[PATCH -next 0/8] vmwgfx compatibility-affecting changes

[PATCH -next 0/8] vmwgfx compatibility-affecting changes

[Thomas Hellstrom]

A set of changes and cleanups that affects the kernel-user-interface in a
non-backwards-compatible way. This is intended to be the last major bump
before we exit staging, but there may be a couple of minor bumps in
upcoming patches, as we add new features.

In-Reply-To: 

[PATCH 1/8] vmwgfx: Remove the fifo debug ioctl

[Thomas Hellstrom]

It was only used for bringup debugging, and probably doesn't work
anymore. Remove it.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c   |    5 -----
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h   |    6 ------
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c  |   21 ---------------------
 drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c |   23 -----------------------
 include/drm/vmwgfx_drm.h              |   30 ++----------------------------
 5 files changed, 2 insertions(+), 83 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index bff7932..26eebdc 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -82,9 +82,6 @@
 #define DRM_IOCTL_VMW_EXECBUF					\
 	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_EXECBUF,		\
 		struct drm_vmw_execbuf_arg)
-#define DRM_IOCTL_VMW_FIFO_DEBUG				\
-	DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FIFO_DEBUG,		\
-		 struct drm_vmw_fifo_debug_arg)
 #define DRM_IOCTL_VMW_FENCE_WAIT				\
 	DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FENCE_WAIT,		\
 		 struct drm_vmw_fence_wait_arg)
@@ -135,8 +132,6 @@ static struct drm_ioctl_desc vmw_ioctls[] = {
 		      DRM_AUTH | DRM_UNLOCKED),
 	VMW_IOCTL_DEF(VMW_EXECBUF, vmw_execbuf_ioctl,
 		      DRM_AUTH | DRM_UNLOCKED),
-	VMW_IOCTL_DEF(VMW_FIFO_DEBUG, vmw_fifo_debug_ioctl,
-		      DRM_AUTH | DRM_ROOT_ONLY | DRM_MASTER | DRM_UNLOCKED),
 	VMW_IOCTL_DEF(VMW_FENCE_WAIT, vmw_fence_wait_ioctl,
 		      DRM_AUTH | DRM_UNLOCKED),
 	VMW_IOCTL_DEF(VMW_UPDATE_LAYOUT, vmw_kms_update_layout_ioctl,
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 323fc10..512b78b 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -116,10 +116,6 @@ struct vmw_fifo_state {
 	unsigned long reserved_size;
 	__le32 *dynamic_buffer;
 	__le32 *static_buffer;
-	__le32 *last_buffer;
-	uint32_t last_data_size;
-	uint32_t last_buffer_size;
-	bool last_buffer_add;
 	unsigned long static_buffer_size;
 	bool using_bounce_buffer;
 	uint32_t capabilities;
@@ -401,8 +397,6 @@ extern int vmw_user_stream_lookup(struct vmw_private *dev_priv,
 
 extern int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 			      struct drm_file *file_priv);
-extern int vmw_fifo_debug_ioctl(struct drm_device *dev, void *data,
-				struct drm_file *file_priv);
 
 /**
  * Fifo utilities - vmwgfx_fifo.c
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index 635c0ff..d031c55 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -72,22 +72,12 @@ 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;
@@ -141,10 +131,6 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 	iowrite32(dev_priv->last_read_sequence, fifo_mem + SVGA_FIFO_FENCE);
 	vmw_fence_queue_init(&fifo->fence_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)
@@ -182,11 +168,6 @@ void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 	mutex_unlock(&dev_priv->hw_mutex);
 	vmw_fence_queue_takedown(&fifo->fence_queue);
 
-	if (likely(fifo->last_buffer != NULL)) {
-		vfree(fifo->last_buffer);
-		fifo->last_buffer = NULL;
-	}
-
 	if (likely(fifo->static_buffer != NULL)) {
 		vfree(fifo->static_buffer);
 		fifo->static_buffer = NULL;
@@ -503,9 +484,7 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence)
 	    ((unsigned long)fm + sizeof(__le32));
 
 	iowrite32(*sequence, &cmd_fence->fence);
-	fifo_state->last_buffer_add = true;
 	vmw_fifo_commit(dev_priv, bytes);
-	fifo_state->last_buffer_add = false;
 	(void) vmw_fence_push(&fifo_state->fence_queue, *sequence);
 	vmw_update_sequence(dev_priv, fifo_state);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
index 570d577..f1b9175 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
@@ -65,26 +65,3 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 
 	return 0;
 }
-
-int vmw_fifo_debug_ioctl(struct drm_device *dev, void *data,
-			 struct drm_file *file_priv)
-{
-	struct vmw_private *dev_priv = vmw_priv(dev);
-	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
-	struct drm_vmw_fifo_debug_arg *arg =
-	    (struct drm_vmw_fifo_debug_arg *)data;
-	__le32 __user *buffer = (__le32 __user *)
-	    (unsigned long)arg->debug_buffer;
-
-	if (unlikely(fifo_state->last_buffer == NULL))
-		return -EINVAL;
-
-	if (arg->debug_buffer_size < fifo_state->last_data_size) {
-		arg->used_size = arg->debug_buffer_size;
-		arg->did_not_fit = 1;
-	} else {
-		arg->used_size = fifo_state->last_data_size;
-		arg->did_not_fit = 0;
-	}
-	return copy_to_user(buffer, fifo_state->last_buffer, arg->used_size);
-}
diff --git a/include/drm/vmwgfx_drm.h b/include/drm/vmwgfx_drm.h
index 5c36432..d233544 100644
--- a/include/drm/vmwgfx_drm.h
+++ b/include/drm/vmwgfx_drm.h
@@ -48,10 +48,9 @@
 #define DRM_VMW_UNREF_SURFACE        10
 #define DRM_VMW_REF_SURFACE          11
 #define DRM_VMW_EXECBUF              12
-#define DRM_VMW_FIFO_DEBUG           13
-#define DRM_VMW_FENCE_WAIT           14
+#define DRM_VMW_FENCE_WAIT           13
 /* guarded by minor version >= 2 */
-#define DRM_VMW_UPDATE_LAYOUT        15
+#define DRM_VMW_UPDATE_LAYOUT        14
 
 
 /*************************************************************************/
@@ -433,31 +432,6 @@ struct drm_vmw_unref_dmabuf_arg {
 	uint32_t pad64;
 };
 
-/*************************************************************************/
-/**
- * DRM_VMW_FIFO_DEBUG - Get last FIFO submission.
- *
- * This IOCTL copies the last FIFO submission directly out of the FIFO buffer.
- */
-
-/**
- * struct drm_vmw_fifo_debug_arg
- *
- * @debug_buffer: User space address of a debug_buffer cast to an uint64_t //In
- * @debug_buffer_size: Size in bytes of debug buffer //In
- * @used_size: Number of bytes copied to the buffer // Out
- * @did_not_fit: Boolean indicating that the fifo contents did not fit. //Out
- *
- * Argument to the DRM_VMW_FIFO_DEBUG Ioctl.
- */
-
-struct drm_vmw_fifo_debug_arg {
-	uint64_t debug_buffer;
-	uint32_t debug_buffer_size;
-	uint32_t used_size;
-	int32_t did_not_fit;
-	uint32_t pad64;
-};
 
 struct drm_vmw_fence_wait_arg {
 	uint64_t sequence;
-- 
1.7.4.4

[PATCH 2/8] vmwgfx: Remove the update layout IOCTL.

