gpu: nvgpu: Split mem_desc from MM code

Split the mem_desc code out from the MM code. This is to help
simplify the MM code and make it easier to abstract the DMA
allocation routines.

JIRA NVGPU-12

Change-Id: I2ccb643efe6bbed80d1360a580ff5593acb407bd
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: http://git-master/r/1323324
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

2 files changed, 7 insertions, 288 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
index 2fe76d80..e78eb941 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
@@ -56,21 +56,18 @@
 static void gk20a_vidmem_clear_mem_worker(struct work_struct *work);
 #endif
 
-static inline void
-set_vidmem_page_alloc(struct scatterlist *sgl, u64 addr)
+void set_vidmem_page_alloc(struct scatterlist *sgl, u64 addr)
 {
         /* set bit 0 to indicate vidmem allocation */
         sg_dma_address(sgl) = (addr | 1ULL);
 }
 
-static inline bool
-is_vidmem_page_alloc(u64 addr)
+bool is_vidmem_page_alloc(u64 addr)
 {
         return !!(addr & 1ULL);
 }
 
-struct nvgpu_page_alloc *
-get_vidmem_page_alloc(struct scatterlist *sgl)
+struct nvgpu_page_alloc *get_vidmem_page_alloc(struct scatterlist *sgl)
 {
         u64 addr;
 
@@ -84,187 +81,6 @@ get_vidmem_page_alloc(struct scatterlist *sgl)
         return (struct nvgpu_page_alloc *)(uintptr_t)addr;
 }
 
