1 files changed, 210 insertions, 80 deletions

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index 01feb48af33..6ccd993e26b 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -26,8 +26,72 @@
  **************************************************************************/
 
 #include "vmwgfx_drv.h"
-#include "drmP.h"
-#include "ttm/ttm_placement.h"
+#include <drm/drmP.h>
+#include <drm/ttm/ttm_placement.h>
+
+bool vmw_fifo_have_3d(struct vmw_private *dev_priv)
+{
+	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	uint32_t fifo_min, hwversion;
+	const struct vmw_fifo_state *fifo = &dev_priv->fifo;
+
+	if (!(dev_priv->capabilities & SVGA_CAP_3D))
+		return false;
+
+	if (dev_priv->capabilities & SVGA_CAP_GBOBJECTS) {
+		uint32_t result;
+
+		if (!dev_priv->has_mob)
+			return false;
+
+		mutex_lock(&dev_priv->hw_mutex);
+		vmw_write(dev_priv, SVGA_REG_DEV_CAP, SVGA3D_DEVCAP_3D);
+		result = vmw_read(dev_priv, SVGA_REG_DEV_CAP);
+		mutex_unlock(&dev_priv->hw_mutex);
+
+		return (result != 0);
+	}
+
+	if (!(dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO))
+		return false;
+
+	fifo_min = ioread32(fifo_mem  + SVGA_FIFO_MIN);
+	if (fifo_min <= SVGA_FIFO_3D_HWVERSION * sizeof(unsigned int))
+		return false;
+
+	hwversion = ioread32(fifo_mem +
+			     ((fifo->capabilities &
+			       SVGA_FIFO_CAP_3D_HWVERSION_REVISED) ?
+			      SVGA_FIFO_3D_HWVERSION_REVISED :
+			      SVGA_FIFO_3D_HWVERSION));
+
+	if (hwversion == 0)
+		return false;
+
+	if (hwversion < SVGA3D_HWVERSION_WS8_B1)
+		return false;
+
+	/* Non-Screen Object path does not support surfaces */
+	if (!dev_priv->sou_priv)
+		return false;
+
+	return true;
+}
+
+bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv)
+{
+	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+	uint32_t caps;
+
+	if (!(dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO))
+		return false;
+
+	caps = ioread32(fifo_mem + SVGA_FIFO_CAPABILITIES);
+	if (caps & SVGA_FIFO_CAP_PITCHLOCK)
+		return true;
+
+	return false;
+}
 
 int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 {
@@ -35,26 +99,17 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 	uint32_t max;
 	uint32_t min;
 	uint32_t dummy;
-	int ret;
 
 	fifo->static_buffer_size = VMWGFX_FIFO_STATIC_SIZE;
 	fifo->static_buffer = vmalloc(fifo->static_buffer_size);
 	if (unlikely(fifo->static_buffer == NULL))
 		return -ENOMEM;
 
-	fifo->last_buffer_size = VMWGFX_FIFO_STATIC_SIZE;
-	fifo->last_data_size = 0;
-	fifo->last_buffer_add = false;
-	fifo->last_buffer = vmalloc(fifo->last_buffer_size);
-	if (unlikely(fifo->last_buffer == NULL)) {
-		ret = -ENOMEM;
-		goto out_err;
-	}
-
 	fifo->dynamic_buffer = NULL;
 	fifo->reserved_size = 0;
 	fifo->using_bounce_buffer = false;
 
+	mutex_init(&fifo->fifo_mutex);
 	init_rwsem(&fifo->rwsem);
 
 	/*
@@ -68,6 +123,7 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 	mutex_lock(&dev_priv->hw_mutex);
 	dev_priv->enable_state = vmw_read(dev_priv, SVGA_REG_ENABLE);
 	dev_priv->config_done_state = vmw_read(dev_priv, SVGA_REG_CONFIG_DONE);
+	dev_priv->traces_state = vmw_read(dev_priv, SVGA_REG_TRACES);
 	vmw_write(dev_priv, SVGA_REG_ENABLE, 1);
 
 	min = 4;
@@ -98,15 +154,10 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 		 (unsigned int) min,
 		 (unsigned int) fifo->capabilities);
 