[Thomas Hellstrom]

It doesn't seem like its needed. If this turns out to be an incorrect
assumption, we can reinstate it.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c |    6 ----
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h |    2 -
 drivers/gpu/drm/vmwgfx/vmwgfx_kms.c |   46 -----------------------------------
 include/drm/vmwgfx_drm.h            |    3 --
 4 files changed, 0 insertions(+), 57 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 26eebdc..9819d0d 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -85,10 +85,6 @@
 #define DRM_IOCTL_VMW_FENCE_WAIT				\
 	DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FENCE_WAIT,		\
 		 struct drm_vmw_fence_wait_arg)
-#define DRM_IOCTL_VMW_UPDATE_LAYOUT				\
-	DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_UPDATE_LAYOUT,	\
-		 struct drm_vmw_update_layout_arg)
-
 
 /**
  * The core DRM version of this macro doesn't account for
@@ -134,8 +130,6 @@ static struct drm_ioctl_desc vmw_ioctls[] = {
 		      DRM_AUTH | DRM_UNLOCKED),
 	VMW_IOCTL_DEF(VMW_FENCE_WAIT, vmw_fence_wait_ioctl,
 		      DRM_AUTH | DRM_UNLOCKED),
-	VMW_IOCTL_DEF(VMW_UPDATE_LAYOUT, vmw_kms_update_layout_ioctl,
-		      DRM_MASTER | DRM_CONTROL_ALLOW | DRM_UNLOCKED)
 };
 
 static struct pci_device_id vmw_pci_id_list[] = {
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 512b78b..a99400f 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -508,8 +508,6 @@ void vmw_kms_cursor_snoop(struct vmw_surface *srf,
 int vmw_kms_write_svga(struct vmw_private *vmw_priv,
 		       unsigned width, unsigned height, unsigned pitch,
 		       unsigned bpp, unsigned depth);
-int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data,
-				struct drm_file *file_priv);
 void vmw_kms_idle_workqueues(struct vmw_master *vmaster);
 bool vmw_kms_validate_mode_vram(struct vmw_private *dev_priv,
 				uint32_t pitch,
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 9ea2f05..1a4c84c 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -1092,52 +1092,6 @@ int vmw_kms_restore_vga(struct vmw_private *vmw_priv)
 	return 0;
 }
 
-int vmw_kms_update_layout_ioctl(struct drm_device *dev, void *data,
-				struct drm_file *file_priv)
-{
-	struct vmw_private *dev_priv = vmw_priv(dev);
-	struct drm_vmw_update_layout_arg *arg =
-		(struct drm_vmw_update_layout_arg *)data;
-	struct vmw_master *vmaster = vmw_master(file_priv->master);
-	void __user *user_rects;
-	struct drm_vmw_rect *rects;
-	unsigned rects_size;
-	int ret;
-
-	ret = ttm_read_lock(&vmaster->lock, true);
-	if (unlikely(ret != 0))
-		return ret;
-
-	if (!arg->num_outputs) {
-		struct drm_vmw_rect def_rect = {0, 0, 800, 600};
-		vmw_kms_ldu_update_layout(dev_priv, 1, &def_rect);
-		goto out_unlock;
-	}
-
-	rects_size = arg->num_outputs * sizeof(struct drm_vmw_rect);
-	rects = kzalloc(rects_size, GFP_KERNEL);
-	if (unlikely(!rects)) {
-		ret = -ENOMEM;
-		goto out_unlock;
-	}
-
-	user_rects = (void __user *)(unsigned long)arg->rects;
-	ret = copy_from_user(rects, user_rects, rects_size);
-	if (unlikely(ret != 0)) {
-		DRM_ERROR("Failed to get rects.\n");
-		ret = -EFAULT;
-		goto out_free;
-	}
-
-	vmw_kms_ldu_update_layout(dev_priv, arg->num_outputs, rects);
-
-out_free:
-	kfree(rects);
-out_unlock:
-	ttm_read_unlock(&vmaster->lock);
-	return ret;
-}
-
 bool vmw_kms_validate_mode_vram(struct vmw_private *dev_priv,
 				uint32_t pitch,
 				uint32_t height)
diff --git a/include/drm/vmwgfx_drm.h b/include/drm/vmwgfx_drm.h
index d233544..c0b226b 100644
--- a/include/drm/vmwgfx_drm.h
+++ b/include/drm/vmwgfx_drm.h
@@ -49,9 +49,6 @@
 #define DRM_VMW_REF_SURFACE          11
 #define DRM_VMW_EXECBUF              12
 #define DRM_VMW_FENCE_WAIT           13
-/* guarded by minor version >= 2 */
-#define DRM_VMW_UPDATE_LAYOUT        14
-
 
 /*************************************************************************/
 /**
-- 
1.7.4.4

[PATCH 3/8] vmwgfx: Remove the possibility to map the fifo from user-space

[Thomas Hellstrom]

