gpu: nvgpu: Move common DMA code to common/mm

This migrates the common DMA code (os agnostic) to the
common directory. This new unit will be the common DMA
allocator that lets users allocate SYSMEM, VIDMEM, or
either. Other units will be responsible for actually
handling the mechanics of allocating VIDMEM or SYSMEM.

Also update the names of the DMA related files so that
tmake doesn't complain about duplicate C file names. To
do this call the common DMA file dma.c and prepend the
OS to the other DMA files. So now we have:

  common/mm/dma.c
  os/posix/posix-dma.c
  os/linux/linux-dma.c

JIRA NVGPU-990

Change-Id: I22d2d41803ad89be7d9c28f87864ce4fedf10836
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1799807
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

6 files changed, 138 insertions, 242 deletions

diff --git a/drivers/gpu/nvgpu/Makefile b/drivers/gpu/nvgpu/Makefile
index 43a36221..a3372cef 100644
--- a/drivers/gpu/nvgpu/Makefile
+++ b/drivers/gpu/nvgpu/Makefile
@@ -65,7 +65,7 @@ nvgpu-y += \
         os/linux/log.o \
         os/linux/cond.o \
         os/linux/nvgpu_mem.o \
-        os/linux/dma.o \
+        os/linux/linux-dma.o \
         os/linux/driver_common.o \
         os/linux/firmware.o \
         os/linux/thread.o \
@@ -180,6 +180,7 @@ nvgpu-y += \
         common/mm/nvgpu_mem.o \
         common/mm/comptags.o \
         common/mm/mm.o \
+        common/mm/dma.o \
         common/enabled.o \
         common/nvlink.o \
         common/pramin.o \
diff --git a/drivers/gpu/nvgpu/Makefile.sources b/drivers/gpu/nvgpu/Makefile.sources
index f09ebf78..7dc0251b 100644
--- a/drivers/gpu/nvgpu/Makefile.sources
+++ b/drivers/gpu/nvgpu/Makefile.sources
@@ -27,7 +27,7 @@ srcs :=	os/posix/nvgpu.c \
         os/posix/kmem.c \
         os/posix/io.c \
         os/posix/posix-nvgpu_mem.c \
-        os/posix/dma.c \
+        os/posix/posix-dma.c \
         os/posix/posix-vm.c \
         os/posix/timers.c \
         os/posix/posix-comptags.c \
@@ -56,6 +56,7 @@ srcs :=	os/posix/nvgpu.c \
         common/mm/nvgpu_mem.c \
         common/mm/comptags.c \
         common/mm/mm.c \
+        common/mm/dma.c \
         common/bus/bus_gk20a.c \
         common/bus/bus_gm20b.c \
         common/bus/bus_gp10b.c \
diff --git a/drivers/gpu/nvgpu/os/posix/dma.c b/drivers/gpu/nvgpu/common/mm/dma.c
index 95bb1a75..f7331f8e 100644
--- a/drivers/gpu/nvgpu/os/posix/dma.c
+++ b/drivers/gpu/nvgpu/common/mm/dma.c
@@ -20,45 +20,12 @@
  * DEALINGS IN THE SOFTWARE.
  */
 
-#include <stdlib.h>
-
 #include <nvgpu/mm.h>
 #include <nvgpu/vm.h>
-#include <nvgpu/bug.h>
 #include <nvgpu/dma.h>
 #include <nvgpu/gmmu.h>
-#include <nvgpu/nvgpu_mem.h>
 #include <nvgpu/enabled.h>
-
-/*
- * In userspace vidmem vs sysmem is just a difference in what is placed in the
- * aperture field.
- */
-static int __nvgpu_do_dma_alloc(struct gk20a *g, unsigned long flags,
-                                size_t size, struct nvgpu_mem *mem,
-                                enum nvgpu_aperture ap)
-{
-        void *memory = malloc(mem->aligned_size);
-
-        if (memory == NULL)
-                return -ENOMEM;
-
-        mem->cpu_va       = memory;
-        mem->aperture     = ap;
-        mem->size         = size;
-        mem->aligned_size = PAGE_ALIGN(size);
-        mem->gpu_va       = 0ULL;
-        mem->skip_wmb     = true;
-        mem->vidmem_alloc = NULL;
-        mem->allocator    = NULL;
-
-        return 0;
-}
-
-bool nvgpu_iommuable(struct gk20a *g)
-{
-        return false;
-}
+#include <nvgpu/nvgpu_mem.h>
 