-int gk20a_mem_begin(struct gk20a *g, struct mem_desc *mem)
-{
-        void *cpu_va;
-
-        if (mem->aperture != APERTURE_SYSMEM || g->mm.force_pramin)
-                return 0;
-
-        if (WARN_ON(mem->cpu_va)) {
-                gk20a_warn(dev_from_gk20a(g), "nested %s", __func__);
-                return -EBUSY;
-        }
-
-        cpu_va = vmap(mem->pages,
-                        PAGE_ALIGN(mem->size) >> PAGE_SHIFT,
-                        0, pgprot_writecombine(PAGE_KERNEL));
-
-        if (WARN_ON(!cpu_va))
-                return -ENOMEM;
-
-        mem->cpu_va = cpu_va;
-        return 0;
-}
-
-void gk20a_mem_end(struct gk20a *g, struct mem_desc *mem)
-{
-        if (mem->aperture != APERTURE_SYSMEM || g->mm.force_pramin)
-                return;
-
-        vunmap(mem->cpu_va);
-        mem->cpu_va = NULL;
-}
-
-u32 gk20a_mem_rd32(struct gk20a *g, struct mem_desc *mem, u32 w)
-{
-        u32 data = 0;
-
-        if (mem->aperture == APERTURE_SYSMEM && !g->mm.force_pramin) {
-                u32 *ptr = mem->cpu_va;
-
-                WARN_ON(!ptr);
-                data = ptr[w];
-#ifdef CONFIG_TEGRA_SIMULATION_PLATFORM
-                gk20a_dbg(gpu_dbg_mem, " %p = 0x%x", ptr + w, data);
-#endif
-        } else if (mem->aperture == APERTURE_VIDMEM || g->mm.force_pramin) {
-                u32 value;
-                u32 *p = &value;
-
-                nvgpu_pramin_access_batched(g, mem, w * sizeof(u32),
-                                sizeof(u32), pramin_access_batch_rd_n, &p);
-
-                data = value;
-
-        } else {
-                WARN_ON("Accessing unallocated mem_desc");
-        }
-
-        return data;
-}
-
-u32 gk20a_mem_rd(struct gk20a *g, struct mem_desc *mem, u32 offset)
-{
-        WARN_ON(offset & 3);
-        return gk20a_mem_rd32(g, mem, offset / sizeof(u32));
-}
-
-void gk20a_mem_rd_n(struct gk20a *g, struct mem_desc *mem,
-                u32 offset, void *dest, u32 size)
-{
-        WARN_ON(offset & 3);
-        WARN_ON(size & 3);
-
-        if (mem->aperture == APERTURE_SYSMEM && !g->mm.force_pramin) {
-                u8 *src = (u8 *)mem->cpu_va + offset;
-
-                WARN_ON(!mem->cpu_va);
-                memcpy(dest, src, size);
-#ifdef CONFIG_TEGRA_SIMULATION_PLATFORM
-                if (size)
-                        gk20a_dbg(gpu_dbg_mem, " %p = 0x%x ... [%d bytes]",
-                                        src, *dest, size);
-#endif
-        } else if (mem->aperture == APERTURE_VIDMEM || g->mm.force_pramin) {
-                u32 *dest_u32 = dest;
-
-                nvgpu_pramin_access_batched(g, mem, offset, size,
-                                pramin_access_batch_rd_n, &dest_u32);
-        } else {
-                WARN_ON("Accessing unallocated mem_desc");
-        }
-}
-
-void gk20a_mem_wr32(struct gk20a *g, struct mem_desc *mem, u32 w, u32 data)
-{
-        if (mem->aperture == APERTURE_SYSMEM && !g->mm.force_pramin) {
-                u32 *ptr = mem->cpu_va;
-
-                WARN_ON(!ptr);
-#ifdef CONFIG_TEGRA_SIMULATION_PLATFORM
-                gk20a_dbg(gpu_dbg_mem, " %p = 0x%x", ptr + w, data);
-#endif
-                ptr[w] = data;
-        } else if (mem->aperture == APERTURE_VIDMEM || g->mm.force_pramin) {
-                u32 value = data;
-                u32 *p = &value;
-
-                nvgpu_pramin_access_batched(g, mem, w * sizeof(u32),
-                                sizeof(u32), pramin_access_batch_wr_n, &p);
-                if (!mem->skip_wmb)
-                        wmb();
-        } else {
-                WARN_ON("Accessing unallocated mem_desc");
-        }
-}
-
-void gk20a_mem_wr(struct gk20a *g, struct mem_desc *mem, u32 offset, u32 data)
-{
-        WARN_ON(offset & 3);
-        gk20a_mem_wr32(g, mem, offset / sizeof(u32), data);
-}
-
-void gk20a_mem_wr_n(struct gk20a *g, struct mem_desc *mem, u32 offset,
-                void *src, u32 size)
-{
-        WARN_ON(offset & 3);
-        WARN_ON(size & 3);
-
-        if (mem->aperture == APERTURE_SYSMEM && !g->mm.force_pramin) {
-                u8 *dest = (u8 *)mem->cpu_va + offset;
-
-                WARN_ON(!mem->cpu_va);
-#ifdef CONFIG_TEGRA_SIMULATION_PLATFORM
-                if (size)
-                        gk20a_dbg(gpu_dbg_mem, " %p = 0x%x ... [%d bytes]",
-                                        dest, *src, size);
-#endif
-                memcpy(dest, src, size);
-        } else if (mem->aperture == APERTURE_VIDMEM || g->mm.force_pramin) {
-                u32 *src_u32 = src;
-
-                nvgpu_pramin_access_batched(g, mem, offset, size,
-                                pramin_access_batch_wr_n, &src_u32);
-                if (!mem->skip_wmb)
-                        wmb();
-        } else {
-                WARN_ON("Accessing unallocated mem_desc");
-        }
-}
-
-void gk20a_memset(struct gk20a *g, struct mem_desc *mem, u32 offset,
-                u32 c, u32 size)
-{
-        WARN_ON(offset & 3);
-        WARN_ON(size & 3);
-        WARN_ON(c & ~0xff);
-
-        c &= 0xff;
-
-        if (mem->aperture == APERTURE_SYSMEM && !g->mm.force_pramin) {
-                u8 *dest = (u8 *)mem->cpu_va + offset;
-
-                WARN_ON(!mem->cpu_va);
-#ifdef CONFIG_TEGRA_SIMULATION_PLATFORM
-                if (size)
-                        gk20a_dbg(gpu_dbg_mem, " %p = 0x%x [times %d]",
-                                dest, c, size);
-#endif
-                memset(dest, c, size);
-        } else if (mem->aperture == APERTURE_VIDMEM || g->mm.force_pramin) {
-                u32 repeat_value = c | (c << 8) | (c << 16) | (c << 24);
-                u32 *p = &repeat_value;
-
-                nvgpu_pramin_access_batched(g, mem, offset, size,
-                                pramin_access_batch_set, &p);
-                if (!mem->skip_wmb)
-                        wmb();
-        } else {
-                WARN_ON("Accessing unallocated mem_desc");
-        }
-}
-
 /*
  * GPU mapping life cycle
  * ======================
@@ -3121,29 +2937,6 @@ static void gk20a_vidmem_clear_mem_worker(struct work_struct *work)
 }
 #endif
 
-u32 __gk20a_aperture_mask(struct gk20a *g, enum gk20a_aperture aperture,
-                u32 sysmem_mask, u32 vidmem_mask)
-{
-        switch (aperture) {
-        case APERTURE_SYSMEM:
-                /* sysmem for dgpus; some igpus consider system memory vidmem */
-                return g->mm.vidmem_is_vidmem ? sysmem_mask : vidmem_mask;
-        case APERTURE_VIDMEM:
-                /* for dgpus only */
-                return vidmem_mask;
-        case APERTURE_INVALID:
-                WARN_ON("Bad aperture");
-        }
-        return 0;
-}
-
-u32 gk20a_aperture_mask(struct gk20a *g, struct mem_desc *mem,
-                u32 sysmem_mask, u32 vidmem_mask)
-{
-        return __gk20a_aperture_mask(g, mem->aperture,
-                        sysmem_mask, vidmem_mask);
-}
-
 int gk20a_gmmu_alloc_map(struct vm_gk20a *vm, size_t size,
                 struct mem_desc *mem)
 {
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
index b425ec5c..da8bbb0a 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
@@ -24,6 +24,7 @@
 #include <asm/dma-iommu.h>
 #include <asm/cacheflush.h>
 
