drm/gma500: Make SGX MMU driver actually do something

Old MMU code never wrote PDs or PTEs to any registers. Now we do, and
that's a good start.

Signed-off-by: Patrik Jakobsson <patrik.r.jakobsson@gmail.com>

1 files changed, 132 insertions, 126 deletions

diff --git a/drivers/gpu/drm/gma500/mmu.c b/drivers/gpu/drm/gma500/mmu.c
index 49bac41beefb..0bfc9f7b343b 100644
--- a/drivers/gpu/drm/gma500/mmu.c
+++ b/drivers/gpu/drm/gma500/mmu.c
@@ -59,15 +59,14 @@ struct psb_mmu_driver {
 	spinlock_t lock;
 
 	atomic_t needs_tlbflush;
-
-	uint8_t __iomem *register_map;
+	atomic_t *msvdx_mmu_invaldc;
 	struct psb_mmu_pd *default_pd;
-	/*uint32_t bif_ctrl;*/
+	uint32_t bif_ctrl;
 	int has_clflush;
 	int clflush_add;
 	unsigned long clflush_mask;
 
-	struct drm_psb_private *dev_priv;
+	struct drm_device *dev;
 };
 
 struct psb_mmu_pd;
@@ -102,13 +101,13 @@ static inline uint32_t psb_mmu_pd_index(uint32_t offset)
 	return offset >> PSB_PDE_SHIFT;
 }
 
+#if defined(CONFIG_X86)
 static inline void psb_clflush(void *addr)
 {
 	__asm__ __volatile__("clflush (%0)\n" : : "r"(addr) : "memory");
 }
 
-static inline void psb_mmu_clflush(struct psb_mmu_driver *driver,
-				   void *addr)
+static inline void psb_mmu_clflush(struct psb_mmu_driver *driver, void *addr)
 {
 	if (!driver->has_clflush)
 		return;
@@ -117,62 +116,77 @@ static inline void psb_mmu_clflush(struct psb_mmu_driver *driver,
 	psb_clflush(addr);
 	mb();
 }
+#else
 
-static void psb_page_clflush(struct psb_mmu_driver *driver, struct page* page)
-{
-	uint32_t clflush_add = driver->clflush_add >> PAGE_SHIFT;
-	uint32_t clflush_count = PAGE_SIZE / clflush_add;
-	int i;
-	uint8_t *clf;
-
-	clf = kmap_atomic(page);
-	mb();
-	for (i = 0; i < clflush_count; ++i) {
-		psb_clflush(clf);
-		clf += clflush_add;
-	}
-	mb();
-	kunmap_atomic(clf);
+static inline void psb_mmu_clflush(struct psb_mmu_driver *driver, void *addr)
+{;
 }
 
-static void psb_pages_clflush(struct psb_mmu_driver *driver,
-				struct page *page[], unsigned long num_pages)
-{
-	int i;
-
-	if (!driver->has_clflush)
-		return ;
-
-	for (i = 0; i < num_pages; i++)
-		psb_page_clflush(driver, *page++);
-}
+#endif
 
-static void psb_mmu_flush_pd_locked(struct psb_mmu_driver *driver,
-				    int force)
+static void psb_mmu_flush_pd_locked(struct psb_mmu_driver *driver, int force)
 {
+	struct drm_device *dev = driver->dev;
+	struct drm_psb_private *dev_priv = dev->dev_private;
+
+	if (atomic_read(&driver->needs_tlbflush) || force) {
+		uint32_t val = PSB_RSGX32(PSB_CR_BIF_CTRL);
+		PSB_WSGX32(val | _PSB_CB_CTRL_INVALDC, PSB_CR_BIF_CTRL);
+
+		/* Make sure data cache is turned off before enabling it */
+		wmb();
+		PSB_WSGX32(val & ~_PSB_CB_CTRL_INVALDC, PSB_CR_BIF_CTRL);
+		(void)PSB_RSGX32(PSB_CR_BIF_CTRL);
+		if (driver->msvdx_mmu_invaldc)
+			atomic_set(driver->msvdx_mmu_invaldc, 1);
+	}
 	atomic_set(&driver->needs_tlbflush, 0);
 }
 