 int nvgpu_dma_alloc(struct gk20a *g, size_t size, struct nvgpu_mem *mem)
 {
@@ -66,15 +33,22 @@ int nvgpu_dma_alloc(struct gk20a *g, size_t size, struct nvgpu_mem *mem)
 }
 
 int nvgpu_dma_alloc_flags(struct gk20a *g, unsigned long flags, size_t size,
-                          struct nvgpu_mem *mem)
+                struct nvgpu_mem *mem)
 {
         if (!nvgpu_is_enabled(g, NVGPU_MM_UNIFIED_MEMORY)) {
                 /*
-                 * First try vidmem. Obviously in userspace there's no such
-                 * thing as vidmem per se but we will mark the aperture as
-                 * vidmem.
+                 * Force the no-kernel-mapping flag on because we don't support
+                 * the lack of it for vidmem - the user should not care when
+                 * using nvgpu_gmmu_alloc_map and it's vidmem, or if there's a
+                 * difference, the user should use the flag explicitly anyway.
+                 *
+                 * Incoming flags are ignored here, since bits other than the
+                 * no-kernel-mapping flag are ignored by the vidmem mapping
+                 * functions anyway.
                  */
-                int err = nvgpu_dma_alloc_flags_vid(g, 0, size, mem);
+                int err = nvgpu_dma_alloc_flags_vid(g,
+                                NVGPU_DMA_NO_KERNEL_MAPPING,
+                                size, mem);
 
                 if (!err)
                         return 0;
@@ -85,7 +59,6 @@ int nvgpu_dma_alloc_flags(struct gk20a *g, unsigned long flags, size_t size,
         }
 
         return nvgpu_dma_alloc_flags_sys(g, flags, size, mem);
-
 }
 
 int nvgpu_dma_alloc_sys(struct gk20a *g, size_t size, struct nvgpu_mem *mem)
@@ -93,45 +66,23 @@ int nvgpu_dma_alloc_sys(struct gk20a *g, size_t size, struct nvgpu_mem *mem)
         return nvgpu_dma_alloc_flags_sys(g, 0, size, mem);
 }
 
-int nvgpu_dma_alloc_flags_sys(struct gk20a *g, unsigned long flags,
-                              size_t size, struct nvgpu_mem *mem)
-{
-        return __nvgpu_do_dma_alloc(g, flags, size, mem, APERTURE_SYSMEM);
-}
-
 int nvgpu_dma_alloc_vid(struct gk20a *g, size_t size, struct nvgpu_mem *mem)
 {
-        return nvgpu_dma_alloc_flags_vid(g, 0, size, mem);
+        return nvgpu_dma_alloc_flags_vid(g,
+                        NVGPU_DMA_NO_KERNEL_MAPPING, size, mem);
 }
 
 int nvgpu_dma_alloc_flags_vid(struct gk20a *g, unsigned long flags,
-                              size_t size, struct nvgpu_mem *mem)
+                size_t size, struct nvgpu_mem *mem)
 {
-        return __nvgpu_do_dma_alloc(g, flags, size, mem, APERTURE_VIDMEM);
+        return nvgpu_dma_alloc_flags_vid_at(g, flags, size, mem, 0);
 }
 
 int nvgpu_dma_alloc_vid_at(struct gk20a *g,
-                                 size_t size, struct nvgpu_mem *mem, u64 at)
+                size_t size, struct nvgpu_mem *mem, u64 at)
 {
-        BUG();
-
-        return 0;
-}
-
-int nvgpu_dma_alloc_flags_vid_at(struct gk20a *g, unsigned long flags,
-                                 size_t size, struct nvgpu_mem *mem, u64 at)
-{
-        BUG();
-
-        return 0;
-}
-
-void nvgpu_dma_free(struct gk20a *g, struct nvgpu_mem *mem)
-{
-        if (!(mem->mem_flags & NVGPU_MEM_FLAG_SHADOW_COPY))
-                free(mem->cpu_va);
-
-        memset(mem, 0, sizeof(*mem));
+        return nvgpu_dma_alloc_flags_vid_at(g,
+                        NVGPU_DMA_NO_KERNEL_MAPPING, size, mem, at);
 }
 
 int nvgpu_dma_alloc_map(struct vm_gk20a *vm, size_t size,
@@ -144,6 +95,12 @@ int nvgpu_dma_alloc_map_flags(struct vm_gk20a *vm, unsigned long flags,
                 size_t size, struct nvgpu_mem *mem)
 {
         if (!nvgpu_is_enabled(gk20a_from_vm(vm), NVGPU_MM_UNIFIED_MEMORY)) {
+                /*
+                 * Force the no-kernel-mapping flag on because we don't support
+                 * the lack of it for vidmem - the user should not care when
+                 * using nvgpu_dma_alloc_map and it's vidmem, or if there's a
+                 * difference, the user should use the flag explicitly anyway.
+                 */
                 int err = nvgpu_dma_alloc_map_flags_vid(vm,
                                 flags | NVGPU_DMA_NO_KERNEL_MAPPING,
                                 size, mem);
@@ -218,6 +175,18 @@ fail_free:
         return err;
 }
 
