diff options
Diffstat (limited to 'drivers/gpu/drm/nouveau/nvkm/subdev/instmem')
-rw-r--r-- | drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c | 301 | ||||
-rw-r--r-- | drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c | 394 | ||||
-rw-r--r-- | drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.c | 240 | ||||
-rw-r--r-- | drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.h | 36 | ||||
-rw-r--r-- | drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c | 247 | ||||
-rw-r--r-- | drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c | 266 | ||||
-rw-r--r-- | drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h | 60 |
7 files changed, 946 insertions, 598 deletions
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c index d16358cc6cbb..895ba74057d4 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c @@ -23,124 +23,291 @@ */ #include "priv.h" -#include <core/engine.h> +#include <core/memory.h> +#include <subdev/bar.h> /****************************************************************************** * instmem object base implementation *****************************************************************************/ +#define nvkm_instobj(p) container_of((p), struct nvkm_instobj, memory) -void -_nvkm_instobj_dtor(struct nvkm_object *object) +struct nvkm_instobj { + struct nvkm_memory memory; + struct nvkm_memory *parent; + struct nvkm_instmem *imem; + struct list_head head; + u32 *suspend; + void __iomem *map; +}; + +static enum nvkm_memory_target +nvkm_instobj_target(struct nvkm_memory *memory) +{ + memory = nvkm_instobj(memory)->parent; + return nvkm_memory_target(memory); +} + +static u64 +nvkm_instobj_addr(struct nvkm_memory *memory) +{ + memory = nvkm_instobj(memory)->parent; + return nvkm_memory_addr(memory); +} + +static u64 +nvkm_instobj_size(struct nvkm_memory *memory) +{ + memory = nvkm_instobj(memory)->parent; + return nvkm_memory_size(memory); +} + +static void +nvkm_instobj_release(struct nvkm_memory *memory) +{ + struct nvkm_instobj *iobj = nvkm_instobj(memory); + nvkm_bar_flush(iobj->imem->subdev.device->bar); +} + +static void __iomem * +nvkm_instobj_acquire(struct nvkm_memory *memory) +{ + return nvkm_instobj(memory)->map; +} + +static u32 +nvkm_instobj_rd32(struct nvkm_memory *memory, u64 offset) +{ + return ioread32_native(nvkm_instobj(memory)->map + offset); +} + +static void +nvkm_instobj_wr32(struct nvkm_memory *memory, u64 offset, u32 data) { - struct nvkm_instmem *imem = nvkm_instmem(object); - struct nvkm_instobj *iobj = (void *)object; + iowrite32_native(data, nvkm_instobj(memory)->map + offset); +} + +static void +nvkm_instobj_map(struct nvkm_memory *memory, struct nvkm_vma *vma, u64 offset) +{ + memory = nvkm_instobj(memory)->parent; + nvkm_memory_map(memory, vma, offset); +} - mutex_lock(&nv_subdev(imem)->mutex); +static void * +nvkm_instobj_dtor(struct nvkm_memory *memory) +{ + struct nvkm_instobj *iobj = nvkm_instobj(memory); list_del(&iobj->head); - mutex_unlock(&nv_subdev(imem)->mutex); + nvkm_memory_del(&iobj->parent); + return iobj; +} + +const struct nvkm_memory_func +nvkm_instobj_func = { + .dtor = nvkm_instobj_dtor, + .target = nvkm_instobj_target, + .addr = nvkm_instobj_addr, + .size = nvkm_instobj_size, + .acquire = nvkm_instobj_acquire, + .release = nvkm_instobj_release, + .rd32 = nvkm_instobj_rd32, + .wr32 = nvkm_instobj_wr32, + .map = nvkm_instobj_map, +}; + +static void +nvkm_instobj_boot(struct nvkm_memory *memory, struct nvkm_vm *vm) +{ + memory = nvkm_instobj(memory)->parent; + nvkm_memory_boot(memory, vm); +} + +static void +nvkm_instobj_release_slow(struct nvkm_memory *memory) +{ + struct nvkm_instobj *iobj = nvkm_instobj(memory); + nvkm_instobj_release(memory); + nvkm_done(iobj->parent); +} + +static void __iomem * +nvkm_instobj_acquire_slow(struct nvkm_memory *memory) +{ + struct nvkm_instobj *iobj = nvkm_instobj(memory); + iobj->map = nvkm_kmap(iobj->parent); + if (iobj->map) + memory->func = &nvkm_instobj_func; + return iobj->map; +} + +static u32 +nvkm_instobj_rd32_slow(struct nvkm_memory *memory, u64 offset) +{ + struct nvkm_instobj *iobj = nvkm_instobj(memory); + return nvkm_ro32(iobj->parent, offset); +} - return nvkm_object_destroy(&iobj->base); +static void +nvkm_instobj_wr32_slow(struct nvkm_memory *memory, u64 offset, u32 data) +{ + struct nvkm_instobj *iobj = nvkm_instobj(memory); + return nvkm_wo32(iobj->parent, offset, data); } +const struct nvkm_memory_func +nvkm_instobj_func_slow = { + .dtor = nvkm_instobj_dtor, + .target = nvkm_instobj_target, + .addr = nvkm_instobj_addr, + .size = nvkm_instobj_size, + .boot = nvkm_instobj_boot, + .acquire = nvkm_instobj_acquire_slow, + .release = nvkm_instobj_release_slow, + .rd32 = nvkm_instobj_rd32_slow, + .wr32 = nvkm_instobj_wr32_slow, + .map = nvkm_instobj_map, +}; + int -nvkm_instobj_create_(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, int length, void **pobject) +nvkm_instobj_new(struct nvkm_instmem *imem, u32 size, u32 align, bool zero, + struct nvkm_memory **pmemory) { - struct nvkm_instmem *imem = nvkm_instmem(parent); + struct nvkm_memory *memory = NULL; struct nvkm_instobj *iobj; + u32 offset; int ret; - ret = nvkm_object_create_(parent, engine, oclass, NV_MEMOBJ_CLASS, - length, pobject); - iobj = *pobject; + ret = imem->func->memory_new(imem, size, align, zero, &memory); if (ret) - return ret; + goto done; - mutex_lock(&imem->base.mutex); - list_add(&iobj->head, &imem->list); - mutex_unlock(&imem->base.mutex); - return 0; + if (!imem->func->persistent) { + if (!(iobj = kzalloc(sizeof(*iobj), GFP_KERNEL))) { + ret = -ENOMEM; + goto done; + } + + nvkm_memory_ctor(&nvkm_instobj_func_slow, &iobj->memory); + iobj->parent = memory; + iobj->imem = imem; + list_add_tail(&iobj->head, &imem->list); + memory = &iobj->memory; + } + + if (!imem->func->zero && zero) { + void __iomem *map = nvkm_kmap(memory); + if (unlikely(!map)) { + for (offset = 0; offset < size; offset += 4) + nvkm_wo32(memory, offset, 0x00000000); + } else { + memset_io(map, 0x00, size); + } + nvkm_done(memory); + } + +done: + if (ret) + nvkm_memory_del(&memory); + *pmemory = memory; + return ret; } /****************************************************************************** * instmem subdev base implementation *****************************************************************************/ -static int -nvkm_instmem_alloc(struct nvkm_instmem *imem, struct nvkm_object *parent, - u32 size, u32 align, struct nvkm_object **pobject) +u32 +nvkm_instmem_rd32(struct nvkm_instmem *imem, u32 addr) { - struct nvkm_instmem_impl *impl = (void *)imem->base.object.oclass; - struct nvkm_instobj_args args = { .size = size, .align = align }; - return nvkm_object_ctor(parent, &parent->engine->subdev.object, - impl->instobj, &args, sizeof(args), pobject); + return imem->func->rd32(imem, addr); } -int -_nvkm_instmem_fini(struct nvkm_object *object, bool suspend) +void +nvkm_instmem_wr32(struct nvkm_instmem *imem, u32 addr, u32 data) +{ + return imem->func->wr32(imem, addr, data); +} + +static int +nvkm_instmem_fini(struct nvkm_subdev *subdev, bool suspend) { - struct nvkm_instmem *imem = (void *)object; + struct nvkm_instmem *imem = nvkm_instmem(subdev); struct