+#include <nvgpu/mem_desc.h>
 #include <nvgpu/allocator.h>
 #include <nvgpu/list.h>
 #include <nvgpu/rbtree.h>
@@ -39,60 +40,12 @@
                 outer_flush_range(pa, pa + (size_t)(size));             \
         } while (0)
 
-/*
- * Real location of a buffer - gk20a_aperture_mask() will deduce what will be
- * told to the gpu about the aperture, but this flag designates where the
- * memory actually was allocated from.
- */
-enum gk20a_aperture {
-        APERTURE_INVALID, /* unallocated or N/A */
-        APERTURE_SYSMEM,
-        APERTURE_VIDMEM
-};
-
 enum gk20a_mem_rw_flag {
         gk20a_mem_flag_none = 0,
         gk20a_mem_flag_read_only = 1,
         gk20a_mem_flag_write_only = 2,
 };
 
-static inline const char *gk20a_aperture_str(enum gk20a_aperture aperture)
-{
-        switch (aperture) {
-                case APERTURE_INVALID: return "invalid";
-                case APERTURE_SYSMEM:  return "sysmem";
-                case APERTURE_VIDMEM:  return "vidmem";
-        };
-        return "UNKNOWN";
-}
-
-struct mem_desc {
-        void *cpu_va; /* sysmem only */
-        struct page **pages; /* sysmem only */
-        struct sg_table *sgt;
-        enum gk20a_aperture aperture;
-        size_t size;
-        u64 gpu_va;
-        bool fixed; /* vidmem only */
-        bool user_mem; /* vidmem only */
-        struct nvgpu_allocator *allocator; /* vidmem only */
-        struct nvgpu_list_node clear_list_entry; /* vidmem only */
-        bool skip_wmb;
-        unsigned long flags;
-};
-
-static inline struct mem_desc *
-mem_desc_from_clear_list_entry(struct nvgpu_list_node *node)
-{
-        return (struct mem_desc *)
-                ((uintptr_t)node - offsetof(struct mem_desc, clear_list_entry));
-};
-
-struct mem_desc_sub {
-        u32 offset;
-        u32 size;
-};
-
 struct gpfifo_desc {
         struct mem_desc mem;
         u32 entry_num;
@@ -511,36 +464,9 @@ enum gmmu_pgsz_gk20a __get_pte_size_fixed_map(struct vm_gk20a *vm,
                                               u64 base, u64 size);
 enum gmmu_pgsz_gk20a __get_pte_size(struct vm_gk20a *vm, u64 base, u64 size);
 
-struct nvgpu_page_alloc *
-get_vidmem_page_alloc(struct scatterlist *sgl);
-
-/*
- * Buffer accessors - wrap between begin() and end() if there is no permanent
- * kernel mapping for this buffer.
- */
-
-int gk20a_mem_begin(struct gk20a *g, struct mem_desc *mem);
-/* nop for null mem, like with free() or vunmap() */
-void gk20a_mem_end(struct gk20a *g, struct mem_desc *mem);
-
-/* word-indexed offset */
-u32 gk20a_mem_rd32(struct gk20a *g, struct mem_desc *mem, u32 w);
-/* byte offset (32b-aligned) */
-u32 gk20a_mem_rd(struct gk20a *g, struct mem_desc *mem, u32 offset);
-/* memcpy to cpu, offset and size in bytes (32b-aligned) */
-void gk20a_mem_rd_n(struct gk20a *g, struct mem_desc *mem, u32 offset,
-                void *dest, u32 size);
-
-/* word-indexed offset */
-void gk20a_mem_wr32(struct gk20a *g, struct mem_desc *mem, u32 w, u32 data);
-/* byte offset (32b-aligned) */
-void gk20a_mem_wr(struct gk20a *g, struct mem_desc *mem, u32 offset, u32 data);
-/* memcpy from cpu, offset and size in bytes (32b-aligned) */
-void gk20a_mem_wr_n(struct gk20a *g, struct mem_desc *mem, u32 offset,
-                void *src, u32 size);
-/* size and offset in bytes (32b-aligned), filled with the constant byte c */
-void gk20a_memset(struct gk20a *g, struct mem_desc *mem, u32 offset,
-                u32 c, u32 size);
+void set_vidmem_page_alloc(struct scatterlist *sgl, u64 addr);
+bool is_vidmem_page_alloc(u64 addr);
+struct nvgpu_page_alloc *get_vidmem_page_alloc(struct scatterlist *sgl);
 
 #if 0 /*related to addr bits above, concern below TBD on which is accurate */
 #define bar1_instance_block_shift_gk20a() (max_physaddr_bits_gk20a() -\