Make the PPGTT code work. Seems to fix the caching issues on Broadwell.

Comments on some of the later Broadwell-related commits in the Linux tree
seem to say that the PPAT flags in for the (global) GTT are simply broken in
the hardware.

4 files changed, 151 insertions, 16 deletions

diff --git a/sys/dev/pci/drm/drm_linux.c b/sys/dev/pci/drm/drm_linux.c
index 97aaa90c0a1..4dcca39fdec 100644
--- a/sys/dev/pci/drm/drm_linux.c
+++ b/sys/dev/pci/drm/drm_linux.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: drm_linux.c,v 1.4 2015/04/08 15:01:33 jsg Exp $	*/
+/*	$OpenBSD: drm_linux.c,v 1.5 2015/09/26 11:17:15 kettenis Exp $	*/
 /*
  * Copyright (c) 2013 Jonathan Gray <jsg@openbsd.org>
  *
@@ -118,6 +118,34 @@ dmi_check_system(const struct dmi_system_id *sysid)
 	return (num);
 }
 
+struct vm_page *
+alloc_pages(unsigned int gfp_mask, unsigned int order)
+{
+	int flags = (gfp_mask & M_NOWAIT) ? UVM_PLA_NOWAIT : UVM_PLA_WAITOK;
+	struct pglist mlist;
+
+	if (gfp_mask & M_CANFAIL)
+		flags |= UVM_PLA_FAILOK;
+
+	TAILQ_INIT(&mlist);
+	if (uvm_pglistalloc(PAGE_SIZE << order, 0, -1, PAGE_SIZE, 0,
+	    &mlist, 1, flags))
+		return NULL;
+	return TAILQ_FIRST(&mlist);
+}
+
+void
+__free_pages(struct vm_page *page, unsigned int order)
+{
+	struct pglist mlist;
+	int i;
+	
+	TAILQ_INIT(&mlist);
+	for (i = 0; i < (1 << order); i++)
+		TAILQ_INSERT_TAIL(&mlist, &page[i], pageq);
+	uvm_pglistfree(&mlist);
+}
+
 void *
 kmap(struct vm_page *pg)
 {
diff --git a/sys/dev/pci/drm/drm_linux.h b/sys/dev/pci/drm/drm_linux.h
index e12188455d8..0819de601c2 100644
--- a/sys/dev/pci/drm/drm_linux.h
+++ b/sys/dev/pci/drm/drm_linux.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: drm_linux.h,v 1.36 2015/09/25 20:27:52 kettenis Exp $	*/
+/*	$OpenBSD: drm_linux.h,v 1.37 2015/09/26 11:17:15 kettenis Exp $	*/
 /*
  * Copyright (c) 2013, 2014, 2015 Mark Kettenis
  *
@@ -934,6 +934,29 @@ typedef enum {
 	PCI_D3cold
 } pci_power_t;
 
+#if defined(__amd64__) || defined(__i386__)
+
+#define PCI_DMA_BIDIRECTIONAL	0
+
+static inline dma_addr_t
+pci_map_page(struct pci_dev *pdev, struct vm_page *page, unsigned long offset, size_t size, int direction)
+{
+	return VM_PAGE_TO_PHYS(page);
+}
+
+static inline void
+pci_unmap_page(struct pci_dev *pdev, dma_addr_t dma_address, size_t size, int direction)
+{
+}
+
+static inline int
+pci_dma_mapping_error(struct pci_dev *pdev, dma_addr_t dma_addr)
+{
+	return 0;
+}
+
+#endif
+
 #define memcpy_toio(d, s, n)	memcpy(d, s, n)
 #define memcpy_fromio(d, s, n)	memcpy(d, s, n)
 #define memset_io(d, b, n)	memset(d, b, n)
@@ -957,6 +980,7 @@ iowrite32(u32 val, volatile void __iomem *addr)
 }
 
 #define readl(p) ioread32(p)
+#define writel(v, p) iowrite32(v, p)
 #define readq(p) ioread64(p)
 
 #define page_to_phys(page)	(VM_PAGE_TO_PHYS(page))
@@ -1096,6 +1120,27 @@ of_machine_is_compatible(const char *model)
 }
 #endif
 
