From 8d91c1e411f55d7ea91b1183a2e9f8088fb4d5be Mon Sep 17 00:00:00 2001 From: André Fabian Silva Delgado Date: Tue, 15 Dec 2015 14:52:16 -0300 Subject: Linux-libre 4.3.2-gnu --- drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c | 234 +++++++++++++++--------- drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c | 138 ++++++-------- drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c | 128 +++++++------ drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h | 15 +- drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c | 136 ++++++-------- drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c | 195 ++++++++++---------- drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c | 174 +++++++++--------- drivers/gpu/drm/nouveau/nvkm/subdev/mmu/priv.h | 39 ++++ 8 files changed, 552 insertions(+), 507 deletions(-) create mode 100644 drivers/gpu/drm/nouveau/nvkm/subdev/mmu/priv.h (limited to 'drivers/gpu/drm/nouveau/nvkm/subdev/mmu') diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c index 277b6ec04..e04a2296e 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c @@ -21,10 +21,10 @@ * * Authors: Ben Skeggs */ -#include -#include +#include "priv.h" #include +#include void nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node) @@ -32,12 +32,12 @@ nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node) struct nvkm_vm *vm = vma->vm; struct nvkm_mmu *mmu = vm->mmu; struct nvkm_mm_node *r; - int big = vma->node->type != mmu->spg_shift; + int big = vma->node->type != mmu->func->spg_shift; u32 offset = vma->node->offset + (delta >> 12); u32 bits = vma->node->type - 12; - u32 pde = (offset >> mmu->pgt_bits) - vm->fpde; - u32 pte = (offset & ((1 << mmu->pgt_bits) - 1)) >> bits; - u32 max = 1 << (mmu->pgt_bits - bits); + u32 pde = (offset >> mmu->func->pgt_bits) - vm->fpde; + u32 pte = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits; + u32 max = 1 << (mmu->func->pgt_bits - bits); u32 end, len; delta = 0; @@ -46,14 +46,14 @@ nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node) u32 num = r->length >> bits; while (num) { - struct nvkm_gpuobj *pgt = vm->pgt[pde].obj[big]; + struct nvkm_memory *pgt = vm->pgt[pde].mem[big]; end = (pte + num); if (unlikely(end >= max)) end = max; len = end - pte; - mmu->map(vma, pgt, node, pte, len, phys, delta); + mmu->func->map(vma, pgt, node, pte, len, phys, delta); num -= len; pte += len; @@ -67,7 +67,7 @@ nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node) } } - mmu->flush(vm); + mmu->func->flush(vm); } static void @@ -76,20 +76,20 @@ nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length, { struct nvkm_vm *vm = vma->vm; struct nvkm_mmu *mmu = vm->mmu; - int big = vma->node->type != mmu->spg_shift; + int big = vma->node->type != mmu->func->spg_shift; u32 offset = vma->node->offset + (delta >> 12); u32 bits = vma->node->type - 12; u32 num = length >> vma->node->type; - u32 pde = (offset >> mmu->pgt_bits) - vm->fpde; - u32 pte = (offset & ((1 << mmu->pgt_bits) - 1)) >> bits; - u32 max = 1 << (mmu->pgt_bits - bits); + u32 pde = (offset >> mmu->func->pgt_bits) - vm->fpde; + u32 pte = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits; + u32 max = 1 << (mmu->func->pgt_bits - bits); unsigned m, sglen; u32 end, len; int i; struct scatterlist *sg; for_each_sg(mem->sg->sgl, sg, mem->sg->nents, i) { - struct nvkm_gpuobj *pgt = vm->pgt[pde].obj[big]; + struct nvkm_memory *pgt = vm->pgt[pde].mem[big]; sglen = sg_dma_len(sg) >> PAGE_SHIFT; end = pte + sglen; @@ -100,7 +100,7 @@ nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length, for (m = 0; m < len; m++) { dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT); - mmu->map_sg(vma, pgt, mem, pte, 1, &addr); + mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr); num--; pte++; @@ -115,7 +115,7 @@ nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length, for (; m < sglen; m++) { dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT); - mmu->map_sg(vma, pgt, mem, pte, 1, &addr); + mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr); num--; pte++; if (num == 0) @@ -125,7 +125,7 @@ nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length, } finish: - mmu->flush(vm); + mmu->func->flush(vm); } static void @@ -135,24 +135,24 @@ nvkm_vm_map_sg(struct nvkm_vma *vma, u64 delta, u64 length, struct nvkm_vm *vm = vma->vm; struct nvkm_mmu *mmu = vm->mmu; dma_addr_t *list = mem->pages; - int big = vma->node->type != mmu->spg_shift; + int big = vma->node->type != mmu->func->spg_shift; u32 offset = vma->node->offset + (delta >> 12); u32 bits = vma->node->type - 12; u32 num = length >> vma->node->type; - u32 pde = (offset >> mmu->pgt_bits) - vm->fpde; - u32 pte = (offset & ((1 << mmu->pgt_bits) - 1)) >> bits; - u32 max = 1 << (mmu->pgt_bits - bits); + u32 pde = (offset >> mmu->func->pgt_bits) - vm->fpde; + u32 pte = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits; + u32 max = 1 << (mmu->func->pgt_bits - bits); u32 end, len; while (num) { - struct nvkm_gpuobj *pgt = vm->pgt[pde].obj[big]; + struct nvkm_memory *pgt = vm->pgt[pde].mem[big]; end = (pte + num); if (unlikely(end >= max)) end = max; len = end - pte; - mmu->map_sg(vma, pgt, mem, pte, len, list); + mmu->func->map_sg(vma, pgt, mem, pte, len, list); num -= len; pte += len; @@ -163,7 +163,7 @@ nvkm_vm_map_sg(struct nvkm_vma *vma, u64 delta, u64 length, } } - mmu->flush(vm); + mmu->func->flush(vm); } void @@ -183,24 +183,24 @@ nvkm_vm_unmap_at(struct nvkm_vma *vma, u64 delta, u64 length) { struct nvkm_vm *vm = vma->vm; struct nvkm_mmu *mmu = vm->mmu; - int big = vma->node->type != mmu->spg_shift; + int big = vma->node->type != mmu->func->spg_shift; u32 offset = vma->node->offset + (delta >> 12); u32 bits = vma->node->type - 12; u32 num = length >> vma->node->type; - u32 pde = (offset >> mmu->pgt_bits) - vm->fpde; - u32 pte = (offset & ((1 << mmu->pgt_bits) - 1)) >> bits; - u32 max = 1 << (mmu->pgt_bits - bits); + u32 pde = (offset >> mmu->func->pgt_bits) - vm->fpde; + u32 pte = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits; + u32 max = 1 << (mmu->func->pgt_bits - bits); u32 end, len; while (num) { - struct nvkm_gpuobj *pgt = vm->pgt[pde].obj[big]; + struct nvkm_memory *pgt = vm->pgt[pde].mem[big]; end = (pte + num); if (unlikely(end >= max)) end = max; len = end - pte; - mmu->unmap(pgt, pte, len); + mmu->func->unmap(vma, pgt, pte, len); num -= len; pte += len; @@ -210,7 +210,7 @@ nvkm_vm_unmap_at(struct nvkm_vma *vma, u64 delta, u64 length) } } - mmu->flush(vm); + mmu->func->flush(vm); } void @@ -225,7 +225,7 @@ nvkm_vm_unmap_pgt(struct nvkm_vm *vm, int big, u32 fpde, u32 lpde) struct nvkm_mmu *mmu = vm->mmu; struct nvkm_vm_pgd *vpgd; struct nvkm_vm_pgt *vpgt; - struct nvkm_gpuobj *pgt; + struct nvkm_memory *pgt; u32 pde; for (pde = fpde; pde <= lpde; pde++) { @@ -233,16 +233,14 @@ nvkm_vm_unmap_pgt(struct nvkm_vm *vm, int big, u32 fpde, u32 lpde) if (--vpgt->refcount[big]) continue; - pgt = vpgt->obj[big]; - vpgt->obj[big] = NULL; + pgt = vpgt->mem[big]; + vpgt->mem[big] = NULL; list_for_each_entry(vpgd, &vm->pgd_list, head) { - mmu->map_pgt(vpgd->obj, pde, vpgt->obj); + mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem); } - mutex_unlock(&nv_subdev(mmu)->mutex); - nvkm_gpuobj_ref(NULL, &pgt); - mutex_lock(&nv_subdev(mmu)->mutex); + nvkm_memory_del(&pgt); } } @@ -252,34 +250,23 @@ nvkm_vm_map_pgt(struct nvkm_vm *vm, u32 pde, u32 type) struct nvkm_mmu *mmu = vm->mmu; struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde]; struct nvkm_vm_pgd *vpgd; - struct nvkm_gpuobj *pgt; - int big = (type != mmu->spg_shift); + int big = (type != mmu->func->spg_shift); u32 pgt_size; int ret; - pgt_size = (1 << (mmu->pgt_bits + 12)) >> type; + pgt_size = (1 << (mmu->func->pgt_bits + 12)) >> type; pgt_size *= 8; - mutex_unlock(&nv_subdev(mmu)->mutex); - ret = nvkm_gpuobj_new(nv_object(vm->mmu), NULL, pgt_size, 0x1000, - NVOBJ_FLAG_ZERO_ALLOC, &pgt); - mutex_lock(&nv_subdev(mmu)->mutex); + ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST, + pgt_size, 0x1000, true, &vpgt->mem[big]); if (unlikely(ret)) return ret; - /* someone beat us to filling the PDE while we didn't have the lock */ - if (unlikely(vpgt->refcount[big]++)) { - mutex_unlock(&nv_subdev(mmu)->mutex); - nvkm_gpuobj_ref(NULL, &pgt); - mutex_lock(&nv_subdev(mmu)->mutex); - return 0; - } - - vpgt->obj[big] = pgt; list_for_each_entry(vpgd, &vm->pgd_list, head) { - mmu->map_pgt(vpgd->obj, pde, vpgt->obj); + mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem); } + vpgt->refcount[big]++; return 0; } @@ -293,20 +280,20 @@ nvkm_vm_get(struct nvkm_vm *vm, u64 size, u32 page_shift, u32 access, u32 fpde, lpde, pde; int ret; - mutex_lock(&nv_subdev(mmu)->mutex); + mutex_lock(&vm->mutex); ret = nvkm_mm_head(&vm->mm, 0, page_shift, msize, msize, align, &vma->node); if (unlikely(ret != 0)) { - mutex_unlock(&nv_subdev(mmu)->mutex); + mutex_unlock(&vm->mutex); return ret; } - fpde = (vma->node->offset >> mmu->pgt_bits); - lpde = (vma->node->offset + vma->node->length - 1) >> mmu->pgt_bits; + fpde = (vma->node->offset >> mmu->func->pgt_bits); + lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits; for (pde = fpde; pde <= lpde; pde++) { struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde]; - int big = (vma->node->type != mmu->spg_shift); + int big = (vma->node->type != mmu->func->spg_shift); if (likely(vpgt->refcount[big])) { vpgt->refcount[big]++; @@ -318,11 +305,11 @@ nvkm_vm_get(struct nvkm_vm *vm, u64 size, u32 page_shift, u32 access, if (pde != fpde) nvkm_vm_unmap_pgt(vm, big, fpde, pde - 1); nvkm_mm_free(&vm->mm, &vma->node); - mutex_unlock(&nv_subdev(mmu)->mutex); + mutex_unlock(&vm->mutex); return ret; } } - mutex_unlock(&nv_subdev(mmu)->mutex); + mutex_unlock(&vm->mutex); vma->vm = NULL; nvkm_vm_ref(vm, &vma->vm, NULL); @@ -334,27 +321,49 @@ nvkm_vm_get(struct nvkm_vm *vm, u64 size, u32 page_shift, u32 access, void nvkm_vm_put(struct nvkm_vma *vma) { - struct nvkm_vm *vm = vma->vm; - struct nvkm_mmu *mmu = vm->mmu; + struct nvkm_mmu *mmu; + struct nvkm_vm *vm; u32 fpde, lpde; if (unlikely(vma->node == NULL)) return; - fpde = (vma->node->offset >> mmu->pgt_bits); - lpde = (vma->node->offset + vma->node->length - 1) >> mmu->pgt_bits; + vm = vma->vm; + mmu = vm->mmu; + + fpde = (vma->node->offset >> mmu->func->pgt_bits); + lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits; - mutex_lock(&nv_subdev(mmu)->mutex); - nvkm_vm_unmap_pgt(vm, vma->node->type != mmu->spg_shift, fpde, lpde); + mutex_lock(&vm->mutex); + nvkm_vm_unmap_pgt(vm, vma->node->type != mmu->func->spg_shift, fpde, lpde); nvkm_mm_free(&vm->mm, &vma->node); - mutex_unlock(&nv_subdev(mmu)->mutex); + mutex_unlock(&vm->mutex); nvkm_vm_ref(NULL, &vma->vm, NULL); } +int +nvkm_vm_boot(struct nvkm_vm *vm, u64 size) +{ + struct nvkm_mmu *mmu = vm->mmu; + struct nvkm_memory *pgt; + int ret; + + ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST, + (size >> mmu->func->spg_shift) * 8, 0x1000, true, &pgt); + if (ret == 0) { + vm->pgt[0].refcount[0] = 1; + vm->pgt[0].mem[0] = pgt; + nvkm_memory_boot(pgt, vm); + } + + return ret; +} + int nvkm_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset, - u32 block, struct nvkm_vm **pvm) + u32 block, struct lock_class_key *key, struct nvkm_vm **pvm) { + static struct lock_class_key _key; struct nvkm_vm *vm; u64 mm_length = (offset + length) - mm_offset; int ret; @@ -363,11 +372,12 @@ nvkm_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset, if (!vm) return -ENOMEM; + __mutex_init(&vm->mutex, "&vm->mutex", key ? key : &_key); INIT_LIST_HEAD(&vm->pgd_list); vm->mmu = mmu; kref_init(&vm->refcount); - vm->fpde = offset >> (mmu->pgt_bits + 12); - vm->lpde = (offset + length - 1) >> (mmu->pgt_bits + 12); + vm->fpde = offset >> (mmu->func->pgt_bits + 12); + vm->lpde = (offset + length - 1) >> (mmu->func->pgt_bits + 12); vm->pgt = vzalloc((vm->lpde - vm->fpde + 1) * sizeof(*vm->pgt)); if (!vm->pgt) { @@ -390,10 +400,12 @@ nvkm_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset, int nvkm_vm_new(struct nvkm_device *device, u64 offset, u64 length, u64 mm_offset, - struct nvkm_vm **pvm) + struct lock_class_key *key, struct nvkm_vm **pvm) { - struct nvkm_mmu *mmu = nvkm_mmu(device); - return mmu->create(mmu, offset, length, mm_offset, pvm); + struct nvkm_mmu *mmu = device->mmu; + if (!mmu->func->create) + return -EINVAL; + return