This was previously used by user-space to check whether a fence
sequence had passed or not.
With fence objects that's not needed anymore.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h      |    1 -
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c     |   48 ------------------------------
 drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c    |    3 --
 drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c |    5 +--
 include/drm/vmwgfx_drm.h                 |    7 ++--
 5 files changed, 5 insertions(+), 59 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index a99400f..39be796 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -411,7 +411,6 @@ extern void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes);
 extern int vmw_fifo_send_fence(struct vmw_private *dev_priv,
 			       uint32_t *sequence);
 extern void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason);
-extern int vmw_fifo_mmap(struct file *filp, struct vm_area_struct *vma);
 extern bool vmw_fifo_have_3d(struct vmw_private *dev_priv);
 extern bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index d031c55..aae01b9 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -491,51 +491,3 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence)
 out_err:
 	return ret;
 }
-
-/**
- * Map the first page of the FIFO read-only to user-space.
- */
-
-static int vmw_fifo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
-{
-	int ret;
-	unsigned long address = (unsigned long)vmf->virtual_address;
-
-	if (address != vma->vm_start)
-		return VM_FAULT_SIGBUS;
-
-	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;
-
-	return VM_FAULT_SIGBUS;
-}
-
-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)
-{
-	struct drm_file *file_priv;
-	struct vmw_private *dev_priv;
-
-	file_priv = 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;
-}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
index f1b9175..74b1dc8 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
@@ -45,9 +45,6 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 	case DRM_VMW_PARAM_3D:
 		param->value = vmw_fifo_have_3d(dev_priv) ? 1 : 0;
 		break;
-	case DRM_VMW_PARAM_FIFO_OFFSET:
-		param->value = dev_priv->mmio_start;
-		break;
 	case DRM_VMW_PARAM_HW_CAPS:
 		param->value = dev_priv->capabilities;
 		break;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c
index 1e8eedd..d3c11f5 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c
@@ -34,9 +34,8 @@ int vmw_mmap(struct file *filp, struct vm_area_struct *vma)
 	struct vmw_private *dev_priv;
 
 	if (unlikely(vma->vm_pgoff < VMWGFX_FILE_PAGE_OFFSET)) {
-		if (vmw_fifo_mmap(filp, vma) == 0)
-			return 0;
-		return drm_mmap(filp, vma);
+		DRM_ERROR("Illegal attempt to mmap old fifo space.\n");
+		return -EINVAL;
 	}
 
 	file_priv = filp->private_data;
diff --git a/include/drm/vmwgfx_drm.h b/include/drm/vmwgfx_drm.h
index c0b226b..490db45 100644
--- a/include/drm/vmwgfx_drm.h
+++ b/include/drm/vmwgfx_drm.h
@@ -65,10 +65,9 @@
 #define DRM_VMW_PARAM_NUM_STREAMS      0
 #define DRM_VMW_PARAM_NUM_FREE_STREAMS 1
 #define DRM_VMW_PARAM_3D               2
-#define DRM_VMW_PARAM_FIFO_OFFSET      3
-#define DRM_VMW_PARAM_HW_CAPS          4
-#define DRM_VMW_PARAM_FIFO_CAPS        5
-#define DRM_VMW_PARAM_MAX_FB_SIZE      6
+#define DRM_VMW_PARAM_HW_CAPS          3
+#define DRM_VMW_PARAM_FIFO_CAPS        4
+#define DRM_VMW_PARAM_MAX_FB_SIZE      5
 
 /**
  * struct drm_vmw_getparam_arg
-- 
1.7.4.4

[PATCH 4/8] vmwgfx: Add functionality to get 3D caps

[Thomas Hellstrom]

Since we don't allow user-space to map the fifo anymore,
add a parameter to get fifo hw version and
an ioctl to copy the 3D capabilities.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecranz <jakob@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c   |    5 +++
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h   |    2 +
 drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c |   48 +++++++++++++++++++++++++++++++++
 include/drm/vmwgfx_drm.h              |   25 +++++++++++++++++
 4 files changed, 80 insertions(+), 0 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 9819d0d..8010254 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -85,6 +85,9 @@
 #define DRM_IOCTL_VMW_FENCE_WAIT				\
 	DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FENCE_WAIT,		\
 		 struct drm_vmw_fence_wait_arg)
+#define DRM_IOCTL_VMW_GET_3D_CAP				\
+	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_GET_3D_CAP,		\
+		struct drm_vmw_get_3d_cap_arg)
 
 /**
  * The core DRM version of this macro doesn't account for
@@ -130,6 +133,8 @@ static struct drm_ioctl_desc vmw_ioctls[] = {
 		      DRM_AUTH | DRM_UNLOCKED),
 	VMW_IOCTL_DEF(VMW_FENCE_WAIT, vmw_fence_wait_ioctl,
 		      DRM_AUTH | DRM_UNLOCKED),
+	VMW_IOCTL_DEF(VMW_GET_3D_CAP, vmw_get_cap_3d_ioctl,
+		      DRM_AUTH | DRM_UNLOCKED),
 };
 
 static struct pci_device_id vmw_pci_id_list[] = {
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 39be796..2374a5c 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -397,6 +397,8 @@ extern int vmw_user_stream_lookup(struct vmw_private *dev_priv,
 
 extern int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 			      struct drm_file *file_priv);
+extern int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data,
+				struct drm_file *file_priv);
 
 /**
  * Fifo utilities - vmwgfx_fifo.c
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
index 74b1dc8..5ecf966 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
@@ -54,6 +54,13 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 	case DRM_VMW_PARAM_MAX_FB_SIZE:
 		param->value = dev_priv->vram_size;
 		break;
+	case DRM_VMW_PARAM_FIFO_HW_VERSION:
+	{
+		__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+
+		param->value = ioread32(fifo_mem + SVGA_FIFO_3D_HWVERSION);
+		break;
+	}
 	default:
 		DRM_ERROR("Illegal vmwgfx get param request: %d\n",
 			  param->param);
@@ -62,3 +69,44 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 
 	return 0;
 }
+
+
+int vmw_get_cap_3d_ioctl(struct drm_device *dev, void *data,
+			 struct drm_file *file_priv)
+{
+	struct drm_vmw_get_3d_cap_arg *arg =
+		(struct drm_vmw_get_3d_cap_arg *) data;
+	struct vmw_private *dev_priv = vmw_priv(dev);
+	uint32_t size;
+	__le32 __iomem *fifo_mem;
+	void __user *buffer = (void __user *)((unsigned long)(arg->buffer));
+	void *bounce;
+	int ret;
+
+	if (unlikely(arg->pad64 != 0)) {
+		DRM_ERROR("Illegal GET_3D_CAP argument.\n");
+		return -EINVAL;
+	}
+
+	size = (SVGA_FIFO_3D_CAPS_LAST - SVGA_FIFO_3D_CAPS + 1) << 2;
+
+	if (arg->max_size < size)
+		size = arg->max_size;
+
+	bounce = vmalloc(size);
+	if (unlikely(bounce == NULL)) {
+		DRM_ERROR("Failed to allocate bounce buffer for 3D caps.\n");
+		return -ENOMEM;
+	}
+
+	fifo_mem = dev_priv->mmio_virt;
+	memcpy_fromio(bounce, &fifo_mem[SVGA_FIFO_3D_CAPS], size);
+
+	ret = copy_to_user(buffer, bounce, size);
+	vfree(bounce);
+
+	if (unlikely(ret != 0))
+		DRM_ERROR("Failed to report 3D caps info.\n");
+
+	return ret;
+}
diff --git a/include/drm/vmwgfx_drm.h b/include/drm/vmwgfx_drm.h
index 490db45..467b80c 100644
--- a/include/drm/vmwgfx_drm.h
+++ b/include/drm/vmwgfx_drm.h
@@ -49,6 +49,7 @@
 #define DRM_VMW_REF_SURFACE          11
 #define DRM_VMW_EXECBUF              12
 #define DRM_VMW_FENCE_WAIT           13
+#define DRM_VMW_GET_3D_CAP           14
 