nvkm_instobj *iobj; - int i, ret = 0; + int i; + + if (imem->func->fini) + imem->func->fini(imem); if (suspend) { - mutex_lock(&imem->base.mutex); list_for_each_entry(iobj, &imem->list, head) { - iobj->suspend = vmalloc(iobj->size); - if (!iobj->suspend) { - ret = -ENOMEM; - break; - } - - for (i = 0; i < iobj->size; i += 4) - iobj->suspend[i / 4] = nv_ro32(iobj, i); + struct nvkm_memory *memory = iobj->parent; + u64 size = nvkm_memory_size(memory); + + iobj->suspend = vmalloc(size); + if (!iobj->suspend) + return -ENOMEM; + + for (i = 0; i < size; i += 4) + iobj->suspend[i / 4] = nvkm_ro32(memory, i); } - mutex_unlock(&imem->base.mutex); - if (ret) - return ret; } - return nvkm_subdev_fini(&imem->base, suspend); + return 0; } -int -_nvkm_instmem_init(struct nvkm_object *object) +static int +nvkm_instmem_oneinit(struct nvkm_subdev *subdev) { - struct nvkm_instmem *imem = (void *)object; - struct nvkm_instobj *iobj; - int ret, i; + struct nvkm_instmem *imem = nvkm_instmem(subdev); + if (imem->func->oneinit) + return imem->func->oneinit(imem); + return 0; +} - ret = nvkm_subdev_init(&imem->base); - if (ret) - return ret; +static int +nvkm_instmem_init(struct nvkm_subdev *subdev) +{ + struct nvkm_instmem *imem = nvkm_instmem(subdev); + struct nvkm_instobj *iobj; + int i; - mutex_lock(&imem->base.mutex); list_for_each_entry(iobj, &imem->list, head) { if (iobj->suspend) { - for (i = 0; i < iobj->size; i += 4) - nv_wo32(iobj, i, iobj->suspend[i / 4]); + struct nvkm_memory *memory = iobj->parent; + u64 size = nvkm_memory_size(memory); + for (i = 0; i < size; i += 4) + nvkm_wo32(memory, i, iobj->suspend[i / 4]); vfree(iobj->suspend); iobj->suspend = NULL; } } - mutex_unlock(&imem->base.mutex); + return 0; } -int -nvkm_instmem_create_(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, int length, void **pobject) +static void * +nvkm_instmem_dtor(struct nvkm_subdev *subdev) { - struct nvkm_instmem *imem; - int ret; + struct nvkm_instmem *imem = nvkm_instmem(subdev); + if (imem->func->dtor) + return imem->func->dtor(imem); + return imem; +} - ret = nvkm_subdev_create_(parent, engine, oclass, 0, "INSTMEM", - "instmem", length, pobject); - imem = *pobject; - if (ret) - return ret; +static const struct nvkm_subdev_func +nvkm_instmem = { + .dtor = nvkm_instmem_dtor, + .oneinit = nvkm_instmem_oneinit, + .init = nvkm_instmem_init, + .fini = nvkm_instmem_fini, +}; +void +nvkm_instmem_ctor(const struct nvkm_instmem_func *func, + struct nvkm_device *device, int index, + struct nvkm_instmem *imem) +{ + nvkm_subdev_ctor(&nvkm_instmem, device, index, 0, &imem->subdev); + imem->func = func; INIT_LIST_HEAD(&imem->list); - imem->alloc = nvkm_instmem_alloc; - return 0; } diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c index dd0994d9ebfc..cd7feb1b25f6 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c @@ -37,32 +37,27 @@ * to use more "relaxed" allocation parameters when using the DMA API, since we * never need a kernel mapping. */ +#define gk20a_instmem(p) container_of((p), struct gk20a_instmem, base) +#include "priv.h" -#include <subdev/fb.h> +#include <core/memory.h> #include <core/mm.h> -#include <core/device.h> +#include <core/tegra.h> +#include <subdev/fb.h> -#ifdef __KERNEL__ -#include <linux/dma-attrs.h> -#include <linux/iommu.h> -#include <nouveau_platform.h> -#endif +#define gk20a_instobj(p) container_of((p), struct gk20a_instobj, memory) -#include "priv.h" - -struct gk20a_instobj_priv { - struct nvkm_instobj base; - /* Must be second member here - see nouveau_gpuobj_map_vm() */ - struct nvkm_mem *mem; - /* Pointed by mem */ - struct nvkm_mem _mem; +struct gk20a_instobj { + struct nvkm_memory memory; + struct gk20a_instmem *imem; + struct nvkm_mem mem; }; /* * Used for objects allocated using the DMA API */ struct gk20a_instobj_dma { - struct gk20a_instobj_priv base; + struct gk20a_instobj base; void *cpuaddr; dma_addr_t handle; @@ -73,14 +68,15 @@ struct gk20a_instobj_dma { * Used for objects flattened using the IOMMU API */ struct gk20a_instobj_iommu { - struct gk20a_instobj_priv base; + struct gk20a_instobj base; /* array of base.mem->size pages */ struct page *pages[]; }; -struct gk20a_instmem_priv { +struct gk20a_instmem { struct nvkm_instmem base; + unsigned long lock_flags; spinlock_t lock; u64 addr; @@ -94,6 +90,42 @@ struct gk20a_instmem_priv { struct dma_attrs attrs; }; +static enum nvkm_memory_target +gk20a_instobj_target(struct nvkm_memory *memory) +{ + return NVKM_MEM_TARGET_HOST; +} + +static u64 +gk20a_instobj_addr(struct nvkm_memory *memory) +{ + return gk20a_instobj(memory)->mem.offset; + +} + +static u64 +gk20a_instobj_size(struct nvkm_memory *memory) +{ + return (u64)gk20a_instobj(memory)->mem.size << 12; +} + +static void __iomem * +gk20a_instobj_acquire(struct nvkm_memory *memory) +{ + struct gk20a_instmem *imem = gk20a_instobj(memory)->imem; + unsigned long flags; + spin_lock_irqsave(&imem->lock, flags); + imem->lock_flags = flags; + return NULL; +} + +static void +gk20a_instobj_release(struct nvkm_memory *memory) +{ + struct gk20a_instmem *imem = gk20a_instobj(memory)->imem; + spin_unlock_irqrestore(&imem->lock, imem->lock_flags); +} + /* * Use PRAMIN to read/write data and avoid coherency issues. * PRAMIN uses the GPU path and ensures data will always be coherent. @@ -104,160 +136,170 @@ struct gk20a_instmem_priv { */ static u32 -gk20a_instobj_rd32(struct nvkm_object *object, u64 offset) +gk20a_instobj_rd32(struct nvkm_memory *memory, u64 offset) { - struct gk20a_instmem_priv *priv = (void *)nvkm_instmem(object); - struct gk20a_instobj_priv *node = (void *)object; - unsigned long flags; - u64 base = (node->mem->offset + offset) & 0xffffff00000ULL; - u64 addr = (node->mem->offset + offset) & 0x000000fffffULL; + struct gk20a_instobj *node = gk20a_instobj(memory); + struct gk20a_instmem *imem = node->imem; + struct nvkm_device *device = imem->base.subdev.device; + u64 base = (node->mem.offset + offset) & 0xffffff00000ULL; + u64 addr = (node->mem.offset + offset) & 0x000000fffffULL; u32 data; - spin_lock_irqsave(&priv->lock, flags); - if (unlikely(priv->addr != base)) { - nv_wr32(priv, 0x001700, base >> 16); - priv->addr = base; + if (unlikely(imem->addr != base)) { + nvkm_wr32(device, 0x001700, base >> 16); + imem->addr = base; } - data = nv_rd32(priv, 0x700000 + addr); - spin_unlock_irqrestore(&priv->lock, flags); + data = nvkm_rd32(device, 0x700000 + addr); return data; } static void -gk20a_instobj_wr32(struct nvkm_object *object, u64 offset, u32 data) +gk20a_instobj_wr32(struct nvkm_memory *memory, u64 offset, u32 data) { - struct gk20a_instmem_priv *priv = (void *)nvkm_instmem(object); - struct gk20a_instobj_priv *node = (void *)object; - unsigned long flags; - u64 base = (node->mem->offset + offset) & 0xffffff00000ULL; - u64 addr = (node->mem->offset + offset) & 0x000000fffffULL; - - spin_lock_irqsave(&priv->lock, flags); - if (unlikely(priv->addr != base)) { - nv_wr32(priv, 0x001700, base >> 16); - priv->addr = base; + struct gk20a_instobj *node = gk20a_instobj(memory); + struct gk20a_instmem *imem = node->imem; + struct nvkm_device *device = imem->base.subdev.device; + u64 base = (node->mem.offset + offset) & 0xffffff00000ULL; + u64 