mmu->func->create(mmu, offset, length, mm_offset, key, pvm); } static int @@ -410,38 +422,33 @@ nvkm_vm_link(struct nvkm_vm *vm, struct nvkm_gpuobj *pgd) if (!vpgd) return -ENOMEM; - nvkm_gpuobj_ref(pgd, &vpgd->obj); + vpgd->obj = pgd; - mutex_lock(&nv_subdev(mmu)->mutex); + mutex_lock(&vm->mutex); for (i = vm->fpde; i <= vm->lpde; i++) - mmu->map_pgt(pgd, i, vm->pgt[i - vm->fpde].obj); + mmu->func->map_pgt(pgd, i, vm->pgt[i - vm->fpde].mem); list_add(&vpgd->head, &vm->pgd_list); - mutex_unlock(&nv_subdev(mmu)->mutex); + mutex_unlock(&vm->mutex); return 0; } static void nvkm_vm_unlink(struct nvkm_vm *vm, struct nvkm_gpuobj *mpgd) { - struct nvkm_mmu *mmu = vm->mmu; struct nvkm_vm_pgd *vpgd, *tmp; - struct nvkm_gpuobj *pgd = NULL; if (!mpgd) return; - mutex_lock(&nv_subdev(mmu)->mutex); + mutex_lock(&vm->mutex); list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) { if (vpgd->obj == mpgd) { - pgd = vpgd->obj; list_del(&vpgd->head); kfree(vpgd); break; } } - mutex_unlock(&nv_subdev(mmu)->mutex); - - nvkm_gpuobj_ref(NULL, &pgd); + mutex_unlock(&vm->mutex); } static void @@ -478,3 +485,58 @@ nvkm_vm_ref(struct nvkm_vm *ref, struct nvkm_vm **ptr, struct nvkm_gpuobj *pgd) *ptr = ref; return 0; } + +static int +nvkm_mmu_oneinit(struct nvkm_subdev *subdev) +{ + struct nvkm_mmu *mmu = nvkm_mmu(subdev); + if (mmu->func->oneinit) + return mmu->func->oneinit(mmu); + return 0; +} + +static int +nvkm_mmu_init(struct nvkm_subdev *subdev) +{ + struct nvkm_mmu *mmu = nvkm_mmu(subdev); + if (mmu->func->init) + mmu->func->init(mmu); + return 0; +} + +static void * +nvkm_mmu_dtor(struct nvkm_subdev *subdev) +{ + struct nvkm_mmu *mmu = nvkm_mmu(subdev); + if (mmu->func->dtor) + return mmu->func->dtor(mmu); + return mmu; +} + +static const struct nvkm_subdev_func +nvkm_mmu = { + .dtor = nvkm_mmu_dtor, + .oneinit = nvkm_mmu_oneinit, + .init = nvkm_mmu_init, +}; + +void +nvkm_mmu_ctor(const struct nvkm_mmu_func *func, struct nvkm_device *device, + int index, struct nvkm_mmu *mmu) +{ + nvkm_subdev_ctor(&nvkm_mmu, device, index, 0, &mmu->subdev); + mmu->func = func; + mmu->limit = func->limit; + mmu->dma_bits = func->dma_bits; + mmu->lpg_shift = func->lpg_shift; +} + +int +nvkm_mmu_new_(const struct nvkm_mmu_func *func, struct nvkm_device *device, + int index, struct nvkm_mmu **pmmu) +{ + if (!(*pmmu = kzalloc(sizeof(**pmmu), GFP_KERNEL))) + return -ENOMEM; + nvkm_mmu_ctor(func, device, index, *pmmu); + return 0; +} diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c index 294cda37f..7ac507c92 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/gf100.c @@ -21,19 +21,14 @@ * * Authors: Ben Skeggs */ -#include -#include +#include "priv.h" + #include #include #include #include -struct gf100_mmu_priv { - struct nvkm_mmu base; -}; - - /* Map from compressed to corresponding uncompressed storage type. * The value 0xff represents an invalid storage type. */ @@ -75,17 +70,19 @@ const u8 gf100_pte_storage_type_map[256] = static void -gf100_vm_map_pgt(struct nvkm_gpuobj *pgd, u32 index, struct nvkm_gpuobj *pgt[2]) +gf100_vm_map_pgt(struct nvkm_gpuobj *pgd, u32 index, struct nvkm_memory *pgt[2]) { u32 pde[2] = { 0, 0 }; if (pgt[0]) - pde[1] = 0x00000001 | (pgt[0]->addr >> 8); + pde[1] = 0x00000001 | (nvkm_memory_addr(pgt[0]) >> 8); if (pgt[1]) - pde[0] = 0x00000001 | (pgt[1]->addr >> 8); + pde[0] = 0x00000001 | (nvkm_memory_addr(pgt[1]) >> 8); - nv_wo32(pgd, (index * 8) + 0, pde[0]); - nv_wo32(pgd, (index * 8) + 4, pde[1]); + nvkm_kmap(pgd); + nvkm_wo32(pgd, (index * 8) + 0, pde[0]); + nvkm_wo32(pgd, (index * 8) + 4, pde[1]); + nvkm_done(pgd); } static inline u64 @@ -103,7 +100,7 @@ gf100_vm_addr(struct nvkm_vma *vma, u64 phys, u32 memtype, u32 target) } static void -gf100_vm_map(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt, +gf100_vm_map(struct nvkm_vma *vma, struct nvkm_memory *pgt, struct nvkm_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta) { u64 next = 1 << (vma->node->type - 8); @@ -112,126 +109,113 @@ gf100_vm_map(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt, pte <<= 3; if (mem->tag) { - struct nvkm_ltc *ltc = nvkm_ltc(vma->vm->mmu); + struct nvkm_ltc *ltc = vma->vm->mmu->subdev.device->ltc; u32 tag = mem->tag->offset + (delta >> 17); phys |= (u64)tag << (32 + 12); next |= (u64)1 << (32 + 12); - ltc->tags_clear(ltc, tag, cnt); + nvkm_ltc_tags_clear(ltc, tag, cnt); } + nvkm_kmap(pgt); while (cnt--) { - nv_wo32(pgt, pte + 0, lower_32_bits(phys)); - nv_wo32(pgt, pte + 4, upper_32_bits(phys)); + nvkm_wo32(pgt, pte + 0, lower_32_bits(phys)); + nvkm_wo32(pgt, pte + 4, upper_32_bits(phys)); phys += next; pte += 8; } + nvkm_done(pgt); } static void -gf100_vm_map_sg(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt, +gf100_vm_map_sg(struct nvkm_vma *vma, struct nvkm_memory *pgt, struct nvkm_mem *mem, u32 pte, u32 cnt, dma_addr_t *list) { u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 7 : 5; /* compressed storage types are invalid for system memory */ u32 memtype = gf100_pte_storage_type_map[mem->memtype & 0xff]; + nvkm_kmap(pgt); pte <<= 3; while (cnt--) { u64 phys = gf100_vm_addr(vma, *list++, memtype, target); - nv_wo32(pgt, pte + 0, lower_32_bits(phys)); - nv_wo32(pgt, pte + 4, upper_32_bits(phys)); + nvkm_wo32(pgt, pte + 0, lower_32_bits(phys)); + nvkm_wo32(pgt, pte + 4, upper_32_bits(phys)); pte += 8; } + nvkm_done(pgt); } static void -gf100_vm_unmap(struct nvkm_gpuobj *pgt, u32 pte, u32 cnt) +gf100_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt) { + nvkm_kmap(pgt); pte <<= 3; while (cnt--) { - nv_wo32(pgt, pte + 0, 0x00000000); - nv_wo32(pgt, pte + 4, 0x00000000); + nvkm_wo32(pgt, pte + 0, 0x00000000); + nvkm_wo32(pgt, pte + 4, 0x00000000); pte += 8; } + nvkm_done(pgt); } static void gf100_vm_flush(struct nvkm_vm *vm) { - struct gf100_mmu_priv *priv = (void *)vm->mmu; - struct nvkm_bar *bar = nvkm_bar(priv); + struct nvkm_mmu *mmu = vm->mmu; + struct nvkm_device *device = mmu->subdev.device; struct nvkm_vm_pgd *vpgd; u32 type; - bar->flush(bar); - type = 0x00000001; /* PAGE_ALL */ - if (atomic_read(&vm->engref[NVDEV_SUBDEV_BAR])) + if (atomic_read(&vm->engref[NVKM_SUBDEV_BAR])) type |= 0x00000004; /* HUB_ONLY */ - mutex_lock(&nv_subdev(priv)->mutex); + mutex_lock(&mmu->subdev.mutex); list_for_each_entry(vpgd, &vm->pgd_list, head) { /* looks like maybe a "free flush slots" counter, the * faster you write to 0x100cbc to more it decreases */ - if (!nv_wait_ne(priv, 0x100c80, 0x00ff0000, 0x00000000)) { - nv_error(priv, "vm timeout 0: 0x%08x %d\n", - nv_rd32(priv, 0x100c80), type); - } + nvkm_msec(device, 2000, + if (nvkm_rd32(device, 0x100c80) & 0x00ff0000) + break; + ); - nv_wr32(priv, 0x100cb8, vpgd->obj->addr >> 8); - nv_wr32(priv, 0x100cbc, 0x80000000 | type); + nvkm_wr32(device, 0x100cb8, vpgd->obj->addr >> 8); + nvkm_wr32(device, 0x100cbc, 0x80000000 | type); /* wait for flush to be queued? */ - if (!nv_wait(priv, 0x100c80, 0x00008000, 0x00008000)) { - nv_error(priv, "vm timeout 1: 0x%08x %d\n", - nv_rd32(priv, 0x100c80), type); - } + nvkm_msec(device, 2000, + if (nvkm_rd32(device, 0x100c80) & 0x00008000) + break; + ); } - mutex_unlock(&nv_subdev(priv)->mutex); + mutex_unlock(&mmu->subdev.mutex); } static int gf100_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset, - struct nvkm_vm **pvm) + struct lock_class_key *key, struct nvkm_vm **pvm) { - return nvkm_vm_create(mmu, offset, length, mm_offset, 4096, pvm); + return nvkm_vm_create(mmu, offset, length, mm_offset, 4096, key, pvm); } -static int -gf100_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +static const struct nvkm_mmu_func +gf100_mmu = { + .limit = (1ULL << 40), + .dma_bits = 40, + .pgt_bits = 27 - 12, + .spg_shift = 12, + .lpg_shift = 17, + .create = gf100_vm_create, + .map_pgt = gf100_vm_map_pgt, + .map = gf100_vm_map, + .map_sg = gf100_vm_map_sg, + .unmap = gf100_vm_unmap, + .flush = gf100_vm_flush, +}; + +int +gf100_mmu_new(struct nvkm_device *device, int index, struct nvkm_mmu **pmmu) { - struct gf100_mmu_priv *priv; - int ret; - - ret = nvkm_mmu_create(parent, engine, oclass, "VM", "vm", &priv); - *pobject = nv_object(priv); - if (ret) - return ret; - - priv->base.limit = 1ULL << 40; - priv->base.dma_bits = 40; - priv->base.pgt_bits = 27 - 12; - priv->base.spg_shift = 12; - priv->base.lpg_shift = 17; - priv->base.create = gf100_vm_create; - priv->base.map_pgt = gf100_vm_map_pgt; - priv->base.map = gf100_vm_map; - priv->base.map_sg = gf100_vm_map_sg; - priv->base.unmap = gf100_vm_unmap; - priv->base.flush = gf100_vm_flush; - return 0; + return nvkm_mmu_new_(&gf100_mmu, device, index, pmmu); } - -struct nvkm_oclass -gf100_mmu_oclass = { - .handle = NV_SUBDEV(MMU, 0xc0), - .ofuncs = &(struct nvkm_ofuncs) { - .ctor = gf100_mmu_ctor, - .dtor = _nvkm_mmu_dtor, - .init = _nvkm_mmu_init, - .fini = _nvkm_mmu_fini, - }, -}; diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c index fe93ea271..37927c3fd 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c @@ -23,7 +23,6 @@ */ #include "nv04.h" -#include #include #define NV04_PDMA_SIZE (128 * 1024 * 1024) @@ -34,30 +33,34 @@ ******************************************************************************/ static void -nv04_vm_map_sg(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt, +nv04_vm_map_sg(struct nvkm_vma *vma, struct nvkm_memory *pgt, struct nvkm_mem *mem, u32 pte, u32 cnt, dma_addr_t *list) { pte = 0x00008 + (pte * 4); + nvkm_kmap(pgt); while (cnt) { u32 page = PAGE_SIZE / NV04_PDMA_PAGE; u32 phys = (u32)*list++; while (cnt && page--) { - nv_wo32(pgt, pte, phys | 3); + nvkm_wo32(pgt, pte, phys | 3); phys += NV04_PDMA_PAGE; pte += 4; cnt -= 1; } } + nvkm_done(pgt); } static void -nv04_vm_unmap(struct nvkm_gpuobj *pgt, u32 pte, u32 cnt) +nv04_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt) { pte = 0x00008 + (pte * 4); + nvkm_kmap(pgt); while (cnt--) { - nv_wo32(pgt, pte, 0x00000000); + nvkm_wo32(pgt, pte, 0x00000000); pte += 4; } + nvkm_done(pgt); } static void @@ -65,87 +68,82 @@ nv04_vm_flush(struct nvkm_vm *vm) { } -/******************************************************************************* - * VM object - ******************************************************************************/ - -int -nv04_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mmstart, - struct nvkm_vm **pvm) -{ - return -EINVAL; -} - /******************************************************************************* * MMU subdev ******************************************************************************/ static int -nv04_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +nv04_mmu_oneinit(struct nvkm_mmu *base) { - struct nv04_mmu_priv *priv; - struct nvkm_gpuobj *dma; + struct nv04_mmu *mmu = nv04_mmu(base); + struct nvkm_device *device = mmu->base.subdev.device; + struct nvkm_memory *dma; int ret; - ret = nvkm_mmu_create(parent, engine, oclass, "PCIGART", - "pcigart", &priv); - *pobject = nv_object(priv); - if (ret) - return ret; - - priv->base.create = nv04_vm_create; - priv->base.limit = NV04_PDMA_SIZE; - priv->base.dma_bits = 32; - priv->base.pgt_bits = 32 - 12; - priv->base.spg_shift = 12; - priv->base.lpg_shift = 12; - priv->base.map_sg = nv04_vm_map_sg; - priv->base.unmap = nv04_vm_unmap; - priv->base.flush = nv04_vm_flush; - - ret = nvkm_vm_create(&priv->base, 0, NV04_PDMA_SIZE, 0, 4096, - &priv->vm); + ret = nvkm_vm_create(&mmu->base, 0, NV04_PDMA_SIZE, 0, 4096, NULL, + &mmu->vm); if (ret) return ret; - ret = nvkm_gpuobj_new(nv_object(priv), NULL, + ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, (NV04_PDMA_SIZE / NV04_PDMA_PAGE) * 4 + 8, - 16, NVOBJ_FLAG_ZERO_ALLOC, - &priv->vm->pgt[0].obj[0]); - dma = priv->vm->pgt[0].obj[0]; - priv->vm->pgt[0].refcount[0] = 1; + 16, true, &dma); + mmu->vm->pgt[0].mem[0] = dma; + mmu->vm->pgt[0].refcount[0] = 1; if (ret) return ret; - nv_wo32(dma, 0x00000, 0x0002103d); /* PCI, RW, PT, !LN */ - nv_wo32(dma, 0x00004, NV04_PDMA_SIZE - 1); + nvkm_kmap(dma); + nvkm_wo32(dma, 0x00000, 0x0002103d); /* PCI, RW, PT, !LN */ + nvkm_wo32(dma, 0x00004, NV04_PDMA_SIZE - 1); + nvkm_done(dma); return 0; } -void -nv04_mmu_dtor(struct nvkm_object *object) +void * +nv04_mmu_dtor(struct nvkm_mmu *base) { - struct nv04_mmu_priv *priv = (void *)object; - if (priv->vm) { - nvkm_gpuobj_ref(NULL, &priv->vm->pgt[0].obj[0]); - nvkm_vm_ref(NULL, &priv->vm, NULL); + struct nv04_mmu *mmu = nv04_mmu(base); + struct nvkm_device *device = mmu->base.subdev.device; + if (mmu->vm) { + nvkm_memory_del(&mmu->vm->pgt[0].mem[0]); + nvkm_vm_ref(NULL, &mmu->vm, NULL); } - if (priv->nullp) { - pci_free_consistent(nv_device(priv)->pdev, 16 * 1024, - priv->nullp, priv->null); + if (mmu->nullp) { + dma_free_coherent(device->dev, 16 * 1024, + mmu->nullp, mmu->null); } - nvkm_mmu_destroy(&priv->base); + return mmu; +} + +int +nv04_mmu_new_(const struct nvkm_mmu_func *func, struct nvkm_device *device, + int index, struct nvkm_mmu **pmmu) +{ + struct nv04_mmu *mmu; + if (!