+void nvgpu_dma_free(struct gk20a *g, struct nvgpu_mem *mem)
+{
+        switch (mem->aperture) {
+        case APERTURE_SYSMEM:
+                return nvgpu_dma_free_sys(g, mem);
+        case APERTURE_VIDMEM:
+                return nvgpu_dma_free_vid(g, mem);
+        default:
+                break; /* like free() on "null" memory */
+        }
+}
+
 void nvgpu_dma_unmap_free(struct vm_gk20a *vm, struct nvgpu_mem *mem)
 {
         if (mem->gpu_va)
diff --git a/drivers/gpu/nvgpu/include/nvgpu/dma.h b/drivers/gpu/nvgpu/include/nvgpu/dma.h
index d83346a3..f6f264cf 100644
--- a/drivers/gpu/nvgpu/include/nvgpu/dma.h
+++ b/drivers/gpu/nvgpu/include/nvgpu/dma.h
@@ -352,4 +352,10 @@ int nvgpu_dma_alloc_map_flags_vid(struct vm_gk20a *vm, unsigned long flags,
  */
 void nvgpu_dma_unmap_free(struct vm_gk20a *vm, struct nvgpu_mem *mem);
 
+/*
+ * Don't use these directly. Instead use nvgpu_dma_free().
+ */
+void nvgpu_dma_free_sys(struct gk20a *g, struct nvgpu_mem *mem);
+void nvgpu_dma_free_vid(struct gk20a *g, struct nvgpu_mem *mem);
+
 #endif
diff --git a/drivers/gpu/nvgpu/os/linux/dma.c b/drivers/gpu/nvgpu/os/linux/linux-dma.c
index 77669493..a42e7cb5 100644
--- a/drivers/gpu/nvgpu/os/linux/dma.c
+++ b/drivers/gpu/nvgpu/os/linux/linux-dma.c
@@ -174,45 +174,6 @@ static void nvgpu_dma_flags_to_attrs(struct dma_attrs *attrs,
 #undef ATTR_ARG
 }
 
-int nvgpu_dma_alloc(struct gk20a *g, size_t size, struct nvgpu_mem *mem)
-{
-        return nvgpu_dma_alloc_flags(g, 0, size, mem);
-}
-
-int nvgpu_dma_alloc_flags(struct gk20a *g, unsigned long flags, size_t size,
-                struct nvgpu_mem *mem)
-{
-        if (!nvgpu_is_enabled(g, NVGPU_MM_UNIFIED_MEMORY)) {
-                /*
-                 * Force the no-kernel-mapping flag on because we don't support
-                 * the lack of it for vidmem - the user should not care when
-                 * using nvgpu_gmmu_alloc_map and it's vidmem, or if there's a
-                 * difference, the user should use the flag explicitly anyway.
-                 *
-                 * Incoming flags are ignored here, since bits other than the
-                 * no-kernel-mapping flag are ignored by the vidmem mapping
-                 * functions anyway.
-                 */
-                int err = nvgpu_dma_alloc_flags_vid(g,
-                                NVGPU_DMA_NO_KERNEL_MAPPING,
-                                size, mem);
-
-                if (!err)
-                        return 0;
-                /*
-                 * Fall back to sysmem (which may then also fail) in case
-                 * vidmem is exhausted.
-                 */
-        }
-
-        return nvgpu_dma_alloc_flags_sys(g, flags, size, mem);
-}
-
-int nvgpu_dma_alloc_sys(struct gk20a *g, size_t size, struct nvgpu_mem *mem)
-{
-        return nvgpu_dma_alloc_flags_sys(g, 0, size, mem);
-}
-
 int nvgpu_dma_alloc_flags_sys(struct gk20a *g, unsigned long flags,
                 size_t size, struct nvgpu_mem *mem)
 {
@@ -302,25 +263,6 @@ fail_free_dma:
         return err;
 }
 