 /*************************************************************************/
 /**
@@ -68,6 +69,7 @@
 #define DRM_VMW_PARAM_HW_CAPS          3
 #define DRM_VMW_PARAM_FIFO_CAPS        4
 #define DRM_VMW_PARAM_MAX_FB_SIZE      5
+#define DRM_VMW_PARAM_FIFO_HW_VERSION  6
 
 /**
  * struct drm_vmw_getparam_arg
@@ -557,6 +559,29 @@ struct drm_vmw_stream_arg {
  * Return a single stream that was claimed by this process. Also makes
  * sure that the stream has been stopped.
  */
+/*************************************************************************/
+/**
+ * DRM_VMW_GET_3D_CAP
+ *
+ * Read 3D capabilities from the FIFO
+ *
+ */
+
+/**
+ * struct drm_vmw_get_3d_cap_arg
+ *
+ * @buffer: Pointer to a buffer for capability data, cast to an uint64_t
+ * @size: Max size to copy
+ *
+ * Input argument to the DRM_VMW_GET_3D_CAP_IOCTL
+ * ioctls.
+ */
+
+struct drm_vmw_get_3d_cap_arg {
+	uint64_t buffer;
+	uint32_t max_size;
+	uint32_t pad64;
+};
 
 /*************************************************************************/
 /**
-- 
1.7.4.4

[PATCH 5/8] vmwgfx: Fix confusion caused by using "fence" in various places

[Thomas Hellstrom]

This is needed before we introduce the fence objects.
Otherwise this will be even more confusing. The plan is to use the following:

seqno: A 32-bit sequence number that may be passed in the fifo.
marker: Objects, carrying a seqno, that track fifo submission time. They
are used for fifo lag based throttling.
fence objects: Kernel space objects, possibly accessible from user-space and
carrying a 32-bit seqno together with signaled status.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
---
 drivers/gpu/drm/vmwgfx/Makefile         |    2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c  |    8 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c     |    2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h     |   42 ++++----
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c |   10 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_fence.c   |  173 -------------------------------
 drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c    |   26 +++---
 drivers/gpu/drm/vmwgfx/vmwgfx_irq.c     |   56 +++++-----
 drivers/gpu/drm/vmwgfx/vmwgfx_marker.c  |  171 ++++++++++++++++++++++++++++++
 include/drm/vmwgfx_drm.h                |    6 +-
 10 files changed, 247 insertions(+), 249 deletions(-)
 delete mode 100644 drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
 create mode 100644 drivers/gpu/drm/vmwgfx/vmwgfx_marker.c

diff --git a/drivers/gpu/drm/vmwgfx/Makefile b/drivers/gpu/drm/vmwgfx/Makefile
index c9281a1..f41e8b4 100644
--- a/drivers/gpu/drm/vmwgfx/Makefile
+++ b/drivers/gpu/drm/vmwgfx/Makefile
@@ -4,6 +4,6 @@ ccflags-y := -Iinclude/drm
 vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \
 	    vmwgfx_fb.o vmwgfx_ioctl.o vmwgfx_resource.o vmwgfx_buffer.o \
 	    vmwgfx_fifo.o vmwgfx_irq.o vmwgfx_ldu.o vmwgfx_ttm_glue.o \
-	    vmwgfx_overlay.o vmwgfx_fence.o vmwgfx_gmrid_manager.o
+	    vmwgfx_overlay.o vmwgfx_marker.o vmwgfx_gmrid_manager.o
 
 obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index 87e43e0..72d9561 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -295,18 +295,18 @@ static int vmw_sync_obj_flush(void *sync_obj, void *sync_arg)
 static bool vmw_sync_obj_signaled(void *sync_obj, void *sync_arg)
 {
 	struct vmw_private *dev_priv = (struct vmw_private *)sync_arg;
-	uint32_t sequence = (unsigned long) sync_obj;
+	uint32_t seqno = (unsigned long) sync_obj;
 
-	return vmw_fence_signaled(dev_priv, sequence);
+	return vmw_seqno_passed(dev_priv, seqno);
 }
 
 static int vmw_sync_obj_wait(void *sync_obj, void *sync_arg,
 			     bool lazy, bool interruptible)
 {
 	struct vmw_private *dev_priv = (struct vmw_private *)sync_arg;
-	uint32_t sequence = (unsigned long) sync_obj;
+	uint32_t seqno = (unsigned long) sync_obj;
 
-	return vmw_wait_fence(dev_priv, false, sequence, false, 3*HZ);
+	return vmw_wait_seqno(dev_priv, false, seqno, false, 3*HZ);
 }
 
 struct ttm_bo_driver vmw_bo_driver = {
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 8010254..c8b5a53 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -280,7 +280,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 
 	dev_priv->dev = dev;
 	dev_priv->vmw_chipset = chipset;
-	dev_priv->last_read_sequence = (uint32_t) -100;
+	dev_priv->last_read_seqno = (uint32_t) -100;
 	mutex_init(&dev_priv->hw_mutex);
 	mutex_init(&dev_priv->cmdbuf_mutex);
 	mutex_init(&dev_priv->release_mutex);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 2374a5c..9c3016b 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -105,7 +105,7 @@ struct vmw_surface {
 	struct vmw_cursor_snooper snooper;
 };
 
-struct vmw_fence_queue {
+struct vmw_marker_queue {
 	struct list_head head;
 	struct timespec lag;
 	struct timespec lag_time;
@@ -121,7 +121,7 @@ struct vmw_fifo_state {
 	uint32_t capabilities;
 	struct mutex fifo_mutex;
 	struct rw_semaphore rwsem;
-	struct vmw_fence_queue fence_queue;
+	struct vmw_marker_queue marker_queue;
 };
 
 struct vmw_relocation {
@@ -238,12 +238,12 @@ struct vmw_private {
 	 * Fencing and IRQs.
 	 */
 
-	atomic_t fence_seq;
+	atomic_t marker_seq;
 	wait_queue_head_t fence_queue;
 	wait_queue_head_t fifo_queue;
 	atomic_t fence_queue_waiters;
 	atomic_t fifo_queue_waiters;
-	uint32_t last_read_sequence;
+	uint32_t last_read_seqno;
 	spinlock_t irq_lock;
 
