *vm) +{ + spin_lock(&vm->free_pages.lock); + if (pagevec_count(&vm->free_pages.pvec)) + vm_free_pages_release(vm, true); + GEM_BUG_ON(pagevec_count(&vm->free_pages.pvec)); + spin_unlock(&vm->free_pages.lock); + + drm_mm_takedown(&vm->mm); + + mutex_destroy(&vm->mutex); +} + +static void ppgtt_destroy_vma(struct i915_address_space *vm) +{ + struct list_head *phases[] = { + &vm->bound_list, + &vm->unbound_list, + NULL, + }, **phase; + + mutex_lock(&vm->i915->drm.struct_mutex); + for (phase = phases; *phase; phase++) { + struct i915_vma *vma, *vn; + + list_for_each_entry_safe(vma, vn, *phase, vm_link) + i915_vma_destroy(vma); + } + mutex_unlock(&vm->i915->drm.struct_mutex); +} + +static void __i915_vm_release(struct work_struct *work) +{ + struct i915_address_space *vm = + container_of(work, struct i915_address_space, rcu.work); + + ppgtt_destroy_vma(vm); + + GEM_BUG_ON(!list_empty(&vm->bound_list)); + GEM_BUG_ON(!list_empty(&vm->unbound_list)); + + vm->cleanup(vm); + i915_address_space_fini(vm); + + kfree(vm); +} + +void i915_vm_release(struct kref *kref) +{ + struct i915_address_space *vm = + container_of(kref, struct i915_address_space, ref); + + GEM_BUG_ON(i915_is_ggtt(vm)); + trace_i915_ppgtt_release(vm); + + vm->closed = true; + queue_rcu_work(vm->i915->wq, &vm->rcu); +} + static void i915_address_space_init(struct i915_address_space *vm, int subclass) { kref_init(&vm->ref); + INIT_RCU_WORK(&vm->rcu, __i915_vm_release); /* * The vm->mutex must be reclaim safe (for use in the shrinker). @@ -505,19 +566,6 @@ static void i915_address_space_init(struct i915_address_space *vm, int subclass) INIT_LIST_HEAD(&vm->bound_list); } -static void i915_address_space_fini(struct i915_address_space *vm) -{ - spin_lock(&vm->free_pages.lock); - if (pagevec_count(&vm->free_pages.pvec)) - vm_free_pages_release(vm, true); - GEM_BUG_ON(pagevec_count(&vm->free_pages.pvec)); - spin_unlock(&vm->free_pages.lock); - - drm_mm_takedown(&vm->mm); - - mutex_destroy(&vm->mutex); -} - static int __setup_page_dma(struct i915_address_space *vm, struct i915_page_dma *p, gfp_t gfp) @@ -554,28 +602,17 @@ static void cleanup_page_dma(struct i915_address_space *vm, #define kmap_atomic_px(px) kmap_atomic(px_base(px)->page) -#define setup_px(vm, px) setup_page_dma((vm), px_base(px)) -#define cleanup_px(vm, px) cleanup_page_dma((vm), px_base(px)) -#define fill_px(vm, px, v) fill_page_dma((vm), px_base(px), (v)) -#define fill32_px(vm, px, v) fill_page_dma_32((vm), px_base(px), (v)) - -static void fill_page_dma(struct i915_address_space *vm, - struct i915_page_dma *p, - const u64 val) +static void +fill_page_dma(const struct i915_page_dma *p, const u64 val, unsigned int count) { - u64 * const vaddr = kmap_atomic(p->page); - - memset64(vaddr, val, PAGE_SIZE / sizeof(val)); - - kunmap_atomic(vaddr); + kunmap_atomic(memset64(kmap_atomic(p->page), val, count)); } -static void fill_page_dma_32(struct i915_address_space *vm, - struct i915_page_dma *p, - const u32 v) -{ - fill_page_dma(vm, p, (u64)v << 32 | v); -} +#define fill_px(px, v) fill_page_dma(px_base(px), (v), PAGE_SIZE / sizeof(u64)) +#define fill32_px(px, v) do { \ + u64 v__ = lower_32_bits(v); \ + fill_px((px), v__ << 32 | v__); \ +} while (0) static int setup_scratch_page(struct i915_address_space *vm, gfp_t gfp) @@ -602,7 +639,7 @@ setup_scratch_page(struct i915_address_space *vm, gfp_t gfp) gfp |= __GFP_ZERO | __GFP_RETRY_MAYFAIL; do { - int order = get_order(size); + unsigned int order = get_order(size); struct page *page; dma_addr_t addr; @@ -621,8 +658,8 @@ setup_scratch_page(struct i915_address_space *vm, gfp_t gfp) if (unlikely(!IS_ALIGNED(addr, size))) goto unmap_page; - vm->scratch_page.page = page; - vm->scratch_page.daddr = addr; + vm->scratch[0].base.page = page; + vm->scratch[0].base.daddr = addr; vm->scratch_order = order; return 0; @@ -641,14 +678,30 @@ skip: static void cleanup_scratch_page(struct i915_address_space *vm) { - struct i915_page_dma *p = &vm->scratch_page; - int order = vm->scratch_order; + struct i915_page_dma *p = px_base(&vm->scratch[0]); + unsigned int order = vm->scratch_order; dma_unmap_page(vm->dma, p->daddr, BIT(order) << PAGE_SHIFT, PCI_DMA_BIDIRECTIONAL); __free_pages(p->page, order); } +static void free_scratch(struct i915_address_space *vm) +{ + int i; + + if (!px_dma(&vm->scratch[0])) /* set to 0 on clones */ + return; + + for (i = 1; i <= vm->top; i++) { + if (!px_dma(&vm->scratch[i])) + break; + cleanup_page_dma(vm, px_base(&vm->scratch[i])); + } + + cleanup_scratch_page(vm); +} + static struct i915_page_table *alloc_pt(struct i915_address_space *vm) { struct i915_page_table *pt; @@ -657,50 +710,24 @@ static struct i915_page_table *alloc_pt(struct i915_address_space *vm) if (unlikely(!pt)) return ERR_PTR(-ENOMEM); - if (unlikely(setup_px(vm, pt))) { + if (unlikely(setup_page_dma(vm, &pt->base))) { kfree(pt); return ERR_PTR(-ENOMEM); } atomic_set(&pt->used, 0); - return pt; } -static void free_pt(struct i915_address_space *vm, struct i915_page_table *pt) -{ - cleanup_px(vm, pt); - kfree(pt); -} - -static void gen8_initialize_pt(struct i915_address_space *vm, - struct i915_page_table *pt) -{ - fill_px(vm, pt, vm->scratch_pte); -} - -static void gen6_initialize_pt(struct i915_address_space *vm, - struct i915_page_table *pt) -{ - fill32_px(vm, pt, vm->scratch_pte); -} - -static struct i915_page_directory *__alloc_pd(void) +static struct i915_page_directory *__alloc_pd(size_t sz) { struct i915_page_directory *pd; - pd = kmalloc(sizeof(*pd), I915_GFP_ALLOW_FAIL); - + pd = kzalloc(sz, I915_GFP_ALLOW_FAIL); if (unlikely(!pd)) return NULL; - memset(&pd->base, 0, sizeof(pd->base)); - atomic_set(&pd->used, 0); spin_lock_init(&pd->lock); - - /* for safety */ - pd->entry[0] = NULL; - return pd; } @@ -708,11 +735,11 @@ static struct i915_page_directory *alloc_pd(struct i915_address_space *vm) { struct i915_page_directory *pd; - pd = __alloc_pd(); + pd = __alloc_pd(sizeof(*pd)); if (unlikely(!pd)) return ERR_PTR(-ENOMEM); - if (unlikely(setup_px(vm, pd))) { + if (unlikely(setup_page_dma(vm, px_base(pd)))) { kfree(pd); return ERR_PTR(-ENOMEM); } @@ -720,36 +747,72 @@ static struct i915_page_directory *alloc_pd(struct i915_address_space *vm) return pd; } -static inline bool pd_has_phys_page(const struct i915_page_directory * const pd) +static void free_pd(struct i915_address_space *vm, struct i915_page_dma *pd) { - return pd->base.page; + cleanup_page_dma(vm, pd); + kfree(pd); } -static void free_pd(struct i915_address_space *vm, - struct i915_page_directory *pd) +#define free_px(vm, px) free_pd(vm, px_base(px)) + +static inline void +write_dma_entry(struct i915_page_dma * const pdma, + const unsigned short idx, + const u64 encoded_entry) { - if (likely(pd_has_phys_page(pd))) - cleanup_px(vm, pd); + u64 * const vaddr = kmap_atomic(pdma->page); - kfree(pd); + vaddr[idx] = encoded_entry; + kunmap_atomic(vaddr); } -static void init_pd_with_page(struct i915_address_space *vm, - struct i915_page_directory * const pd, - struct i915_page_table *pt) +static inline void +__set_pd_entry(struct i915_page_directory * const pd, + const unsigned short idx, + struct i915_page_dma * const to, + u64 (*encode)(const dma_addr_t, const enum i915_cache_level)) +{ + GEM_BUG_ON(atomic_read(px_used(pd)) > ARRAY_SIZE(pd->entry)); + + atomic_inc(px_used(pd)); + pd->entry[idx] = to; + write_dma_entry(px_base(pd), idx, encode(to->daddr, I915_CACHE_LLC)); +} + +#define set_pd_entry(pd, idx, to) \ + __set_pd_entry((pd), (idx), px_base(to), gen8_pde_encode) + +static inline void +clear_pd_entry(struct i915_page_directory * const pd, + const unsigned short idx, + const struct i915_page_scratch * const scratch) { - fill_px(vm, pd, gen8_pde_encode(px_dma(pt), I915_CACHE_LLC)); - memset_p(pd->entry, pt, 512); + GEM_BUG_ON(atomic_read(px_used(pd)) == 0); + + write_dma_entry(px_base(pd), idx, scratch->encode); + pd->entry[idx] = NULL; + atomic_dec(px_used(pd)); } -static void init_pd(struct i915_address_space *vm, - struct i915_page_directory * const pd, - struct i915_page_directory * const to) +static bool +release_pd_entry(struct i915_page_directory * const pd, + const unsigned short idx, + struct i915_page_table * const pt, + const struct i915_page_scratch * const scratch) { - GEM_DEBUG_BUG_ON(!pd_has_phys_page(pd)); + bool free = false; + + if (atomic_add_unless(&pt->used, -1, 1)) + return false; + + spin_lock(&pd->lock); + if (atomic_dec_and_test(&pt->used)) { + clear_pd_entry(pd, idx, scratch); + free = true; + } + spin_unlock(&pd->lock); - fill_px(vm, pd, gen8_pdpe_encode(px_dma(to), I915_CACHE_LLC)); - memset_p(pd->entry, to, 512); + return free; } /* @@ -763,165 +826,305 @@ static void mark_tlbs_dirty(struct i915_ppgtt *ppgtt) ppgtt->pd_dirty_engines = ALL_ENGINES; } -/* Removes entries from a single page table, releasing it if it's empty. - * Caller can use the return value to update higher-level entries. - */ -static bool gen8_ppgtt_clear_pt(const struct i915_address_space *vm, - struct i915_page_table *pt, - u64 start, u64 length) +static int gen8_ppgtt_notify_vgt(struct i915_ppgtt *ppgtt, bool create) { - unsigned int num_entries = gen8_pte_count(start, length); - gen8_pte_t *vaddr; + struct i915_address_space *vm = &ppgtt->vm; + struct drm_i915_private *dev_priv = vm->i915; + enum vgt_g2v_type msg; + int i; - vaddr = kmap_atomic_px(pt); - memset64(vaddr + gen8_pte_index(start), vm->scratch_pte, num_entries); - kunmap_atomic(vaddr); + if (create) + atomic_inc(px_used(ppgtt->pd)); /* never remove */ + else + atomic_dec(px_used(ppgtt->pd)); + + if (i915_vm_is_4lvl(vm)) { + const u64 daddr = px_dma(ppgtt->pd); - GEM_BUG_ON(num_entries > atomic_read(&pt->used)); - return !atomic_sub_return(num_entries, &pt->used); + I915_WRITE(vgtif_reg(pdp[0].lo), lower_32_bits(daddr)); + I915_WRITE(vgtif_reg(pdp[0].hi), upper_32_bits(daddr)); + + msg = (create ? VGT_G2V_PPGTT_L4_PAGE_TABLE_CREATE : + VGT_G2V_PPGTT_L4_PAGE_TABLE_DESTROY); + } else { + for (i = 0; i < GEN8_3LVL_PDPES; i++) { + const u64 daddr = i915_page_dir_dma_addr(ppgtt, i); + + I915_WRITE(vgtif_reg(pdp[i].lo), lower_32_bits(daddr)); + I915_WRITE(vgtif_reg(pdp[i].hi), upper_32_bits(daddr)); + } + + msg = (create ? VGT_G2V_PPGTT_L3_PAGE_TABLE_CREATE : + VGT_G2V_PPGTT_L3_PAGE_TABLE_DESTROY); + } + + I915_WRITE(vgtif_reg(g2v_notify), msg); + + return 0; } -static void gen8_ppgtt_set_pde(struct i915_address_space *vm, - struct i915_page_directory *pd, - struct i915_page_table *pt, - unsigned int pde) +/* Index shifts into the pagetable are offset by GEN8_PTE_SHIFT [12] */ +#define GEN8_PAGE_SIZE (SZ_4K) /* page and page-directory sizes are the same */ +#define GEN8_PTE_SHIFT (ilog2(GEN8_PAGE_SIZE)) +#define GEN8_PDES (GEN8_PAGE_SIZE / sizeof(u64)) +#define gen8_pd_shift(lvl) ((lvl) * ilog2(GEN8_PDES)) +#define gen8_pd_index(i, lvl) i915_pde_index((i), gen8_pd_shift(lvl)) +#define __gen8_pte_shift(lvl) (GEN8_PTE_SHIFT + gen8_pd_shift(lvl)) +#define __gen8_pte_index(a, lvl) i915_pde_index((a), __gen8_pte_shift(lvl)) + +static inline unsigned int +gen8_pd_range(u64 start, u64 end, int lvl, unsigned int *idx) { - gen8_pde_t *vaddr; + const int shift = gen8_pd_shift(lvl); + const u64 mask = ~0ull << gen8_pd_shift(lvl + 1); - vaddr = kmap_atomic_px(pd); - vaddr[pde] = gen8_pde_encode(px_dma(pt), I915_CACHE_LLC); - kunmap_atomic(vaddr); + GEM_BUG_ON(start >= end); + end += ~mask >> gen8_pd_shift(1); + + *idx = i915_pde_index(start, shift); + if ((start ^ end) & mask) + return GEN8_PDES - *idx; + else + return i915_pde_index(end, shift) - *idx; } -static bool gen8_ppgtt_clear_pd(struct i915_address_space *vm, - struct i915_page_directory *pd, - u64 start, u64 length) +static inline bool gen8_pd_contains(u64 start, u64 end, int lvl) { - struct i915_page_table *pt; - u32 pde; + const u64 mask = ~0ull << gen8_pd_shift(lvl + 1); - gen8_for_each_pde(pt, pd, start, length, pde) { - bool free = false; + GEM_BUG_ON(start >= end); + return (start ^ end) & mask && (start & ~mask) == 0; +} - GEM_BUG_ON(pt == vm->scratch_pt); +static inline unsigned int gen8_pt_count(u64 start, u64 end) +{ + GEM_BUG_ON(start >= end); + if ((start ^ end) >> gen8_pd_shift(1)) + return GEN8_PDES - (start & (GEN8_PDES - 1)); + else + return end - start; +} - if (!gen8_ppgtt_clear_pt(vm, pt, start, length)) - continue; +static inline unsigned int gen8_pd_top_count(const struct i915_address_space *vm) +{ + unsigned int shift = __gen8_pte_shift(vm->top); + return (vm->total + (1ull << shift) - 1) >> shift; +} - spin_lock(&pd->lock); - if (!atomic_read(&pt->used)) { - gen8_ppgtt_set_pde(vm, pd, vm->scratch_pt, pde); - pd->entry[pde] = vm->scratch_pt; +static void __gen8_ppgtt_cleanup(struct i915_address_space *vm, + struct i915_page_directory *pd, + int count, int lvl) +{ + if (lvl) { + void **pde = pd->entry; - GEM_BUG_ON(!atomic_read(&pd->used)); - atomic_dec(&pd->used); - free = true; - } - spin_unlock(&pd->lock); - if (free) - free_pt(vm, pt); + do { + if (!*pde) + continue; + + __gen8_ppgtt_cleanup(vm, *pde, GEN8_PDES, lvl - 1); + } while (pde++, --count); } - return !atomic_read(&pd->used); + free_px(vm, pd); } -static void gen8_ppgtt_set_pdpe(struct i915_page_directory *pdp, - struct i915_page_directory *pd, - unsigned int pdpe) +static void gen8_ppgtt_cleanup(struct i915_address_space *vm) { - gen8_ppgtt_pdpe_t *vaddr; + struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm); - if (!pd_has_phys_page(pdp)) - return; + if (intel_vgpu_active(vm->i915)) + gen8_ppgtt_notify_vgt(ppgtt, false); - vaddr = kmap_atomic_px(pdp); - vaddr[pdpe] = gen8_pdpe_encode(px_dma(pd), I915_CACHE_LLC); - kunmap_atomic(vaddr); + __gen8_ppgtt_cleanup(vm, ppgtt->pd, gen8_pd_top_count(vm), vm->top); + free_scratch(vm); } -/* Removes entries from a single page dir pointer, releasing it if it's empty. - * Caller can use the return value to update higher-level entries - */ -static bool gen8_ppgtt_clear_pdp(struct i915_address_space *vm, - struct i915_page_directory * const pdp, - u64 start, u64 length) +static u64 __gen8_ppgtt_clear(struct i915_address_space * const vm, + struct i915_page_directory * const pd, + u64 start, const u64 end, int lvl) { - struct i915_page_directory *pd; - unsigned int pdpe; + const struct i915_page_scratch * const scratch = &vm->scratch[lvl]; + unsigned int idx, len; - gen8_for_each_pdpe(pd, pdp, start, length, pdpe) { - bool free = false; + len = gen8_pd_range(start, end, lvl--, &idx); + DBG("%s(%p):{ lvl:%d, start:%llx, end:%llx, idx:%d, len:%d, used:%d}\n", + __func__, vm, lvl + 1, start, end, + idx, len, atomic_read(px_used(pd))); + GEM_BUG_ON(!len || len >= atomic_read(px_used(pd))); - GEM_BUG_ON(pd == vm->scratch_pd); - - if (!gen8_ppgtt_clear_pd(vm, pd, start, length)) + do { + struct i915_page_table *pt = pd->entry[idx]; + + if (atomic_fetch_inc(&pt->used) >> gen8_pd_shift(1) && + gen8_pd_contains(start, end, lvl)) { + DBG("%s(%p):{ lvl:%d, idx:%d, start:%llx, end:%llx } removing pd\n", + __func__, vm, lvl + 1, idx, start, end); + clear_pd_entry(pd, idx, scratch); + __gen8_ppgtt_cleanup(vm, as_pd(pt), I915_PDES, lvl); + start += (u64)I915_PDES << gen8_pd_shift(lvl); continue; + } + + if (lvl) { + start = __gen8_ppgtt_clear(vm, as_pd(pt), + start, end, lvl); + } else { + unsigned int count; + u64 *vaddr; - spin_lock(&pdp->lock); - if (!atomic_read(&pd->used)) { - gen8_ppgtt_set_pdpe(pdp, vm->scratch_pd, pdpe); - pdp->entry[pdpe] = vm->scratch_pd; + count = gen8_pt_count(start, end); + DBG("%s(%p):{ lvl:%d, start:%llx, end:%llx, idx:%d, len:%d, used:%d} removing pte\n", + __func__, vm, lvl, start, end, + gen8_pd_index(start, 0), count, + atomic_read(&pt->used)); + GEM_BUG_ON(!count || count >= atomic_read(&pt->used)); + + vaddr = kmap_atomic_px(pt); + memset64(vaddr + gen8_pd_index(start, 0), + vm->scratch[0].encode, + count); + kunmap_atomic(vaddr); - GEM_BUG_ON(!atomic_read(&pdp->used)); - atomic_dec(&pdp->used); - free = true; + atomic_sub(count, &pt->used); + start += count; } - spin_unlock(&pdp->lock); - if (free) - free_pd(vm, pd); - } - return !atomic_read(&pdp->used); + if (release_pd_entry(pd, idx, pt, scratch)) + free_px(vm, pt); + } while (idx++, --len); + + return start; } -static void gen8_ppgtt_clear_3lvl(struct i915_address_space *vm, - u64 start, u64 length) +static void gen8_ppgtt_clear(struct i915_address_space *vm, + u64 start, u64 length) { - gen8_ppgtt_clear_pdp(vm, i915_vm_to_ppgtt(vm)->pd, start, length); + GEM_BUG_ON(!IS_ALIGNED(start, BIT_ULL(GEN8_PTE_SHIFT))); + GEM_BUG_ON(!IS_ALIGNED(length, BIT_ULL(GEN8_PTE_SHIFT))); + + start >>= GEN8_PTE_SHIFT; + length >>= GEN8_PTE_SHIFT; + GEM_BUG_ON(length == 0); + + __gen8_ppgtt_clear(vm, i915_vm_to_ppgtt(vm)->pd, + start, start + length, vm->top); } -static void gen8_ppgtt_set_pml4e(struct i915_page_directory *pml4, - struct i915_page_directory *pdp, - unsigned int pml4e) +static int __gen8_ppgtt_alloc(struct i915_address_space * const vm, + struct i915_page_directory * const pd, + u64 * const start, u64 end, int lvl) { - gen8_ppgtt_pml4e_t *vaddr; + const struct i915_page_scratch * const scratch = &vm->scratch[lvl]; + struct i915_page_table *alloc = NULL; + unsigned int idx, len; + int ret = 0; - vaddr = kmap_atomic_px(pml4); - vaddr[pml4e] = gen8_pml4e_encode(px_dma(pdp), I915_CACHE_LLC); - kunmap_atomic(vaddr); + len = gen8_pd_range(*start, end, lvl--, &idx); + DBG("%s(%p):{lvl:%d, start:%llx, end:%llx, idx:%d, len:%d, used:%d}\n", + __func__, vm, lvl + 1, *start, end, + idx, len, atomic_read(px_used(pd))); + GEM_BUG_ON(!len || (idx + len - 1) >> gen8_pd_shift(1)); + + spin_lock(&pd->lock); + GEM_BUG_ON(!atomic_read(px_used(pd))); /* Must be pinned! */ + do { + struct i915_page_table *pt = pd->entry[idx]; + + if (!pt) { + spin_unlock(&pd->lock); + + DBG("%s(%p):{ lvl:%d, idx:%d } allocating new tree\n", + __func__, vm, lvl + 1, idx); + + pt = fetch_and_zero(&alloc); + if (lvl) { + if (!pt) { + pt = &alloc_pd(vm)->pt; + if (IS_ERR(pt)) { + ret = PTR_ERR(pt); + goto out; + } + } + + fill_px(pt, vm->scratch[lvl].encode); + } else { + if (!pt) { + pt = alloc_pt(vm); + if (IS_ERR(pt)) { + ret = PTR_ERR(pt); + goto out; + } + } + + if (intel_vgpu_active(vm->i915) || + gen8_pt_count(*start, end) < I915_PDES) + fill_px(pt, vm->scratch[lvl].encode); + } + + spin_lock(&pd->lock); + if (likely(!pd->entry[idx])) + set_pd_entry(pd, idx, pt); + else + alloc = pt, pt = pd->entry[idx]; + } + + if (lvl) { + atomic_inc(&pt->used); + spin_unlock(&pd->lock); + + ret = __gen8_ppgtt_alloc(vm, as_pd(pt), + start, end, lvl); + if (unlikely(ret)) { + if (release_pd_entry(pd, idx, pt, scratch)) + free_px(vm, pt); + goto out; + } + + spin_lock(&pd->lock); + atomic_dec(&pt->used); + GEM_BUG_ON(!atomic_read(&pt->used)); + } else { + unsigned int count = gen8_pt_count(*start, end); + + DBG("%s(%p):{lvl:%d, start:%llx, end:%llx, idx:%d, len:%d, used:%d} inserting pte\n", + __func__, vm, lvl, *start, end, + gen8_pd_index(*start, 0), count, + atomic_read(&pt->used)); + + atomic_add(count, &pt->used); + GEM_BUG_ON(atomic_read(&pt->used) > I915_PDES); + *start += count; + } + } while (idx++, --len); + spin_unlock(&pd->lock); +out: + if (alloc) + free_px(vm, alloc); + return ret; } -/* Removes entries from a single pml4. - * This is the top-level structure in 4-level page tables used on gen8+. - * Empty entries are always scratch pml4e. - */ -static void gen8_ppgtt_clear_4lvl(struct i915_address_space *vm, - u64 start, u64 length) +static int gen8_ppgtt_alloc(struct i915_address_space *vm, + u64 start, u64 length) { - struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm); - struct i915_page_directory * const pml4 = ppgtt->pd; - struct i915_page_directory *pdp; - unsigned int pml4e; + u64 from; + int err; - GEM_BUG_ON(!i915_vm_is_4lvl(vm)); + GEM_BUG_ON(!IS_ALIGNED(start, BIT_ULL(GEN8_PTE_SHIFT))); + GEM_BUG_ON(!IS_ALIGNED(length, BIT_ULL(GEN8_PTE_SHIFT))); - gen8_for_each_pml4e(pdp, pml4, start, length, pml4e) { - bool free = false; - GEM_BUG_ON(pdp == vm->scratch_pdp); + start >>= GEN8_PTE_SHIFT; + length >>= GEN8_PTE_SHIFT; + GEM_BUG_ON(length == 0); + from = start; - if (!gen8_ppgtt_clear_pdp(vm, pdp, start, length)) - continue; + err = __gen8_ppgtt_alloc(vm, i915_vm_to_ppgtt(vm)->pd, + &start, start + length, vm->top); + if (unlikely(err && from != start)) + __gen8_ppgtt_clear(vm, i915_vm_to_ppgtt(vm)->pd, + from, start, vm->top); - spin_lock(&pml4->lock); - if (!atomic_read(&pdp->used)) { - gen8_ppgtt_set_pml4e(pml4, vm->scratch_pdp, pml4e); - pml4->entry[pml4e] = vm->scratch_pdp; - free = true; - } - spin_unlock(&pml4->lock); - if (free) - free_pd(vm, pdp); - } + return err; } static inline struct sgt_dma { @@ -933,47 +1136,28 @@ static inline struct sgt_dma { return (struct sgt_dma) { sg, addr, addr + sg->length }; } -struct gen8_insert_pte { - u16 pml4e; - u16 pdpe; - u16 pde; - u16 pte; -}; - -static __always_inline struct gen8_insert_pte gen8_insert_pte(u64 start) -{ - return (struct gen8_insert_pte) { - gen8_pml4e_index(start), - gen8_pdpe_index(start), - gen8_pde_index(start), - gen8_pte_index(start), - }; -} - -static __always_inline bool +static __always_inline u64 gen8_ppgtt_insert_pte_entries(struct i915_ppgtt *ppgtt, struct i915_page_directory *pdp, struct sgt_dma *iter, - struct gen8_insert_pte *idx, + u64 idx, enum i915_cache_level cache_level, u32 flags) { struct i915_page_directory *pd; const gen8_pte_t pte_encode = gen8_pte_encode(0, cache_level, flags); gen8_pte_t *vaddr; - bool ret; - GEM_BUG_ON(idx->pdpe >= i915_pdpes_per_pdp(&ppgtt->vm)); - pd = i915_pd_entry(pdp, idx->pdpe); - vaddr = kmap_atomic_px(i915_pt_entry(pd, idx->pde)); + pd = i915_pd_entry(pdp, gen8_pd_index(idx, 2)); + vaddr = kmap_atomic_px(i915_pt_entry(pd, gen8_pd_index(idx, 1))); do { - vaddr[idx->pte] = pte_encode | iter->dma; + vaddr[gen8_pd_index(idx, 0)] = pte_encode | iter->dma; iter->dma += I915_GTT_PAGE_SIZE; if (iter->dma >= iter->max) { iter->sg = __sg_next(iter->sg); if (!iter->sg) { - ret = false; + idx = 0; break; } @@ -981,30 +1165,22 @@ gen8_ppgtt_insert_pte_entries(struct i915_ppgtt *ppgtt, iter->max = iter->dma + iter->sg->length; } - if (++idx->pte == GEN8_PTES) { - idx->pte = 0; - - if (++idx->pde == I915_PDES) { - idx->pde = 0; - + if (gen8_pd_index(++idx, 0) == 0) { + if (gen8_pd_index(idx, 1) == 0) { /* Limited by sg length for 3lvl */ - if (++idx->pdpe == GEN8_PML4ES_PER_PML4) { - idx->pdpe = 0; - ret = true; + if (gen8_pd_index(idx, 2) == 0) break; - } - GEM_BUG_ON(idx->pdpe >= i915_pdpes_per_pdp(&ppgtt->vm)); - pd = pdp->entry[idx->pdpe]; + pd = pdp->entry[gen8_pd_index(idx, 2)]; } kunmap_atomic(vaddr); - vaddr = kmap_atomic_px(i915_pt_entry(pd, idx->pde)); + vaddr = kmap_atomic_px(i915_pt_entry(pd, gen8_pd_index(idx, 1))); } } while (1); kunmap_atomic(vaddr); - return ret; + return idx; } static void gen8_ppgtt_insert_3lvl(struct i915_address_space *vm, @@ -1014,9 +1190,9 @@ static void gen8_ppgtt_insert_3lvl(struct i915_address_space *vm, { struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm); struct sgt_dma iter = sgt_dma(vma); - struct gen8_insert_pte idx = gen8_insert_pte(vma->node.start); - gen8_ppgtt_insert_pte_entries(ppgtt, ppgtt->pd, &iter, &idx, + gen8_ppgtt_insert_pte_entries(ppgtt, ppgtt->pd, &iter, + vma->node.start >> GEN8_PTE_SHIFT, cache_level, flags); vma->page_sizes.gtt = I915_GTT_PAGE_SIZE; @@ -1033,39 +1209,38 @@ static void gen8_ppgtt_insert_huge_entries(struct i915_vma *vma, dma_addr_t rem = iter->sg->length; do { - struct gen8_insert_pte idx = gen8_insert_pte(start); struct i915_page_directory *pdp = - i915_pdp_entry(pml4, idx.pml4e); - struct i915_page_directory *pd = i915_pd_entry(pdp, idx.pdpe); - unsigned int page_size; - bool maybe_64K = false; + i915_pd_entry(pml4, __gen8_pte_index(start, 3)); + struct i915_page_directory *pd = + i915_pd_entry(pdp, __gen8_pte_index(start, 2)); gen8_pte_t encode = pte_encode; + unsigned int maybe_64K = -1; + unsigned int page_size; gen8_pte_t *vaddr; - u16 index, max; + u16 index; if (vma->page_sizes.sg & I915_GTT_PAGE_SIZE_2M && IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_2M) && - rem >= I915_GTT_PAGE_SIZE_2M && !idx.pte) { - index = idx.pde; - max = I915_PDES; - page_size = I915_GTT_PAGE_SIZE_2M; - + rem >= I915_GTT_PAGE_SIZE_2M && + !__gen8_pte_index(start, 0)) { + index = __gen8_pte_index(start, 1); encode |= GEN8_PDE_PS_2M; + page_size = I915_GTT_PAGE_SIZE_2M; vaddr = kmap_atomic_px(pd); } else { - struct i915_page_table *pt = i915_pt_entry(pd, idx.pde); + struct i915_page_table *pt = + i915_pt_entry(pd, __gen8_pte_index(start, 1)); - index = idx.pte; - max = GEN8_PTES; + index = __gen8_pte_index(start, 0); page_size = I915_GTT_PAGE_SIZE; if (!index && vma->page_sizes.sg & I915_GTT_PAGE_SIZE_64K && IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_64K) && (IS_ALIGNED(rem, I915_GTT_PAGE_SIZE_64K) || - rem >= (max - index) * I915_GTT_PAGE_SIZE)) - maybe_64K = true; + rem >= (I915_PDES - index) * I915_GTT_PAGE_SIZE)) + maybe_64K = __gen8_pte_index(start, 1); vaddr = kmap_atomic_px(pt); } @@ -1086,16 +1261,16 @@ static void gen8_ppgtt_insert_huge_entries(struct i915_vma *vma, iter->dma = sg_dma_address(iter->sg); iter->max = iter->dma + rem; - if (maybe_64K && index < max && + if (maybe_64K != -1 && index < I915_PDES && !(IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_64K) && (IS_ALIGNED(rem, I915_GTT_PAGE_SIZE_64K) || - rem >= (max - index) * I915_GTT_PAGE_SIZE))) - maybe_64K = false; + rem >= (I915_PDES - index) * I915_GTT_PAGE_SIZE))) + maybe_64K = -1; if (unlikely(!IS_ALIGNED(iter->dma, page_size))) break; } - } while (rem >= page_size && index < max); + } while (rem >= page_size && index < I915_PDES); kunmap_atomic(vaddr); @@ -1105,14 +1280,14 @@ static void gen8_ppgtt_insert_huge_entries(struct i915_vma *vma, * it and have reached the end of the sg table and we have * enough padding. */ - if (maybe_64K && - (index == max || + if (maybe_64K != -1 && + (index == I915_PDES || (i915_vm_has_scratch_64K(vma->vm) && !iter->sg && IS_ALIGNED(vma->node.start + vma->node.size, I915_GTT_PAGE_SIZE_2M)))) { vaddr = kmap_atomic_px(pd); - vaddr[idx.pde] |= GEN8_PDE_IPS_64K; + vaddr[maybe_64K] |= GEN8_PDE_IPS_64K; kunmap_atomic(vaddr); page_size = I915_GTT_PAGE_SIZE_64K; @@ -1128,9 +1303,8 @@ static void gen8_ppgtt_insert_huge_entries(struct i915_vma *vma, if (I915_SELFTEST_ONLY(vma->vm->scrub_64K)) { u16 i; - encode = vma->vm->scratch_pte; - vaddr = kmap_atomic_px(i915_pt_entry(pd, - idx.pde)); + encode = vma->vm->scratch[0].encode; + vaddr = kmap_atomic_px(i915_pt_entry(pd, maybe_64K)); for (i = 1; i < index; i += 16) memset64(vaddr + i, encode, 15); @@ -1156,32 +1330,22 @@ static void gen8_ppgtt_insert_4lvl(struct i915_address_space *vm, gen8_ppgtt_insert_huge_entries(vma, pml4, &iter, cache_level, flags); } else { - struct gen8_insert_pte idx = gen8_insert_pte(vma->node.start); + u64 idx = vma->node.start >> GEN8_PTE_SHIFT; - while (gen8_ppgtt_insert_pte_entries(ppgtt, - i915_pdp_entry(pml4, idx.pml4e++), - &iter, &idx, cache_level, - flags)) - GEM_BUG_ON(idx.pml4e >= GEN8_PML4ES_PER_PML4); + while ((idx = gen8_ppgtt_insert_pte_entries(ppgtt, + i915_pd_entry(pml4, gen8_pd_index(idx, 3)), + &iter, idx, cache_level, + flags))) + ; vma->page_sizes.gtt = I915_GTT_PAGE_SIZE; } } -static void gen8_free_page_tables(struct i915_address_space *vm, - struct i915_page_directory *pd) -{ - int i; - - for (i = 0; i < I915_PDES; i++) { - if (pd->entry[i] != vm->scratch_pt) - free_pt(vm, pd->entry[i]); - } -} - static int gen8_init_scratch(struct i915_address_space *vm) { int ret; + int i; /* * If everybody agrees to not to write into the scratch page, @@ -1195,10 +1359,8 @@ static int gen8_init_scratch(struct i915_address_space *vm) GEM_BUG_ON(!clone->has_read_only); vm->scratch_order = clone->scratch_order; - vm->scratch_pte = clone->scratch_pte; - vm->scratch_pt = clone->scratch_pt; - vm->scratch_pd = clone->scratch_pd; - vm->scratch_pdp = clone->scratch_pdp; + memcpy(vm->scratch, clone->scratch, sizeof(vm->scratch)); + px_dma(&vm->scratch[0]) = 0; /* no xfer of ownership */ return 0; } @@ -1206,377 +1368,88 @@ static int gen8_init_scratch(struct i915_address_space *vm) if (ret) return ret; - vm->scratch_pte = - gen8_pte_encode(vm->scratch_page.daddr, - I915_CACHE_LLC, - vm->has_read_only); + vm->scratch[0].encode = + gen8_pte_encode(px_dma(&vm->scratch[0]), + I915_CACHE_LLC, vm->has_read_only); - vm->scratch_pt = alloc_pt(vm); - if (IS_ERR(vm->scratch_pt)) { - ret = PTR_ERR(vm->scratch_pt); - goto free_scratch_page; - } + for (i = 1; i <= vm->top; i++) { + if (unlikely(setup_page_dma(vm, px_base(&vm->scratch[i])))) + goto free_scratch; - vm->scratch_pd = alloc_pd(vm); - if (IS_ERR(vm->scratch_pd)) { - ret = PTR_ERR(vm->scratch_pd); - goto free_pt; - } - - if (i915_vm_is_4lvl(vm)) { - vm->scratch_pdp = alloc_pd(vm); - if (IS_ERR(vm->scratch_pdp)) { - ret = PTR_ERR(vm->scratch_pdp); - goto free_pd; - } + fill_px(&vm->scratch[i], vm->scratch[i - 1].encode); + vm->scratch[i].encode = + gen8_pde_encode(px_dma(&vm->scratch[i]), + I915_CACHE_LLC); } - gen8_initialize_pt(vm, vm->scratch_pt); - init_pd_with_page(vm, vm->scratch_pd, vm->scratch_pt); - if (i915_vm_is_4lvl(vm)) - init_pd(vm, vm->scratch_pdp, vm->scratch_pd); - return 0; -free_pd: - free_pd(vm, vm->scratch_pd); -free_pt: - free_pt(vm, vm->scratch_pt); -free_scratch_page: - cleanup_scratch_page(vm); - - return ret; +free_scratch: + free_scratch(vm); + return -ENOMEM; } -static int gen8_ppgtt_notify_vgt(struct i915_ppgtt *ppgtt, bool create) +static int gen8_preallocate_top_level_pdp(struct i915_ppgtt *ppgtt) { struct i915_address_space *vm = &ppgtt->vm; - struct drm_i915_private *dev_priv = vm->i915; - enum vgt_g2v_type msg; - int i; + struct i915_page_directory *pd = ppgtt->pd; + unsigned int idx; - if (i915_vm_is_4lvl(vm)) { - const u64 daddr = px_dma(ppgtt->pd); + GEM_BUG_ON(vm->top != 2); + GEM_BUG_ON(gen8_pd_top_count(vm) != GEN8_3LVL_PDPES); - I915_WRITE(vgtif_reg(pdp[0].lo), lower_32_bits(daddr)); - I915_WRITE(vgtif_reg(pdp[0].hi), upper_32_bits(daddr)); + for (idx = 0; idx < GEN8_3LVL_PDPES; idx++) { + struct i915_page_directory *pde; - msg = (create ? VGT_G2V_PPGTT_L4_PAGE_TABLE_CREATE : - VGT_G2V_PPGTT_L4_PAGE_TABLE_DESTROY); - } else { - for (i = 0; i < GEN8_3LVL_PDPES; i++) { - const u64 daddr = i915_page_dir_dma_addr(ppgtt, i); + pde = alloc_pd(vm); + if (IS_ERR(pde)) + return PTR_ERR(pde); - I915_WRITE(vgtif_reg(pdp[i].lo), lower_32_bits(daddr)); - I915_WRITE(vgtif_reg(pdp[i].hi), upper_32_bits(daddr)); - } - - msg = (create ? VGT_G2V_PPGTT_L3_PAGE_TABLE_CREATE : - VGT_G2V_PPGTT_L3_PAGE_TABLE_DESTROY); + fill_px(pde, vm->scratch[1].encode); + set_pd_entry(pd, idx, pde); + atomic_inc(px_used(pde)); /* keep pinned */ } - I915_WRITE(vgtif_reg(g2v_notify), msg); - return 0; } -static void gen8_free_scratch(struct i915_address_space *vm) -{ - if (!vm->scratch_page.daddr) - return; - - if (i915_vm_is_4lvl(vm)) - free_pd(vm, vm->scratch_pdp); - free_pd(vm, vm->scratch_pd); - free_pt(vm, vm->scratch_pt); - cleanup_scratch_page(vm); -} - -static void gen8_ppgtt_cleanup_3lvl(struct i915_address_space *vm, - struct i915_page_directory *pdp) +static void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt) { - const unsigned int pdpes = i915_pdpes_per_pdp(vm); - int i; - - for (i = 0; i < pdpes; i++) { - if (pdp->entry[i] == vm->scratch_pd) - continue; - - gen8_free_page_tables(vm, pdp->entry[i]); - free_pd(vm, pdp->entry[i]); - } - - free_pd(vm, pdp); -} - -static void gen8_ppgtt_cleanup_4lvl(struct i915_ppgtt *ppgtt) -{ - struct i915_page_directory * const pml4 = ppgtt->pd; - int i; - - for (i = 0; i < GEN8_PML4ES_PER_PML4; i++) { - struct i915_page_directory *pdp = i915_pdp_entry(pml4, i); - - if (pdp == ppgtt->vm.scratch_pdp) - continue; - - gen8_ppgtt_cleanup_3lvl(&ppgtt->vm, pdp); - } - - free_pd(&ppgtt->vm, pml4); -} - -static void gen8_ppgtt_cleanup(struct i915_address_space *vm) -{ - struct drm_i915_private *i915 = vm->i915; - struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm); + struct drm_i915_private *i915 = gt->i915; - if (intel_vgpu_active(i915)) - gen8_ppgtt_notify_vgt(ppgtt, false); - - if (i915_vm_is_4lvl(vm)) - gen8_ppgtt_cleanup_4lvl(ppgtt); - else - gen8_ppgtt_cleanup_3lvl(&ppgtt->vm, ppgtt->pd); - - gen8_free_scratch(vm); -} - -static int gen8_ppgtt_alloc_pd(struct i915_address_space *vm, - struct i915_page_directory *pd, - u64 start, u64 length) -{ - struct i915_page_table *pt, *alloc = NULL; - u64 from = start; - unsigned int pde; - int ret = 0; - - spin_lock(&pd->lock); - gen8_for_each_pde(pt, pd, start, length, pde) { - const int count = gen8_pte_count(start, length); - - if (pt == vm->scratch_pt) { - spin_unlock(&pd->lock); - - pt = fetch_and_zero(&alloc); - if (!pt) - pt = alloc_pt(vm); - if (IS_ERR(pt)) { - ret = PTR_ERR(pt); - goto unwind; - } - - if (count < GEN8_PTES || intel_vgpu_active(vm->i915)) - gen8_initialize_pt(vm, pt); - - spin_lock(&pd->lock); - if (pd->entry[pde] == vm->scratch_pt) { - gen8_ppgtt_set_pde(vm, pd, pt, pde); - pd->entry[pde] = pt; - atomic_inc(&pd->used); - } else { - alloc = pt; - pt = pd->entry[pde]; - } - } - - atomic_add(count, &pt->used); - } - spin_unlock(&pd->lock); - goto out; - -unwind: - gen8_ppgtt_clear_pd(vm, pd, from, start - from); -out: - if (alloc) - free_pt(vm, alloc); - return ret; -} - -static int gen8_ppgtt_alloc_pdp(struct i915_address_space *vm, - struct i915_page_directory *pdp, - u64 start, u64 length) -{ - struct i915_page_directory *pd, *alloc = NULL; - u64 from = start; - unsigned int pdpe; - int ret = 0; - - spin_lock(&pdp->lock); - gen8_for_each_pdpe(pd, pdp, start, length, pdpe) { - if (pd == vm->scratch_pd) { - spin_unlock(&pdp->lock); - - pd = fetch_and_zero(&alloc); - if (!pd) - pd = alloc_pd(vm); - if (IS_ERR(pd)) { - ret = PTR_ERR(pd); - goto unwind; - } - - init_pd_with_page(vm, pd, vm->scratch_pt); - - spin_lock(&pdp->lock); - if (pdp->entry[pdpe] == vm->scratch_pd) { - gen8_ppgtt_set_pdpe(pdp, pd, pdpe); - pdp->entry[pdpe] = pd; - atomic_inc(&pdp->used); - } else { - alloc = pd; - pd = pdp->entry[pdpe]; - } - } - atomic_inc(&pd->used); - spin_unlock(&pdp->lock); - - ret = gen8_ppgtt_alloc_pd(vm, pd, start, length); - if (unlikely(ret)) - goto unwind_pd; - - spin_lock(&pdp->lock); - atomic_dec(&pd->used); - } - spin_unlock(&pdp->lock); - goto out; - -unwind_pd: - spin_lock(&pdp->lock); - if (atomic_dec_and_test(&pd->used)) { - gen8_ppgtt_set_pdpe(pdp, vm->scratch_pd, pdpe); - pdp->entry[pdpe] = vm->scratch_pd; - GEM_BUG_ON(!atomic_read(&pdp->used)); - atomic_dec(&pdp->used); - GEM_BUG_ON(alloc); - alloc = pd; /* defer the free to after the lock */ - } - spin_unlock(&pdp->lock); -unwind: - gen8_ppgtt_clear_pdp(vm, pdp, from, start - from); -out: - if (alloc) - free_pd(vm, alloc); - return ret; -} - -static int gen8_ppgtt_alloc_3lvl(struct i915_address_space *vm, - u64 start, u64 length) -{ - return gen8_ppgtt_alloc_pdp(vm, - i915_vm_to_ppgtt(vm)->pd, start, length); -} - -static int gen8_ppgtt_alloc_4lvl(struct i915_address_space *vm, - u64 start, u64 length) -{ - struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm); - struct i915_page_directory * const pml4 = ppgtt->pd; - struct i915_page_directory *pdp, *alloc = NULL; - u64 from = start; - int ret = 0; - u32 pml4e; - - spin_lock(&pml4->lock); - gen8_for_each_pml4e(pdp, pml4, start, length, pml4e) { - if (pdp == vm->scratch_pdp) { - spin_unlock(&pml4->lock); - - pdp = fetch_and_zero(&alloc); - if (!pdp) - pdp = alloc_pd(vm); - if (IS_ERR(pdp)) { - ret = PTR_ERR(pdp); - goto unwind; - } - - init_pd(vm, pdp, vm->scratch_pd); - - spin_lock(&pml4->lock); - if (pml4->entry[pml4e] == vm->scratch_pdp) { - gen8_ppgtt_set_pml4e(pml4, pdp, pml4e); - pml4->entry[pml4e] = pdp; - } else { - alloc = pdp; - pdp = pml4->entry[pml4e]; - } - } - atomic_inc(&pdp->used); - spin_unlock(&pml4->lock); - - ret = gen8_ppgtt_alloc_pdp(vm, pdp, start, length); - if (unlikely(ret)) - goto unwind_pdp; + ppgtt->vm.gt = gt; + ppgtt->vm.i915 = i915; + ppgtt->vm.dma = &i915->drm.pdev->dev; + ppgtt->vm.total = BIT_ULL(INTEL_INFO(i915)->ppgtt_size); - spin_lock(&pml4->lock); - atomic_dec(&pdp->used); - } - spin_unlock(&pml4->lock); - goto out; + i915_address_space_init(&ppgtt->vm, VM_CLASS_PPGTT); -unwind_pdp: - spin_lock(&pml4->lock); - if (atomic_dec_and_test(&pdp->used)) { - gen8_ppgtt_set_pml4e(pml4, vm->scratch_pdp, pml4e); - pml4->entry[pml4e] = vm->scratch_pdp; - GEM_BUG_ON(alloc); - alloc = pdp; /* defer the free until after the lock */ - } - spin_unlock(&pml4->lock); -unwind: - gen8_ppgtt_clear_4lvl(vm, from, start - from); -out: - if (alloc) - free_pd(vm, alloc); - return ret; + ppgtt->vm.vma_ops.bind_vma = ppgtt_bind_vma; + ppgtt->vm.vma_ops.unbind_vma = ppgtt_unbind_vma; + ppgtt->vm.vma_ops.set_pages = ppgtt_set_pages; + ppgtt->vm.vma_ops.clear_pages = clear_pages; } -static int gen8_preallocate_top_level_pdp(struct i915_ppgtt *ppgtt) +static struct i915_page_directory * +gen8_alloc_top_pd(struct i915_address_space *vm) { - struct i915_address_space *vm = &ppgtt->vm; - struct i915_page_directory *pdp = ppgtt->pd; + const unsigned int count = gen8_pd_top_count(vm); struct i915_page_directory *pd; - u64 start = 0, length = ppgtt->vm.total; - u64 from = start; - unsigned int pdpe; - gen8_for_each_pdpe(pd, pdp, start, length, pdpe) { - pd = alloc_pd(vm); - if (IS_ERR(pd)) - goto unwind; + GEM_BUG_ON(count > ARRAY_SIZE(pd->entry)); - init_pd_with_page(vm, pd, vm->scratch_pt); - gen8_ppgtt_set_pdpe(pdp, pd, pdpe); - - atomic_inc(&pdp->used); - } - - atomic_inc(&pdp->used); /* never remove */ - - return 0; + pd = __alloc_pd(offsetof(typeof(*pd), entry[count])); + if (unlikely(!pd)) + return ERR_PTR(-ENOMEM); -unwind: - start -= from; - gen8_for_each_pdpe(pd, pdp, from, start, pdpe) { - gen8_ppgtt_set_pdpe(pdp, vm->scratch_pd, pdpe); - free_pd(vm, pd); + if (unlikely(setup_page_dma(vm, px_base(pd)))) { + kfree(pd); + return ERR_PTR(-ENOMEM); } - atomic_set(&pdp->used, 0); - return -ENOMEM; -} -static void ppgtt_init(struct drm_i915_private *i915, - struct i915_ppgtt *ppgtt) -{ - ppgtt->vm.i915 = i915; - ppgtt->vm.dma = &i915->drm.pdev->dev; - ppgtt->vm.total = BIT_ULL(INTEL_INFO(i915)->ppgtt_size); - - i915_address_space_init(&ppgtt->vm, VM_CLASS_PPGTT); - - ppgtt->vm.vma_ops.bind_vma = ppgtt_bind_vma; - ppgtt->vm.vma_ops.unbind_vma = ppgtt_unbind_vma; - ppgtt->vm.vma_ops.set_pages = ppgtt_set_pages; - ppgtt->vm.vma_ops.clear_pages = clear_pages; + fill_page_dma(px_base(pd), vm->scratch[vm->top].encode, count); + atomic_inc(px_used(pd)); /* mark as pinned */ + return pd; } /* @@ -1595,7 +1468,8 @@ static struct i915_ppgtt *gen8_ppgtt_create(struct drm_i915_private *i915) if (!ppgtt) return ERR_PTR(-ENOMEM); - ppgtt_init(i915, ppgtt); + ppgtt_init(ppgtt, &i915->gt); + ppgtt->vm.top = i915_vm_is_4lvl(&ppgtt->vm) ? 3 : 2; /* * From bdw, there is hw support for read-only pages in the PPGTT. @@ -1615,41 +1489,27 @@ static struct i915_ppgtt *gen8_ppgtt_create(struct drm_i915_private *i915) if (err) goto err_free; - ppgtt->pd = __alloc_pd(); - if (!ppgtt->pd) { - err = -ENOMEM; + ppgtt->pd = gen8_alloc_top_pd(&ppgtt->vm); + if (IS_ERR(ppgtt->pd)) { + err = PTR_ERR(ppgtt->pd); goto err_free_scratch; } if (i915_vm_is_4lvl(&ppgtt->vm)) { - err = setup_px(&ppgtt->vm, ppgtt->pd); - if (err) - goto err_free_pdp; - - init_pd(&ppgtt->vm, ppgtt->pd, ppgtt->vm.scratch_pdp); - - ppgtt->vm.allocate_va_range = gen8_ppgtt_alloc_4lvl; ppgtt->vm.insert_entries = gen8_ppgtt_insert_4lvl; - ppgtt->vm.clear_range = gen8_ppgtt_clear_4lvl; } else { - /* - * We don't need to setup dma for top level pdp, only - * for entries. So point entries to scratch. - */ - memset_p(ppgtt->pd->entry, ppgtt->vm.scratch_pd, - GEN8_3LVL_PDPES); - if (intel_vgpu_active(i915)) { err = gen8_preallocate_top_level_pdp(ppgtt); if (err) - goto err_free_pdp; + goto err_free_pd; } - ppgtt->vm.allocate_va_range = gen8_ppgtt_alloc_3lvl; ppgtt->vm.insert_entries = gen8_ppgtt_insert_3lvl; - ppgtt->vm.clear_range = gen8_ppgtt_clear_3lvl; } + ppgtt->vm.allocate_va_range = gen8_ppgtt_alloc; + ppgtt->vm.clear_range = gen8_ppgtt_clear; + if (intel_vgpu_active(i915)) gen8_ppgtt_notify_vgt(ppgtt, true); @@ -1657,10 +1517,11 @@ static struct i915_ppgtt *gen8_ppgtt_create(struct drm_i915_private *i915) return ppgtt; -err_free_pdp: - free_pd(&ppgtt->vm, ppgtt->pd); +err_free_pd: + __gen8_ppgtt_cleanup(&ppgtt->vm, ppgtt->pd, + gen8_pd_top_count(&ppgtt->vm), ppgtt->vm.top); err_free_scratch: - gen8_free_scratch(&ppgtt->vm); + free_scratch(&ppgtt->vm); err_free: kfree(ppgtt); return ERR_PTR(err); @@ -1676,25 +1537,26 @@ static inline void gen6_write_pde(const struct gen6_ppgtt *ppgtt, ppgtt->pd_addr + pde); } -static void gen7_ppgtt_enable(struct drm_i915_private *dev_priv) +static void gen7_ppgtt_enable(struct intel_gt *gt) { + struct drm_i915_private *i915 = gt->i915; + struct intel_uncore *uncore = gt->uncore; struct intel_engine_cs *engine; - u32 ecochk, ecobits; enum intel_engine_id id; + u32 ecochk; - ecobits = I915_READ(GAC_ECO_BITS); - I915_WRITE(GAC_ECO_BITS, ecobits | ECOBITS_PPGTT_CACHE64B); + intel_uncore_rmw(uncore, GAC_ECO_BITS, 0, ECOBITS_PPGTT_CACHE64B); - ecochk = I915_READ(GAM_ECOCHK); - if (IS_HASWELL(dev_priv)) { + ecochk = intel_uncore_read(uncore, GAM_ECOCHK); + if (IS_HASWELL(i915)) { ecochk |= ECOCHK_PPGTT_WB_HSW; } else { ecochk |= ECOCHK_PPGTT_LLC_IVB; ecochk &= ~ECOCHK_PPGTT_GFDT_IVB; } - I915_WRITE(GAM_ECOCHK, ecochk); + intel_uncore_write(uncore, GAM_ECOCHK, ecochk); - for_each_engine(engine, dev_priv, id) { + for_each_engine(engine, i915, id) { /* GFX_MODE is per-ring on gen7+ */ ENGINE_WRITE(engine, RING_MODE_GEN7, @@ -1702,22 +1564,29 @@ static void gen7_ppgtt_enable(struct drm_i915_private *dev_priv) } } -static void gen6_ppgtt_enable(struct drm_i915_private *dev_priv) +static void gen6_ppgtt_enable(struct intel_gt *gt) { - u32 ecochk, gab_ctl, ecobits; + struct intel_uncore *uncore = gt->uncore; - ecobits = I915_READ(GAC_ECO_BITS); - I915_WRITE(GAC_ECO_BITS, ecobits | ECOBITS_SNB_BIT | - ECOBITS_PPGTT_CACHE64B); + intel_uncore_rmw(uncore, + GAC_ECO_BITS, + 0, + ECOBITS_SNB_BIT | ECOBITS_PPGTT_CACHE64B); - gab_ctl = I915_READ(GAB_CTL); - I915_WRITE(GAB_CTL, gab_ctl | GAB_CTL_CONT_AFTER_PAGEFAULT); + intel_uncore_rmw(uncore, + GAB_CTL, + 0, + GAB_CTL_CONT_AFTER_PAGEFAULT); - ecochk = I915_READ(GAM_ECOCHK); - I915_WRITE(GAM_ECOCHK, ecochk | ECOCHK_SNB_BIT | ECOCHK_PPGTT_CACHE64B); + intel_uncore_rmw(uncore, + GAM_ECOCHK, + 0, + ECOCHK_SNB_BIT | ECOCHK_PPGTT_CACHE64B); - if (HAS_PPGTT(dev_priv)) /* may be disabled for VT-d */ - I915_WRITE(GFX_MODE, _MASKED_BIT_ENABLE(GFX_PPGTT_ENABLE)); + if (HAS_PPGTT(uncore->i915)) /* may be disabled for VT-d */ + intel_uncore_write(uncore, + GFX_MODE, + _MASKED_BIT_ENABLE(GFX_PPGTT_ENABLE)); } /* PPGTT support for Sandybdrige/Gen6 and later */ @@ -1726,7 +1595,7 @@ static void gen6_ppgtt_clear_range(struct i915_address_space *vm, { struct gen6_ppgtt * const ppgtt = to_gen6_ppgtt(i915_vm_to_ppgtt(vm)); const unsigned int first_entry = start / I915_GTT_PAGE_SIZE; - const gen6_pte_t scratch_pte = vm->scratch_pte; + const gen6_pte_t scratch_pte = vm->scratch[0].encode; unsigned int pde = first_entry / GEN6_PTES; unsigned int pte = first_entry % GEN6_PTES; unsigned int num_entries = length / I915_GTT_PAGE_SIZE; @@ -1737,7 +1606,7 @@ static void gen6_ppgtt_clear_range(struct i915_address_space *vm, const unsigned int count = min(num_entries, GEN6_PTES - pte); gen6_pte_t *vaddr; - GEM_BUG_ON(pt == vm->scratch_pt); + GEM_BUG_ON(px_base(pt) == px_base(&vm->scratch[1])); num_entries -= count; @@ -1774,7 +1643,7 @@ static void gen6_ppgtt_insert_entries(struct i915_address_space *vm, struct sgt_dma iter = sgt_dma(vma); gen6_pte_t *vaddr; - GEM_BUG_ON(i915_pt_entry(pd, act_pt) == vm->scratch_pt); + GEM_BUG_ON(pd->entry[act_pt] == &vm->scratch[1]); vaddr = kmap_atomic_px(i915_pt_entry(pd, act_pt)); do { @@ -1819,7 +1688,7 @@ static int gen6_alloc_va_range(struct i915_address_space *vm, gen6_for_each_pde(pt, pd, start, length, pde) { const unsigned int count = gen6_pte_count(start, length); - if (pt == vm->scratch_pt) { + if (px_base(pt) == px_base(&vm->scratch[1])) { spin_unlock(&pd->lock); pt = fetch_and_zero(&alloc); @@ -1830,10 +1699,10 @@ static int gen6_alloc_va_range(struct i915_address_space *vm, goto unwind_out; } - gen6_initialize_pt(vm, pt); + fill32_px(pt, vm->scratch[0].encode); spin_lock(&pd->lock); - if (pd->entry[pde] == vm->scratch_pt) { + if (pd->entry[pde] == &vm->scratch[1]) { pd->entry[pde] = pt; if (i915_vma_is_bound(ppgtt->vma, I915_VMA_GLOBAL_BIND)) { @@ -1852,7 +1721,7 @@ static int gen6_alloc_va_range(struct i915_address_space *vm, if (flush) { mark_tlbs_dirty(&ppgtt->base); - gen6_ggtt_invalidate(vm->i915); + gen6_ggtt_invalidate(vm->gt->ggtt); } goto out; @@ -1861,7 +1730,7 @@ unwind_out: gen6_ppgtt_clear_range(vm, from, start - from); out: if (alloc) - free_pt(vm, alloc); + free_px(vm, alloc); intel_runtime_pm_put(&vm->i915->runtime_pm, wakeref); return ret; } @@ -1870,108 +1739,52 @@ static int gen6_ppgtt_init_scratch(struct gen6_ppgtt *ppgtt) { struct i915_address_space * const vm = &ppgtt->base.vm; struct i915_page_directory * const pd = ppgtt->base.pd; - struct i915_page_table *unused; - u32 pde; int ret; ret = setup_scratch_page(vm, __GFP_HIGHMEM); if (ret) return ret; - vm->scratch_pte = vm->pte_encode(vm->scratch_page.daddr, - I915_CACHE_NONE, - PTE_READ_ONLY); + vm->scratch[0].encode = + vm->pte_encode(px_dma(&vm->scratch[0]), + I915_CACHE_NONE, PTE_READ_ONLY); - vm->scratch_pt = alloc_pt(vm); - if (IS_ERR(vm->scratch_pt)) { + if (unlikely(setup_page_dma(vm, px_base(&vm->scratch[1])))) { cleanup_scratch_page(vm); - return PTR_ERR(vm->scratch_pt); + return -ENOMEM; } - gen6_initialize_pt(vm, vm->scratch_pt); - - gen6_for_all_pdes(unused, pd, pde) - pd->entry[pde] = vm->scratch_pt; + fill32_px(&vm->scratch[1], vm->scratch[0].encode); + memset_p(pd->entry, &vm->scratch[1], I915_PDES); return 0; } -static void gen6_ppgtt_free_scratch(struct i915_address_space *vm) -{ - free_pt(vm, vm->scratch_pt); - cleanup_scratch_page(vm); -} - static void gen6_ppgtt_free_pd(struct gen6_ppgtt *ppgtt) { struct i915_page_directory * const pd = ppgtt->base.pd; + struct i915_page_dma * const scratch = + px_base(&ppgtt->base.vm.scratch[1]); struct i915_page_table *pt; u32 pde; gen6_for_all_pdes(pt, pd, pde) - if (pt != ppgtt->base.vm.scratch_pt) - free_pt(&ppgtt->base.vm, pt); -} - -struct gen6_ppgtt_cleanup_work { - struct work_struct base; - struct i915_vma *vma; -}; - -static void gen6_ppgtt_cleanup_work(struct work_struct *wrk) -{ - struct gen6_ppgtt_cleanup_work *work = - container_of(wrk, typeof(*work), base); - /* Side note, vma->vm is the GGTT not the ppgtt we just destroyed! */ - struct drm_i915_private *i915 = work->vma->vm->i915; - - mutex_lock(&i915->drm.struct_mutex); - i915_vma_destroy(work->vma); - mutex_unlock(&i915->drm.struct_mutex); - - kfree(work); -} - -static int nop_set_pages(struct i915_vma *vma) -{ - return -ENODEV; + if (px_base(pt) != scratch) + free_px(&ppgtt->base.vm, pt); } -static void nop_clear_pages(struct i915_vma *vma) -{ -} - -static int nop_bind(struct i915_vma *vma, - enum i915_cache_level cache_level, - u32 unused) -{ - return -ENODEV; -} - -static void nop_unbind(struct i915_vma *vma) -{ -} - -static const struct i915_vma_ops nop_vma_ops = { - .set_pages = nop_set_pages, - .clear_pages = nop_clear_pages, - .bind_vma = nop_bind, - .unbind_vma = nop_unbind, -}; - static void gen6_ppgtt_cleanup(struct i915_address_space *vm) { struct gen6_ppgtt *ppgtt = to_gen6_ppgtt(i915_vm_to_ppgtt(vm)); - struct gen6_ppgtt_cleanup_work *work = ppgtt->work; + struct drm_i915_private *i915 = vm->i915; /* FIXME remove the struct_mutex to bring the locking under control */ - INIT_WORK(&work->base, gen6_ppgtt_cleanup_work); - work->vma = ppgtt->vma; - work->vma->ops = &nop_vma_ops; - schedule_work(&work->base); + mutex_lock(&i915->drm.struct_mutex); + i915_vma_destroy(ppgtt->vma); + mutex_unlock(&i915->drm.struct_mutex); gen6_ppgtt_free_pd(ppgtt); - gen6_ppgtt_free_scratch(vm); + free_scratch(vm); kfree(ppgtt->base.pd); } @@ -1998,14 +1811,14 @@ static int pd_vma_bind(struct i915_vma *vma, struct i915_page_table *pt; unsigned int pde; - ppgtt->base.pd->base.ggtt_offset = ggtt_offset * sizeof(gen6_pte_t); + px_base(ppgtt->base.pd)->ggtt_offset = ggtt_offset * sizeof(gen6_pte_t); ppgtt->pd_addr = (gen6_pte_t __iomem *)ggtt->gsm + ggtt_offset; gen6_for_all_pdes(pt, ppgtt->base.pd, pde) gen6_write_pde(ppgtt, pde, pt); mark_tlbs_dirty(&ppgtt->base); - gen6_ggtt_invalidate(ppgtt->base.vm.i915); + gen6_ggtt_invalidate(ggtt); return 0; } @@ -2014,7 +1827,8 @@ static void pd_vma_unbind(struct i915_vma *vma) { struct gen6_ppgtt *ppgtt = vma->private; struct i915_page_directory * const pd = ppgtt->base.pd; - struct i915_page_table * const scratch_pt = ppgtt->base.vm.scratch_pt; + struct i915_page_dma * const scratch = + px_base(&ppgtt->base.vm.scratch[1]); struct i915_page_table *pt; unsigned int pde; @@ -2023,11 +1837,11 @@ static void pd_vma_unbind(struct i915_vma *vma) /* Free all no longer used page tables */ gen6_for_all_pdes(pt, ppgtt->base.pd, pde) { - if (atomic_read(&pt->used) || pt == scratch_pt) + if (px_base(pt) == scratch || atomic_read(&pt->used)) continue; - free_pt(&ppgtt->base.vm, pt); - pd->entry[pde] = scratch_pt; + free_px(&ppgtt->base.vm, pt); + pd->entry[pde] = scratch; } ppgtt->scan_for_unused_pt = false; @@ -2043,7 +1857,7 @@ static const struct i915_vma_ops pd_vma_ops = { static struct i915_vma *pd_vma_create(struct gen6_ppgtt *ppgtt, int size) { struct drm_i915_private *i915 = ppgtt->base.vm.i915; - struct i915_ggtt *ggtt = &i915->ggtt; + struct i915_ggtt *ggtt = ppgtt->base.vm.gt->ggtt; struct i915_vma *vma; GEM_BUG_ON(!IS_ALIGNED(size, I915_GTT_PAGE_SIZE)); @@ -2053,7 +1867,7 @@ static struct i915_vma *pd_vma_create(struct gen6_ppgtt *ppgtt, int size) if (!vma) return ERR_PTR(-ENOMEM); - i915_active_init(i915, &vma->active, NULL); + i915_active_init(i915, &vma->active, NULL, NULL); INIT_ACTIVE_REQUEST(&vma->last_fence); vma->vm = &ggtt->vm; @@ -2141,7 +1955,8 @@ static struct i915_ppgtt *gen6_ppgtt_create(struct drm_i915_private *i915) if (!ppgtt) return ERR_PTR(-ENOMEM); - ppgtt_init(i915, &ppgtt->base); + ppgtt_init(&ppgtt->base, &i915->gt); + ppgtt->base.vm.top = 1; ppgtt->base.vm.allocate_va_range = gen6_alloc_va_range; ppgtt->base.vm.clear_range = gen6_ppgtt_clear_range; @@ -2150,16 +1965,10 @@ static struct i915_ppgtt *gen6_ppgtt_create(struct drm_i915_private *i915) ppgtt->base.vm.pte_encode = ggtt->vm.pte_encode; - ppgtt->work = kmalloc(sizeof(*ppgtt->work), GFP_KERNEL); - if (!ppgtt->work) { - err = -ENOMEM; - goto err_free; - } - - ppgtt->base.pd = __alloc_pd(); + ppgtt->base.pd = __alloc_pd(sizeof(*ppgtt->base.pd)); if (!ppgtt->base.pd) { err = -ENOMEM; - goto err_work; + goto err_free; } err = gen6_ppgtt_init_scratch(ppgtt); @@ -2175,31 +1984,40 @@ static struct i915_ppgtt *gen6_ppgtt_create(struct drm_i915_private *i915) return &ppgtt->base; err_scratch: - gen6_ppgtt_free_scratch(&ppgtt->base.vm); + free_scratch(&ppgtt->base.vm); err_pd: kfree(ppgtt->base.pd); -err_work: - kfree(ppgtt->work); err_free: kfree(ppgtt); return