-int nvgpu_dma_alloc_vid(struct gk20a *g, size_t size, struct nvgpu_mem *mem)
-{
-        return nvgpu_dma_alloc_flags_vid(g,
-                        NVGPU_DMA_NO_KERNEL_MAPPING, size, mem);
-}
-
-int nvgpu_dma_alloc_flags_vid(struct gk20a *g, unsigned long flags,
-                size_t size, struct nvgpu_mem *mem)
-{
-        return nvgpu_dma_alloc_flags_vid_at(g, flags, size, mem, 0);
-}
-
-int nvgpu_dma_alloc_vid_at(struct gk20a *g,
-                size_t size, struct nvgpu_mem *mem, u64 at)
-{
-        return nvgpu_dma_alloc_flags_vid_at(g,
-                        NVGPU_DMA_NO_KERNEL_MAPPING, size, mem, at);
-}
-
 int nvgpu_dma_alloc_flags_vid_at(struct gk20a *g, unsigned long flags,
                 size_t size, struct nvgpu_mem *mem, u64 at)
 {
@@ -405,97 +347,7 @@ fail_physfree:
 #endif
 }
 
-int nvgpu_dma_alloc_map(struct vm_gk20a *vm, size_t size,
-                struct nvgpu_mem *mem)
-{
-        return nvgpu_dma_alloc_map_flags(vm, 0, size, mem);
-}
-
-int nvgpu_dma_alloc_map_flags(struct vm_gk20a *vm, unsigned long flags,
-                size_t size, struct nvgpu_mem *mem)
-{
-        if (!nvgpu_is_enabled(gk20a_from_vm(vm), NVGPU_MM_UNIFIED_MEMORY)) {
-                /*
-                 * Force the no-kernel-mapping flag on because we don't support
-                 * the lack of it for vidmem - the user should not care when
-                 * using nvgpu_dma_alloc_map and it's vidmem, or if there's a
-                 * difference, the user should use the flag explicitly anyway.
-                 */
-                int err = nvgpu_dma_alloc_map_flags_vid(vm,
-                                flags | NVGPU_DMA_NO_KERNEL_MAPPING,
-                                size, mem);
-
-                if (!err)
-                        return 0;
-                /*
-                 * Fall back to sysmem (which may then also fail) in case
-                 * vidmem is exhausted.
-                 */
-        }
-
-        return nvgpu_dma_alloc_map_flags_sys(vm, flags, size, mem);
-}
-
-int nvgpu_dma_alloc_map_sys(struct vm_gk20a *vm, size_t size,
-                struct nvgpu_mem *mem)
-{
-        return nvgpu_dma_alloc_map_flags_sys(vm, 0, size, mem);
-}
-
-int nvgpu_dma_alloc_map_flags_sys(struct vm_gk20a *vm, unsigned long flags,
-                size_t size, struct nvgpu_mem *mem)
-{
-        int err = nvgpu_dma_alloc_flags_sys(vm->mm->g, flags, size, mem);
-
-        if (err)
-                return err;
-
-        mem->gpu_va = nvgpu_gmmu_map(vm, mem, size, 0,
-                                     gk20a_mem_flag_none, false,
-                                     mem->aperture);
-        if (!mem->gpu_va) {
-                err = -ENOMEM;
-                goto fail_free;
-        }
-
-        return 0;
-
-fail_free:
-        nvgpu_dma_free(vm->mm->g, mem);
-        return err;
-}
-
-int nvgpu_dma_alloc_map_vid(struct vm_gk20a *vm, size_t size,
-                struct nvgpu_mem *mem)
-{
-        return nvgpu_dma_alloc_map_flags_vid(vm,
-                        NVGPU_DMA_NO_KERNEL_MAPPING, size, mem);
-}
-
-int nvgpu_dma_alloc_map_flags_vid(struct vm_gk20a *vm, unsigned long flags,
-                size_t size, struct nvgpu_mem *mem)
-{
-        int err = nvgpu_dma_alloc_flags_vid(vm->mm->g, flags, size, mem);
-
-        if (err)
-                return err;
-
-        mem->gpu_va = nvgpu_gmmu_map(vm, mem, size, 0,
-                                     gk20a_mem_flag_none, false,
-                                     mem->aperture);
-        if (!mem->gpu_va) {
-                err = -ENOMEM;
-                goto fail_free;
-        }
-
-        return 0;
-
-fail_free:
-        nvgpu_dma_free(vm->mm->g, mem);
-        return err;
-}
-
-static void nvgpu_dma_free_sys(struct gk20a *g, struct nvgpu_mem *mem)
+void nvgpu_dma_free_sys(struct gk20a *g, struct nvgpu_mem *mem)
 {
         struct device *d = dev_from_gk20a(g);
 
