drm/nouveau/instmem/gk20a: move memory allocation to instmem
authorAlexandre Courbot <acourbot@nvidia.com>
Fri, 20 Feb 2015 09:23:00 +0000 (18:23 +0900)
committerBen Skeggs <bskeggs@redhat.com>
Tue, 14 Apr 2015 07:00:42 +0000 (17:00 +1000)
GK20A does not have dedicated RAM, thus having a RAM device for it does
not make sense. Move the contiguous physical memory allocation to
instmem.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
drivers/gpu/drm/nouveau/include/nvkm/subdev/instmem.h
drivers/gpu/drm/nouveau/nvkm/engine/device/gk104.c
drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgk20a.c
drivers/gpu/drm/nouveau/nvkm/subdev/instmem/Kbuild
drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c [new file with mode: 0644]

index d104c1aac807f1cd37e7dbd92bf5c073c7f80106..1bcb763cfca0f7b5f8559114bd8c186ef4f15974 100644 (file)
@@ -45,4 +45,5 @@ nvkm_instmem(void *obj)
 extern struct nvkm_oclass *nv04_instmem_oclass;
 extern struct nvkm_oclass *nv40_instmem_oclass;
 extern struct nvkm_oclass *nv50_instmem_oclass;
+extern struct nvkm_oclass *gk20a_instmem_oclass;
 #endif
index bf5893458a477792b1da4b1100cdb022e7bebc81..8f266a9a34a68ddc68a52f02f0d17bde1e2294be 100644 (file)
@@ -171,7 +171,7 @@ gk104_identify(struct nvkm_device *device)
                device->oclass[NVDEV_SUBDEV_FB     ] =  gk20a_fb_oclass;
                device->oclass[NVDEV_SUBDEV_LTC    ] =  gk104_ltc_oclass;
                device->oclass[NVDEV_SUBDEV_IBUS   ] = &gk20a_ibus_oclass;
-               device->oclass[NVDEV_SUBDEV_INSTMEM] = nv50_instmem_oclass;
+               device->oclass[NVDEV_SUBDEV_INSTMEM] = gk20a_instmem_oclass;
                device->oclass[NVDEV_SUBDEV_MMU    ] = &gf100_mmu_oclass;
                device->oclass[NVDEV_SUBDEV_BAR    ] = &gk20a_bar_oclass;
                device->oclass[NVDEV_ENGINE_DMAOBJ ] =  gf110_dmaeng_oclass;
index 5f30db140b473208e8f579d4b1f9ba3ca184b45d..60d8e1cead6129cccb8db32f23d289a7e0c74d6b 100644 (file)
 
 #include <core/device.h>
 
-struct gk20a_mem {
-       struct nvkm_mem base;
-       void *cpuaddr;
-       dma_addr_t handle;
-};
-#define to_gk20a_mem(m) container_of(m, struct gk20a_mem, base)
-
 static void
 gk20a_ram_put(struct nvkm_fb *pfb, struct nvkm_mem **pmem)
 {
-       struct device *dev = nv_device_base(nv_device(pfb));
-       struct gk20a_mem *mem = to_gk20a_mem(*pmem);
-
-       *pmem = NULL;
-       if (unlikely(mem == NULL))
-               return;
-
-       if (likely(mem->cpuaddr))
-               dma_free_coherent(dev, mem->base.size << PAGE_SHIFT,
-                                 mem->cpuaddr, mem->handle);
-
-       kfree(mem->base.pages);
-       kfree(mem);
+       BUG();
 }
 
 static int
 gk20a_ram_get(struct nvkm_fb *pfb, u64 size, u32 align, u32 ncmin,
             u32 memtype, struct nvkm_mem **pmem)
 {
-       struct device *dev = nv_device_base(nv_device(pfb));
-       struct gk20a_mem *mem;
-       u32 type = memtype & 0xff;
-       u32 npages, order;
-       int i;
-
-       nv_debug(pfb, "%s: size: %llx align: %x, ncmin: %x\n", __func__, size,
-                align, ncmin);
-
-       npages = size >> PAGE_SHIFT;
-       if (npages == 0)
-               npages = 1;
-
-       if (align == 0)
-               align = PAGE_SIZE;
-       align >>= PAGE_SHIFT;
-
-       /* round alignment to the next power of 2, if needed */
-       order = fls(align);
-       if ((align & (align - 1)) == 0)
-               order--;
-       align = BIT(order);
-
-       /* ensure returned address is correctly aligned */
-       npages = max(align, npages);
-
-       mem = kzalloc(sizeof(*mem), GFP_KERNEL);
-       if (!mem)
-               return -ENOMEM;
-
-       mem->base.size = npages;
-       mem->base.memtype = type;
-
-       mem->base.pages = kzalloc(sizeof(dma_addr_t) * npages, GFP_KERNEL);
-       if (!mem->base.pages) {
-               kfree(mem);
-               return -ENOMEM;
-       }
-
-       *pmem = &mem->base;
-
-       mem->cpuaddr = dma_alloc_coherent(dev, npages << PAGE_SHIFT,
-                                         &mem->handle, GFP_KERNEL);
-       if (!mem->cpuaddr) {
-               nv_error(pfb, "%s: cannot allocate memory!\n", __func__);
-               gk20a_ram_put(pfb, pmem);
-               return -ENOMEM;
-       }
-
-       align <<= PAGE_SHIFT;
-
-       /* alignment check */
-       if (unlikely(mem->handle & (align - 1)))
-               nv_warn(pfb, "memory not aligned as requested: %pad (0x%x)\n",
-                       &mem->handle, align);
-
-       nv_debug(pfb, "alloc size: 0x%x, align: 0x%x, paddr: %pad, vaddr: %p\n",
-                npages << PAGE_SHIFT, align, &mem->handle, mem->cpuaddr);
-
-       for (i = 0; i < npages; i++)
-               mem->base.pages[i] = mem->handle + (PAGE_SIZE * i);
-
-       mem->base.offset = (u64)mem->base.pages[0];
-       return 0;
+       BUG();
 }
 