+#if 0
 static void psb_mmu_flush_pd(struct psb_mmu_driver *driver, int force)
 {
 	down_write(&driver->sem);
 	psb_mmu_flush_pd_locked(driver, force);
 	up_write(&driver->sem);
 }
+#endif
 
-void psb_mmu_flush(struct psb_mmu_driver *driver, int rc_prot)
+void psb_mmu_flush(struct psb_mmu_driver *driver)
 {
-	if (rc_prot)
-		down_write(&driver->sem);
-	if (rc_prot)
-		up_write(&driver->sem);
+	struct drm_device *dev = driver->dev;
+	struct drm_psb_private *dev_priv = dev->dev_private;
+	uint32_t val;
+
+	down_write(&driver->sem);
+	val = PSB_RSGX32(PSB_CR_BIF_CTRL);
+	if (atomic_read(&driver->needs_tlbflush))
+		PSB_WSGX32(val | _PSB_CB_CTRL_INVALDC, PSB_CR_BIF_CTRL);
+	else
+		PSB_WSGX32(val | _PSB_CB_CTRL_FLUSH, PSB_CR_BIF_CTRL);
+
+	/* Make sure data cache is turned off and MMU is flushed before
+	   restoring bank interface control register */
+	wmb();
+	PSB_WSGX32(val & ~(_PSB_CB_CTRL_FLUSH | _PSB_CB_CTRL_INVALDC),
+		   PSB_CR_BIF_CTRL);
+	(void)PSB_RSGX32(PSB_CR_BIF_CTRL);
+
+	atomic_set(&driver->needs_tlbflush, 0);
+	if (driver->msvdx_mmu_invaldc)
+		atomic_set(driver->msvdx_mmu_invaldc, 1);
+	up_write(&driver->sem);
 }
 
 void psb_mmu_set_pd_context(struct psb_mmu_pd *pd, int hw_context)
 {
-	/*ttm_tt_cache_flush(&pd->p, 1);*/
-	psb_pages_clflush(pd->driver, &pd->p, 1);
+	struct drm_device *dev = pd->driver->dev;
+	struct drm_psb_private *dev_priv = dev->dev_private;
+	uint32_t offset = (hw_context == 0) ? PSB_CR_BIF_DIR_LIST_BASE0 :
+			  PSB_CR_BIF_DIR_LIST_BASE1 + hw_context * 4;
+
 	down_write(&pd->driver->sem);
+	PSB_WSGX32(page_to_pfn(pd->p) << PAGE_SHIFT, offset);
 	wmb();
 	psb_mmu_flush_pd_locked(pd->driver, 1);
 	pd->hw_context = hw_context;
@@ -183,7 +197,6 @@ void psb_mmu_set_pd_context(struct psb_mmu_pd *pd, int hw_context)
 static inline unsigned long psb_pd_addr_end(unsigned long addr,
 					    unsigned long end)
 {
-
 	addr = (addr + PSB_PDE_MASK + 1) & ~PSB_PDE_MASK;
 	return (addr < end) ? addr : end;
 }
@@ -223,12 +236,10 @@ struct psb_mmu_pd *psb_mmu_alloc_pd(struct psb_mmu_driver *driver,
 		goto out_err3;
 
 	if (!trap_pagefaults) {
-		pd->invalid_pde =
-		    psb_mmu_mask_pte(page_to_pfn(pd->dummy_pt),
-				     invalid_type);
-		pd->invalid_pte =
-		    psb_mmu_mask_pte(page_to_pfn(pd->dummy_page),
-				     invalid_type);
+		pd->invalid_pde = psb_mmu_mask_pte(page_to_pfn(pd->dummy_pt),
+						   invalid_type);
+		pd->invalid_pte = psb_mmu_mask_pte(page_to_pfn(pd->dummy_page),
+						   invalid_type);
 	} else {
 		pd->invalid_pde = 0;
 		pd->invalid_pte = 0;
@@ -279,12 +290,16 @@ static void psb_mmu_free_pt(struct psb_mmu_pt *pt)
 void psb_mmu_free_pagedir(struct psb_mmu_pd *pd)
 {
 	struct psb_mmu_driver *driver = pd->driver;
+	struct drm_device *dev = driver->dev;
+	struct drm_psb_private *dev_priv = dev->dev_private;
 	struct psb_mmu_pt *pt;
 	int i;
 