 	/*
@@ -411,7 +411,7 @@ extern void vmw_fifo_release(struct vmw_private *dev_priv,
 extern void *vmw_fifo_reserve(struct vmw_private *dev_priv, uint32_t bytes);
 extern void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes);
 extern int vmw_fifo_send_fence(struct vmw_private *dev_priv,
-			       uint32_t *sequence);
+			       uint32_t *seqno);
 extern void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason);
 extern bool vmw_fifo_have_3d(struct vmw_private *dev_priv);
 extern bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv);
@@ -448,39 +448,39 @@ extern int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
  */
 
 extern irqreturn_t vmw_irq_handler(DRM_IRQ_ARGS);
-extern int vmw_wait_fence(struct vmw_private *dev_priv, bool lazy,
-			  uint32_t sequence, bool interruptible,
-			  unsigned long timeout);
+extern int vmw_wait_seqno(struct vmw_private *dev_priv, bool lazy,
+			     uint32_t seqno, bool interruptible,
+			     unsigned long timeout);
 extern void vmw_irq_preinstall(struct drm_device *dev);
 extern int vmw_irq_postinstall(struct drm_device *dev);
 extern void vmw_irq_uninstall(struct drm_device *dev);
-extern bool vmw_fence_signaled(struct vmw_private *dev_priv,
-			       uint32_t sequence);
+extern bool vmw_seqno_passed(struct vmw_private *dev_priv,
+				uint32_t seqno);
 extern int vmw_fence_wait_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *file_priv);
 extern int vmw_fallback_wait(struct vmw_private *dev_priv,
 			     bool lazy,
 			     bool fifo_idle,
-			     uint32_t sequence,
+			     uint32_t seqno,
 			     bool interruptible,
 			     unsigned long timeout);
-extern void vmw_update_sequence(struct vmw_private *dev_priv,
+extern void vmw_update_seqno(struct vmw_private *dev_priv,
 				struct vmw_fifo_state *fifo_state);
 
 
 /**
- * Rudimentary fence objects currently used only for throttling -
- * vmwgfx_fence.c
+ * Rudimentary fence-like objects currently used only for throttling -
+ * vmwgfx_marker.c
  */
 
-extern void vmw_fence_queue_init(struct vmw_fence_queue *queue);
-extern void vmw_fence_queue_takedown(struct vmw_fence_queue *queue);
-extern int vmw_fence_push(struct vmw_fence_queue *queue,
-			  uint32_t sequence);
-extern int vmw_fence_pull(struct vmw_fence_queue *queue,
-			  uint32_t signaled_sequence);
+extern void vmw_marker_queue_init(struct vmw_marker_queue *queue);
+extern void vmw_marker_queue_takedown(struct vmw_marker_queue *queue);
+extern int vmw_marker_push(struct vmw_marker_queue *queue,
+			  uint32_t seqno);
+extern int vmw_marker_pull(struct vmw_marker_queue *queue,
+			  uint32_t signaled_seqno);
 extern int vmw_wait_lag(struct vmw_private *dev_priv,
-			struct vmw_fence_queue *queue, uint32_t us);
+			struct vmw_marker_queue *queue, uint32_t us);
 
 /**
  * Kernel framebuffer - vmwgfx_fb.c
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 3dac8ab..4811c6b 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -685,7 +685,7 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
 	int ret;
 	void *user_cmd;
 	void *cmd;
-	uint32_t sequence;
+	uint32_t seqno;
 	struct vmw_sw_context *sw_context = &dev_priv->ctx;
 	struct vmw_master *vmaster = vmw_master(file_priv->master);
 
@@ -737,7 +737,7 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
 	vmw_apply_relocations(sw_context);
 
 	if (arg->throttle_us) {
-		ret = vmw_wait_lag(dev_priv, &dev_priv->fifo.fence_queue,
+		ret = vmw_wait_lag(dev_priv, &dev_priv->fifo.marker_queue,
 				   arg->throttle_us);
 
 		if (unlikely(ret != 0))
@@ -754,10 +754,10 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
 	memcpy(cmd, sw_context->cmd_bounce, arg->command_size);
 	vmw_fifo_commit(dev_priv, arg->command_size);
 
-	ret = vmw_fifo_send_fence(dev_priv, &sequence);
+	ret = vmw_fifo_send_fence(dev_priv, &seqno);
 
 	ttm_eu_fence_buffer_objects(&sw_context->validate_nodes,
-				    (void *)(unsigned long) sequence);
+				    (void *)(unsigned long) seqno);
 	vmw_clear_validations(sw_context);
 	mutex_unlock(&dev_priv->cmdbuf_mutex);
 
@@ -770,7 +770,7 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
 		DRM_ERROR("Fence submission error. Syncing.\n");
 
 	fence_rep.error = ret;
-	fence_rep.fence_seq = (uint64_t) sequence;
+	fence_rep.fence_seq = (uint64_t) seqno;
 	fence_rep.pad64 = 0;
 