ERR_PTR(err); } -static void gtt_write_workarounds(struct drm_i915_private *dev_priv) +static void gtt_write_workarounds(struct intel_gt *gt) { + struct drm_i915_private *i915 = gt->i915; + struct intel_uncore *uncore = gt->uncore; + /* This function is for gtt related workarounds. This function is * called on driver load and after a GPU reset, so you can place * workarounds here even if they get overwritten by GPU reset. */ /* WaIncreaseDefaultTLBEntries:chv,bdw,skl,bxt,kbl,glk,cfl,cnl,icl */ - if (IS_BROADWELL(dev_priv)) - I915_WRITE(GEN8_L3_LRA_1_GPGPU, GEN8_L3_LRA_1_GPGPU_DEFAULT_VALUE_BDW); - else if (IS_CHERRYVIEW(dev_priv)) - I915_WRITE(GEN8_L3_LRA_1_GPGPU, GEN8_L3_LRA_1_GPGPU_DEFAULT_VALUE_CHV); - else if (IS_GEN9_LP(dev_priv)) - I915_WRITE(GEN8_L3_LRA_1_GPGPU, GEN9_L3_LRA_1_GPGPU_DEFAULT_VALUE_BXT); - else if (INTEL_GEN(dev_priv) >= 9) - I915_WRITE(GEN8_L3_LRA_1_GPGPU, GEN9_L3_LRA_1_GPGPU_DEFAULT_VALUE_SKL); + if (IS_BROADWELL(i915)) + intel_uncore_write(uncore, + GEN8_L3_LRA_1_GPGPU, + GEN8_L3_LRA_1_GPGPU_DEFAULT_VALUE_BDW); + else if (IS_CHERRYVIEW(i915)) + intel_uncore_write(uncore, + GEN8_L3_LRA_1_GPGPU, + GEN8_L3_LRA_1_GPGPU_DEFAULT_VALUE_CHV); + else if (IS_GEN9_LP(i915)) + intel_uncore_write(uncore, + GEN8_L3_LRA_1_GPGPU, + GEN9_L3_LRA_1_GPGPU_DEFAULT_VALUE_BXT); + else if (INTEL_GEN(i915) >= 9) + intel_uncore_write(uncore, + GEN8_L3_LRA_1_GPGPU, + GEN9_L3_LRA_1_GPGPU_DEFAULT_VALUE_SKL); /* * To support 64K PTEs we need to first enable the use of the @@ -2212,21 +2030,24 @@ static void gtt_write_workarounds(struct drm_i915_private *dev_priv) * 32K pages, but we don't currently have any support for it in our * driver. */ - if (HAS_PAGE_SIZES(dev_priv, I915_GTT_PAGE_SIZE_64K) && - INTEL_GEN(dev_priv) <= 10) - I915_WRITE(GEN8_GAMW_ECO_DEV_RW_IA, - I915_READ(GEN8_GAMW_ECO_DEV_RW_IA) | - GAMW_ECO_ENABLE_64K_IPS_FIELD); + if (HAS_PAGE_SIZES(i915, I915_GTT_PAGE_SIZE_64K) && + INTEL_GEN(i915) <= 10) + intel_uncore_rmw(uncore, + GEN8_GAMW_ECO_DEV_RW_IA, + 0, + GAMW_ECO_ENABLE_64K_IPS_FIELD); } -int i915_ppgtt_init_hw(struct drm_i915_private *dev_priv) +int i915_ppgtt_init_hw(struct intel_gt *gt) { - gtt_write_workarounds(dev_priv); + struct drm_i915_private *i915 = gt->i915; - if (IS_GEN(dev_priv, 6)) - gen6_ppgtt_enable(dev_priv); - else if (IS_GEN(dev_priv, 7)) - gen7_ppgtt_enable(dev_priv); + gtt_write_workarounds(gt); + + if (IS_GEN(i915, 6)) + gen6_ppgtt_enable(gt); + else if (IS_GEN(i915, 7)) + gen7_ppgtt_enable(gt); return 0; } @@ -2254,42 +2075,6 @@ i915_ppgtt_create(struct drm_i915_private *i915) return ppgtt; } -static void ppgtt_destroy_vma(struct i915_address_space *vm) -{ - struct list_head *phases[] = { - &vm->bound_list, - &vm->unbound_list, - NULL, - }, **phase; - - vm->closed = true; - for (phase = phases; *phase; phase++) { - struct i915_vma *vma, *vn; - - list_for_each_entry_safe(vma, vn, *phase, vm_link) - i915_vma_destroy(vma); - } -} - -void i915_vm_release(struct kref *kref) -{ - struct i915_address_space *vm = - container_of(kref, struct i915_address_space, ref); - - GEM_BUG_ON(i915_is_ggtt(vm)); - trace_i915_ppgtt_release(vm); - - ppgtt_destroy_vma(vm); - - GEM_BUG_ON(!list_empty(&vm->bound_list)); - GEM_BUG_ON(!list_empty(&vm->unbound_list)); - - vm->cleanup(vm); - i915_address_space_fini(vm); - - kfree(vm); -} - /* Certain Gen5 chipsets require require idling the GPU before * unmapping anything from the GTT when VT-d is enabled. */ @@ -2301,21 +2086,26 @@ static bool needs_idle_maps(struct drm_i915_private *dev_priv) return IS_GEN(dev_priv, 5) && IS_MOBILE(dev_priv) && intel_vtd_active(); } -void i915_gem_suspend_gtt_mappings(struct drm_i915_private *dev_priv) +static void ggtt_suspend_mappings(struct i915_ggtt *ggtt) { - struct i915_ggtt *ggtt = &dev_priv->ggtt; + struct drm_i915_private *i915 = ggtt->vm.i915; /* Don't bother messing with faults pre GEN6 as we have little * documentation supporting that it's a good idea. */ - if (INTEL_GEN(dev_priv) < 6) + if (INTEL_GEN(i915) < 6) return; - i915_check_and_clear_faults(dev_priv); + intel_gt_check_and_clear_faults(ggtt->vm.gt); ggtt->vm.clear_range(&ggtt->vm, 0, ggtt->vm.total); - i915_ggtt_invalidate(dev_priv); + ggtt->invalidate(ggtt); +} + +void i915_gem_suspend_gtt_mappings(struct drm_i915_private *i915) +{ + ggtt_suspend_mappings(&i915->ggtt); } int i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj, @@ -2361,7 +2151,7 @@ static void gen8_ggtt_insert_page(struct i915_address_space *vm, gen8_set_pte(pte, gen8_pte_encode(addr, level, 0)); - ggtt->invalidate(vm->i915); + ggtt->invalidate(ggtt); } static void gen8_ggtt_insert_entries(struct i915_address_space *vm, @@ -2389,7 +2179,7 @@ static void gen8_ggtt_insert_entries(struct i915_address_space *vm, * We want to flush the TLBs only after we're certain all the PTE * updates have finished. */ - ggtt->invalidate(vm->i915); + ggtt->invalidate(ggtt); } static void gen6_ggtt_insert_page(struct i915_address_space *vm, @@ -2404,7 +2194,7 @@ static void gen6_ggtt_insert_page(struct i915_address_space *vm, iowrite32(vm->pte_encode(addr, level, flags), pte); - ggtt->invalidate(vm->i915); + ggtt->invalidate(ggtt); } /* @@ -2430,7 +2220,7 @@ static void gen6_ggtt_insert_entries(struct i915_address_space *vm, * We want to flush the TLBs only after we're certain all the PTE * updates have finished. */ - ggtt->invalidate(vm->i915); + ggtt->invalidate(ggtt); } static void nop_clear_range(struct i915_address_space *vm, @@ -2444,7 +2234,7 @@ static void gen8_ggtt_clear_range(struct i915_address_space *vm, struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm); unsigned first_entry = start / I915_GTT_PAGE_SIZE; unsigned num_entries = length / I915_GTT_PAGE_SIZE; - const gen8_pte_t scratch_pte = vm->scratch_pte; + const gen8_pte_t scratch_pte = vm->scratch[0].encode; gen8_pte_t __iomem *gtt_base = (gen8_pte_t __iomem *)ggtt->gsm + first_entry; const int max_entries = ggtt_total_entries(ggtt) - first_entry; @@ -2569,8 +2359,7 @@ static void gen6_ggtt_clear_range(struct i915_address_space *vm, first_entry, num_entries, max_entries)) num_entries = max_entries; - scratch_pte = vm->scratch_pte; - + scratch_pte = vm->scratch[0].encode; for (i = 0; i < num_entries; i++) iowrite32(scratch_pte, >t_base[i]); } @@ -2657,18 +2446,18 @@ static int aliasing_gtt_bind_vma(struct i915_vma *vma, pte_flags |= PTE_READ_ONLY; if (flags & I915_VMA_LOCAL_BIND) { - struct i915_ppgtt *appgtt = i915->mm.aliasing_ppgtt; + struct i915_ppgtt *alias = i915_vm_to_ggtt(vma->vm)->alias; if (!(vma->flags & I915_VMA_LOCAL_BIND)) { - ret = appgtt->vm.allocate_va_range(&appgtt->vm, - vma->node.start, - vma->size); + ret = alias->vm.allocate_va_range(&alias->vm, + vma->node.start, + vma->size); if (ret) return ret; } - appgtt->vm.insert_entries(&appgtt->vm, vma, cache_level, - pte_flags); + alias->vm.insert_entries(&alias->vm, vma, + cache_level, pte_flags); } if (flags & I915_VMA_GLOBAL_BIND) { @@ -2696,7 +2485,8 @@ static void aliasing_gtt_unbind_vma(struct i915_vma *vma) } if (vma->flags & I915_VMA_LOCAL_BIND) { - struct i915_address_space *vm = &i915->mm.aliasing_ppgtt->vm; + struct i915_address_space *vm = + &i915_vm_to_ggtt(vma->vm)->alias->vm; vm->clear_range(vm, vma->node.start, vma->size); } @@ -2753,13 +2543,12 @@ static void i915_gtt_color_adjust(const struct drm_mm_node *node, *end -= I915_GTT_PAGE_SIZE; } -static int init_aliasing_ppgtt(struct drm_i915_private *i915) +static int init_aliasing_ppgtt(struct i915_ggtt *ggtt) { - struct i915_ggtt *ggtt = &i915->ggtt; struct i915_ppgtt *ppgtt; int err; - ppgtt = i915_ppgtt_create(i915); + ppgtt = i915_ppgtt_create(ggtt->vm.i915); if (IS_ERR(ppgtt)) return PTR_ERR(ppgtt); @@ -2778,7 +2567,7 @@ static int init_aliasing_ppgtt(struct drm_i915_private *i915) if (err) goto err_ppgtt; - i915->mm.aliasing_ppgtt = ppgtt; + ggtt->alias = ppgtt; GEM_BUG_ON(ggtt->vm.vma_ops.bind_vma != ggtt_bind_vma); ggtt->vm.vma_ops.bind_vma = aliasing_gtt_bind_vma; @@ -2793,19 +2582,24 @@ err_ppgtt: return err; } -static void fini_aliasing_ppgtt(struct drm_i915_private *i915) +static void fini_aliasing_ppgtt(struct i915_ggtt *ggtt) { - struct i915_ggtt *ggtt = &i915->ggtt; + struct drm_i915_private *i915 = ggtt->vm.i915; struct i915_ppgtt *ppgtt; - ppgtt = fetch_and_zero(&i915->mm.aliasing_ppgtt); + mutex_lock(&i915->drm.struct_mutex); + + ppgtt = fetch_and_zero(&ggtt->alias); if (!ppgtt) - return; + goto out; i915_vm_put(&ppgtt->vm); ggtt->vm.vma_ops.bind_vma = ggtt_bind_vma; ggtt->vm.vma_ops.unbind_vma = ggtt_unbind_vma; + +out: + mutex_unlock(&i915->drm.struct_mutex); } static int ggtt_reserve_guc_top(struct i915_ggtt *ggtt) @@ -2834,7 +2628,13 @@ static void ggtt_release_guc_top(struct i915_ggtt *ggtt) drm_mm_remove_node(&ggtt->uc_fw); } -int i915_gem_init_ggtt(struct drm_i915_private *dev_priv) +static void cleanup_init_ggtt(struct i915_ggtt *ggtt) +{ + ggtt_release_guc_top(ggtt); + drm_mm_remove_node(&ggtt->error_capture); +} + +static int init_ggtt(struct i915_ggtt *ggtt) { /* Let GEM Manage all of the aperture. * @@ -2845,7 +2645,6 @@ int i915_gem_init_ggtt(struct drm_i915_private *dev_priv) * aperture. One page should be enough to keep any prefetching inside * of the aperture. */ - struct i915_ggtt *ggtt = &dev_priv->ggtt; unsigned long hole_start, hole_end; struct drm_mm_node *entry; int ret; @@ -2857,9 +2656,9 @@ int i915_gem_init_ggtt(struct drm_i915_private *dev_priv) * why. */ ggtt->pin_bias = max_t(u32, I915_GTT_PAGE_SIZE, - intel_wopcm_guc_size(&dev_priv->wopcm)); + intel_wopcm_guc_size(&ggtt->vm.i915->wopcm)); - ret = intel_vgt_balloon(dev_priv); + ret = intel_vgt_balloon(ggtt); if (ret) return ret; @@ -2878,7 +2677,7 @@ int i915_gem_init_ggtt(struct drm_i915_private *dev_priv) */ ret = ggtt_reserve_guc_top(ggtt); if (ret) - goto err_reserve; + goto err; /* Clear any non-preallocated blocks */ drm_mm_for_each_hole(entry, &ggtt->vm.mm, hole_start, hole_end) { @@ -2891,35 +2690,41 @@ int i915_gem_init_ggtt(struct drm_i915_private *dev_priv) /* And finally clear the reserved guard page */ ggtt->vm.clear_range(&ggtt->vm, ggtt->vm.total - PAGE_SIZE, PAGE_SIZE); - if (INTEL_PPGTT(dev_priv) == INTEL_PPGTT_ALIASING) { - ret = init_aliasing_ppgtt(dev_priv); + return 0; + +err: + cleanup_init_ggtt(ggtt); + return