 	down_write(&driver->sem);
-	if (pd->hw_context != -1)
+	if (pd->hw_context != -1) {
+		PSB_WSGX32(0, PSB_CR_BIF_DIR_LIST_BASE0 + pd->hw_context * 4);
 		psb_mmu_flush_pd_locked(driver, 1);
+	}
 
 	/* Should take the spinlock here, but we don't need to do that
 	   since we have the semaphore in write mode. */
@@ -331,7 +346,7 @@ static struct psb_mmu_pt *psb_mmu_alloc_pt(struct psb_mmu_pd *pd)
 	for (i = 0; i < (PAGE_SIZE / sizeof(uint32_t)); ++i)
 		*ptes++ = pd->invalid_pte;
 
-
+#if defined(CONFIG_X86)
 	if (pd->driver->has_clflush && pd->hw_context != -1) {
 		mb();
 		for (i = 0; i < clflush_count; ++i) {
@@ -340,7 +355,7 @@ static struct psb_mmu_pt *psb_mmu_alloc_pt(struct psb_mmu_pd *pd)
 		}
 		mb();
 	}
-
+#endif
 	kunmap_atomic(v);
 	spin_unlock(lock);
 
@@ -351,7 +366,7 @@ static struct psb_mmu_pt *psb_mmu_alloc_pt(struct psb_mmu_pd *pd)
 	return pt;
 }
 
-static struct psb_mmu_pt *psb_mmu_pt_alloc_map_lock(struct psb_mmu_pd *pd,
+struct psb_mmu_pt *psb_mmu_pt_alloc_map_lock(struct psb_mmu_pd *pd,
 					     unsigned long addr)
 {
 	uint32_t index = psb_mmu_pd_index(addr);
@@ -383,7 +398,7 @@ static struct psb_mmu_pt *psb_mmu_pt_alloc_map_lock(struct psb_mmu_pd *pd,
 		kunmap_atomic((void *) v);
 
 		if (pd->hw_context != -1) {
-			psb_mmu_clflush(pd->driver, (void *) &v[index]);
+			psb_mmu_clflush(pd->driver, (void *)&v[index]);
 			atomic_set(&pd->driver->needs_tlbflush, 1);
 		}
 	}
@@ -420,8 +435,7 @@ static void psb_mmu_pt_unmap_unlock(struct psb_mmu_pt *pt)
 		pd->tables[pt->index] = NULL;
 
 		if (pd->hw_context != -1) {
-			psb_mmu_clflush(pd->driver,
-					(void *) &v[pt->index]);
+			psb_mmu_clflush(pd->driver, (void *)&v[pt->index]);
 			atomic_set(&pd->driver->needs_tlbflush, 1);
 		}
 		kunmap_atomic(pt->v);
@@ -432,8 +446,8 @@ static void psb_mmu_pt_unmap_unlock(struct psb_mmu_pt *pt)
 	spin_unlock(&pd->driver->lock);
 }
 
-static inline void psb_mmu_set_pte(struct psb_mmu_pt *pt,
-				   unsigned long addr, uint32_t pte)
+static inline void psb_mmu_set_pte(struct psb_mmu_pt *pt, unsigned long addr,
+				   uint32_t pte)
 {
 	pt->v[psb_mmu_pt_index(addr)] = pte;
 }
@@ -444,69 +458,50 @@ static inline void psb_mmu_invalidate_pte(struct psb_mmu_pt *pt,
 	pt->v[psb_mmu_pt_index(addr)] = pt->pd->invalid_pte;
 }
 