-	dev_priv->fence_seq = (uint32_t) -100;
-	dev_priv->last_read_sequence = (uint32_t) -100;
-	iowrite32(dev_priv->last_read_sequence, fifo_mem + SVGA_FIFO_FENCE);
-
+	atomic_set(&dev_priv->marker_seq, dev_priv->last_read_seqno);
+	iowrite32(dev_priv->last_read_seqno, fifo_mem + SVGA_FIFO_FENCE);
+	vmw_marker_queue_init(&fifo->marker_queue);
 	return vmw_fifo_send_fence(dev_priv, &dummy);
-out_err:
-	vfree(fifo->static_buffer);
-	fifo->static_buffer = NULL;
-	return ret;
 }
 
 void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason)
@@ -132,19 +183,17 @@ void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 	while (vmw_read(dev_priv, SVGA_REG_BUSY) != 0)
 		vmw_write(dev_priv, SVGA_REG_SYNC, SVGA_SYNC_GENERIC);
 
-	dev_priv->last_read_sequence = ioread32(fifo_mem + SVGA_FIFO_FENCE);
+	dev_priv->last_read_seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
 
 	vmw_write(dev_priv, SVGA_REG_CONFIG_DONE,
 		  dev_priv->config_done_state);
 	vmw_write(dev_priv, SVGA_REG_ENABLE,
 		  dev_priv->enable_state);
+	vmw_write(dev_priv, SVGA_REG_TRACES,
+		  dev_priv->traces_state);
 
 	mutex_unlock(&dev_priv->hw_mutex);
-
-	if (likely(fifo->last_buffer != NULL)) {
-		vfree(fifo->last_buffer);
-		fifo->last_buffer = NULL;
-	}
+	vmw_marker_queue_takedown(&fifo->marker_queue);
 
 	if (likely(fifo->static_buffer != NULL)) {
 		vfree(fifo->static_buffer);
@@ -221,9 +270,8 @@ static int vmw_fifo_wait(struct vmw_private *dev_priv,
 		spin_lock_irqsave(&dev_priv->irq_lock, irq_flags);
 		outl(SVGA_IRQFLAG_FIFO_PROGRESS,
 		     dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
-		vmw_write(dev_priv, SVGA_REG_IRQMASK,
-			  vmw_read(dev_priv, SVGA_REG_IRQMASK) |
-			  SVGA_IRQFLAG_FIFO_PROGRESS);
+		dev_priv->irq_mask |= SVGA_IRQFLAG_FIFO_PROGRESS;
+		vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask);
 		spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags);
 	}
 	mutex_unlock(&dev_priv->hw_mutex);
@@ -245,9 +293,8 @@ static int vmw_fifo_wait(struct vmw_private *dev_priv,
 	mutex_lock(&dev_priv->hw_mutex);
 	if (atomic_dec_and_test(&dev_priv->fifo_queue_waiters)) {
 		spin_lock_irqsave(&dev_priv->irq_lock, irq_flags);
-		vmw_write(dev_priv, SVGA_REG_IRQMASK,
-			  vmw_read(dev_priv, SVGA_REG_IRQMASK) &
-			  ~SVGA_IRQFLAG_FIFO_PROGRESS);
+		dev_priv->irq_mask &= ~SVGA_IRQFLAG_FIFO_PROGRESS;
+		vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask);
 		spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags);
 	}
 	mutex_unlock(&dev_priv->hw_mutex);
@@ -255,6 +302,16 @@ static int vmw_fifo_wait(struct vmw_private *dev_priv,
 	return ret;
 }
 
+/**
+ * Reserve @bytes number of bytes in the fifo.
+ *
+ * This function will return NULL (error) on two conditions:
+ *  If it timeouts waiting for fifo space, or if @bytes is larger than the
+ *   available fifo space.
+ *
+ * Returns:
+ *   Pointer to the fifo, or null on error (possible hardware hang).
+ */
 void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes)
 {
 	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
@@ -265,7 +322,7 @@ void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes)
 	uint32_t reserveable = fifo_state->capabilities & SVGA_FIFO_CAP_RESERVE;
 	int ret;
 