ret; +} + +int i915_init_ggtt(struct drm_i915_private *i915) +{ + int ret; + + ret = init_ggtt(&i915->ggtt); + if (ret) + return ret; + + if (INTEL_PPGTT(i915) == INTEL_PPGTT_ALIASING) { + ret = init_aliasing_ppgtt(&i915->ggtt); if (ret) - goto err_appgtt; + cleanup_init_ggtt(&i915->ggtt); } return 0; - -err_appgtt: - ggtt_release_guc_top(ggtt); -err_reserve: - drm_mm_remove_node(&ggtt->error_capture); - return ret; } -/** - * i915_ggtt_cleanup_hw - Clean up GGTT hardware initialization - * @dev_priv: i915 device - */ -void i915_ggtt_cleanup_hw(struct drm_i915_private *dev_priv) +static void ggtt_cleanup_hw(struct i915_ggtt *ggtt) { - struct i915_ggtt *ggtt = &dev_priv->ggtt; + struct drm_i915_private *i915 = ggtt->vm.i915; struct i915_vma *vma, *vn; - struct pagevec *pvec; ggtt->vm.closed = true; - mutex_lock(&dev_priv->drm.struct_mutex); - fini_aliasing_ppgtt(dev_priv); + rcu_barrier(); /* flush the RCU'ed__i915_vm_release */ + flush_workqueue(i915->wq); + + mutex_lock(&i915->drm.struct_mutex); list_for_each_entry_safe(vma, vn, &ggtt->vm.bound_list, vm_link) WARN_ON(i915_vma_unbind(vma)); @@ -2930,24 +2735,37 @@ void i915_ggtt_cleanup_hw(struct drm_i915_private *dev_priv) ggtt_release_guc_top(ggtt); if (drm_mm_initialized(&ggtt->vm.mm)) { - intel_vgt_deballoon(dev_priv); + intel_vgt_deballoon(ggtt); i915_address_space_fini(&ggtt->vm); } ggtt->vm.cleanup(&ggtt->vm); - pvec = &dev_priv->mm.wc_stash.pvec; + mutex_unlock(&i915->drm.struct_mutex); + + arch_phys_wc_del(ggtt->mtrr); + io_mapping_fini(&ggtt->iomap); +} + +/** + * i915_ggtt_driver_release - Clean up GGTT hardware initialization + * @i915: i915 device + */ +void i915_ggtt_driver_release(struct drm_i915_private *i915) +{ + struct pagevec *pvec; + + fini_aliasing_ppgtt(&i915->ggtt); + + ggtt_cleanup_hw(&i915->ggtt); + + pvec = &i915->mm.wc_stash.pvec; if (pvec->nr) { set_pages_array_wb(pvec->pages, pvec->nr); __pagevec_release(pvec); } - mutex_unlock(&dev_priv->drm.struct_mutex); - - arch_phys_wc_del(ggtt->mtrr); - io_mapping_fini(&ggtt->iomap); - - i915_gem_cleanup_stolen(dev_priv); + i915_gem_cleanup_stolen(i915); } static unsigned int gen6_get_total_gtt_size(u16 snb_gmch_ctl) @@ -3018,243 +2836,48 @@ static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size) return ret; } - ggtt->vm.scratch_pte = - ggtt->vm.pte_encode(ggtt->vm.scratch_page.daddr, + ggtt->vm.scratch[0].encode = + ggtt->vm.pte_encode(px_dma(&ggtt->vm.scratch[0]), I915_CACHE_NONE, 0); return 0; } -static struct intel_ppat_entry * -__alloc_ppat_entry(struct intel_ppat *ppat, unsigned int index, u8 value) +static void cnl_setup_private_ppat(struct drm_i915_private *dev_priv) { - struct intel_ppat_entry *entry = &ppat->entries[index]; - - GEM_BUG_ON(index >= ppat->max_entries); - GEM_BUG_ON(test_bit(index, ppat->used)); - - entry->ppat = ppat; - entry->value = value; - kref_init(&entry->ref); - set_bit(index, ppat->used); - set_bit(index, ppat->dirty); - - return entry; + I915_WRITE(GEN10_PAT_INDEX(0), GEN8_PPAT_WB | GEN8_PPAT_LLC); + I915_WRITE(GEN10_PAT_INDEX(1), GEN8_PPAT_WC | GEN8_PPAT_LLCELLC); + I915_WRITE(GEN10_PAT_INDEX(2), GEN8_PPAT_WT | GEN8_PPAT_LLCELLC); + I915_WRITE(GEN10_PAT_INDEX(3), GEN8_PPAT_UC); + I915_WRITE(GEN10_PAT_INDEX(4), GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(0)); + I915_WRITE(GEN10_PAT_INDEX(5), GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(1)); + I915_WRITE(GEN10_PAT_INDEX(6), GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(2)); + I915_WRITE(GEN10_PAT_INDEX(7), GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(3)); } -static void __free_ppat_entry(struct intel_ppat_entry *entry) -{ - struct intel_ppat *ppat = entry->ppat; - unsigned int index = entry - ppat->entries; - - GEM_BUG_ON(index >= ppat->max_entries); - GEM_BUG_ON(!test_bit(index, ppat->used)); - - entry->value = ppat->clear_value; - clear_bit(index, ppat->used); - set_bit(index, ppat->dirty); -} - -/** - * intel_ppat_get - get a usable PPAT entry - * @i915: i915 device instance - * @value: the PPAT value required by the caller - * - * The function tries to search if there is an existing PPAT entry which - * matches with the required value. If perfectly matched, the existing PPAT - * entry will be used. If only partially matched, it will try to check if - * there is any available PPAT index. If yes, it will allocate a new PPAT - * index for the required entry and update the HW. If not, the partially - * matched entry will be used. - */ -const struct intel_ppat_entry * -intel_ppat_get(struct drm_i915_private *i915, u8 value) -{ - struct intel_ppat *ppat = &i915->ppat; - struct intel_ppat_entry *entry = NULL; - unsigned int scanned, best_score; - int i; - - GEM_BUG_ON(!ppat->max_entries); - - scanned = best_score = 0; - for_each_set_bit(i, ppat->used, ppat->max_entries) { - unsigned int score; - - score = ppat->match(ppat->entries[i].value, value); - if (score > best_score) { - entry = &ppat->entries[i]; - if (score == INTEL_PPAT_PERFECT_MATCH) { - kref_get(&entry->ref); - return entry; - } - best_score = score; - } - scanned++; - } - - if (scanned == ppat->max_entries) { - if (!entry) - return ERR_PTR(-ENOSPC); - - kref_get(&entry->ref); - return entry; - } - - i = find_first_zero_bit(ppat->used, ppat->max_entries); - entry = __alloc_ppat_entry(ppat, i, value); - ppat->update_hw(i915); - return entry; -} - -static void release_ppat(struct kref *kref) -{ - struct intel_ppat_entry *entry = - container_of(kref, struct intel_ppat_entry, ref); - struct drm_i915_private *i915 = entry->ppat->i915; - - __free_ppat_entry(entry); - entry->ppat->update_hw(i915); -} - -/** - * intel_ppat_put - put back the PPAT entry got from intel_ppat_get() - * @entry: an intel PPAT entry - * - * Put back the PPAT entry got from intel_ppat_get(). If the PPAT index of the - * entry is dynamically allocated, its reference count will be decreased. Once - * the reference count becomes into zero, the PPAT index becomes free again. - */ -void intel_ppat_put(const struct intel_ppat_entry *entry) -{ - struct intel_ppat *ppat = entry->ppat; - unsigned int index = entry - ppat->entries; - - GEM_BUG_ON(!ppat->max_entries); - - kref_put(&ppat->entries[index].ref, release_ppat); -} - -static void cnl_private_pat_update_hw(struct drm_i915_private *dev_priv) -{ - struct intel_ppat *ppat = &dev_priv->ppat; - int i; - - for_each_set_bit(i, ppat->dirty, ppat->max_entries) { - I915_WRITE(GEN10_PAT_INDEX(i), ppat->entries[i].value); - clear_bit(i, ppat->dirty); - } -} - -static void bdw_private_pat_update_hw(struct drm_i915_private *dev_priv) +/* The GGTT and PPGTT need a private PPAT setup in order to handle cacheability + * bits. When using advanced contexts each context stores its own PAT, but + * writing this data shouldn't be harmful even in those cases. */ +static void bdw_setup_private_ppat(struct drm_i915_private *dev_priv) { - struct intel_ppat *ppat = &dev_priv->ppat; - u64 pat = 0; - int i; - - for (i = 0; i < ppat->max_entries; i++) - pat |= GEN8_PPAT(i, ppat->entries[i].value); + u64 pat; - bitmap_clear(ppat->dirty, 0, ppat->max_entries); + pat = GEN8_PPAT(0, GEN8_PPAT_WB | GEN8_PPAT_LLC) | /* for normal objects, no eLLC */ + GEN8_PPAT(1, GEN8_PPAT_WC | GEN8_PPAT_LLCELLC) | /* for something pointing to ptes? */ + GEN8_PPAT(2, GEN8_PPAT_WT | GEN8_PPAT_LLCELLC) | /* for scanout with eLLC */ + GEN8_PPAT(3, GEN8_PPAT_UC) | /* Uncached objects, mostly for scanout */ + GEN8_PPAT(4, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(0)) | + GEN8_PPAT(5, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(1)) | + GEN8_PPAT(6, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(2)) | + GEN8_PPAT(7, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(3)); I915_WRITE(GEN8_PRIVATE_PAT_LO, lower_32_bits(pat)); I915_WRITE(GEN8_PRIVATE_PAT_HI, upper_32_bits(pat)); } -static unsigned int bdw_private_pat_match(u8 src, u8 dst) +static void chv_setup_private_ppat(struct drm_i915_private *dev_priv) { - unsigned int score = 0; - enum { - AGE_MATCH = BIT(0), - TC_MATCH = BIT(1), - CA_MATCH = BIT(2), - }; - - /* Cache attribute has to be matched. */ - if (GEN8_PPAT_GET_CA(src) != GEN8_PPAT_GET_CA(dst)) - return 0; - - score |= CA_MATCH; - - if (GEN8_PPAT_GET_TC(src) == GEN8_PPAT_GET_TC(dst)) - score |= TC_MATCH; - - if (GEN8_PPAT_GET_AGE(src) == GEN8_PPAT_GET_AGE(dst)) - score |= AGE_MATCH; - - if (score == (AGE_MATCH | TC_MATCH | CA_MATCH)) - return INTEL_PPAT_PERFECT_MATCH; - - return score; -} - -static unsigned int chv_private_pat_match(u8 src, u8 dst) -{ - return (CHV_PPAT_GET_SNOOP(src) == CHV_PPAT_GET_SNOOP(dst)) ? - INTEL_PPAT_PERFECT_MATCH : 0; -} - -static void cnl_setup_private_ppat(struct intel_ppat *ppat) -{ - ppat->max_entries = 8; - ppat->update_hw = cnl_private_pat_update_hw; - ppat->match = bdw_private_pat_match; - ppat->clear_value = GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(3); - - __alloc_ppat_entry(ppat, 0, GEN8_PPAT_WB | GEN8_PPAT_LLC); - __alloc_ppat_entry(ppat, 1, GEN8_PPAT_WC | GEN8_PPAT_LLCELLC); - __alloc_ppat_entry(ppat, 2, GEN8_PPAT_WT | GEN8_PPAT_LLCELLC); - __alloc_ppat_entry(ppat, 3, GEN8_PPAT_UC); - __alloc_ppat_entry(ppat, 4, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(0)); - __alloc_ppat_entry(ppat, 5, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(1)); - __alloc_ppat_entry(ppat, 6, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(2)); - __alloc_ppat_entry(ppat, 7, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(3)); -} - -/* The GGTT and PPGTT need a private PPAT setup in order to handle cacheability - * bits. When using advanced contexts each context stores its own PAT, but - * writing this data shouldn't be harmful even in those cases. */ -static void bdw_setup_private_ppat(struct intel_ppat *ppat) -{ - ppat->max_entries = 8; - ppat->update_hw = bdw_private_pat_update_hw; - ppat->match = bdw_private_pat_match; - ppat->clear_value = GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(3); - - if (!HAS_PPGTT(ppat->i915)) { - /* Spec: "For GGTT, there is NO pat_sel[2:0] from the entry, - * so RTL will always use the value corresponding to - * pat_sel = 000". - * So let's disable cache for GGTT to avoid screen corruptions. - * MOCS still can be used