-
-void psb_mmu_mirror_gtt(struct psb_mmu_pd *pd,
-			uint32_t mmu_offset, uint32_t gtt_start,
-			uint32_t gtt_pages)
+struct psb_mmu_pd *psb_mmu_get_default_pd(struct psb_mmu_driver *driver)
 {
-	uint32_t *v;
-	uint32_t start = psb_mmu_pd_index(mmu_offset);
-	struct psb_mmu_driver *driver = pd->driver;
-	int num_pages = gtt_pages;
+	struct psb_mmu_pd *pd;
 
 	down_read(&driver->sem);
-	spin_lock(&driver->lock);
-
-	v = kmap_atomic(pd->p);
-	v += start;
-
-	while (gtt_pages--) {
-		*v++ = gtt_start | pd->pd_mask;
-		gtt_start += PAGE_SIZE;
-	}
-
-	/*ttm_tt_cache_flush(&pd->p, num_pages);*/
-	psb_pages_clflush(pd->driver, &pd->p, num_pages);
-	kunmap_atomic(v);
-	spin_unlock(&driver->lock);
-
-	if (pd->hw_context != -1)
-		atomic_set(&pd->driver->needs_tlbflush, 1);
+	pd = driver->default_pd;
+	up_read(&driver->sem);
 
-	up_read(&pd->driver->sem);
-	psb_mmu_flush_pd(pd->driver, 0);
+	return pd;
 }
 
-struct psb_mmu_pd *psb_mmu_get_default_pd(struct psb_mmu_driver *driver)
+/* Returns the physical address of the PD shared by sgx/msvdx */
+uint32_t psb_get_default_pd_addr(struct psb_mmu_driver *driver)
 {
 	struct psb_mmu_pd *pd;
 
-	/* down_read(&driver->sem); */
-	pd = driver->default_pd;
-	/* up_read(&driver->sem); */
-
-	return pd;
+	pd = psb_mmu_get_default_pd(driver);
+	return page_to_pfn(pd->p) << PAGE_SHIFT;
 }
 
 void psb_mmu_driver_takedown(struct psb_mmu_driver *driver)
 {
+	struct drm_device *dev = driver->dev;
+	struct drm_psb_private *dev_priv = dev->dev_private;
+
+	PSB_WSGX32(driver->bif_ctrl, PSB_CR_BIF_CTRL);
 	psb_mmu_free_pagedir(driver->default_pd);
 	kfree(driver);
 }
 
-struct psb_mmu_driver *psb_mmu_driver_init(uint8_t __iomem * registers,
-					int trap_pagefaults,
-					int invalid_type,
-					struct drm_psb_private *dev_priv)
+struct psb_mmu_driver *psb_mmu_driver_init(struct drm_device *dev,
+					   int trap_pagefaults,
+					   int invalid_type,
+					   atomic_t *msvdx_mmu_invaldc)
 {
 	struct psb_mmu_driver *driver;
+	struct drm_psb_private *dev_priv = dev->dev_private;
 
 	driver = kmalloc(sizeof(*driver), GFP_KERNEL);
 
 	if (!driver)
 		return NULL;
-	driver->dev_priv = dev_priv;
 
+	driver->dev = dev;
 	driver->default_pd = psb_mmu_alloc_pd(driver, trap_pagefaults,
 					      invalid_type);
 	if (!driver->default_pd)
@@ -515,17 +510,24 @@ struct psb_mmu_driver *psb_mmu_driver_init(uint8_t __iomem * registers,
 	spin_lock_init(&driver->lock);
 	init_rwsem(&driver->sem);
 	down_write(&driver->sem);
-	driver->register_map = registers;
 	atomic_set(&driver->needs_tlbflush, 1);
+	driver->msvdx_mmu_invaldc = msvdx_mmu_invaldc;
+
+	driver->bif_ctrl = PSB_RSGX32(PSB_CR_BIF_CTRL);
+	PSB_WSGX32(driver->bif_ctrl | _PSB_CB_CTRL_CLEAR_FAULT,
+		   PSB_CR_BIF_CTRL);
+	PSB_WSGX32(driver->bif_ctrl & ~_PSB_CB_CTRL_CLEAR_FAULT,
+		   PSB_CR_BIF_CTRL);
 
 	driver->has_clflush = 0;
 