addr = (node->mem.offset + offset) & 0x000000fffffULL; + + if (unlikely(imem->addr != base)) { + nvkm_wr32(device, 0x001700, base >> 16); + imem->addr = base; } - nv_wr32(priv, 0x700000 + addr, data); - spin_unlock_irqrestore(&priv->lock, flags); + nvkm_wr32(device, 0x700000 + addr, data); +} + +static void +gk20a_instobj_map(struct nvkm_memory *memory, struct nvkm_vma *vma, u64 offset) +{ + struct gk20a_instobj *node = gk20a_instobj(memory); + nvkm_vm_map_at(vma, offset, &node->mem); } static void -gk20a_instobj_dtor_dma(struct gk20a_instobj_priv *_node) +gk20a_instobj_dtor_dma(struct gk20a_instobj *_node) { struct gk20a_instobj_dma *node = (void *)_node; - struct gk20a_instmem_priv *priv = (void *)nvkm_instmem(node); - struct device *dev = nv_device_base(nv_device(priv)); + struct gk20a_instmem *imem = _node->imem; + struct device *dev = imem->base.subdev.device->dev; if (unlikely(!node->cpuaddr)) return; - dma_free_attrs(dev, _node->mem->size << PAGE_SHIFT, node->cpuaddr, - node->handle, &priv->attrs); + dma_free_attrs(dev, _node->mem.size << PAGE_SHIFT, node->cpuaddr, + node->handle, &imem->attrs); } static void -gk20a_instobj_dtor_iommu(struct gk20a_instobj_priv *_node) +gk20a_instobj_dtor_iommu(struct gk20a_instobj *_node) { struct gk20a_instobj_iommu *node = (void *)_node; - struct gk20a_instmem_priv *priv = (void *)nvkm_instmem(node); + struct gk20a_instmem *imem = _node->imem; struct nvkm_mm_node *r; int i; - if (unlikely(list_empty(&_node->mem->regions))) + if (unlikely(list_empty(&_node->mem.regions))) return; - r = list_first_entry(&_node->mem->regions, struct nvkm_mm_node, + r = list_first_entry(&_node->mem.regions, struct nvkm_mm_node, rl_entry); /* clear bit 34 to unmap pages */ - r->offset &= ~BIT(34 - priv->iommu_pgshift); + r->offset &= ~BIT(34 - imem->iommu_pgshift); /* Unmap pages from GPU address space and free them */ - for (i = 0; i < _node->mem->size; i++) { - iommu_unmap(priv->domain, - (r->offset + i) << priv->iommu_pgshift, PAGE_SIZE); + for (i = 0; i < _node->mem.size; i++) { + iommu_unmap(imem->domain, + (r->offset + i) << imem->iommu_pgshift, PAGE_SIZE); __free_page(node->pages[i]); } /* Release area from GPU address space */ - mutex_lock(priv->mm_mutex); - nvkm_mm_free(priv->mm, &r); - mutex_unlock(priv->mm_mutex); + mutex_lock(imem->mm_mutex); + nvkm_mm_free(imem->mm, &r); + mutex_unlock(imem->mm_mutex); } -static void -gk20a_instobj_dtor(struct nvkm_object *object) +static void * +gk20a_instobj_dtor(struct nvkm_memory *memory) { - struct gk20a_instobj_priv *node = (void *)object; - struct gk20a_instmem_priv *priv = (void *)nvkm_instmem(node); + struct gk20a_instobj *node = gk20a_instobj(memory); + struct gk20a_instmem *imem = node->imem; - if (priv->domain) + if (imem->domain) gk20a_instobj_dtor_iommu(node); else gk20a_instobj_dtor_dma(node); - nvkm_instobj_destroy(&node->base); + return node; } +static const struct nvkm_memory_func +gk20a_instobj_func = { + .dtor = gk20a_instobj_dtor, + .target = gk20a_instobj_target, + .addr = gk20a_instobj_addr, + .size = gk20a_instobj_size, + .acquire = gk20a_instobj_acquire, + .release = gk20a_instobj_release, + .rd32 = gk20a_instobj_rd32, + .wr32 = gk20a_instobj_wr32, + .map = gk20a_instobj_map, +}; + static int -gk20a_instobj_ctor_dma(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, u32 npages, u32 align, - struct gk20a_instobj_priv **_node) +gk20a_instobj_ctor_dma(struct gk20a_instmem *imem, u32 npages, u32 align, + struct gk20a_instobj **_node) { struct gk20a_instobj_dma *node; - struct gk20a_instmem_priv *priv = (void *)nvkm_instmem(parent); - struct device *dev = nv_device_base(nv_device(parent)); - int ret; + struct nvkm_subdev *subdev = &imem->base.subdev; + struct device *dev = subdev->device->dev; - ret = nvkm_instobj_create_(parent, engine, oclass, sizeof(*node), - (void **)&node); + if (!(node = kzalloc(sizeof(*node), GFP_KERNEL))) + return -ENOMEM; *_node = &node->base; - if (ret) - return ret; node->cpuaddr = dma_alloc_attrs(dev, npages << PAGE_SHIFT, &node->handle, GFP_KERNEL, - &priv->attrs); + &imem->attrs); if (!node->cpuaddr) { - nv_error(priv, "cannot allocate DMA memory\n"); + nvkm_error(subdev, "cannot allocate DMA memory\n"); return -ENOMEM; } /* alignment check */ if (unlikely(node->handle & (align - 1))) - nv_warn(priv, "memory not aligned as requested: %pad (0x%x)\n", - &node->handle, align); + nvkm_warn(subdev, + "memory not aligned as requested: %pad (0x%x)\n", + &node->handle, align); /* present memory for being mapped using small pages */ node->r.type = 12; node->r.offset = node->handle >> 12; node->r.length = (npages << PAGE_SHIFT) >> 12; - node->base._mem.offset = node->handle; + node->base.mem.offset = node->handle; - INIT_LIST_HEAD(&node->base._mem.regions); - list_add_tail(&node->r.rl_entry, &node->base._mem.regions); + INIT_LIST_HEAD(&node->base.mem.regions); + list_add_tail(&node->r.rl_entry, &node->base.mem.regions); return 0; } static int -gk20a_instobj_ctor_iommu(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, u32 npages, u32 align, - struct gk20a_instobj_priv **_node) +gk20a_instobj_ctor_iommu(struct gk20a_instmem *imem, u32 npages, u32 align, + struct gk20a_instobj **_node) { struct gk20a_instobj_iommu *node; - struct gk20a_instmem_priv *priv = (void *)nvkm_instmem(parent); + struct nvkm_subdev *subdev = &imem->base.subdev; struct nvkm_mm_node *r; int ret; int i; - ret = nvkm_instobj_create_(parent, engine, oclass, - sizeof(*node) + sizeof(node->pages[0]) * npages, - (void **)&node); + if (!(node = kzalloc(sizeof(*node) + + sizeof( node->pages[0]) * npages, GFP_KERNEL))) + return -ENOMEM; *_node = &node->base; - if (ret) - return ret; /* Allocate backing memory */ for (i = 0; i < npages; i++) { @@ -270,48 +312,48 @@ gk20a_instobj_ctor_iommu(struct nvkm_object *parent, struct nvkm_object *engine, node->pages[i] = p; } - mutex_lock(priv->mm_mutex); + mutex_lock(imem->mm_mutex); /* Reserve area from GPU address space */ - ret = nvkm_mm_head(priv->mm, 0, 1, npages, npages, - align >> priv->iommu_pgshift, &r); - mutex_unlock(priv->mm_mutex); + ret = nvkm_mm_head(imem->mm, 0, 1, npages, npages, + align >> imem->iommu_pgshift, &r); + mutex_unlock(imem->mm_mutex); if (ret) { - nv_error(priv, "virtual space is full!\n"); + nvkm_error(subdev, "virtual space is full!