drm/nvc0: import initial vm backend

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

7 files changed, 169 insertions, 7 deletions

diff --git a/drivers/gpu/drm/nouveau/Makefile b/drivers/gpu/drm/nouveau/Makefile
index b1d8941..e89d895 100644
--- a/drivers/gpu/drm/nouveau/Makefile
+++ b/drivers/gpu/drm/nouveau/Makefile
@@ -28,7 +28,7 @@ nouveau-y := nouveau_drv.o nouveau_state.o nouveau_channel.o nouveau_mem.o \
              nv10_gpio.o nv50_gpio.o \
 	     nv50_calc.o \
 	     nv04_pm.o nv50_pm.o nva3_pm.o \
-	     nv50_vram.o nv50_vm.o
+	     nv50_vram.o nv50_vm.o nvc0_vm.o
 
 nouveau-$(CONFIG_DRM_NOUVEAU_DEBUG) += nouveau_debugfs.o
 nouveau-$(CONFIG_COMPAT) += nouveau_ioc32.o
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 42d1ad6..d17ffea 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -909,8 +909,9 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 			break;
 		}
 
-		ret = nouveau_vm_get(dev_priv->bar1_vm, mem->bus.size, 12,
-				     NV_MEM_ACCESS_RW, &vram->bar_vma);
+		ret = nouveau_vm_get(dev_priv->bar1_vm, mem->bus.size,
+				     vram->page_shift, NV_MEM_ACCESS_RW,
+				     &vram->bar_vma);
 		if (ret)
 			return ret;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 8f13906..57da219 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -69,6 +69,7 @@ struct nouveau_vram {
 	struct drm_device *dev;
 
 	struct nouveau_vma bar_vma;
+	u8  page_shift;
 
 	struct list_head regions;
 	u32 memtype;
diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c
index 2241811..07be1dd 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -731,6 +731,10 @@ nouveau_vram_manager_new(struct ttm_mem_type_manager *man,
 	if (ret)
 		return ret;
 
+	node->page_shift = 12;
+	if (nvbo->vma.node)
+		node->page_shift = nvbo->vma.node->type;
+
 	mem->mm_node = node;
 	mem->start   = node->offset >> PAGE_SHIFT;
 	return 0;
diff --git a/drivers/gpu/drm/nouveau/nouveau_vm.c b/drivers/gpu/drm/nouveau/nouveau_vm.c
index b023a64..97d82ae 100644
--- a/drivers/gpu/drm/nouveau/nouveau_vm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_vm.c
@@ -295,7 +295,34 @@ nouveau_vm_new(struct drm_device *dev, u64 offset, u64 length, u64 mm_offset,
 		vm->flush = nv50_vm_flush;
 		vm->spg_shift = 12;
 		vm->lpg_shift = 16;
+
 		pgt_bits = 29;
+		block = (1 << pgt_bits);
+		if (length < block)
+			block = length;
+
+	} else
+	if (dev_priv->card_type == NV_C0) {
+		vm->map_pgt = nvc0_vm_map_pgt;
+		vm->map = nvc0_vm_map;
+		vm->map_sg = nvc0_vm_map_sg;
+		vm->unmap = nvc0_vm_unmap;
+		vm->flush = nvc0_vm_flush;
+		vm->spg_shift = 12;
+		vm->lpg_shift = 17;
+		pgt_bits = 27;
+
+		/* Should be 4096 everywhere, this is a hack that's
+		 * currently necessary to avoid an elusive bug that
+		 * causes corruption when mixing small/large pages
+		 */
+		if (length < (1ULL << 40))
+			block = 4096;
+		else {
+			block = (1 << pgt_bits);
+			if (length < block)
+				block = length;
+		}
 	} else {
 		kfree(vm);
 		return -ENOSYS;
@@ -314,10 +341,6 @@ nouveau_vm_new(struct drm_device *dev, u64 offset, u64 length, u64 mm_offset,
 	vm->refcount = 1;
 	vm->pgt_bits = pgt_bits - 12;
 