+#if defined(CONFIG_X86)
 	if (boot_cpu_has(X86_FEATURE_CLFLSH)) {
 		uint32_t tfms, misc, cap0, cap4, clflush_size;
 
 		/*
-		 * clflush size is determined at kernel setup for x86_64
-		 *  but not for i386. We have to do it here.
+		 * clflush size is determined at kernel setup for x86_64 but not
+		 * for i386. We have to do it here.
 		 */
 
 		cpuid(0x00000001, &tfms, &misc, &cap0, &cap4);
@@ -536,6 +538,7 @@ struct psb_mmu_driver *psb_mmu_driver_init(uint8_t __iomem * registers,
 		driver->clflush_mask = driver->clflush_add - 1;
 		driver->clflush_mask = ~driver->clflush_mask;
 	}
+#endif
 
 	up_write(&driver->sem);
 	return driver;
@@ -545,9 +548,9 @@ out_err1:
 	return NULL;
 }
 
-static void psb_mmu_flush_ptes(struct psb_mmu_pd *pd,
-			       unsigned long address, uint32_t num_pages,
-			       uint32_t desired_tile_stride,
+#if defined(CONFIG_X86)
+static void psb_mmu_flush_ptes(struct psb_mmu_pd *pd, unsigned long address,
+			       uint32_t num_pages, uint32_t desired_tile_stride,
 			       uint32_t hw_tile_stride)
 {
 	struct psb_mmu_pt *pt;
@@ -561,11 +564,8 @@ static void psb_mmu_flush_ptes(struct psb_mmu_pd *pd,
 	unsigned long clflush_add = pd->driver->clflush_add;
 	unsigned long clflush_mask = pd->driver->clflush_mask;
 
-	if (!pd->driver->has_clflush) {
-		/*ttm_tt_cache_flush(&pd->p, num_pages);*/
-		psb_pages_clflush(pd->driver, &pd->p, num_pages);
+	if (!pd->driver->has_clflush)
 		return;
-	}
 
 	if (hw_tile_stride)
 		rows = num_pages / desired_tile_stride;
@@ -586,10 +586,8 @@ static void psb_mmu_flush_ptes(struct psb_mmu_pd *pd,
 			if (!pt)
 				continue;
 			do {
-				psb_clflush(&pt->v
-					    [psb_mmu_pt_index(addr)]);
-			} while (addr +=
-				 clflush_add,
+				psb_clflush(&pt->v[psb_mmu_pt_index(addr)]);
+			} while (addr += clflush_add,
 				 (addr & clflush_mask) < next);
 
 			psb_mmu_pt_unmap_unlock(pt);
@@ -598,6 +596,14 @@ static void psb_mmu_flush_ptes(struct psb_mmu_pd *pd,
 	}
 	mb();
 }
+#else
+static void psb_mmu_flush_ptes(struct psb_mmu_pd *pd, unsigned long address,
+			       uint32_t num_pages, uint32_t desired_tile_stride,
+			       uint32_t hw_tile_stride)
+{
+	drm_ttm_cache_flush();
+}
+#endif
 
 void psb_mmu_remove_pfn_sequence(struct psb_mmu_pd *pd,
 				 unsigned long address, uint32_t num_pages)
@@ -633,7 +639,7 @@ out:
 	up_read(&pd->driver->sem);
 
 	if (pd->hw_context != -1)
-		psb_mmu_flush(pd->driver, 0);
+		psb_mmu_flush(pd->driver);
 
 	return;
 }
@@ -660,7 +666,7 @@ void psb_mmu_remove_pages(struct psb_mmu_pd *pd, unsigned long address,
 	add = desired_tile_stride << PAGE_SHIFT;
 	row_add = hw_tile_stride << PAGE_SHIFT;
 