 static int
index e6f35abe787973499cd212fc1ee40b40ddc40344..13bb7fc0a569c9c217a3ca5711f49896b176fde0 100644 (file)
@@ -2,3 +2,4 @@ nvkm-y += nvkm/subdev/instmem/base.o
 nvkm-y += nvkm/subdev/instmem/nv04.o
 nvkm-y += nvkm/subdev/instmem/nv40.o
 nvkm-y += nvkm/subdev/instmem/nv50.o
+nvkm-y += nvkm/subdev/instmem/gk20a.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gk20a.c
new file mode 100644 (file)
index 0000000..9a1fc84
--- /dev/null
@@ -0,0 +1,211 @@
+/*
+ * Copyright (c) 2015, 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 <subdev/fb.h>
+#include <core/mm.h>
+#include <core/device.h>
+
+#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;
+       void *cpuaddr;
+       dma_addr_t handle;
+       struct nvkm_mm_node r;
+};
+
+struct gk20a_instmem_priv {
+       struct nvkm_instmem base;
+       spinlock_t lock;
+       u64 addr;
+};
+
+static u32
+gk20a_instobj_rd32(struct nvkm_object *object, 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;
+       u32 data;
+
+       spin_lock_irqsave(&priv->lock, flags);
+       if (unlikely(priv->addr != base)) {
+               nv_wr32(priv, 0x001700, base >> 16);
+               priv->addr = base;
+       }
+       data = nv_rd32(priv, 0x700000 + addr);
+       spin_unlock_irqrestore(&priv->lock, flags);
+       return data;
+}
+
+static void
+gk20a_instobj_wr32(struct nvkm_object *object, 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;
+       }
+       nv_wr32(priv, 0x700000 + addr, data);
+       spin_unlock_irqrestore(&priv->lock, flags);
+}
+
+static void
+gk20a_instobj_dtor(struct nvkm_object *object)
+{
+       struct gk20a_instobj_priv *node = (void *)object;
+       struct gk20a_instmem_priv *priv = (void *)nvkm_instmem(node);
+       struct device *dev = nv_device_base(nv_device(priv));
+
+       if (unlikely(!node->handle))
+               return;
+
+       dma_free_coherent(dev, node->mem->size << PAGE_SHIFT, node->cpuaddr,
+                         node->handle);
+
+       nvkm_instobj_destroy(&node->base);
+}
+
+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)
+{
+       struct nvkm_instobj_args *args = data;
+       struct gk20a_instmem_priv *priv = (void *)nvkm_instmem(parent);
+       struct device *dev = nv_device_base(nv_device(priv));
+       struct gk20a_instobj_priv *node;
+       u32 size, align;
+       u32 npages;
+       int ret;
+
+       nv_debug(parent, "%s: size: %x align: %x\n", __func__,
+                args->size, args->align);
+
+       size  = max((args->size  + 4095) & ~4095, (u32)4096);
+       align = max((args->align + 4095) & ~4095, (u32)4096);
+
+       npages = size >> PAGE_SHIFT;
+
+       ret = nvkm_instobj_create_(parent, engine, oclass, sizeof(*node),
+                                     (void **)&node);
+       *pobject = nv_object(node);
+       if (ret)
+               return ret;
+
+       node->mem = &node->_mem;
+
+       node->cpuaddr = dma_alloc_coherent(dev, npages << PAGE_SHIFT,
+                                          &node->handle, GFP_KERNEL);
+       if (!node->cpuaddr) {
+               nv_error(priv, "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);
+
+       node->mem->offset = node->handle;
+       node->mem->size = size >> 12;
+       node->mem->memtype = 0;
+       node->mem->page_shift = 12;
+       INIT_LIST_HEAD(&node->mem->regions);
+
+       node->r.type = 12;
+       node->r.offset = node->handle >> 12;
+       node->r.length = npages;
+       list_add_tail(&node->r.rl_entry, &node->mem->regions);
+
+       node->base.addr = node->mem->offset;
+       node->base.size = size;
+
+       nv_debug(parent, "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)
+{
+       struct gk20a_instmem_priv *priv = (void *)object;
+       priv->addr = ~0ULL;
+       return nvkm_instmem_fini(&priv->base, suspend);
+}
+
+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;
+       int ret;
+
+       ret = nvkm_instmem_create(parent, engine, oclass, &priv);
+       *pobject = nv_object(priv);
+       if (ret)
+               return ret;
+
+       spin_lock_init(&priv->lock);
+
+       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;