@@ -551,7 +403,7 @@ static void nvgpu_dma_free_sys(struct gk20a *g, struct nvgpu_mem *mem)
         mem->aperture = APERTURE_INVALID;
 }
 
-static void nvgpu_dma_free_vid(struct gk20a *g, struct nvgpu_mem *mem)
+void nvgpu_dma_free_vid(struct gk20a *g, struct nvgpu_mem *mem)
 {
 #if defined(CONFIG_GK20A_VIDMEM)
         size_t mem_size = mem->size;
@@ -590,27 +442,6 @@ static void nvgpu_dma_free_vid(struct gk20a *g, struct nvgpu_mem *mem)
 #endif
 }
 
-void nvgpu_dma_free(struct gk20a *g, struct nvgpu_mem *mem)
-{
-        switch (mem->aperture) {
-        case APERTURE_SYSMEM:
-                return nvgpu_dma_free_sys(g, mem);
-        case APERTURE_VIDMEM:
-                return nvgpu_dma_free_vid(g, mem);
-        default:
-                break; /* like free() on "null" memory */
-        }
-}
-
-void nvgpu_dma_unmap_free(struct vm_gk20a *vm, struct nvgpu_mem *mem)
-{
-        if (mem->gpu_va)
-                nvgpu_gmmu_unmap(vm, mem, mem->gpu_va);
-        mem->gpu_va = 0;
-
-        nvgpu_dma_free(vm->mm->g, mem);
-}
-
 int nvgpu_get_sgtable_attrs(struct gk20a *g, struct sg_table **sgt,
                       void *cpuva, u64 iova, size_t size, unsigned long flags)
 {
diff --git a/drivers/gpu/nvgpu/os/posix/posix-dma.c b/drivers/gpu/nvgpu/os/posix/posix-dma.c
new file mode 100644
index 00000000..e8c5c9d4
--- /dev/null
+++ b/drivers/gpu/nvgpu/os/posix/posix-dma.c
@@ -0,0 +1,88 @@
+/*
+ * Copyright (c) 2018, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include <stdlib.h>
+
+#include <nvgpu/mm.h>
+#include <nvgpu/vm.h>
+#include <nvgpu/bug.h>
+#include <nvgpu/dma.h>
+#include <nvgpu/gmmu.h>
+#include <nvgpu/nvgpu_mem.h>
+#include <nvgpu/enabled.h>
+
+/*
+ * In userspace vidmem vs sysmem is just a difference in what is placed in the
+ * aperture field.
+ */
+static int __nvgpu_do_dma_alloc(struct gk20a *g, unsigned long flags,
+                                size_t size, struct nvgpu_mem *mem,
+                                enum nvgpu_aperture ap)
+{
+        void *memory = malloc(mem->aligned_size);
+
+        if (memory == NULL)
+                return -ENOMEM;
+
+        mem->cpu_va       = memory;
+        mem->aperture     = ap;
+        mem->size         = size;
+        mem->aligned_size = PAGE_ALIGN(size);
+        mem->gpu_va       = 0ULL;
+        mem->skip_wmb     = true;
+        mem->vidmem_alloc = NULL;
+        mem->allocator    = NULL;
+
+        return 0;
+}
+
+bool nvgpu_iommuable(struct gk20a *g)
+{
+        return false;
+}
+
+int nvgpu_dma_alloc_flags_sys(struct gk20a *g, unsigned long flags,
+                              size_t size, struct nvgpu_mem *mem)
+{
+        return __nvgpu_do_dma_alloc(g, flags, size, mem, APERTURE_SYSMEM);
+}
+
+int nvgpu_dma_alloc_flags_vid_at(struct gk20a *g, unsigned long flags,
+                                 size_t size, struct nvgpu_mem *mem, u64 at)
+{
+        BUG();
+
+        return 0;
+}
+
+void nvgpu_dma_free_sys(struct gk20a *g, struct nvgpu_mem *mem)
+{
+        if (!(mem->mem_flags & NVGPU_MEM_FLAG_SHADOW_COPY))
+                free(mem->cpu_va);
+
+        memset(mem, 0, sizeof(*mem));
+}
+
+void nvgpu_dma_free_vid(struct gk20a *g, struct nvgpu_mem *mem)
+{
+        BUG();
+}