(mmu = kzalloc(sizeof(*mmu), GFP_KERNEL))) + return -ENOMEM; + *pmmu = &mmu->base; + nvkm_mmu_ctor(func, device, index, &mmu->base); + return 0; } -struct nvkm_oclass -nv04_mmu_oclass = { - .handle = NV_SUBDEV(MMU, 0x04), - .ofuncs = &(struct nvkm_ofuncs) { - .ctor = nv04_mmu_ctor, - .dtor = nv04_mmu_dtor, - .init = _nvkm_mmu_init, - .fini = _nvkm_mmu_fini, - }, +const struct nvkm_mmu_func +nv04_mmu = { + .oneinit = nv04_mmu_oneinit, + .dtor = nv04_mmu_dtor, + .limit = NV04_PDMA_SIZE, + .dma_bits = 32, + .pgt_bits = 32 - 12, + .spg_shift = 12, + .lpg_shift = 12, + .map_sg = nv04_vm_map_sg, + .unmap = nv04_vm_unmap, + .flush = nv04_vm_flush, }; + +int +nv04_mmu_new(struct nvkm_device *device, int index, struct nvkm_mmu **pmmu) +{ + return nv04_mmu_new_(&nv04_mmu, device, index, pmmu); +} diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h index 7bf6f4b38..363e33b29 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h @@ -1,19 +1,18 @@ #ifndef __NV04_MMU_PRIV__ #define __NV04_MMU_PRIV__ +#define nv04_mmu(p) container_of((p), struct nv04_mmu, base) +#include "priv.h" -#include - -struct nv04_mmu_priv { +struct nv04_mmu { struct nvkm_mmu base; struct nvkm_vm *vm; dma_addr_t null; void *nullp; }; -static inline struct nv04_mmu_priv * -nv04_mmu(void *obj) -{ - return (void *)nvkm_mmu(obj); -} +int nv04_mmu_new_(const struct nvkm_mmu_func *, struct nvkm_device *, + int index, struct nvkm_mmu **); +void *nv04_mmu_dtor(struct nvkm_mmu *); +extern const struct nvkm_mmu_func nv04_mmu; #endif diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c index 61ee3ab11..c6a26f907 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c @@ -23,7 +23,6 @@ */ #include "nv04.h" -#include #include #include #include @@ -36,45 +35,50 @@ ******************************************************************************/ static void -nv41_vm_map_sg(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt, +nv41_vm_map_sg(struct nvkm_vma *vma, struct nvkm_memory *pgt, struct nvkm_mem *mem, u32 pte, u32 cnt, dma_addr_t *list) { pte = pte * 4; + nvkm_kmap(pgt); while (cnt) { u32 page = PAGE_SIZE / NV41_GART_PAGE; u64 phys = (u64)*list++; while (cnt && page--) { - nv_wo32(pgt, pte, (phys >> 7) | 1); + nvkm_wo32(pgt, pte, (phys >> 7) | 1); phys += NV41_GART_PAGE; pte += 4; cnt -= 1; } } + nvkm_done(pgt); } static void -nv41_vm_unmap(struct nvkm_gpuobj *pgt, u32 pte, u32 cnt) +nv41_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt) { pte = pte * 4; + nvkm_kmap(pgt); while (cnt--) { - nv_wo32(pgt, pte, 0x00000000); + nvkm_wo32(pgt, pte, 0x00000000); pte += 4; } + nvkm_done(pgt); } static void nv41_vm_flush(struct nvkm_vm *vm) { - struct nv04_mmu_priv *priv = (void *)vm->mmu; - - mutex_lock(&nv_subdev(priv)->mutex); - nv_wr32(priv, 0x100810, 0x00000022); - if (!nv_wait(priv, 0x100810, 0x00000020, 0x00000020)) { - nv_warn(priv, "flush timeout, 0x%08x\n", - nv_rd32(priv, 0x100810)); - } - nv_wr32(priv, 0x100810, 0x00000000); - mutex_unlock(&nv_subdev(priv)->mutex); + struct nv04_mmu *mmu = nv04_mmu(vm->mmu); + struct nvkm_device *device = mmu->base.subdev.device; + + mutex_lock(&mmu->base.subdev.mutex); + nvkm_wr32(device, 0x100810, 0x00000022); + nvkm_msec(device, 2000, + if (nvkm_rd32(device, 0x100810) & 0x00000020) + break; + ); + nvkm_wr32(device, 0x100810, 0x00000000); + mutex_unlock(&mmu->base.subdev.mutex); } /******************************************************************************* @@ -82,76 +86,56 @@ nv41_vm_flush(struct nvkm_vm *vm) ******************************************************************************/ static int -nv41_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +nv41_mmu_oneinit(struct nvkm_mmu *base) { - struct nvkm_device *device = nv_device(parent); - struct nv04_mmu_priv *priv; + struct nv04_mmu *mmu = nv04_mmu(base); + struct nvkm_device *device = mmu->base.subdev.device; int ret; - if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) || - !nvkm_boolopt(device->cfgopt, "NvPCIE", true)) { - return nvkm_object_ctor(parent, engine, &nv04_mmu_oclass, - data, size, pobject); - } - - ret = nvkm_mmu_create(parent, engine, oclass, "PCIEGART", - "pciegart", &priv); - *pobject = nv_object(priv); - if (ret) - return ret; - - priv->base.create = nv04_vm_create; - priv->base.limit = NV41_GART_SIZE; - priv->base.dma_bits = 39; - priv->base.pgt_bits = 32 - 12; - priv->base.spg_shift = 12; - priv->base.lpg_shift = 12; - priv->base.map_sg = nv41_vm_map_sg; - priv->base.unmap = nv41_vm_unmap; - priv->base.flush = nv41_vm_flush; - - ret = nvkm_vm_create(&priv->base, 0, NV41_GART_SIZE, 0, 4096, - &priv->vm); - if (ret) - return ret; - - ret = nvkm_gpuobj_new(nv_object(priv), NULL, - (NV41_GART_SIZE / NV41_GART_PAGE) * 4, 16, - NVOBJ_FLAG_ZERO_ALLOC, - &priv->vm->pgt[0].obj[0]); - priv->vm->pgt[0].refcount[0] = 1; + ret = nvkm_vm_create(&mmu->base, 0, NV41_GART_SIZE, 0, 4096, NULL, + &mmu->vm); if (ret) return ret; - return 0; + ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, + (NV41_GART_SIZE / NV41_GART_PAGE) * 4, 16, true, + &mmu->vm->pgt[0].mem[0]); + mmu->vm->pgt[0].refcount[0] = 1; + return ret; } -static int -nv41_mmu_init(struct nvkm_object *object) +static void +nv41_mmu_init(struct nvkm_mmu *base) { - struct nv04_mmu_priv *priv = (void *)object; - struct nvkm_gpuobj *dma = priv->vm->pgt[0].obj[0]; - int ret; - - ret = nvkm_mmu_init(&priv->base); - if (ret) - return ret; - - nv_wr32(priv, 0x100800, dma->addr | 0x00000002); - nv_mask(priv, 0x10008c, 0x00000100, 0x00000100); - nv_wr32(priv, 0x100820, 0x00000000); - return 0; + struct nv04_mmu *mmu = nv04_mmu(base); + struct nvkm_device *device = mmu->base.subdev.device; + struct nvkm_memory *dma = mmu->vm->pgt[0].mem[0]; + nvkm_wr32(device, 0x100800, 