-	down_write(&fifo_state->rwsem);
+	mutex_lock(&fifo_state->fifo_mutex);
 	max = ioread32(fifo_mem + SVGA_FIFO_MAX);
 	min = ioread32(fifo_mem + SVGA_FIFO_MIN);
 	next_cmd = ioread32(fifo_mem + SVGA_FIFO_NEXT_CMD);
@@ -333,7 +390,7 @@ void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes)
 	}
 out_err:
 	fifo_state->reserved_size = 0;
-	up_write(&fifo_state->rwsem);
+	mutex_unlock(&fifo_state->fifo_mutex);
 	return NULL;
 }
 
@@ -408,6 +465,7 @@ void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
 
 	}
 
+	down_write(&fifo_state->rwsem);
 	if (fifo_state->using_bounce_buffer || reserveable) {
 		next_cmd += bytes;
 		if (next_cmd >= max)
@@ -419,11 +477,12 @@ void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
 	if (reserveable)
 		iowrite32(0, fifo_mem + SVGA_FIFO_RESERVED);
 	mb();
-	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
 	up_write(&fifo_state->rwsem);
+	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
+	mutex_unlock(&fifo_state->fifo_mutex);
 }
 
-int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence)
+int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *seqno)
 {
 	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
 	struct svga_fifo_cmd_fence *cmd_fence;
@@ -433,18 +492,16 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence)
 
 	fm = vmw_fifo_reserve(dev_priv, bytes);
 	if (unlikely(fm == NULL)) {
-		down_write(&fifo_state->rwsem);
-		*sequence = dev_priv->fence_seq;
-		up_write(&fifo_state->rwsem);
+		*seqno = atomic_read(&dev_priv->marker_seq);
 		ret = -ENOMEM;
-		(void)vmw_fallback_wait(dev_priv, false, true, *sequence,
+		(void)vmw_fallback_wait(dev_priv, false, true, *seqno,
 					false, 3*HZ);
 		goto out_err;
 	}
 
 	do {
-		*sequence = dev_priv->fence_seq++;
-	} while (*sequence == 0);
+		*seqno = atomic_add_return(1, &dev_priv->marker_seq);
+	} while (*seqno == 0);
 
 	if (!(fifo_state->capabilities & SVGA_FIFO_CAP_FENCE)) {
 
@@ -461,59 +518,132 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence)
 	cmd_fence = (struct svga_fifo_cmd_fence *)
 	    ((unsigned long)fm + sizeof(__le32));
 
-	iowrite32(*sequence, &cmd_fence->fence);
-	fifo_state->last_buffer_add = true;
+	iowrite32(*seqno, &cmd_fence->fence);
 	vmw_fifo_commit(dev_priv, bytes);
-	fifo_state->last_buffer_add = false;
+	(void) vmw_marker_push(&fifo_state->marker_queue, *seqno);
+	vmw_update_seqno(dev_priv, fifo_state);
 
 out_err:
 	return ret;
 }
 
 /**
- * Map the first page of the FIFO read-only to user-space.
+ * vmw_fifo_emit_dummy_legacy_query - emits a dummy query to the fifo using
+ * legacy query commands.
+ *
+ * @dev_priv: The device private structure.
+ * @cid: The hardware context id used for the query.
+ *
+ * See the vmw_fifo_emit_dummy_query documentation.
  */
+static int vmw_fifo_emit_dummy_legacy_query(struct vmw_private *dev_priv,
+					    uint32_t cid)
+{
+	/*
+	 * A query wait without a preceding query end will
+	 * actually finish all queries for this cid
+	 * without writing to the query result structure.
+	 */
+
+	struct ttm_buffer_object *bo = dev_priv->dummy_query_bo;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdWaitForQuery body;
+	} *cmd;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
 