-	block = (1 << pgt_bits);
-	if (length < block)
-		block = length;
-
 	ret = nouveau_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12,
 			      block >> 12);
 	if (ret) {
diff --git a/drivers/gpu/drm/nouveau/nouveau_vm.h b/drivers/gpu/drm/nouveau/nouveau_vm.h
index 105b6f6..e119351 100644
--- a/drivers/gpu/drm/nouveau/nouveau_vm.h
+++ b/drivers/gpu/drm/nouveau/nouveau_vm.h
@@ -100,4 +100,14 @@ void nv50_vm_unmap(struct nouveau_gpuobj *, u32 pte, u32 cnt);
 void nv50_vm_flush(struct nouveau_vm *);
 void nv50_vm_flush_engine(struct drm_device *, int engine);
 
+/* nvc0_vm.c */
+void nvc0_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 pde,
+		     struct nouveau_gpuobj *pgt[2]);
+void nvc0_vm_map(struct nouveau_vma *, struct nouveau_gpuobj *,
+		 struct nouveau_vram *, u32 pte, u32 cnt, u64 phys);
+void nvc0_vm_map_sg(struct nouveau_vma *, struct nouveau_gpuobj *,
+		    u32 pte, dma_addr_t *, u32 cnt);
+void nvc0_vm_unmap(struct nouveau_gpuobj *, u32 pte, u32 cnt);
+void nvc0_vm_flush(struct nouveau_vm *);
+
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvc0_vm.c b/drivers/gpu/drm/nouveau/nvc0_vm.c
new file mode 100644
index 0000000..4b9251b
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvc0_vm.c
@@ -0,0 +1,123 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "drmP.h"
+
+#include "nouveau_drv.h"
+#include "nouveau_vm.h"
+
+void
+nvc0_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 index,
+		struct nouveau_gpuobj *pgt[2])
+{
+	u32 pde[2] = { 0, 0 };
+
+	if (pgt[0])
+		pde[1] = 0x00000001 | (pgt[0]->vinst >> 8);
+	if (pgt[1])
+		pde[0] = 0x00000001 | (pgt[1]->vinst >> 8);
+
+	nv_wo32(pgd, (index * 8) + 0, pde[0]);
+	nv_wo32(pgd, (index * 8) + 4, pde[1]);
+}
+
+static inline u64
+nvc0_vm_addr(struct nouveau_vma *vma, u64 phys, u32 memtype, u32 target)
+{
+	phys >>= 8;
+
+	phys |= 0x00000001; /* present */
+//	if (vma->access & NV_MEM_ACCESS_SYS)
+//		phys |= 0x00000002;
+
+	phys |= ((u64)target  << 32);
+	phys |= ((u64)memtype << 36);
+
+	return phys;
+}
+
+void
+nvc0_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	    struct nouveau_vram *mem, u32 pte, u32 cnt, u64 phys)
+{
+	u32 next = 1 << (vma->node->type - 8);
+
+	phys  = nvc0_vm_addr(vma, phys, mem->memtype, 0);
+	pte <<= 3;
+	while (cnt--) {
+		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
+		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
+		phys += next;
+		pte  += 8;
+	}
+}
+
+void
+nvc0_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       u32 pte, dma_addr_t *list, u32 cnt)
+{
+	pte <<= 3;
+	while (cnt--) {
+		u64 phys = nvc0_vm_addr(vma, *list++, 0, 5);
+		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
+		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
+		pte += 8;
+	}
+}
+
+void
+nvc0_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	pte <<= 3;
+	while (cnt--) {
+		nv_wo32(pgt, pte + 0, 0x00000000);
+		nv_wo32(pgt, pte + 4, 0x00000000);
+		pte += 8;
+	}
+}
+
+void
+nvc0_vm_flush(struct nouveau_vm *vm)
+{
+	struct drm_nouveau_private *dev_priv = vm->dev->dev_private;
+	struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem;
+	struct drm_device *dev = vm->dev;
+	struct nouveau_vm_pgd *vpgd;
+	u32 r100c80, engine;
+
+	pinstmem->flush(vm->dev);
+
+	if (vm == dev_priv->chan_vm)
+		engine = 1;
+	else
+		engine = 5;
+
+	list_for_each_entry(vpgd, &vm->pgd_list, head) {
+		r100c80 = nv_rd32(dev, 0x100c80);
+		nv_wr32(dev, 0x100cb8, vpgd->obj->vinst >> 8);
+		nv_wr32(dev, 0x100cbc, 0x80000000 | engine);
+		if (!nv_wait(dev, 0x100c80, 0xffffffff, r100c80))
+			NV_ERROR(dev, "vm flush timeout eng %d\n", engine);
+	}
+}