\n"); goto free_pages; } /* Map into GPU address space */ for (i = 0; i < npages; i++) { struct page *p = node->pages[i]; - u32 offset = (r->offset + i) << priv->iommu_pgshift; + u32 offset = (r->offset + i) << imem->iommu_pgshift; - ret = iommu_map(priv->domain, offset, page_to_phys(p), + ret = iommu_map(imem->domain, offset, page_to_phys(p), PAGE_SIZE, IOMMU_READ | IOMMU_WRITE); if (ret < 0) { - nv_error(priv, "IOMMU mapping failure: %d\n", ret); + nvkm_error(subdev, "IOMMU mapping failure: %d\n", ret); while (i-- > 0) { offset -= PAGE_SIZE; - iommu_unmap(priv->domain, offset, PAGE_SIZE); + iommu_unmap(imem->domain, offset, PAGE_SIZE); } goto release_area; } } /* Bit 34 tells that an address is to be resolved through the IOMMU */ - r->offset |= BIT(34 - priv->iommu_pgshift); + r->offset |= BIT(34 - imem->iommu_pgshift); - node->base._mem.offset = ((u64)r->offset) << priv->iommu_pgshift; + node->base.mem.offset = ((u64)r->offset) << imem->iommu_pgshift; - INIT_LIST_HEAD(&node->base._mem.regions); - list_add_tail(&r->rl_entry, &node->base._mem.regions); + INIT_LIST_HEAD(&node->base.mem.regions); + list_add_tail(&r->rl_entry, &node->base.mem.regions); return 0; release_area: - mutex_lock(priv->mm_mutex); - nvkm_mm_free(priv->mm, &r); - mutex_unlock(priv->mm_mutex); + mutex_lock(imem->mm_mutex); + nvkm_mm_free(imem->mm, &r); + mutex_unlock(imem->mm_mutex); free_pages: for (i = 0; i < npages && node->pages[i] != NULL; i++) @@ -321,120 +363,92 @@ free_pages: } static int -gk20a_instobj_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 _size, - struct nvkm_object **pobject) +gk20a_instobj_new(struct nvkm_instmem *base, u32 size, u32 align, bool zero, + struct nvkm_memory **pmemory) { - struct nvkm_instobj_args *args = data; - struct gk20a_instmem_priv *priv = (void *)nvkm_instmem(parent); - struct gk20a_instobj_priv *node; - u32 size, align; + struct gk20a_instmem *imem = gk20a_instmem(base); + struct gk20a_instobj *node = NULL; + struct nvkm_subdev *subdev = &imem->base.subdev; int ret; - nv_debug(parent, "%s (%s): size: %x align: %x\n", __func__, - priv->domain ? "IOMMU" : "DMA", args->size, args->align); + nvkm_debug(subdev, "%s (%s): size: %x align: %x\n", __func__, + imem->domain ? "IOMMU" : "DMA", size, align); /* Round size and align to page bounds */ - size = max(roundup(args->size, PAGE_SIZE), PAGE_SIZE); - align = max(roundup(args->align, PAGE_SIZE), PAGE_SIZE); + size = max(roundup(size, PAGE_SIZE), PAGE_SIZE); + align = max(roundup(align, PAGE_SIZE), PAGE_SIZE); - if (priv->domain) - ret = gk20a_instobj_ctor_iommu(parent, engine, oclass, - size >> PAGE_SHIFT, align, &node); + if (imem->domain) + ret = gk20a_instobj_ctor_iommu(imem, size >> PAGE_SHIFT, + align, &node); else - ret = gk20a_instobj_ctor_dma(parent, engine, oclass, - size >> PAGE_SHIFT, align, &node); - *pobject = nv_object(node); + ret = gk20a_instobj_ctor_dma(imem, size >> PAGE_SHIFT, + align, &node); + *pmemory = node ? &node->memory : NULL; if (ret) return ret; - node->mem = &node->_mem; + nvkm_memory_ctor(&gk20a_instobj_func, &node->memory); + node->imem = imem; /* present memory for being mapped using small pages */ - node->mem->size = size >> 12; - node->mem->memtype = 0; - node->mem->page_shift = 12; - - node->base.addr = node->mem->offset; - node->base.size = size; + node->mem.size = size >> 12; + node->mem.memtype = 0; + node->mem.page_shift = 12; - nv_debug(parent, "alloc size: 0x%x, align: 0x%x, gaddr: 0x%llx\n", - size, align, node->mem->offset); + nvkm_debug(subdev, "alloc size: 0x%x, align: 0x%x, gaddr: 0x%llx\n", + size, align, node->mem.offset); return 0; } -static struct nvkm_instobj_impl -gk20a_instobj_oclass = { - .base.ofuncs = &(struct nvkm_ofuncs) { - .ctor = gk20a_instobj_ctor, - .dtor = gk20a_instobj_dtor, - .init = _nvkm_instobj_init, - .fini = _nvkm_instobj_fini, - .rd32 = gk20a_instobj_rd32, - .wr32 = gk20a_instobj_wr32, - }, -}; - - - -static int -gk20a_instmem_fini(struct nvkm_object *object, bool suspend) +static void +gk20a_instmem_fini(struct nvkm_instmem *base) { - struct gk20a_instmem_priv *priv = (void *)object; - priv->addr = ~0ULL; - return nvkm_instmem_fini(&priv->base, suspend); + gk20a_instmem(base)->addr = ~0ULL; } -static int -gk20a_instmem_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) -{ - struct gk20a_instmem_priv *priv; - struct nouveau_platform_device *plat; - int ret; +static const struct nvkm_instmem_func +gk20a_instmem = { + .fini = gk20a_instmem_fini, + .memory_new = gk20a_instobj_new, + .persistent = true, + .zero = false, +}; - ret = nvkm_instmem_create(parent, engine, oclass, &priv); - *pobject = nv_object(priv); - if (ret) - return ret; +int +gk20a_instmem_new(struct nvkm_device *device, int index, + struct nvkm_instmem **pimem) +{ + struct nvkm_device_tegra *tdev = device->func->tegra(device); + struct gk20a_instmem *imem; - spin_lock_init(&priv->lock); + if (!(imem = kzalloc(sizeof(*imem), GFP_KERNEL))) + return -ENOMEM; + nvkm_instmem_ctor(&gk20a_instmem, device, index, &imem->base); + spin_lock_init(&imem->lock); + *pimem = &imem->base; - plat = nv_device_to_platform(nv_device(parent)); - if (plat->gpu->iommu.domain) { - priv->domain = plat->gpu->iommu.domain; - priv->mm = plat->gpu->iommu.mm; - priv->iommu_pgshift = plat->gpu->iommu.pgshift; - priv->mm_mutex = &plat->gpu->iommu.mutex; + if (tdev->iommu.domain) { + imem->domain = tdev->iommu.domain; + imem->mm = &tdev->iommu.mm; + imem->iommu_pgshift = tdev->iommu.pgshift; + imem->mm_mutex = &tdev->iommu.mutex; - nv_info(priv, "using IOMMU\n"); + nvkm_info(&imem->base.subdev, "using IOMMU\n"); } else { - init_dma_attrs(&priv->attrs); + init_dma_attrs(&imem->attrs); /* * We will access instmem through PRAMIN and thus do not need a * consistent CPU pointer or kernel mapping */ - dma_set_attr(DMA_ATTR_NON_CONSISTENT, &priv->attrs); - dma_set_attr(DMA_ATTR_WEAK_ORDERING, &priv->attrs); - dma_set_attr(DMA_ATTR_WRITE_COMBINE, &priv->attrs); - dma_set_attr(DMA_ATTR_NO_KERNEL_MAPPING, &priv->attrs); + dma_set_attr(DMA_ATTR_NON_CONSISTENT, &imem->attrs); + dma_set_attr(DMA_ATTR_WEAK_ORDERING, &imem->attrs); + dma_set_attr(DMA_ATTR_WRITE_COMBINE, &imem->attrs); + dma_set_attr(DMA_ATTR_NO_KERNEL_MAPPING, &imem->attrs); - nv_info(priv, "using DMA API\n"); + nvkm_info(&imem->base.subdev, "using DMA API\n"); } return 0; } - -struct nvkm_oclass * -gk20a_instmem_oclass = &(struct nvkm_instmem_impl) { - .base.handle = NV_SUBDEV(INSTMEM, 0xea), - .base.ofuncs = &(struct nvkm_ofuncs) { - .ctor = gk20a_instmem_ctor, - .dtor = _nvkm_instmem_dtor, - .init = _nvkm_instmem_init, - .fini = gk20a_instmem_fini, - }, - .instobj = &gk20a_instobj_oclass.base, -}.base; diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.c index 282143f49d72..6133c8bb2d42 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.c @@ -21,173 +21,207 @@ * * Authors: Ben Skeggs */ -#include "nv04.h" +#define nv04_instmem(p) container_of((p), struct nv04_instmem, base) +#include "priv.h" +#include <core/memory.h> #include <core/ramht.h> +struct nv04_instmem { + struct nvkm_instmem base; + struct nvkm_mm heap; +}; + /****************************************************************************** * instmem object implementation *****************************************************************************/ +#define nv04_instobj(p) container_of((p), struct nv04_instobj, memory) -static u32 -nv04_instobj_rd32(struct nvkm_object *object, u64 addr) +struct nv04_instobj { + struct nvkm_memory memory; + struct nv04_instmem *imem; + struct nvkm_mm_node *node; +}; + +static enum nvkm_memory_target +nv04_instobj_target(struct nvkm_memory *memory) { - struct nv04_instmem_priv *priv = (void *)nvkm_instmem(object); - struct nv04_instobj_priv *node = (void *)object; - return nv_ro32(priv, node->mem->offset + addr); + return NVKM_MEM_TARGET_INST; } -static void -nv04_instobj_wr32(struct nvkm_object *object, u64 addr, u32 data) +static u64 +nv04_instobj_addr(struct nvkm_memory *memory) +{ + return nv04_instobj(memory)->node->offset; +} + +static u64 +nv04_instobj_size(struct nvkm_memory *memory) +{ + return