0x00000002 | nvkm_memory_addr(dma)); + nvkm_mask(device, 0x10008c, 0x00000100, 0x00000100); + nvkm_wr32(device, 0x100820, 0x00000000); } -struct nvkm_oclass -nv41_mmu_oclass = { - .handle = NV_SUBDEV(MMU, 0x41), - .ofuncs = &(struct nvkm_ofuncs) { - .ctor = nv41_mmu_ctor, - .dtor = nv04_mmu_dtor, - .init = nv41_mmu_init, - .fini = _nvkm_mmu_fini, - }, +static const struct nvkm_mmu_func +nv41_mmu = { + .dtor = nv04_mmu_dtor, + .oneinit = nv41_mmu_oneinit, + .init = nv41_mmu_init, + .limit = NV41_GART_SIZE, + .dma_bits = 39, + .pgt_bits = 32 - 12, + .spg_shift = 12, + .lpg_shift = 12, + .map_sg = nv41_vm_map_sg, + .unmap = nv41_vm_unmap, + .flush = nv41_vm_flush, }; + +int +nv41_mmu_new(struct nvkm_device *device, int index, struct nvkm_mmu **pmmu) +{ + if (device->type == NVKM_DEVICE_AGP || + !nvkm_boolopt(device->cfgopt, "NvPCIE", true)) + return nv04_mmu_new(device, index, pmmu); + + return nv04_mmu_new_(&nv41_mmu, device, index, pmmu); +} diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c index b90ded188..a648c2395 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c @@ -23,7 +23,6 @@ */ #include "nv04.h" -#include #include #include #include @@ -36,16 +35,16 @@ ******************************************************************************/ static void -nv44_vm_fill(struct nvkm_gpuobj *pgt, dma_addr_t null, +nv44_vm_fill(struct nvkm_memory *pgt, dma_addr_t null, dma_addr_t *list, u32 pte, u32 cnt) { u32 base = (pte << 2) & ~0x0000000f; u32 tmp[4]; - tmp[0] = nv_ro32(pgt, base + 0x0); - tmp[1] = nv_ro32(pgt, base + 0x4); - tmp[2] = nv_ro32(pgt, base + 0x8); - tmp[3] = nv_ro32(pgt, base + 0xc); + tmp[0] = nvkm_ro32(pgt, base + 0x0); + tmp[1] = nvkm_ro32(pgt, base + 0x4); + tmp[2] = nvkm_ro32(pgt, base + 0x8); + tmp[3] = nvkm_ro32(pgt, base + 0xc); while (cnt--) { u32 addr = list ? (*list++ >> 12) : (null >> 12); @@ -75,24 +74,25 @@ nv44_vm_fill(struct nvkm_gpuobj *pgt, dma_addr_t null, } } - nv_wo32(pgt, base + 0x0, tmp[0]); - nv_wo32(pgt, base + 0x4, tmp[1]); - nv_wo32(pgt, base + 0x8, tmp[2]); - nv_wo32(pgt, base + 0xc, tmp[3] | 0x40000000); + nvkm_wo32(pgt, base + 0x0, tmp[0]); + nvkm_wo32(pgt, base + 0x4, tmp[1]); + nvkm_wo32(pgt, base + 0x8, tmp[2]); + nvkm_wo32(pgt, base + 0xc, tmp[3] | 0x40000000); } static void -nv44_vm_map_sg(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt, +nv44_vm_map_sg(struct nvkm_vma *vma, struct nvkm_memory *pgt, struct nvkm_mem *mem, u32 pte, u32 cnt, dma_addr_t *list) { - struct nv04_mmu_priv *priv = (void *)vma->vm->mmu; + struct nv04_mmu *mmu = nv04_mmu(vma->vm->mmu); u32 tmp[4]; int i; + nvkm_kmap(pgt); if (pte & 3) { u32 max = 4 - (pte & 3); u32 part = (cnt > max) ? max : cnt; - nv44_vm_fill(pgt, priv->null, list, pte, part); + nv44_vm_fill(pgt, mmu->null, list, pte, part); pte += part; list += part; cnt -= part; @@ -101,51 +101,57 @@ nv44_vm_map_sg(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt, while (cnt >= 4) { for (i = 0; i < 4; i++) tmp[i] = *list++ >> 12; - nv_wo32(pgt, pte++ * 4, tmp[0] >> 0 | tmp[1] << 27); - nv_wo32(pgt, pte++ * 4, tmp[1] >> 5 | tmp[2] << 22); - nv_wo32(pgt, pte++ * 4, tmp[2] >> 10 | tmp[3] << 17); - nv_wo32(pgt, pte++ * 4, tmp[3] >> 15 | 0x40000000); + nvkm_wo32(pgt, pte++ * 4, tmp[0] >> 0 | tmp[1] << 27); + nvkm_wo32(pgt, pte++ * 4, tmp[1] >> 5 | tmp[2] << 22); + nvkm_wo32(pgt, pte++ * 4, tmp[2] >> 10 | tmp[3] << 17); + nvkm_wo32(pgt, pte++ * 4, tmp[3] >> 15 | 0x40000000); cnt -= 4; } if (cnt) - nv44_vm_fill(pgt, priv->null, list, pte, cnt); + nv44_vm_fill(pgt, mmu->null, list, pte, cnt); + nvkm_done(pgt); } static void -nv44_vm_unmap(struct nvkm_gpuobj *pgt, u32 pte, u32 cnt) +nv44_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt) { - struct nv04_mmu_priv *priv = (void *)nvkm_mmu(pgt); + struct nv04_mmu *mmu = nv04_mmu(vma->vm->mmu); + nvkm_kmap(pgt); if (pte & 3) { u32 max = 4 - (pte & 3); u32 part = (cnt > max) ? max : cnt; - nv44_vm_fill(pgt, priv->null, NULL, pte, part); + nv44_vm_fill(pgt, mmu->null, NULL, pte, part); pte += part; cnt -= part; } while (cnt >= 4) { - nv_wo32(pgt, pte++ * 4, 0x00000000); - nv_wo32(pgt, pte++ * 4, 0x00000000); - nv_wo32(pgt, pte++ * 4, 0x00000000); - nv_wo32(pgt, pte++ * 4, 0x00000000); + nvkm_wo32(pgt, pte++ * 4, 0x00000000); + nvkm_wo32(pgt, pte++ * 4, 0x00000000); + nvkm_wo32(pgt, pte++ * 4, 0x00000000); + nvkm_wo32(pgt, pte++ * 4, 0x00000000); cnt -= 4; } if (cnt) - nv44_vm_fill(pgt, priv->null, NULL, pte, cnt); + nv44_vm_fill(pgt, mmu->null, NULL, pte, cnt); + nvkm_done(pgt); } static void nv44_vm_flush(struct nvkm_vm *vm) { - struct nv04_mmu_priv *priv = (void *)vm->mmu; - nv_wr32(priv, 0x100814, priv->base.limit - NV44_GART_PAGE); - nv_wr32(priv, 0x100808, 0x00000020); - if (!nv_wait(priv, 0x100808, 0x00000001, 0x00000001)) - nv_error(priv, "timeout: 0x%08x\n", nv_rd32(priv, 0x100808)); - nv_wr32(priv, 0x100808, 0x00000000); + struct nv04_mmu *mmu = nv04_mmu(vm->mmu); + struct nvkm_device *device = mmu->base.subdev.device; + nvkm_wr32(device, 0x100814, mmu->base.limit - NV44_GART_PAGE); + nvkm_wr32(device, 0x100808, 0x00000020); + nvkm_msec(device, 2000, + if (nvkm_rd32(device, 0x100808) & 0x00000001) + break; + ); + nvkm_wr32(device, 0x100808, 0x00000000); } /******************************************************************************* @@ -153,95 +159,78 @@ nv44_vm_flush(struct nvkm_vm *vm) ******************************************************************************/ static int -nv44_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +nv44_mmu_oneinit(struct nvkm_mmu *base) { - struct nvkm_device *device = nv_device(parent); - struct nv04_mmu_priv *priv; + struct nv04_mmu *mmu = nv04_mmu(base); + struct nvkm_device *device = mmu->base.subdev.device; int ret; - if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) || - !nvkm_boolopt(device->cfgopt, "NvPCIE", true)) { - return nvkm_object_ctor(parent, engine, &nv04_mmu_oclass, - data, size, pobject); - } - - ret = nvkm_mmu_create(parent, engine, oclass, "PCIEGART", - "pciegart", &priv); - *pobject = nv_object(priv); - if (ret) - return ret; - - priv->base.create = nv04_vm_create; - priv->base.limit = NV44_GART_SIZE; - priv->base.dma_bits = 39; - priv->base.pgt_bits = 32 - 