-	/* down_read(&pd->driver->sem); */
+	down_read(&pd->driver->sem);
 
 	/* Make sure we only need to flush this processor's cache */
 
@@ -688,10 +694,10 @@ void psb_mmu_remove_pages(struct psb_mmu_pd *pd, unsigned long address,
 		psb_mmu_flush_ptes(pd, f_address, num_pages,
 				   desired_tile_stride, hw_tile_stride);
 
-	/* up_read(&pd->driver->sem); */
+	up_read(&pd->driver->sem);
 
 	if (pd->hw_context != -1)
-		psb_mmu_flush(pd->driver, 0);
+		psb_mmu_flush(pd->driver);
 }
 
 int psb_mmu_insert_pfn_sequence(struct psb_mmu_pd *pd, uint32_t start_pfn,
@@ -704,7 +710,7 @@ int psb_mmu_insert_pfn_sequence(struct psb_mmu_pd *pd, uint32_t start_pfn,
 	unsigned long end;
 	unsigned long next;
 	unsigned long f_address = address;
-	int ret = 0;
+	int ret = -ENOMEM;
 
 	down_read(&pd->driver->sem);
 
@@ -726,6 +732,7 @@ int psb_mmu_insert_pfn_sequence(struct psb_mmu_pd *pd, uint32_t start_pfn,
 		psb_mmu_pt_unmap_unlock(pt);
 
 	} while (addr = next, next != end);
+	ret = 0;
 
 out:
 	if (pd->hw_context != -1)
@@ -734,15 +741,15 @@ out:
 	up_read(&pd->driver->sem);
 
 	if (pd->hw_context != -1)
-		psb_mmu_flush(pd->driver, 1);
+		psb_mmu_flush(pd->driver);
 
-	return ret;
+	return 0;
 }
 
 int psb_mmu_insert_pages(struct psb_mmu_pd *pd, struct page **pages,
 			 unsigned long address, uint32_t num_pages,
-			 uint32_t desired_tile_stride,
-			 uint32_t hw_tile_stride, int type)
+			 uint32_t desired_tile_stride, uint32_t hw_tile_stride,
+			 int type)
 {
 	struct psb_mmu_pt *pt;
 	uint32_t rows = 1;
@@ -754,7 +761,7 @@ int psb_mmu_insert_pages(struct psb_mmu_pd *pd, struct page **pages,
 	unsigned long add;
 	unsigned long row_add;
 	unsigned long f_address = address;
-	int ret = 0;
+	int ret = -ENOMEM;
 
 	if (hw_tile_stride) {
 		if (num_pages % desired_tile_stride != 0)
@@ -777,14 +784,11 @@ int psb_mmu_insert_pages(struct psb_mmu_pd *pd, struct page **pages,
 		do {
 			next = psb_pd_addr_end(addr, end);
 			pt = psb_mmu_pt_alloc_map_lock(pd, addr);
-			if (!pt) {
-				ret = -ENOMEM;
+			if (!pt)
 				goto out;
-			}
 			do {
-				pte =
-				    psb_mmu_mask_pte(page_to_pfn(*pages++),
-						     type);
+				pte = psb_mmu_mask_pte(page_to_pfn(*pages++),
+						       type);
 				psb_mmu_set_pte(pt, addr, pte);
 				pt->count++;
 			} while (addr += PAGE_SIZE, addr < next);
@@ -794,6 +798,8 @@ int psb_mmu_insert_pages(struct psb_mmu_pd *pd, struct page **pages,
 
 		address += row_add;
 	}
+
+	ret = 0;
 out:
 	if (pd->hw_context != -1)
 		psb_mmu_flush_ptes(pd, f_address, num_pages,
@@ -802,7 +808,7 @@ out:
 	up_read(&pd->driver->sem);
 
 	if (pd->hw_context != -1)
-		psb_mmu_flush(pd->driver, 1);
+		psb_mmu_flush(pd->driver);
 
 	return ret;
 }