though. - * - System agent ggtt writes (i.e. cpu gtt mmaps) already work - * before this patch, i.e. the same uncached + snooping access - * like on gen6/7 seems to be in effect. - * - So this just fixes blitter/render access. Again it looks - * like it's not just uncached access, but uncached + snooping. - * So we can still hold onto all our assumptions wrt cpu - * clflushing on LLC machines. - */ - __alloc_ppat_entry(ppat, 0, GEN8_PPAT_UC); - return; - } - - __alloc_ppat_entry(ppat, 0, GEN8_PPAT_WB | GEN8_PPAT_LLC); /* for normal objects, no eLLC */ - __alloc_ppat_entry(ppat, 1, GEN8_PPAT_WC | GEN8_PPAT_LLCELLC); /* for something pointing to ptes? */ - __alloc_ppat_entry(ppat, 2, GEN8_PPAT_WT | GEN8_PPAT_LLCELLC); /* for scanout with eLLC */ - __alloc_ppat_entry(ppat, 3, GEN8_PPAT_UC); /* Uncached objects, mostly for scanout */ - __alloc_ppat_entry(ppat, 4, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(0)); - __alloc_ppat_entry(ppat, 5, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(1)); - __alloc_ppat_entry(ppat, 6, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(2)); - __alloc_ppat_entry(ppat, 7, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(3)); -} - -static void chv_setup_private_ppat(struct intel_ppat *ppat) -{ - ppat->max_entries = 8; - ppat->update_hw = bdw_private_pat_update_hw; - ppat->match = chv_private_pat_match; - ppat->clear_value = CHV_PPAT_SNOOP; + u64 pat; /* * Map WB on BDW to snooped on CHV. @@ -3275,14 +2898,17 @@ static void chv_setup_private_ppat(struct intel_ppat *ppat) * in order to keep the global status page working. */ - __alloc_ppat_entry(ppat, 0, CHV_PPAT_SNOOP); - __alloc_ppat_entry(ppat, 1, 0); - __alloc_ppat_entry(ppat, 2, 0); - __alloc_ppat_entry(ppat, 3, 0); - __alloc_ppat_entry(ppat, 4, CHV_PPAT_SNOOP); - __alloc_ppat_entry(ppat, 5, CHV_PPAT_SNOOP); - __alloc_ppat_entry(ppat, 6, CHV_PPAT_SNOOP); - __alloc_ppat_entry(ppat, 7, CHV_PPAT_SNOOP); + pat = GEN8_PPAT(0, CHV_PPAT_SNOOP) | + GEN8_PPAT(1, 0) | + GEN8_PPAT(2, 0) | + GEN8_PPAT(3, 0) | + GEN8_PPAT(4, CHV_PPAT_SNOOP) | + GEN8_PPAT(5, CHV_PPAT_SNOOP) | + GEN8_PPAT(6, CHV_PPAT_SNOOP) | + GEN8_PPAT(7, CHV_PPAT_SNOOP); + + I915_WRITE(GEN8_PRIVATE_PAT_LO, lower_32_bits(pat)); + I915_WRITE(GEN8_PRIVATE_PAT_HI, upper_32_bits(pat)); } static void gen6_gmch_remove(struct i915_address_space *vm) @@ -3295,27 +2921,14 @@ static void gen6_gmch_remove(struct i915_address_space *vm) static void setup_private_pat(struct drm_i915_private *dev_priv) { - struct intel_ppat *ppat = &dev_priv->ppat; - int i; - - ppat->i915 = dev_priv; + GEM_BUG_ON(INTEL_GEN(dev_priv) < 8); if (INTEL_GEN(dev_priv) >= 10) - cnl_setup_private_ppat(ppat); + cnl_setup_private_ppat(dev_priv); else if (IS_CHERRYVIEW(dev_priv) || IS_GEN9_LP(dev_priv)) - chv_setup_private_ppat(ppat); + chv_setup_private_ppat(dev_priv); else - bdw_setup_private_ppat(ppat); - - GEM_BUG_ON(ppat->max_entries > INTEL_MAX_PPAT_ENTRIES); - - for_each_clear_bit(i, ppat->used, ppat->max_entries) { - ppat->entries[i].value = ppat->clear_value; - ppat->entries[i].ppat = ppat; - set_bit(i, ppat->dirty); - } - - ppat->update_hw(dev_priv); + bdw_setup_private_ppat(dev_priv); } static int gen8_gmch_probe(struct i915_ggtt *ggtt) @@ -3360,11 +2973,6 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt) ggtt->vm.insert_page = bxt_vtd_ggtt_insert_page__BKL; if (ggtt->vm.clear_range != nop_clear_range) ggtt->vm.clear_range = bxt_vtd_ggtt_clear_range__BKL; - - /* Prevent recursively calling stop_machine() and deadlocks. */ - dev_info(dev_priv->drm.dev, - "Disabling error capture for VT-d workaround\n"); - i915_disable_error_state(dev_priv, -ENODEV); } ggtt->invalidate = gen6_ggtt_invalidate; @@ -3482,21 +3090,18 @@ static int i915_gmch_probe(struct i915_ggtt *ggtt) return 0; } -/** - * i915_ggtt_probe_hw - Probe GGTT hardware location - * @dev_priv: i915 device - */ -int i915_ggtt_probe_hw(struct drm_i915_private *dev_priv) +static int ggtt_probe_hw(struct i915_ggtt *ggtt, struct intel_gt *gt) { - struct i915_ggtt *ggtt = &dev_priv->ggtt; + struct drm_i915_private *i915 = gt->i915; int ret; - ggtt->vm.i915 = dev_priv; - ggtt->vm.dma = &dev_priv->drm.pdev->dev; + ggtt->vm.gt = gt; + ggtt->vm.i915 = i915; + ggtt->vm.dma = &i915->drm.pdev->dev; - if (INTEL_GEN(dev_priv) <= 5) + if (INTEL_GEN(i915) <= 5) ret = i915_gmch_probe(ggtt); - else if (INTEL_GEN(dev_priv) < 8) + else if (INTEL_GEN(i915) < 8) ret = gen6_gmch_probe(ggtt); else ret = gen8_gmch_probe(ggtt); @@ -3524,51 +3129,82 @@ int i915_ggtt_probe_hw(struct drm_i915_private *dev_priv) DRM_DEBUG_DRIVER("GMADR size = %lluM\n", (u64)ggtt->mappable_end >> 20); DRM_DEBUG_DRIVER("DSM size = %lluM\n", (u64)resource_size(&intel_graphics_stolen_res) >> 20); - if (intel_vtd_active()) - DRM_INFO("VT-d active for gfx access\n"); return 0; } /** - * i915_ggtt_init_hw - Initialize GGTT hardware - * @dev_priv: i915 device + * i915_ggtt_probe_hw - Probe GGTT hardware location + * @i915: i915 device */ -int i915_ggtt_init_hw(struct drm_i915_private *dev_priv) +int i915_ggtt_probe_hw(struct drm_i915_private *i915) { - struct i915_ggtt *ggtt = &dev_priv->ggtt; int ret; - stash_init(&dev_priv->mm.wc_stash); + ret = ggtt_probe_hw(&i915->ggtt, &i915->gt); + if (ret) + return ret; + + if (intel_vtd_active()) + DRM_INFO("VT-d active for gfx access\n"); + + return 0; +} + +static int ggtt_init_hw(struct i915_ggtt *ggtt) +{ + struct drm_i915_private *i915 = ggtt->vm.i915; + int ret = 0; + + mutex_lock(&i915->drm.struct_mutex); - /* Note that we use page colouring to enforce a guard page at the - * end of the address space. This is required as the CS may prefetch - * beyond the end of the batch buffer, across the page boundary, - * and beyond the end of the GTT if we do not provide a guard. - */ - mutex_lock(&dev_priv->drm.struct_mutex); i915_address_space_init(&ggtt->vm, VM_CLASS_GGTT); ggtt->vm.is_ggtt = true; /* Only VLV supports read-only GGTT mappings */ - ggtt->vm.has_read_only = IS_VALLEYVIEW(dev_priv); + ggtt->vm.has_read_only = IS_VALLEYVIEW(i915); - if (!HAS_LLC(dev_priv) && !HAS_PPGTT(dev_priv)) + if (!HAS_LLC(i915) && !HAS_PPGTT(i915)) ggtt->vm.mm.color_adjust = i915_gtt_color_adjust; - mutex_unlock(&dev_priv->drm.struct_mutex); - if (!io_mapping_init_wc(&dev_priv->ggtt.iomap, - dev_priv->ggtt.gmadr.start, - dev_priv->ggtt.mappable_end)) { + if (!io_mapping_init_wc(&ggtt->iomap, + ggtt->gmadr.start, + ggtt->mappable_end)) { + ggtt->vm.cleanup(&ggtt->vm); ret = -EIO; - goto out_gtt_cleanup; + goto out; } ggtt->mtrr = arch_phys_wc_add(ggtt->gmadr.start, ggtt->mappable_end); i915_ggtt_init_fences(ggtt); +out: + mutex_unlock(&i915->drm.struct_mutex); + + return ret; +} + +/** + * i915_ggtt_init_hw - Initialize GGTT hardware + * @dev_priv: i915 device + */ +int i915_ggtt_init_hw(struct drm_i915_private *dev_priv) +{ + int ret; + + stash_init(&dev_priv->mm.wc_stash); + + /* Note that we use page colouring to enforce a guard page at the + * end of the address space. This is required as the CS may prefetch + * beyond the end of the batch buffer, across the page boundary, + * and beyond the end of the GTT if we do not provide a guard. + */ + ret = ggtt_init_hw(&dev_priv->ggtt); + if (ret) + return ret; + /* * Initialise stolen early so that we may reserve preallocated * objects for the BIOS to KMS transition. @@ -3580,7 +3216,7 @@ int i915_ggtt_init_hw(struct drm_i915_private *dev_priv) return 0; out_gtt_cleanup: - ggtt->vm.cleanup(&ggtt->vm); + dev_priv->ggtt.vm.cleanup(&dev_priv->ggtt.vm); return ret; } @@ -3592,35 +3228,34 @@ int i915_ggtt_enable_hw(struct drm_i915_private *dev_priv) return 0; } -void i915_ggtt_enable_guc(struct drm_i915_private *i915) +void i915_ggtt_enable_guc(struct i915_ggtt *ggtt) { - GEM_BUG_ON(i915->ggtt.invalidate != gen6_ggtt_invalidate); + GEM_BUG_ON(ggtt->invalidate != gen6_ggtt_invalidate); - i915->ggtt.invalidate = guc_ggtt_invalidate; + ggtt->invalidate = guc_ggtt_invalidate; - i915_ggtt_invalidate(i915); + ggtt->invalidate(ggtt); } -void i915_ggtt_disable_guc(struct drm_i915_private *i915) +void i915_ggtt_disable_guc(struct i915_ggtt *ggtt) { /* XXX Temporary pardon for error unload */ - if (i915->ggtt.invalidate == gen6_ggtt_invalidate) + if (ggtt->invalidate == gen6_ggtt_invalidate) return; /* We should only be called after i915_ggtt_enable_guc() */ - GEM_BUG_ON(i915->ggtt.invalidate != guc_ggtt_invalidate); + GEM_BUG_ON(ggtt->invalidate != guc_ggtt_invalidate); - i915->ggtt.invalidate = gen6_ggtt_invalidate; + ggtt->invalidate = gen6_ggtt_invalidate; - i915_ggtt_invalidate(i915); + ggtt->invalidate(ggtt); } -void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv) +static void ggtt_restore_mappings(struct i915_ggtt *ggtt) { - struct i915_ggtt *ggtt = &dev_priv->ggtt; struct i915_vma *vma, *vn; - i915_check_and_clear_faults(dev_priv); + intel_gt_check_and_clear_faults(ggtt->vm.gt); mutex_lock(&ggtt->vm.mutex); @@ -3654,17 +3289,17 @@ lock: } ggtt->vm.closed = false; - i915_ggtt_invalidate(dev_priv); + ggtt->invalidate(ggtt); mutex_unlock(&ggtt->vm.mutex); +} - if (INTEL_GEN(dev_priv) >= 8) { - struct intel_ppat *ppat = &dev_priv->ppat; +void i915_gem_restore_gtt_mappings(struct drm_i915_private *i915) +{ + ggtt_restore_mappings(&i915->ggtt); - bitmap_set(ppat->dirty, 0, ppat->max_entries); - dev_priv->ppat.update_hw(dev_priv); - return; - } + if (INTEL_GEN(i915) >= 8) + setup_private_pat(i915); } static struct scatterlist * @@ -3953,7 +3588,7 @@ int i915_gem_gtt_reserve(struct i915_address_space *vm, GEM_BUG_ON(!IS_ALIGNED(size, I915_GTT_PAGE_SIZE)); GEM_BUG_ON(!IS_ALIGNED(offset, I915_GTT_MIN_ALIGNMENT)); GEM_BUG_ON(range_overflows(offset, size, vm->total)); - GEM_BUG_ON(vm == &vm->i915->mm.aliasing_ppgtt->vm); + GEM_BUG_ON(vm == &vm->i915->ggtt.alias->vm); GEM_BUG_ON(drm_mm_node_allocated(node)); node->size = size; @@ -4050,7 +3685,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm, GEM_BUG_ON(start >= end); GEM_BUG_ON(start > 0 && !IS_ALIGNED(start, I915_GTT_PAGE_SIZE)); GEM_BUG_ON(end < U64_MAX && !IS_ALIGNED(end, I915_GTT_PAGE_SIZE)); - GEM_BUG_ON(vm == &vm->i915->mm.aliasing_ppgtt->vm); + GEM_BUG_ON(vm == &vm->i915->ggtt.alias->vm); GEM_BUG_ON(drm_mm_node_allocated(node)); if (unlikely(range_overflows(start, size, end))) |