nv04_instobj(memory)->node->length; +} + +static void __iomem * +nv04_instobj_acquire(struct nvkm_memory *memory) { - struct nv04_instmem_priv *priv = (void *)nvkm_instmem(object); - struct nv04_instobj_priv *node = (void *)object; - nv_wo32(priv, node->mem->offset + addr, data); + struct nv04_instobj *iobj = nv04_instobj(memory); + struct nvkm_device *device = iobj->imem->base.subdev.device; + return device->pri + 0x700000 + iobj->node->offset; } static void -nv04_instobj_dtor(struct nvkm_object *object) +nv04_instobj_release(struct nvkm_memory *memory) { - struct nv04_instmem_priv *priv = (void *)nvkm_instmem(object); - struct nv04_instobj_priv *node = (void *)object; - struct nvkm_subdev *subdev = (void *)priv; +} - mutex_lock(&subdev->mutex); - nvkm_mm_free(&priv->heap, &node->mem); - mutex_unlock(&subdev->mutex); +static u32 +nv04_instobj_rd32(struct nvkm_memory *memory, u64 offset) +{ + struct nv04_instobj *iobj = nv04_instobj(memory); + struct nvkm_device *device = iobj->imem->base.subdev.device; + return nvkm_rd32(device, 0x700000 + iobj->node->offset + offset); +} - nvkm_instobj_destroy(&node->base); +static void +nv04_instobj_wr32(struct nvkm_memory *memory, u64 offset, u32 data) +{ + struct nv04_instobj *iobj = nv04_instobj(memory); + struct nvkm_device *device = iobj->imem->base.subdev.device; + nvkm_wr32(device, 0x700000 + iobj->node->offset + offset, data); } +static void * +nv04_instobj_dtor(struct nvkm_memory *memory) +{ + struct nv04_instobj *iobj = nv04_instobj(memory); + mutex_lock(&iobj->imem->base.subdev.mutex); + nvkm_mm_free(&iobj->imem->heap, &iobj->node); + mutex_unlock(&iobj->imem->base.subdev.mutex); + return iobj; +} + +static const struct nvkm_memory_func +nv04_instobj_func = { + .dtor = nv04_instobj_dtor, + .target = nv04_instobj_target, + .size = nv04_instobj_size, + .addr = nv04_instobj_addr, + .acquire = nv04_instobj_acquire, + .release = nv04_instobj_release, + .rd32 = nv04_instobj_rd32, + .wr32 = nv04_instobj_wr32, +}; + static int -nv04_instobj_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +nv04_instobj_new(struct nvkm_instmem *base, u32 size, u32 align, bool zero, + struct nvkm_memory **pmemory) { - struct nv04_instmem_priv *priv = (void *)nvkm_instmem(parent); - struct nv04_instobj_priv *node; - struct nvkm_instobj_args *args = data; - struct nvkm_subdev *subdev = (void *)priv; + struct nv04_instmem *imem = nv04_instmem(base); + struct nv04_instobj *iobj; int ret; - if (!args->align) - args->align = 1; + if (!(iobj = kzalloc(sizeof(*iobj), GFP_KERNEL))) + return -ENOMEM; + *pmemory = &iobj->memory; - ret = nvkm_instobj_create(parent, engine, oclass, &node); - *pobject = nv_object(node); - if (ret) - return ret; - - mutex_lock(&subdev->mutex); - ret = nvkm_mm_head(&priv->heap, 0, 1, args->size, args->size, - args->align, &node->mem); - mutex_unlock(&subdev->mutex); - if (ret) - return ret; + nvkm_memory_ctor(&nv04_instobj_func, &iobj->memory); + iobj->imem = imem; - node->base.addr = node->mem->offset; - node->base.size = node->mem->length; - return 0; + mutex_lock(&imem->base.subdev.mutex); + ret = nvkm_mm_head(&imem->heap, 0, 1, size, size, + align ? align : 1, &iobj->node); + mutex_unlock(&imem->base.subdev.mutex); + return ret; } -struct nvkm_instobj_impl -nv04_instobj_oclass = { - .base.ofuncs = &(struct nvkm_ofuncs) { - .ctor = nv04_instobj_ctor, - .dtor = nv04_instobj_dtor, - .init = _nvkm_instobj_init, - .fini = _nvkm_instobj_fini, - .rd32 = nv04_instobj_rd32, - .wr32 = nv04_instobj_wr32, - }, -}; - /****************************************************************************** * instmem subdev implementation *****************************************************************************/ static u32 -nv04_instmem_rd32(struct nvkm_object *object, u64 addr) +nv04_instmem_rd32(struct nvkm_instmem *imem, u32 addr) { - return nv_rd32(object, 0x700000 + addr); + return nvkm_rd32(imem->subdev.device, 0x700000 + addr); } static void -nv04_instmem_wr32(struct nvkm_object *object, u64 addr, u32 data) -{ - return nv_wr32(object, 0x700000 + addr, data); -} - -void -nv04_instmem_dtor(struct nvkm_object *object) +nv04_instmem_wr32(struct nvkm_instmem *imem, u32 addr, u32 data) { - struct nv04_instmem_priv *priv = (void *)object; - nvkm_gpuobj_ref(NULL, &priv->ramfc); - nvkm_gpuobj_ref(NULL, &priv->ramro); - nvkm_ramht_ref(NULL, &priv->ramht); - nvkm_gpuobj_ref(NULL, &priv->vbios); - nvkm_mm_fini(&priv->heap); - if (priv->iomem) - iounmap(priv->iomem); - nvkm_instmem_destroy(&priv->base); + nvkm_wr32(imem->subdev.device, 0x700000 + addr, data); } static int -nv04_instmem_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +nv04_instmem_oneinit(struct nvkm_instmem *base) { - struct nv04_instmem_priv *priv; + struct nv04_instmem *imem = nv04_instmem(base); + struct nvkm_device *device = imem->base.subdev.device; int ret; - ret = nvkm_instmem_create(parent, engine, oclass, &priv); - *pobject = nv_object(priv); - if (ret) - return ret; - /* PRAMIN aperture maps over the end of VRAM, reserve it */ - priv->base.reserved = 512 * 1024; + imem->base.reserved = 512 * 1024; - ret = nvkm_mm_init(&priv->heap, 0, priv->base.reserved, 1); + ret = nvkm_mm_init(&imem->heap, 0, imem->base.reserved, 1); if (ret) return ret; /* 0x00000-0x10000: reserve for probable vbios image */ - ret = nvkm_gpuobj_new(nv_object(priv), NULL, 0x10000, 0, 0, - &priv->vbios); + ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x10000, 0, false, + &imem->base.vbios); if (ret) return ret; /* 0x10000-0x18000: reserve for RAMHT */ - ret = nvkm_ramht_new(nv_object(priv), NULL, 0x08000, 0, &priv->ramht); + ret = nvkm_ramht_new(device, 0x08000, 0, NULL, &imem->base.ramht); if (ret) return ret; /* 0x18000-0x18800: reserve for RAMFC (enough for 32 nv30 channels) */ - ret = nvkm_gpuobj_new(nv_object(priv), NULL, 0x00800, 0, - NVOBJ_FLAG_ZERO_ALLOC, &priv->ramfc); + ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x00800, 0, true, + &imem->base.ramfc); if (ret) return ret; /* 0x18800-0x18a00: reserve for RAMRO */ - ret = nvkm_gpuobj_new(nv_object(priv), NULL, 0x00200, 0, 0, - &priv->ramro); + ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x00200, 0, false, + &imem->base.ramro); if (ret) return ret; return 0; } -struct nvkm_oclass * -nv04_instmem_oclass = &(struct nvkm_instmem_impl) { - .base.handle = NV_SUBDEV(INSTMEM, 0x04), - .base.ofuncs = &(struct nvkm_ofuncs) { - .ctor = nv04_instmem_ctor, - .dtor = nv04_instmem_dtor, - .init = _nvkm_instmem_init, - .fini = _nvkm_instmem_fini, - .rd32 = nv04_instmem_rd32, - .wr32 = nv04_instmem_wr32, - }, - .instobj = &nv04_instobj_oclass.base, -}.base; +static void * +nv04_instmem_dtor(struct nvkm_instmem *base) +{ + struct nv04_instmem *imem = nv04_instmem(base); + nvkm_memory_del(&imem->base.ramfc); + nvkm_memory_del(&imem->base.ramro); + nvkm_ramht_del(&imem->base.ramht); + nvkm_memory_del(&imem->base.vbios); + nvkm_mm_fini(&imem->heap); + return imem; +} + +static const struct nvkm_instmem_func +nv04_instmem = { + .dtor = nv04_instmem_dtor, + .oneinit = nv04_instmem_oneinit, + .rd32 = nv04_instmem_rd32, + .wr32 = nv04_instmem_wr32, + .memory_new = nv04_instobj_new, + .persistent = false, + .zero = false, +}; + +int +nv04_instmem_new(struct nvkm_device *device, int index, + struct nvkm_instmem **pimem) +{ + struct nv04_instmem *imem; + + if (!