 	user_fence_rep = (struct drm_vmw_fence_rep __user *)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
deleted file mode 100644
index 61eacc1..0000000
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
+++ /dev/null
@@ -1,173 +0,0 @@
-/**************************************************************************
- *
- * Copyright (C) 2010 VMware, Inc., Palo Alto, CA., USA
- * 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, sub license, 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 (including the
- * next paragraph) 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 NON-INFRINGEMENT. IN NO EVENT SHALL
- * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS 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 "vmwgfx_drv.h"
-
-struct vmw_fence {
-	struct list_head head;
-	uint32_t sequence;
-	struct timespec submitted;
-};
-
-void vmw_fence_queue_init(struct vmw_fence_queue *queue)
-{
-	INIT_LIST_HEAD(&queue->head);
-	queue->lag = ns_to_timespec(0);
-	getrawmonotonic(&queue->lag_time);
-	spin_lock_init(&queue->lock);
-}
-
-void vmw_fence_queue_takedown(struct vmw_fence_queue *queue)
-{
-	struct vmw_fence *fence, *next;
-
-	spin_lock(&queue->lock);
-	list_for_each_entry_safe(fence, next, &queue->head, head) {
-		kfree(fence);
-	}
-	spin_unlock(&queue->lock);
-}
-
-int vmw_fence_push(struct vmw_fence_queue *queue,
-		   uint32_t sequence)
-{
-	struct vmw_fence *fence = kmalloc(sizeof(*fence), GFP_KERNEL);
-
-	if (unlikely(!fence))
-		return -ENOMEM;
-
-	fence->sequence = sequence;
-	getrawmonotonic(&fence->submitted);
-	spin_lock(&queue->lock);
-	list_add_tail(&fence->head, &queue->head);
-	spin_unlock(&queue->lock);
-
-	return 0;
-}
-
-int vmw_fence_pull(struct vmw_fence_queue *queue,
-		   uint32_t signaled_sequence)
-{
-	struct vmw_fence *fence, *next;
-	struct timespec now;
-	bool updated = false;
-
-	spin_lock(&queue->lock);
-	getrawmonotonic(&now);
-
-	if (list_empty(&queue->head)) {
-		queue->lag = ns_to_timespec(0);
-		queue->lag_time = now;
-		updated = true;
-		goto out_unlock;
-	}
-
-	list_for_each_entry_safe(fence, next, &queue->head, head) {
-		if (signaled_sequence - fence->sequence > (1 << 30))
-			continue;
-
-		queue->lag = timespec_sub(now, fence->submitted);
-		queue->lag_time = now;
-		updated = true;
-		list_del(&fence->head);
-		kfree(fence);
-	}
-
-out_unlock:
-	spin_unlock(&queue->lock);
-
-	return (updated) ? 0 : -EBUSY;
-}
-
-static struct timespec vmw_timespec_add(struct timespec t1,
-					struct timespec t2)
-{
-	t1.tv_sec += t2.tv_sec;
-	t1.tv_nsec += t2.tv_nsec;
-	if (t1.tv_nsec >= 1000000000L) {
-		t1.tv_sec += 1;
-		t1.tv_nsec -= 1000000000L;
-	}
-
-	return t1;
-}
-
-static struct timespec vmw_fifo_lag(struct vmw_fence_queue *queue)
-{
-	struct timespec now;
-
-	spin_lock(&queue->lock);
-	getrawmonotonic(&now);
-	queue->lag = vmw_timespec_add(queue->lag,
-				      timespec_sub(now, queue->lag_time));
-	queue->lag_time = now;
-	spin_unlock(&queue->lock);
-	return queue->lag;
-}
-
-
-static bool vmw_lag_lt(struct vmw_fence_queue *queue,
-		       uint32_t us)
-{
-	struct timespec lag, cond;
-
-	cond = ns_to_timespec((s64) us * 1000);
-	lag = vmw_fifo_lag(queue);
-	return (timespec_compare(&lag, &cond) < 1);
-}
-
-int vmw_wait_lag(struct vmw_private *dev_priv,
-		 struct vmw_fence_queue *queue, uint32_t us)
-{
-	struct vmw_fence *fence;
-	uint32_t sequence;
-	int ret;
-
-	while (!vmw_lag_lt(queue, us)) {
-		spin_lock(&queue->lock);
-		if (list_empty(&queue->head))
-			sequence = atomic_read(&dev_priv->fence_seq);
-		else {
-			fence = list_first_entry(&queue->head,
-						 struct vmw_fence, head);
-			sequence = fence->sequence;
-		}
-		spin_unlock(&queue->lock);
-
-		ret = vmw_wait_fence(dev_priv, false, sequence, true,
-				     3*HZ);
-
-		if (unlikely(ret != 0))
-			return ret;
-
-		(void) vmw_fence_pull(queue, sequence);
-	}
-	return 0;
-}
-
-
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index aae01b9..3ba9cac 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -127,9 +127,9 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 		 (unsigned int) min,
 		 (unsigned int) fifo->capabilities);
 
-	atomic_set(&dev_priv->fence_seq, dev_priv->last_read_sequence);
-	iowrite32(dev_priv->last_read_sequence, fifo_mem + SVGA_FIFO_FENCE);
-	vmw_fence_queue_init(&fifo->fence_queue);
+	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);
 }
 
@@ -156,7 +156,7 @@ 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);
@@ -166,7 +166,7 @@ void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 		  dev_priv->traces_state);
 
 	mutex_unlock(&dev_priv->hw_mutex);
-	vmw_fence_queue_takedown(&fifo->fence_queue);
+	vmw_marker_queue_takedown(&fifo->marker_queue);
 
 	if (likely(fifo->static_buffer != NULL)) {
 		vfree(fifo->static_buffer);
@@ -447,7 +447,7 @@ void vmw_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
 	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;
@@ -457,16 +457,16 @@ int vmw_fifo_send_fence(struct vmw_private *dev_priv, uint32_t *sequence)
 
 	fm = vmw_fifo_reserve(dev_priv, bytes);
 	if (unlikely(fm == NULL)) {
-		*sequence = atomic_read(&dev_priv->fence_seq);
+		*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 = atomic_add_return(1, &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)) {
 
@@ -483,10 +483,10 @@ 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);
+	iowrite32(*seqno, &cmd_fence->fence);
 	vmw_fifo_commit(dev_priv, bytes);
-	(void) vmw_fence_push(&fifo_state->fence_queue, *sequence);
-	vmw_update_sequence(dev_priv, fifo_state);
+	(void) vmw_marker_push(&fifo_state->marker_queue, *seqno);
+	vmw_update_seqno(dev_priv, fifo_state);
 
 out_err:
 	return ret;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
index e92298a..48701d2 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
@@ -53,7 +53,7 @@ irqreturn_t vmw_irq_handler(DRM_IRQ_ARGS)
 	return IRQ_NONE;
 }
 
-static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t sequence)
+static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t seqno)
 {
 	uint32_t busy;
 
@@ -64,43 +64,43 @@ static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t sequence)
 	return (busy == 0);
 }
 
-void vmw_update_sequence(struct vmw_private *dev_priv,
+void vmw_update_seqno(struct vmw_private *dev_priv,
 			 struct vmw_fifo_state *fifo_state)
 {
 	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
 
-	uint32_t sequence = ioread32(fifo_mem + SVGA_FIFO_FENCE);
+	uint32_t seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
 
-	if (dev_priv->last_read_sequence != sequence) {
-		dev_priv->last_read_sequence = sequence;
-		vmw_fence_pull(&fifo_state->fence_queue, sequence);
+	if (dev_priv->last_read_seqno != seqno) {
+		dev_priv->last_read_seqno = seqno;
+		vmw_marker_pull(&fifo_state->marker_queue, seqno);
 	}
 }
 
-bool vmw_fence_signaled(struct vmw_private *dev_priv,
-			uint32_t sequence)
+bool vmw_seqno_passed(struct vmw_private *dev_priv,
+			 uint32_t seqno)
 {
 	struct vmw_fifo_state *fifo_state;
 	bool ret;
 