-static int vmw_fifo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Out of fifo space for dummy query.\n");
+		return -ENOMEM;
+	}
+
+	cmd->header.id = SVGA_3D_CMD_WAIT_FOR_QUERY;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = cid;
+	cmd->body.type = SVGA3D_QUERYTYPE_OCCLUSION;
+
+	if (bo->mem.mem_type == TTM_PL_VRAM) {
+		cmd->body.guestResult.gmrId = SVGA_GMR_FRAMEBUFFER;
+		cmd->body.guestResult.offset = bo->offset;
+	} else {
+		cmd->body.guestResult.gmrId = bo->mem.start;
+		cmd->body.guestResult.offset = 0;
+	}
+
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
+
+	return 0;
+}
+
+/**
+ * vmw_fifo_emit_dummy_gb_query - emits a dummy query to the fifo using
+ * guest-backed resource query commands.
+ *
+ * @dev_priv: The device private structure.
+ * @cid: The hardware context id used for the query.
+ *
+ * See the vmw_fifo_emit_dummy_query documentation.
+ */
+static int vmw_fifo_emit_dummy_gb_query(struct vmw_private *dev_priv,
+					uint32_t cid)
 {
-	int ret;
-	unsigned long address = (unsigned long)vmf->virtual_address;
+	/*
+	 * A query wait without a preceding query end will
+	 * actually finish all queries for this cid
+	 * without writing to the query result structure.
+	 */
+
+	struct ttm_buffer_object *bo = dev_priv->dummy_query_bo;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdWaitForGBQuery body;
+	} *cmd;
+
+	cmd = vmw_fifo_reserve(dev_priv, sizeof(*cmd));
+
+	if (unlikely(cmd == NULL)) {
+		DRM_ERROR("Out of fifo space for dummy query.\n");
+		return -ENOMEM;
+	}
 
-	if (address != vma->vm_start)
-		return VM_FAULT_SIGBUS;
+	cmd->header.id = SVGA_3D_CMD_WAIT_FOR_GB_QUERY;
+	cmd->header.size = sizeof(cmd->body);
+	cmd->body.cid = cid;
+	cmd->body.type = SVGA3D_QUERYTYPE_OCCLUSION;
+	BUG_ON(bo->mem.mem_type != VMW_PL_MOB);
+	cmd->body.mobid = bo->mem.start;
+	cmd->body.offset = 0;
 
-	ret = vm_insert_pfn(vma, address, vma->vm_pgoff);
-	if (likely(ret == -EBUSY || ret == 0))
-		return VM_FAULT_NOPAGE;
-	else if (ret == -ENOMEM)
-		return VM_FAULT_OOM;
+	vmw_fifo_commit(dev_priv, sizeof(*cmd));
 
-	return VM_FAULT_SIGBUS;
+	return 0;
 }
 
-static struct vm_operations_struct vmw_fifo_vm_ops = {
-	.fault = vmw_fifo_vm_fault,
-	.open = NULL,
-	.close = NULL
-};
 
-int vmw_fifo_mmap(struct file *filp, struct vm_area_struct *vma)
+/**
+ * vmw_fifo_emit_dummy_gb_query - emits a dummy query to the fifo using
+ * appropriate resource query commands.
+ *
+ * @dev_priv: The device private structure.
+ * @cid: The hardware context id used for the query.
+ *
+ * This function is used to emit a dummy occlusion query with
+ * no primitives rendered between query begin and query end.
+ * It's used to provide a query barrier, in order to know that when
+ * this query is finished, all preceding queries are also finished.
+ *
+ * A Query results structure should have been initialized at the start
+ * of the dev_priv->dummy_query_bo buffer object. And that buffer object
+ * must also be either reserved or pinned when this function is called.
+ *
+ * Returns -ENOMEM on failure to reserve fifo space.
+ */
+int vmw_fifo_emit_dummy_query(struct vmw_private *dev_priv,
+			      uint32_t cid)
 {
-	struct drm_file *file_priv;
-	struct vmw_private *dev_priv;
-
-	file_priv = (struct drm_file *)filp->private_data;
-	dev_priv = vmw_priv(file_priv->minor->dev);
-
-	if (vma->vm_pgoff != (dev_priv->mmio_start >> PAGE_SHIFT) ||
-	    (vma->vm_end - vma->vm_start) != PAGE_SIZE)
-		return -EINVAL;
-
-	vma->vm_flags &= ~(VM_WRITE | VM_MAYWRITE);
-	vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_SHARED;
-	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
-	vma->vm_page_prot = ttm_io_prot(TTM_PL_FLAG_UNCACHED,
-					vma->vm_page_prot);
-	vma->vm_ops = &vmw_fifo_vm_ops;
-	return 0;
+	if (dev_priv->has_mob)
+		return vmw_fifo_emit_dummy_gb_query(dev_priv, cid);
+
+	return vmw_fifo_emit_dummy_legacy_query(dev_priv, cid);
 }