+struct vm_page *alloc_pages(unsigned int, unsigned int);
+void	__free_pages(struct vm_page *, unsigned int);
+
+static inline struct vm_page *
+alloc_page(unsigned int gfp_mask)
+{
+	return alloc_pages(gfp_mask, 0);
+}
+
+static inline void
+__free_page(struct vm_page *page)
+{
+	return __free_pages(page, 0);
+}
+
+static inline unsigned int
+get_order(size_t size)
+{
+	return flsl((size - 1) >> PAGE_SHIFT);
+}
+
 #if defined(__i386__) || defined(__amd64__)
 
 static inline void
diff --git a/sys/dev/pci/drm/i915/i915_drv.h b/sys/dev/pci/drm/i915/i915_drv.h
index 9084c6ee487..da813eef5c8 100644
--- a/sys/dev/pci/drm/i915/i915_drv.h
+++ b/sys/dev/pci/drm/i915/i915_drv.h
@@ -1,4 +1,4 @@
-/* $OpenBSD: i915_drv.h,v 1.67 2015/09/23 23:12:12 kettenis Exp $ */
+/* $OpenBSD: i915_drv.h,v 1.68 2015/09/26 11:17:15 kettenis Exp $ */
 /* i915_drv.h -- Private header for the I915 driver -*- linux-c -*-
  */
 /*
@@ -710,10 +710,10 @@ struct i915_hw_ppgtt {
 	struct i915_address_space base;
 	unsigned num_pd_entries;
 	union {
-		struct page **pt_pages;
-		struct page *gen8_pt_pages;
+		struct vm_page **pt_pages;
+		struct vm_page *gen8_pt_pages;
 	};
-	struct page *pd_pages;
+	struct vm_page *pd_pages;
 	int num_pd_pages;
 	int num_pt_pages;
 	union {
diff --git a/sys/dev/pci/drm/i915/i915_gem_gtt.c b/sys/dev/pci/drm/i915/i915_gem_gtt.c
index 836d20ee5af..537b50f9fb4 100644
--- a/sys/dev/pci/drm/i915/i915_gem_gtt.c
+++ b/sys/dev/pci/drm/i915/i915_gem_gtt.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: i915_gem_gtt.c,v 1.11 2015/09/25 16:15:19 jsg Exp $	*/
+/*	$OpenBSD: i915_gem_gtt.c,v 1.12 2015/09/26 11:17:15 kettenis Exp $	*/
 /*
  * Copyright © 2010 Daniel Vetter
  *
@@ -207,7 +207,6 @@ static gen6_gtt_pte_t iris_pte_encode(dma_addr_t addr,
 	return pte;
 }
 
-#ifdef notyet
 /* Broadwell Page Directory Pointer Descriptors */
 static int gen8_write_pdp(struct intel_ring_buffer *ring, unsigned entry,
 			   uint64_t val)
@@ -279,7 +278,7 @@ static void gen8_ppgtt_clear_range(struct i915_address_space *vm,
 				      I915_CACHE_LLC, use_scratch);
 
 	while (num_entries) {
-		struct page *page_table = &ppgtt->gen8_pt_pages[act_pt];
+		struct vm_page *page_table = &ppgtt->gen8_pt_pages[act_pt];
 
 		last_pte = first_pte + num_entries;
 		if (last_pte > GEN8_PTES_PER_PAGE)
@@ -298,6 +297,7 @@ static void gen8_ppgtt_clear_range(struct i915_address_space *vm,
 	}
 }
 
+#ifdef __linux__
 static void gen8_ppgtt_insert_entries(struct i915_address_space *vm,
 				      struct sg_table *pages,
 				      unsigned first_entry,
@@ -328,6 +328,39 @@ static void gen8_ppgtt_insert_entries(struct i915_address_space *vm,
 	if (pt_vaddr)
 		kunmap_atomic(pt_vaddr);
 }
+#else
+static void gen8_ppgtt_insert_entries(struct i915_address_space *vm,
+				      struct vm_page **pages,
+				      unsigned int num_entries,
+				      unsigned first_entry,
+				      enum i915_cache_level cache_level)
+{
+	struct i915_hw_ppgtt *ppgtt =
+		container_of(vm, struct i915_hw_ppgtt, base);
+	gen8_gtt_pte_t *pt_vaddr;
+	unsigned act_pt = first_entry / GEN8_PTES_PER_PAGE;
+	unsigned act_pte = first_entry % GEN8_PTES_PER_PAGE;
+	int i;
+
+	pt_vaddr = NULL;
+	for (i = 0; i < num_entries; i++) {
+		if (pt_vaddr == NULL)
+			pt_vaddr = kmap_atomic(&ppgtt->gen8_pt_pages[act_pt]);
+
+		pt_vaddr[act_pte] =
+			gen8_pte_encode(VM_PAGE_TO_PHYS(pages[i]),
+					cache_level, true);
+		if (++act_pte == GEN8_PTES_PER_PAGE) {
+			kunmap_atomic(pt_vaddr);
+			pt_vaddr = NULL;
+			act_pt++;
+			act_pte = 0;
+		}
+	}
+	if (pt_vaddr)
+		kunmap_atomic(pt_vaddr);
+}
+#endif
 