12; - priv->base.spg_shift = 12; - priv->base.lpg_shift = 12; - priv->base.map_sg = nv44_vm_map_sg; - priv->base.unmap = nv44_vm_unmap; - priv->base.flush = nv44_vm_flush; - - priv->nullp = pci_alloc_consistent(device->pdev, 16 * 1024, &priv->null); - if (!priv->nullp) { - nv_error(priv, "unable to allocate dummy pages\n"); - return -ENOMEM; + mmu->nullp = dma_alloc_coherent(device->dev, 16 * 1024, + &mmu->null, GFP_KERNEL); + if (!mmu->nullp) { + nvkm_warn(&mmu->base.subdev, "unable to allocate dummy pages\n"); + mmu->null = 0; } - ret = nvkm_vm_create(&priv->base, 0, NV44_GART_SIZE, 0, 4096, - &priv->vm); + ret = nvkm_vm_create(&mmu->base, 0, NV44_GART_SIZE, 0, 4096, NULL, + &mmu->vm); if (ret) return ret; - ret = nvkm_gpuobj_new(nv_object(priv), NULL, + ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, (NV44_GART_SIZE / NV44_GART_PAGE) * 4, - 512 * 1024, NVOBJ_FLAG_ZERO_ALLOC, - &priv->vm->pgt[0].obj[0]); - priv->vm->pgt[0].refcount[0] = 1; - if (ret) - return ret; - - return 0; + 512 * 1024, true, + &mmu->vm->pgt[0].mem[0]); + mmu->vm->pgt[0].refcount[0] = 1; + return ret; } -static int -nv44_mmu_init(struct nvkm_object *object) +static void +nv44_mmu_init(struct nvkm_mmu *base) { - struct nv04_mmu_priv *priv = (void *)object; - struct nvkm_gpuobj *gart = priv->vm->pgt[0].obj[0]; + struct nv04_mmu *mmu = nv04_mmu(base); + struct nvkm_device *device = mmu->base.subdev.device; + struct nvkm_memory *gart = mmu->vm->pgt[0].mem[0]; u32 addr; - int ret; - - ret = nvkm_mmu_init(&priv->base); - if (ret) - return ret; /* calculate vram address of this PRAMIN block, object must be * allocated on 512KiB alignment, and not exceed a total size * of 512KiB for this to work correctly */ - addr = nv_rd32(priv, 0x10020c); - addr -= ((gart->addr >> 19) + 1) << 19; - - nv_wr32(priv, 0x100850, 0x80000000); - nv_wr32(priv, 0x100818, priv->null); - nv_wr32(priv, 0x100804, NV44_GART_SIZE); - nv_wr32(priv, 0x100850, 0x00008000); - nv_mask(priv, 0x10008c, 0x00000200, 0x00000200); - nv_wr32(priv, 0x100820, 0x00000000); - nv_wr32(priv, 0x10082c, 0x00000001); - nv_wr32(priv, 0x100800, addr | 0x00000010); - return 0; + addr = nvkm_rd32(device, 0x10020c); + addr -= ((nvkm_memory_addr(gart) >> 19) + 1) << 19; + + nvkm_wr32(device, 0x100850, 0x80000000); + nvkm_wr32(device, 0x100818, mmu->null); + nvkm_wr32(device, 0x100804, NV44_GART_SIZE); + nvkm_wr32(device, 0x100850, 0x00008000); + nvkm_mask(device, 0x10008c, 0x00000200, 0x00000200); + nvkm_wr32(device, 0x100820, 0x00000000); + nvkm_wr32(device, 0x10082c, 0x00000001); + nvkm_wr32(device, 0x100800, addr | 0x00000010); } -struct nvkm_oclass -nv44_mmu_oclass = { - .handle = NV_SUBDEV(MMU, 0x44), - .ofuncs = &(struct nvkm_ofuncs) { - .ctor = nv44_mmu_ctor, - .dtor = nv04_mmu_dtor, - .init = nv44_mmu_init, - .fini = _nvkm_mmu_fini, - }, +static const struct nvkm_mmu_func +nv44_mmu = { + .dtor = nv04_mmu_dtor, + .oneinit = nv44_mmu_oneinit, + .init = nv44_mmu_init, + .limit = NV44_GART_SIZE, + .dma_bits = 39, + .pgt_bits = 32 - 12, + .spg_shift = 12, + .lpg_shift = 12, + .map_sg = nv44_vm_map_sg, + .unmap = nv44_vm_unmap, + .flush = nv44_vm_flush, }; + +int +nv44_mmu_new(struct nvkm_device *device, int index, struct nvkm_mmu **pmmu) +{ + if (device->type == NVKM_DEVICE_AGP || + !nvkm_boolopt(device->cfgopt, "NvPCIE", true)) + return nv04_mmu_new(device, index, pmmu); + + return nv04_mmu_new_(&nv44_mmu, device, index, pmmu); +} diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c index b83550fa7..a1f8d65f0 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv50.c @@ -21,31 +21,28 @@ * * Authors: Ben Skeggs */ -#include -#include -#include -#include +#include "priv.h" -#include #include - -struct nv50_mmu_priv { - struct nvkm_mmu base; -}; +#include +#include +#include static void -nv50_vm_map_pgt(struct nvkm_gpuobj *pgd, u32 pde, struct nvkm_gpuobj *pgt[2]) +nv50_vm_map_pgt(struct nvkm_gpuobj *pgd, u32 pde, struct nvkm_memory *pgt[2]) { u64 phys = 0xdeadcafe00000000ULL; u32 coverage = 0; if (pgt[0]) { - phys = 0x00000003 | pgt[0]->addr; /* present, 4KiB pages */ - coverage = (pgt[0]->size >> 3) << 12; + /* present, 4KiB pages */ + phys = 0x00000003 | nvkm_memory_addr(pgt[0]); + coverage = (nvkm_memory_size(pgt[0]) >> 3) << 12; } else if (pgt[1]) { - phys = 0x00000001 | pgt[1]->addr; /* present */ - coverage = (pgt[1]->size >> 3) << 16; + /* present, 64KiB pages */ + phys = 0x00000001 | nvkm_memory_addr(pgt[1]); + coverage = (nvkm_memory_size(pgt[1]) >> 3) << 16; } if (phys & 1) { @@ -57,8 +54,10 @@ nv50_vm_map_pgt(struct nvkm_gpuobj *pgd, u32 pde, struct nvkm_gpuobj *pgt[2]) phys |= 0x20; } - nv_wo32(pgd, (pde * 8) + 0, lower_32_bits(phys)); - nv_wo32(pgd, (pde * 8) + 4, upper_32_bits(phys)); + nvkm_kmap(pgd); + nvkm_wo32(pgd, (pde * 8) + 0, lower_32_bits(phys)); + nvkm_wo32(pgd, (pde * 8) + 4, upper_32_bits(phys)); + nvkm_done(pgd); } static inline u64 @@ -75,17 +74,18 @@ vm_addr(struct nvkm_vma *vma, u64 phys, u32 memtype, u32 target) } static void -nv50_vm_map(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt, +nv50_vm_map(struct nvkm_vma *vma, struct nvkm_memory *pgt, struct nvkm_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta) { + struct nvkm_ram *ram = vma->vm->mmu->subdev.device->fb->ram; u32 comp = (mem->memtype & 0x180) >> 7; u32 block, target; int i; /* IGPs don't have real VRAM, re-target to stolen system memory */ target = 0; - if (nvkm_fb(vma->vm->mmu)->ram->stolen) { - phys += nvkm_fb(vma->vm->mmu)->ram->stolen; + if (ram->stolen) { + phys += ram->stolen; target = 3; } @@ -93,6 +93,7 @@ nv50_vm_map(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt, pte <<= 3; cnt <<= 3; + nvkm_kmap(pgt); while (cnt) { u32 offset_h = upper_32_bits(phys); u32 offset_l = lower_32_bits(phys); @@ -113,129 +114,118 @@ nv50_vm_map(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt, } while (block) { - nv_wo32(pgt, pte + 0, offset_l); - nv_wo32(pgt, pte + 4, offset_h); + nvkm_wo32(pgt, pte + 0, offset_l); + nvkm_wo32(pgt, pte + 4, offset_h); pte += 8; block -= 8; } } + nvkm_done(pgt); } static void -nv50_vm_map_sg(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt, +nv50_vm_map_sg(struct nvkm_vma *vma, struct nvkm_memory *pgt, struct