-	if (likely(dev_priv->last_read_sequence - sequence < VMW_FENCE_WRAP))
+	if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP))
 		return true;
 
 	fifo_state = &dev_priv->fifo;
-	vmw_update_sequence(dev_priv, fifo_state);
-	if (likely(dev_priv->last_read_sequence - sequence < VMW_FENCE_WRAP))
+	vmw_update_seqno(dev_priv, fifo_state);
+	if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP))
 		return true;
 
 	if (!(fifo_state->capabilities & SVGA_FIFO_CAP_FENCE) &&
-	    vmw_fifo_idle(dev_priv, sequence))
+	    vmw_fifo_idle(dev_priv, seqno))
 		return true;
 
 	/**
-	 * Then check if the sequence is higher than what we've actually
+	 * Then check if the seqno is higher than what we've actually
 	 * emitted. Then the fence is stale and signaled.
 	 */
 
-	ret = ((atomic_read(&dev_priv->fence_seq) - sequence)
+	ret = ((atomic_read(&dev_priv->marker_seq) - seqno)
 	       > VMW_FENCE_WRAP);
 
 	return ret;
@@ -109,7 +109,7 @@ bool vmw_fence_signaled(struct vmw_private *dev_priv,
 int vmw_fallback_wait(struct vmw_private *dev_priv,
 		      bool lazy,
 		      bool fifo_idle,
-		      uint32_t sequence,
+		      uint32_t seqno,
 		      bool interruptible,
 		      unsigned long timeout)
 {
@@ -123,7 +123,7 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
 	DEFINE_WAIT(__wait);
 
 	wait_condition = (fifo_idle) ? &vmw_fifo_idle :
-		&vmw_fence_signaled;
+		&vmw_seqno_passed;
 
 	/**
 	 * Block command submission while waiting for idle.
@@ -131,14 +131,14 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
 
 	if (fifo_idle)
 		down_read(&fifo_state->rwsem);
-	signal_seq = atomic_read(&dev_priv->fence_seq);
+	signal_seq = atomic_read(&dev_priv->marker_seq);
 	ret = 0;
 
 	for (;;) {
 		prepare_to_wait(&dev_priv->fence_queue, &__wait,
 				(interruptible) ?
 				TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE);
-		if (wait_condition(dev_priv, sequence))
+		if (wait_condition(dev_priv, seqno))
 			break;
 		if (time_after_eq(jiffies, end_jiffies)) {
 			DRM_ERROR("SVGA device lockup.\n");
@@ -175,28 +175,28 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
 	return ret;
 }
 
-int vmw_wait_fence(struct vmw_private *dev_priv,
-		   bool lazy, uint32_t sequence,
-		   bool interruptible, unsigned long timeout)
+int vmw_wait_seqno(struct vmw_private *dev_priv,
+		      bool lazy, uint32_t seqno,
+		      bool interruptible, unsigned long timeout)
 {
 	long ret;
 	unsigned long irq_flags;
 	struct vmw_fifo_state *fifo = &dev_priv->fifo;
 
-	if (likely(dev_priv->last_read_sequence - sequence < VMW_FENCE_WRAP))
+	if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP))
 		return 0;
 
-	if (likely(vmw_fence_signaled(dev_priv, sequence)))
+	if (likely(vmw_seqno_passed(dev_priv, seqno)))
 		return 0;
 
 	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
 
 	if (!(fifo->capabilities & SVGA_FIFO_CAP_FENCE))
-		return vmw_fallback_wait(dev_priv, lazy, true, sequence,
+		return vmw_fallback_wait(dev_priv, lazy, true, seqno,
 					 interruptible, timeout);
 
 	if (!(dev_priv->capabilities & SVGA_CAP_IRQMASK))
-		return vmw_fallback_wait(dev_priv, lazy, false, sequence,
+		return vmw_fallback_wait(dev_priv, lazy, false, seqno,
 					 interruptible, timeout);
 
 	mutex_lock(&dev_priv->hw_mutex);
@@ -214,12 +214,12 @@ int vmw_wait_fence(struct vmw_private *dev_priv,
 	if (interruptible)
 		ret = wait_event_interruptible_timeout
 		    (dev_priv->fence_queue,
-		     vmw_fence_signaled(dev_priv, sequence),
+		     vmw_seqno_passed(dev_priv, seqno),
 		     timeout);
 	else
 		ret = wait_event_timeout
 		    (dev_priv->fence_queue,
-		     vmw_fence_signaled(dev_priv, sequence),
+		     vmw_seqno_passed(dev_priv, seqno),
 		     timeout);
 
 	if (unlikely(ret == 0))
@@ -293,5 +293,5 @@ int vmw_fence_wait_ioctl(struct drm_device *dev, void *data,
 		return -EBUSY;
 