(imem = kzalloc(sizeof(*imem), GFP_KERNEL))) + return -ENOMEM; + nvkm_instmem_ctor(&nv04_instmem, device, index, &imem->base); + *pimem = &imem->base; + return 0; +} diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.h b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.h deleted file mode 100644 index 42b6c928047c..000000000000 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv04.h +++ /dev/null @@ -1,36 +0,0 @@ -#ifndef __NV04_INSTMEM_H__ -#define __NV04_INSTMEM_H__ -#include "priv.h" - -#include <core/mm.h> - -extern struct nvkm_instobj_impl nv04_instobj_oclass; - -struct nv04_instmem_priv { - struct nvkm_instmem base; - - void __iomem *iomem; - struct nvkm_mm heap; - - struct nvkm_gpuobj *vbios; - struct nvkm_ramht *ramht; - struct nvkm_gpuobj *ramro; - struct nvkm_gpuobj *ramfc; -}; - -static inline struct nv04_instmem_priv * -nv04_instmem(void *obj) -{ - return (void *)nvkm_instmem(obj); -} - -struct nv04_instobj_priv { - struct nvkm_instobj base; - struct nvkm_mm_node *mem; -}; - -void nv04_instmem_dtor(struct nvkm_object *); - -int nv04_instmem_alloc(struct nvkm_instmem *, struct nvkm_object *, - u32 size, u32 align, struct nvkm_object **pobject); -#endif diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c index b42b8588fc0e..c0543875e490 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c @@ -21,116 +21,239 @@ * * Authors: Ben Skeggs */ -#include "nv04.h" +#define nv40_instmem(p) container_of((p), struct nv40_instmem, base) +#include "priv.h" +#include <core/memory.h> #include <core/ramht.h> #include <engine/gr/nv40.h> +struct nv40_instmem { + struct nvkm_instmem base; + struct nvkm_mm heap; + void __iomem *iomem; +}; + /****************************************************************************** - * instmem subdev implementation + * instmem object implementation *****************************************************************************/ +#define nv40_instobj(p) container_of((p), struct nv40_instobj, memory) + +struct nv40_instobj { + struct nvkm_memory memory; + struct nv40_instmem *imem; + struct nvkm_mm_node *node; +}; + +static enum nvkm_memory_target +nv40_instobj_target(struct nvkm_memory *memory) +{ + return NVKM_MEM_TARGET_INST; +} + +static u64 +nv40_instobj_addr(struct nvkm_memory *memory) +{ + return nv40_instobj(memory)->node->offset; +} + +static u64 +nv40_instobj_size(struct nvkm_memory *memory) +{ + return nv40_instobj(memory)->node->length; +} + +static void __iomem * +nv40_instobj_acquire(struct nvkm_memory *memory) +{ + struct nv40_instobj *iobj = nv40_instobj(memory); + return iobj->imem->iomem + iobj->node->offset; +} + +static void +nv40_instobj_release(struct nvkm_memory *memory) +{ +} static u32 -nv40_instmem_rd32(struct nvkm_object *object, u64 addr) +nv40_instobj_rd32(struct nvkm_memory *memory, u64 offset) { - struct nv04_instmem_priv *priv = (void *)object; - return ioread32_native(priv->iomem + addr); + struct nv40_instobj *iobj = nv40_instobj(memory); + return ioread32_native(iobj->imem->iomem + iobj->node->offset + offset); } static void -nv40_instmem_wr32(struct nvkm_object *object, u64 addr, u32 data) +nv40_instobj_wr32(struct nvkm_memory *memory, u64 offset, u32 data) +{ + struct nv40_instobj *iobj = nv40_instobj(memory); + iowrite32_native(data, iobj->imem->iomem + iobj->node->offset + offset); +} + +static void * +nv40_instobj_dtor(struct nvkm_memory *memory) { - struct nv04_instmem_priv *priv = (void *)object; - iowrite32_native(data, priv->iomem + addr); + struct nv40_instobj *iobj = nv40_instobj(memory); + mutex_lock(&iobj->imem->base.subdev.mutex); + nvkm_mm_free(&iobj->imem->heap, &iobj->node); + mutex_unlock(&iobj->imem->base.subdev.mutex); + return iobj; } +static const struct nvkm_memory_func +nv40_instobj_func = { + .dtor = nv40_instobj_dtor, + .target = nv40_instobj_target, + .size = nv40_instobj_size, + .addr = nv40_instobj_addr, + .acquire = nv40_instobj_acquire, + .release = nv40_instobj_release, + .rd32 = nv40_instobj_rd32, + .wr32 = nv40_instobj_wr32, +}; + static int -nv40_instmem_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +nv40_instobj_new(struct nvkm_instmem *base, u32 size, u32 align, bool zero, + struct nvkm_memory **pmemory) { - struct nvkm_device *device = nv_device(parent); - struct nv04_instmem_priv *priv; - int ret, bar, vs; + struct nv40_instmem *imem = nv40_instmem(base); + struct nv40_instobj *iobj; + int ret; - ret = nvkm_instmem_create(parent, engine, oclass, &priv); - *pobject = nv_object(priv); - if (ret) - return ret; + if (!(iobj = kzalloc(sizeof(*iobj), GFP_KERNEL))) + return -ENOMEM; + *pmemory = &iobj->memory; - /* map bar */ - if (nv_device_resource_len(device, 2)) - bar = 2; - else - bar = 3; + nvkm_memory_ctor(&nv40_instobj_func, &iobj->memory); + iobj->imem = imem; - priv->iomem = ioremap(nv_device_resource_start(device, bar), - nv_device_resource_len(device, bar)); - if (!priv->iomem) { - nv_error(priv, "unable to map PRAMIN BAR\n"); - return -EFAULT; - } + mutex_lock(&imem->base.subdev.mutex); + ret = nvkm_mm_head(&imem->heap, 0, 1, size, size, + align ? align : 1, &iobj->node); + mutex_unlock(&imem->base.subdev.mutex); + return ret; +} + +/****************************************************************************** + * instmem subdev implementation + *****************************************************************************/ + +static u32 +nv40_instmem_rd32(struct nvkm_instmem *base, u32 addr) +{ + return ioread32_native(nv40_instmem(base)->iomem + addr); +} + +static void +nv40_instmem_wr32(struct nvkm_instmem *base, u32 addr, u32 data) +{ + iowrite32_native(data, nv40_instmem(base)->iomem + addr); +} + +static int +nv40_instmem_oneinit(struct nvkm_instmem *base) +{ + struct nv40_instmem *imem = nv40_instmem(base); + struct nvkm_device *device = imem->base.subdev.device; + int ret, vs; /* PRAMIN aperture maps over the end of vram, reserve enough space * to fit graphics contexts for every channel, the magics come * from engine/gr/nv40.c */ - vs = hweight8((nv_rd32(priv, 0x001540) & 0x0000ff00) >> 8); - if (device->chipset == 0x40) priv->base.reserved = 0x6aa0 * vs; - else if (device->chipset < 0x43) priv->base.reserved = 0x4f00 * vs; - else if (nv44_gr_class(priv)) priv->base.reserved = 0x4980 * vs; - else priv->base.reserved = 0x4a40 * vs; - priv->base.reserved += 16 * 1024; - priv->base.reserved *= 32; /* per-channel */ - priv->base.reserved += 512 * 1024; /* pci(e)gart table */ - priv->base.reserved += 512 * 1024; /* object storage */ - - priv->base.reserved = round_up(priv->base.reserved, 4096); - - ret = nvkm_mm_init(&priv->heap, 0, priv->base.reserved, 1); + vs = hweight8((nvkm_rd32(device, 0x001540) & 0x0000ff00) >> 8); + if (device->chipset == 0x40) imem->base.reserved = 0x6aa0 * vs; + else if (device->chipset < 0x43) imem->base.reserved = 0x4f00 * vs; + else if (nv44_gr_class(device)) imem->base.reserved = 0x4980 * vs; + else imem->base.reserved = 0x4a40 * vs; + imem->base.reserved += 16 * 1024; + imem->base.reserved *= 32; /* per-channel */ + imem->base.reserved += 512 * 1024; /* pci(e)gart