nvkm_mem *mem, u32 pte, u32 cnt, dma_addr_t *list) { u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 3 : 2; pte <<= 3; + nvkm_kmap(pgt); while (cnt--) { u64 phys = vm_addr(vma, (u64)*list++, mem->memtype, target); - nv_wo32(pgt, pte + 0, lower_32_bits(phys)); - nv_wo32(pgt, pte + 4, upper_32_bits(phys)); + nvkm_wo32(pgt, pte + 0, lower_32_bits(phys)); + nvkm_wo32(pgt, pte + 4, upper_32_bits(phys)); pte += 8; } + nvkm_done(pgt); } static void -nv50_vm_unmap(struct nvkm_gpuobj *pgt, u32 pte, u32 cnt) +nv50_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt) { pte <<= 3; + nvkm_kmap(pgt); while (cnt--) { - nv_wo32(pgt, pte + 0, 0x00000000); - nv_wo32(pgt, pte + 4, 0x00000000); + nvkm_wo32(pgt, pte + 0, 0x00000000); + nvkm_wo32(pgt, pte + 4, 0x00000000); pte += 8; } + nvkm_done(pgt); } static void nv50_vm_flush(struct nvkm_vm *vm) { - struct nv50_mmu_priv *priv = (void *)vm->mmu; - struct nvkm_bar *bar = nvkm_bar(priv); - struct nvkm_engine *engine; + struct nvkm_mmu *mmu = vm->mmu; + struct nvkm_subdev *subdev = &mmu->subdev; + struct nvkm_device *device = subdev->device; int i, vme; - bar->flush(bar); - - mutex_lock(&nv_subdev(priv)->mutex); - for (i = 0; i < NVDEV_SUBDEV_NR; i++) { + mutex_lock(&subdev->mutex); + for (i = 0; i < NVKM_SUBDEV_NR; i++) { if (!atomic_read(&vm->engref[i])) continue; /* unfortunate hw bug workaround... */ - engine = nvkm_engine(priv, i); - if (engine && engine->tlb_flush) { - engine->tlb_flush(engine); - continue; + if (i == NVKM_ENGINE_GR && device->gr) { + int ret = nvkm_gr_tlb_flush(device->gr); + if (ret != -ENODEV) + continue; } switch (i) { - case NVDEV_ENGINE_GR : vme = 0x00; break; - case NVDEV_ENGINE_VP : - case NVDEV_ENGINE_MSPDEC: vme = 0x01; break; - case NVDEV_SUBDEV_BAR : vme = 0x06; break; - case NVDEV_ENGINE_MSPPP : - case NVDEV_ENGINE_MPEG : vme = 0x08; break; - case NVDEV_ENGINE_BSP : - case NVDEV_ENGINE_MSVLD : vme = 0x09; break; - case NVDEV_ENGINE_CIPHER: - case NVDEV_ENGINE_SEC : vme = 0x0a; break; - case NVDEV_ENGINE_CE0 : vme = 0x0d; break; + case NVKM_ENGINE_GR : vme = 0x00; break; + case NVKM_ENGINE_VP : + case NVKM_ENGINE_MSPDEC: vme = 0x01; break; + case NVKM_SUBDEV_BAR : vme = 0x06; break; + case NVKM_ENGINE_MSPPP : + case NVKM_ENGINE_MPEG : vme = 0x08; break; + case NVKM_ENGINE_BSP : + case NVKM_ENGINE_MSVLD : vme = 0x09; break; + case NVKM_ENGINE_CIPHER: + case NVKM_ENGINE_SEC : vme = 0x0a; break; + case NVKM_ENGINE_CE0 : vme = 0x0d; break; default: continue; } - nv_wr32(priv, 0x100c80, (vme << 16) | 1); - if (!nv_wait(priv, 0x100c80, 0x00000001, 0x00000000)) - nv_error(priv, "vm flush timeout: engine %d\n", vme); + nvkm_wr32(device, 0x100c80, (vme << 16) | 1); + if (nvkm_msec(device, 2000, + if (!(nvkm_rd32(device, 0x100c80) & 0x00000001)) + break; + ) < 0) + nvkm_error(subdev, "vm flush timeout: engine %d\n", vme); } - mutex_unlock(&nv_subdev(priv)->mutex); + mutex_unlock(&subdev->mutex); } static int -nv50_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, - u64 mm_offset, struct nvkm_vm **pvm) +nv50_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset, + struct lock_class_key *key, struct nvkm_vm **pvm) { - u32 block = (1 << (mmu->pgt_bits + 12)); + u32 block = (1 << (mmu->func->pgt_bits + 12)); if (block > length) block = length; - return nvkm_vm_create(mmu, offset, length, mm_offset, block, pvm); + return nvkm_vm_create(mmu, offset, length, mm_offset, block, key, pvm); } -static int -nv50_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +static const struct nvkm_mmu_func +nv50_mmu = { + .limit = (1ULL << 40), + .dma_bits = 40, + .pgt_bits = 29 - 12, + .spg_shift = 12, + .lpg_shift = 16, + .create = nv50_vm_create, + .map_pgt = nv50_vm_map_pgt, + .map = nv50_vm_map, + .map_sg = nv50_vm_map_sg, + .unmap = nv50_vm_unmap, + .flush = nv50_vm_flush, +}; + +int +nv50_mmu_new(struct nvkm_device *device, int index, struct nvkm_mmu **pmmu) { - struct nv50_mmu_priv *priv; - int ret; - - ret = nvkm_mmu_create(parent, engine, oclass, "VM", "vm", &priv); - *pobject = nv_object(priv); - if (ret) - return ret; - - priv->base.limit = 1ULL << 40; - priv->base.dma_bits = 40; - priv->base.pgt_bits = 29 - 12; - priv->base.spg_shift = 12; - priv->base.lpg_shift = 16; - priv->base.create = nv50_vm_create; - priv->base.map_pgt = nv50_vm_map_pgt; - priv->base.map = nv50_vm_map; - priv->base.map_sg = nv50_vm_map_sg; - priv->base.unmap = nv50_vm_unmap; - priv->base.flush = nv50_vm_flush; - return 0; + return nvkm_mmu_new_(&nv50_mmu, device, index, pmmu); } - -struct nvkm_oclass -nv50_mmu_oclass = { - .handle = NV_SUBDEV(MMU, 0x50), - .ofuncs = &(struct nvkm_ofuncs) { - .ctor = nv50_mmu_ctor, - .dtor = _nvkm_mmu_dtor, - .init = _nvkm_mmu_init, - .fini = _nvkm_mmu_fini, - }, -}; diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/priv.h new file mode 100644 index 000000000..27cedc60b --- /dev/null +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/priv.h @@ -0,0 +1,39 @@ +#ifndef __NVKM_MMU_PRIV_H__ +#define __NVKM_MMU_PRIV_H__ +#define nvkm_mmu(p) container_of((p), struct nvkm_mmu, subdev) +#include + +void nvkm_mmu_ctor(const struct nvkm_mmu_func *, struct nvkm_device *, + int index, struct nvkm_mmu *); +int nvkm_mmu_new_(const struct nvkm_mmu_func *, struct nvkm_device *, + int index, struct nvkm_mmu **); + +struct nvkm_mmu_func { + void *(*dtor)(struct nvkm_mmu *); + int (*oneinit)(struct nvkm_mmu *); + void (*init)(struct nvkm_mmu *); + + u64 limit; + u8 dma_bits; + u32 pgt_bits; + u8 spg_shift; + u8 lpg_shift; + + int (*create)(struct nvkm_mmu *, u64 offset, u64 length, u64 mm_offset, + struct lock_class_key *, struct nvkm_vm **); + + void (*map_pgt)(struct nvkm_gpuobj *pgd, u32 pde, + struct nvkm_memory *pgt[2]); + void (*map)(struct nvkm_vma *, struct nvkm_memory *, + struct nvkm_mem *, u32 pte, u32 cnt, + u64 phys, u64 delta); + void (*map_sg)(struct nvkm_vma *, struct nvkm_memory *, + struct nvkm_mem *, u32 pte, u32 cnt, dma_addr_t *); + void (*unmap)(struct nvkm_vma *, struct nvkm_memory *pgt, + u32 pte, u32 cnt); + void (*flush)(struct nvkm_vm *); +}; + +int nvkm_vm_create(struct nvkm_mmu *, u64, u64, u64, u32, + struct lock_class_key *, struct nvkm_vm **); +#endif -- cgit v1.2.3-54-g00ecf