 static void gen8_ppgtt_cleanup(struct i915_address_space *vm)
 {
@@ -371,7 +404,7 @@ static void gen8_ppgtt_cleanup(struct i915_address_space *vm)
 static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt, uint64_t size)
 {
 	struct drm_i915_private *dev_priv = ppgtt->base.dev->dev_private;
-	struct page *pt_pages;
+	struct vm_page *pt_pages;
 	int i, j, ret = -ENOMEM;
 	const int max_pdp = DIV_ROUND_UP(size, 1 << 30);
 	const int num_pt_pages = GEN8_PDES_PER_PAGE * max_pdp;
@@ -428,7 +461,7 @@ static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt, uint64_t size)
 			goto err_out;
 
 		for (j = 0; j < GEN8_PDES_PER_PAGE; j++) {
-			struct page *p = &pt_pages[i * GEN8_PDES_PER_PAGE + j];
+			struct vm_page *p = &pt_pages[i * GEN8_PDES_PER_PAGE + j];
 			temp = pci_map_page(ppgtt->base.dev->pdev,
 					    p, 0, PAGE_SIZE,
 					    PCI_DMA_BIDIRECTIONAL);
@@ -584,6 +617,7 @@ static void gen6_ppgtt_clear_range(struct i915_address_space *vm,
 	}
 }
 
+#ifdef __linux__
 static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,
 				      struct sg_table *pages,
 				      unsigned first_entry,
@@ -614,6 +648,39 @@ static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,
 	if (pt_vaddr)
 		kunmap_atomic(pt_vaddr);
 }
+#else
+static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,
+				      struct vm_page **pages,
+				      unsigned int num_entries,
+				      unsigned first_entry,
+				      enum i915_cache_level cache_level)
+{
+	struct i915_hw_ppgtt *ppgtt =
+		container_of(vm, struct i915_hw_ppgtt, base);
+	gen6_gtt_pte_t *pt_vaddr;
+	unsigned act_pt = first_entry / I915_PPGTT_PT_ENTRIES;
+	unsigned act_pte = first_entry % I915_PPGTT_PT_ENTRIES;
+	int i;
+
+	pt_vaddr = NULL;
+	for (i = 0; i < num_entries; i++) {
+		if (pt_vaddr == NULL)
+			pt_vaddr = kmap_atomic(ppgtt->pt_pages[act_pt]);
+
+		pt_vaddr[act_pte] =
+			vm->pte_encode(VM_PAGE_TO_PHYS(pages[i]),
+				       cache_level, true);
+		if (++act_pte == I915_PPGTT_PT_ENTRIES) {
+			kunmap_atomic(pt_vaddr);
+			pt_vaddr = NULL;
+			act_pt++;
+			act_pte = 0;
+		}
+	}
+	if (pt_vaddr)
+		kunmap_atomic(pt_vaddr);
+}
+#endif
 
 static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
 {
@@ -712,11 +779,9 @@ err_pt_alloc:
 
 	return ret;
 }
-#endif
 
 static int i915_gem_init_aliasing_ppgtt(struct drm_device *dev)
 {
-#ifdef notyet
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct i915_hw_ppgtt *ppgtt;
 	int ret;
@@ -743,9 +808,6 @@ static int i915_gem_init_aliasing_ppgtt(struct drm_device *dev)
 	}
 
 	return ret;
-#else
-	return 0;
-#endif
 }
 
 void i915_gem_cleanup_aliasing_ppgtt(struct drm_device *dev)