table */ + imem->base.reserved += 512 * 1024; /* object storage */ + imem->base.reserved = round_up(imem->base.reserved, 4096); + + ret = nvkm_mm_init(&imem->heap, 0, imem->base.reserved, 1); if (ret) return ret; /* 0x00000-0x10000: reserve for probable vbios image */ - ret = nvkm_gpuobj_new(nv_object(priv), NULL, 0x10000, 0, 0, - &priv->vbios); + ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x10000, 0, false, + &imem->base.vbios); if (ret) return ret; /* 0x10000-0x18000: reserve for RAMHT */ - ret = nvkm_ramht_new(nv_object(priv), NULL, 0x08000, 0, &priv->ramht); + ret = nvkm_ramht_new(device, 0x08000, 0, NULL, &imem->base.ramht); if (ret) return ret; /* 0x18000-0x18200: reserve for RAMRO * 0x18200-0x20000: padding */ - ret = nvkm_gpuobj_new(nv_object(priv), NULL, 0x08000, 0, 0, - &priv->ramro); + ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x08000, 0, false, + &imem->base.ramro); if (ret) return ret; /* 0x20000-0x21000: reserve for RAMFC * 0x21000-0x40000: padding and some unknown crap */ - ret = nvkm_gpuobj_new(nv_object(priv), NULL, 0x20000, 0, - NVOBJ_FLAG_ZERO_ALLOC, &priv->ramfc); + ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x20000, 0, true, + &imem->base.ramfc); if (ret) return ret; return 0; } -struct nvkm_oclass * -nv40_instmem_oclass = &(struct nvkm_instmem_impl) { - .base.handle = NV_SUBDEV(INSTMEM, 0x40), - .base.ofuncs = &(struct nvkm_ofuncs) { - .ctor = nv40_instmem_ctor, - .dtor = nv04_instmem_dtor, - .init = _nvkm_instmem_init, - .fini = _nvkm_instmem_fini, - .rd32 = nv40_instmem_rd32, - .wr32 = nv40_instmem_wr32, - }, - .instobj = &nv04_instobj_oclass.base, -}.base; +static void * +nv40_instmem_dtor(struct nvkm_instmem *base) +{ + struct nv40_instmem *imem = nv40_instmem(base); + nvkm_memory_del(&imem->base.ramfc); + nvkm_memory_del(&imem->base.ramro); + nvkm_ramht_del(&imem->base.ramht); + nvkm_memory_del(&imem->base.vbios); + nvkm_mm_fini(&imem->heap); + if (imem->iomem) + iounmap(imem->iomem); + return imem; +} + +static const struct nvkm_instmem_func +nv40_instmem = { + .dtor = nv40_instmem_dtor, + .oneinit = nv40_instmem_oneinit, + .rd32 = nv40_instmem_rd32, + .wr32 = nv40_instmem_wr32, + .memory_new = nv40_instobj_new, + .persistent = false, + .zero = false, +}; + +int +nv40_instmem_new(struct nvkm_device *device, int index, + struct nvkm_instmem **pimem) +{ + struct nv40_instmem *imem; + int bar; + + if (!(imem = kzalloc(sizeof(*imem), GFP_KERNEL))) + return -ENOMEM; + nvkm_instmem_ctor(&nv40_instmem, device, index, &imem->base); + *pimem = &imem->base; + + /* map bar */ + if (device->func->resource_size(device, 2)) + bar = 2; + else + bar = 3; + + imem->iomem = ioremap(device->func->resource_addr(device, bar), + device->func->resource_size(device, bar)); + if (!imem->iomem) { + nvkm_error(&imem->base.subdev, "unable to map PRAMIN BAR\n"); + return -EFAULT; + } + + return 0; +} diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c index 8404143f93ee..6d512c062ae3 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c @@ -21,149 +21,229 @@ * * Authors: Ben Skeggs */ +#define nv50_instmem(p) container_of((p), struct nv50_instmem, base) #include "priv.h" +#include <core/memory.h> +#include <subdev/bar.h> #include <subdev/fb.h> +#include <subdev/mmu.h> -struct nv50_instmem_priv { +struct nv50_instmem { struct nvkm_instmem base; + unsigned long lock_flags; spinlock_t lock; u64 addr; }; -struct nv50_instobj_priv { - struct nvkm_instobj base; - struct nvkm_mem *mem; -}; - /****************************************************************************** * instmem object implementation *****************************************************************************/ +#define nv50_instobj(p) container_of((p), struct nv50_instobj, memory) -static u32 -nv50_instobj_rd32(struct nvkm_object *object, u64 offset) +struct nv50_instobj { + struct nvkm_memory memory; + struct nv50_instmem *imem; + struct nvkm_mem *mem; + struct nvkm_vma bar; + void *map; +}; + +static enum nvkm_memory_target +nv50_instobj_target(struct nvkm_memory *memory) +{ + return NVKM_MEM_TARGET_VRAM; +} + +static u64 +nv50_instobj_addr(struct nvkm_memory *memory) +{ + return nv50_instobj(memory)->mem->offset; +} + +static u64 +nv50_instobj_size(struct nvkm_memory *memory) +{ + return (u64)nv50_instobj(memory)->mem->size << NVKM_RAM_MM_SHIFT; +} + +static void +nv50_instobj_boot(struct nvkm_memory *memory, struct nvkm_vm *vm) +{ + struct nv50_instobj *iobj = nv50_instobj(memory); + struct nvkm_subdev *subdev = &iobj->imem->base.subdev; + struct nvkm_device *device = subdev->device; + u64 size = nvkm_memory_size(memory); + void __iomem *map; + int ret; + + iobj->map = ERR_PTR(-ENOMEM); + + ret = nvkm_vm_get(vm, size, 12, NV_MEM_ACCESS_RW, &iobj->bar); + if (ret == 0) { + map = ioremap(device->func->resource_addr(device, 3) + + (u32)iobj->bar.offset, size); + if (map) { + nvkm_memory_map(memory, &iobj->bar, 0); + iobj->map = map; + } else { + nvkm_warn(subdev, "PRAMIN ioremap failed\n"); + nvkm_vm_put(&iobj->bar); + } + } else { + nvkm_warn(subdev, "PRAMIN exhausted\n"); + } +} + +static void +nv50_instobj_release(struct nvkm_memory *memory) { - struct nv50_instmem_priv *priv = (void *)nvkm_instmem(object); - struct nv50_instobj_priv *node = (void *)object; + struct nv50_instmem *imem = nv50_instobj(memory)->imem; + spin_unlock_irqrestore(&imem->lock, imem->lock_flags); +} + +static void __iomem * +nv50_instobj_acquire(struct nvkm_memory *memory) +{ + struct nv50_instobj *iobj = nv50_instobj(memory); + struct nv50_instmem *imem = iobj->imem; + struct nvkm_bar *bar = imem->base.subdev.device->bar; + struct nvkm_vm *vm; unsigned long flags; - u64 base = (node->mem->offset + offset) & 0xffffff00000ULL; - u64 addr = (node->mem->offset + offset) & 0x000000fffffULL; + + if (!iobj->map && (vm = nvkm_bar_kmap(bar))) + nvkm_memory_boot(memory, vm); + if (!IS_ERR_OR_NULL(iobj->map)) + return iobj->map; + + spin_lock_irqsave(&imem->lock, flags); + imem->lock_flags = flags; + return NULL; +} + +static u32 +nv50_instobj_rd32(struct nvkm_memory *memory, u64 offset) +{ + struct nv50_instobj *iobj = nv50_instobj(memory); + struct nv50_instmem *imem = iobj->imem; + struct nvkm_device *device = imem->base.subdev.device; + u64 base = (iobj->mem->offset + offset) & 0xffffff00000ULL; + u64 addr = (iobj->mem->offset + offset) & 0x000000fffffULL; u32 data; - spin_lock_irqsave(&priv->lock, flags); - if (unlikely(priv->addr != base)) { - nv_wr32(priv, 0x001700, base >> 16); - priv->addr = base; + if (unlikely(imem->addr != base)) { + nvkm_wr32(device, 0x001700, base >> 16); + imem->addr = base; } - data = nv_rd32(priv, 0x700000 + addr); - spin_unlock_irqrestore(&priv->lock, flags); + data = nvkm_rd32(device, 0x700000 + addr); return data; } static void -nv50_instobj_wr32(struct nvkm_object *object, u64 offset, u32 data) +nv50_instobj_wr32(struct nvkm_memory *memory, u64 offset, u32 data) { - struct nv50_instmem_priv *priv = (void *)nvkm_instmem(object); - struct nv50_instobj_priv *node = (void *)object; - unsigned long flags; - u64 base = (node->mem->offset + offset) & 0xffffff00000ULL; - u64 addr = (node->mem->offset + offset) & 0x000000fffffULL; - - spin_lock_irqsave(&priv->lock, flags); - if (unlikely(priv->addr != base)) { - nv_wr32(priv, 0x001700, base >> 16); - priv->addr = base; + struct nv50_instobj *iobj = nv50_instobj(memory); + struct nv50_instmem *imem = iobj->imem; + struct nvkm_device *device = imem->base.subdev.device; + u64 base = (iobj->mem->offset + offset) & 0xffffff00000ULL; + u64 addr = (iobj->mem->offset + offset) & 0x000000fffffULL; + + if (unlikely(imem->addr != base)) { + nvkm_wr32(device, 0x001700, base >> 16); + imem->addr = base; } - nv_wr32(priv, 0x700000 + addr, data); - spin_unlock_irqrestore(&priv->lock, flags); + nvkm_wr32(device, 0x700000 + addr, data); } static void -nv50_instobj_dtor(struct nvkm_object *object) +nv50_instobj_map(struct nvkm_memory *memory, struct nvkm_vma *vma, u64 offset) { - struct nv50_instobj_priv *node = (void *)object; - struct nvkm_fb *pfb = nvkm_fb(object); - pfb->ram->put(pfb, &node->mem); - nvkm_instobj_destroy(&node->base); + struct nv50_instobj *iobj = nv50_instobj(memory); + nvkm_vm_map_at(vma, offset, iobj->mem); } +static void * +nv50_instobj_dtor(struct nvkm_memory *memory) +{ + struct nv50_instobj *iobj = nv50_instobj(memory); + struct nvkm_ram *ram = iobj->imem->base.subdev.device->fb->ram; + if (!IS_ERR_OR_NULL(iobj->map)) { + nvkm_vm_put(&iobj->bar); + iounmap(iobj->map); + } + ram->func->put(ram, &iobj->mem); + return iobj; +} + +static const struct nvkm_memory_func +nv50_instobj_func = { + .dtor = nv50_instobj_dtor, + .target = nv50_instobj_target, + .size = nv50_instobj_size, + .addr = nv50_instobj_addr, + .boot = nv50_instobj_boot, + .acquire = nv50_instobj_acquire, + .release = nv50_instobj_release, + .rd32 = nv50_instobj_rd32, + .wr32 = nv50_instobj_wr32, + .map = nv50_instobj_map, +}; + static int -nv50_instobj_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +nv50_instobj_new(struct nvkm_instmem *base, u32 size, u32 align, bool zero, + struct nvkm_memory **pmemory) { - struct nvkm_fb *pfb = nvkm_fb(parent); - struct nvkm_instobj_args *args = data; - struct nv50_instobj_priv *node; + struct nv50_instmem *imem = nv50_instmem(base); + struct nv50_instobj *iobj; + struct nvkm_ram *ram = imem->base.subdev.device->fb->ram; int ret; - args->size = max((args->size + 4095) & ~4095, (u32)4096); - args->align = max((args->align + 4095) & ~4095, (u32)4096); + if (!(iobj = kzalloc(sizeof(*iobj), GFP_KERNEL))) + return -ENOMEM; + *pmemory = &iobj->memory; - ret = nvkm_instobj_create(parent, engine, oclass, &node); - *pobject = nv_object(node); - if (ret) - return ret; + nvkm_memory_ctor(&nv50_instobj_func, &iobj->memory); + iobj->imem = imem; + + size = max((size + 4095) & ~4095, (u32)4096); + align = max((align + 4095) & ~4095, (u32)4096); - ret = pfb->ram->get(pfb, args->size, args->align, 0, 0x800, &node->mem); + ret = ram->func->get(ram, size, align, 0, 0x800, &iobj->mem); if (ret) return ret; - node->base.addr = node->mem->offset; - node->base.size = node->mem->size << 12; - node->mem->page_shift = 12; + iobj->mem->page_shift = 12; return 0; } -static struct nvkm_instobj_impl -nv50_instobj_oclass = { - .base.ofuncs = &(struct nvkm_ofuncs) { - .ctor = nv50_instobj_ctor, - .dtor = nv50_instobj_dtor, - .init = _nvkm_instobj_init, - .fini = _nvkm_instobj_fini, - .rd32 = nv50_instobj_rd32, - .wr32 = nv50_instobj_wr32, - }, -}; - /****************************************************************************** * instmem subdev implementation *****************************************************************************/ -static int -nv50_instmem_fini(struct nvkm_object *object, bool suspend) +static void +nv50_instmem_fini(struct nvkm_instmem *base) { - struct nv50_instmem_priv *priv = (void *)object; - priv->addr = ~0ULL; - return nvkm_instmem_fini(&priv->base, suspend); + nv50_instmem(base)->addr = ~0ULL; } -static int -nv50_instmem_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) -{ - struct nv50_instmem_priv *priv; - int ret; +static const struct nvkm_instmem_func +nv50_instmem = { + .fini = nv50_instmem_fini, + .memory_new = nv50_instobj_new, + .persistent = false, + .zero = false, +}; - ret = nvkm_instmem_create(parent, engine, oclass, &priv); - *pobject = nv_object(priv); - if (ret) - return ret; +int +nv50_instmem_new(struct nvkm_device *device, int index, + struct nvkm_instmem **pimem) +{ + struct nv50_instmem *imem; - spin_lock_init(&priv->lock); + if (!(imem = kzalloc(sizeof(*imem), GFP_KERNEL))) + return -ENOMEM; + nvkm_instmem_ctor(&nv50_instmem, device, index, &imem->base); + spin_lock_init(&imem->lock); + *pimem = &imem->base; return 0; } - -struct nvkm_oclass * -nv50_instmem_oclass = &(struct nvkm_instmem_impl) { - .base.handle = NV_SUBDEV(INSTMEM, 0x50), - .base.ofuncs = &(struct nvkm_ofuncs) { - .ctor = nv50_instmem_ctor, - .dtor = _nvkm_instmem_dtor, - .init = _nvkm_instmem_init, - .fini = nv50_instmem_fini, - }, - .instobj = &nv50_instobj_oclass.base, -}.base; diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h index b10e292e5607..ace4471864a3 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h @@ -1,54 +1,20 @@ #ifndef __NVKM_INSTMEM_PRIV_H__ #define __NVKM_INSTMEM_PRIV_H__ +#define nvkm_instmem(p) container_of((p), struct nvkm_instmem, subdev) #include <subdev/instmem.h> -struct nvkm_instobj_impl { - struct nvkm_oclass base; +struct nvkm_instmem_func { + void *(*dtor)(struct nvkm_instmem *); + int (*oneinit)(struct nvkm_instmem *); + void (*fini)(struct nvkm_instmem *); + u32 (*rd32)(struct nvkm_instmem *, u32 addr); + void (*wr32)(struct nvkm_instmem *, u32 addr, u32 data); + int (*memory_new)(struct nvkm_instmem *, u32 size, u32 align, + bool zero, struct nvkm_memory **); + bool persistent; + bool zero; }; -struct nvkm_instobj_args { - u32 size; - u32 align; -}; - -#define nvkm_instobj_create(p,e,o,d) \ - nvkm_instobj_create_((p), (e), (o), sizeof(**d), (void **)d) -#define nvkm_instobj_destroy(p) ({ \ - struct nvkm_instobj *iobj = (p); \ - _nvkm_instobj_dtor(nv_object(iobj)); \ -}) -#define nvkm_instobj_init(p) \ - nvkm_object_init(&(p)->base) -#define nvkm_instobj_fini(p,s) \ - nvkm_object_fini(&(p)->base, (s)) - -int nvkm_instobj_create_(struct nvkm_object *, struct nvkm_object *, - struct nvkm_oclass *, int, void **); -void _nvkm_instobj_dtor(struct nvkm_object *); -#define _nvkm_instobj_init nvkm_object_init -#define _nvkm_instobj_fini nvkm_object_fini - -struct nvkm_instmem_impl { - struct nvkm_oclass base; - struct nvkm_oclass *instobj; -}; - -#define nvkm_instmem_create(p,e,o,d) \ - nvkm_instmem_create_((p), (e), (o), sizeof(**d), (void **)d) -#define nvkm_instmem_destroy(p) \ - nvkm_subdev_destroy(&(p)->base) -#define nvkm_instmem_init(p) ({ \ - struct nvkm_instmem *imem = (p); \ - _nvkm_instmem_init(nv_object(imem)); \ -}) -#define nvkm_instmem_fini(p,s) ({ \ - struct nvkm_instmem *imem = (p); \ - _nvkm_instmem_fini(nv_object(imem), (s)); \ -}) - -int nvkm_instmem_create_(struct nvkm_object *, struct nvkm_object *, - struct nvkm_oclass *, int, void **); -#define _nvkm_instmem_dtor _nvkm_subdev_dtor -int _nvkm_instmem_init(struct nvkm_object *); -int _nvkm_instmem_fini(struct nvkm_object *, bool); +void nvkm_instmem_ctor(const struct nvkm_instmem_func *, struct nvkm_device *, + int index, struct nvkm_instmem *); #endif |