 	timeout = (unsigned long)arg->kernel_cookie - timeout;
-	return vmw_wait_fence(vmw_priv(dev), true, arg->sequence, true, timeout);
+	return vmw_wait_seqno(vmw_priv(dev), true, arg->seqno, true, timeout);
 }
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_marker.c b/drivers/gpu/drm/vmwgfx/vmwgfx_marker.c
new file mode 100644
index 0000000..8a8725c
--- /dev/null
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_marker.c
@@ -0,0 +1,171 @@
+/**************************************************************************
+ *
+ * Copyright (C) 2010 VMware, Inc., Palo Alto, CA., USA
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS 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 "vmwgfx_drv.h"
+
+struct vmw_marker {
+	struct list_head head;
+	uint32_t seqno;
+	struct timespec submitted;
+};
+
+void vmw_marker_queue_init(struct vmw_marker_queue *queue)
+{
+	INIT_LIST_HEAD(&queue->head);
+	queue->lag = ns_to_timespec(0);
+	getrawmonotonic(&queue->lag_time);
+	spin_lock_init(&queue->lock);
+}
+
+void vmw_marker_queue_takedown(struct vmw_marker_queue *queue)
+{
+	struct vmw_marker *marker, *next;
+
+	spin_lock(&queue->lock);
+	list_for_each_entry_safe(marker, next, &queue->head, head) {
+		kfree(marker);
+	}
+	spin_unlock(&queue->lock);
+}
+
+int vmw_marker_push(struct vmw_marker_queue *queue,
+		   uint32_t seqno)
+{
+	struct vmw_marker *marker = kmalloc(sizeof(*marker), GFP_KERNEL);
+
+	if (unlikely(!marker))
+		return -ENOMEM;
+
+	marker->seqno = seqno;
+	getrawmonotonic(&marker->submitted);
+	spin_lock(&queue->lock);
+	list_add_tail(&marker->head, &queue->head);
+	spin_unlock(&queue->lock);
+
+	return 0;
+}
+
+int vmw_marker_pull(struct vmw_marker_queue *queue,
+		   uint32_t signaled_seqno)
+{
+	struct vmw_marker *marker, *next;
+	struct timespec now;
+	bool updated = false;
+
+	spin_lock(&queue->lock);
+	getrawmonotonic(&now);
+
+	if (list_empty(&queue->head)) {
+		queue->lag = ns_to_timespec(0);
+		queue->lag_time = now;
+		updated = true;
+		goto out_unlock;
+	}
+
+	list_for_each_entry_safe(marker, next, &queue->head, head) {
+		if (signaled_seqno - marker->seqno > (1 << 30))
+			continue;
+
+		queue->lag = timespec_sub(now, marker->submitted);
+		queue->lag_time = now;
+		updated = true;
+		list_del(&marker->head);
+		kfree(marker);
+	}
+
+out_unlock:
+	spin_unlock(&queue->lock);
+
+	return (updated) ? 0 : -EBUSY;
+}
+
+static struct timespec vmw_timespec_add(struct timespec t1,
+					struct timespec t2)
+{
+	t1.tv_sec += t2.tv_sec;
+	t1.tv_nsec += t2.tv_nsec;
+	if (t1.tv_nsec >= 1000000000L) {
+		t1.tv_sec += 1;
+		t1.tv_nsec -= 1000000000L;
+	}
+
+	return t1;
+}
+
+static struct timespec vmw_fifo_lag(struct vmw_marker_queue *queue)
+{
+	struct timespec now;
+
+	spin_lock(&queue->lock);
+	getrawmonotonic(&now);
+	queue->lag = vmw_timespec_add(queue->lag,
+				      timespec_sub(now, queue->lag_time));
+	queue->lag_time = now;
+	spin_unlock(&queue->lock);
+	return queue->lag;
+}
+
+
+static bool vmw_lag_lt(struct vmw_marker_queue *queue,
+		       uint32_t us)
+{
+	struct timespec lag, cond;
+
+	cond = ns_to_timespec((s64) us * 1000);
+	lag = vmw_fifo_lag(queue);
+	return (timespec_compare(&lag, &cond) < 1);
+}
+
+int vmw_wait_lag(struct vmw_private *dev_priv,
+		 struct vmw_marker_queue *queue, uint32_t us)
+{
+	struct vmw_marker *marker;
+	uint32_t seqno;
+	int ret;
+
+	while (!vmw_lag_lt(queue, us)) {
+		spin_lock(&queue->lock);
+		if (list_empty(&queue->head))
+			seqno = atomic_read(&dev_priv->marker_seq);
+		else {
+			marker = list_first_entry(&queue->head,
+						 struct vmw_marker, head);
+			seqno = marker->seqno;
+		}
+		spin_unlock(&queue->lock);
+
+		ret = vmw_wait_seqno(dev_priv, false, seqno, true,
+					3*HZ);
+
+		if (unlikely(ret != 0))
+			return ret;
+
+		(void) vmw_marker_pull(queue, seqno);
+	}
+	return 0;
+}
diff --git a/include/drm/vmwgfx_drm.h b/include/drm/vmwgfx_drm.h
index 467b80c..c2b3909 100644
--- a/include/drm/vmwgfx_drm.h
+++ b/include/drm/vmwgfx_drm.h
@@ -289,7 +289,7 @@ union drm_vmw_surface_reference_arg {
  * DRM_VMW_EXECBUF
  *
  * Submit a command buffer for execution on the host, and return a
- * fence sequence that when signaled, indicates that the command buffer has
+ * fence seqno that when signaled, indicates that the command buffer has
  * executed.
  */
 
@@ -325,7 +325,7 @@ struct drm_vmw_execbuf_arg {
 /**
  * struct drm_vmw_fence_rep
  *
- * @fence_seq: Fence sequence associated with a command submission.
+ * @fence_seq: Fence seqno associated with a command submission.
  * @error: This member should've been set to -EFAULT on submission.
  * The following actions should be take on completion:
  * error == -EFAULT: Fence communication failed. The host is synchronized.
@@ -432,7 +432,7 @@ struct drm_vmw_unref_dmabuf_arg {
 
 
 struct drm_vmw_fence_wait_arg {
-	uint64_t sequence;
+	uint64_t seqno;
 	uint64_t kernel_cookie;
 	int32_t cookie_valid;
 	int32_t pad64;
-- 
1.7.4.4

[PATCH 6/8] vmwgfx: Make vmw_wait_seqno a bit more readable

[Thomas Hellstrom]

Break out on-demand enabling and disabling of fence irqs to make
the function more readable. Also make dev_priv->fence_queue_waiters an int
instead of an atomic_t since we only manipulate it with dev_priv->hw_mutex
held.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c |    2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h |    2 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_irq.c |   57 +++++++++++++++++++++-------------
 3 files changed, 37 insertions(+), 24 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index c8b5a53..4f65f1e 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -291,7 +291,7 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	mutex_init(&dev_priv->init_mutex);
 	init_waitqueue_head(&dev_priv->fence_queue);
 	init_waitqueue_head(&dev_priv->fifo_queue);
-	atomic_set(&dev_priv->fence_queue_waiters, 0);
+	dev_priv->fence_queue_waiters = 0;
 	atomic_set(&dev_priv->fifo_queue_waiters, 0);
 
 	dev_priv->io_start = pci_resource_start(dev->pdev, 0);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 9c3016b..3018871 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -241,7 +241,7 @@ struct vmw_private {
 	atomic_t marker_seq;
 	wait_queue_head_t fence_queue;
 	wait_queue_head_t fifo_queue;
-	atomic_t fence_queue_waiters;
+	int fence_queue_waiters; /* Protected by hw_mutex */
 	atomic_t fifo_queue_waiters;
 	uint32_t last_read_seqno;
 	spinlock_t irq_lock;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
index 48701d2..13dde06 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
@@ -175,12 +175,43 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
 	return ret;
 }
 
+static void vmw_seqno_waiter_add(struct vmw_private *dev_priv)
+{
+	mutex_lock(&dev_priv->hw_mutex);
+	if (dev_priv->fence_queue_waiters++ == 0) {
+		unsigned long irq_flags;
+
+		spin_lock_irqsave(&dev_priv->irq_lock, irq_flags);
+		outl(SVGA_IRQFLAG_ANY_FENCE,
+		     dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
+		vmw_write(dev_priv, SVGA_REG_IRQMASK,
+			  vmw_read(dev_priv, SVGA_REG_IRQMASK) |
+			  SVGA_IRQFLAG_ANY_FENCE);
+		spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags);
+	}
+	mutex_unlock(&dev_priv->hw_mutex);
+}
+
+static void vmw_seqno_waiter_remove(struct vmw_private *dev_priv)
+{
+	mutex_lock(&dev_priv->hw_mutex);
+	if (--dev_priv->fence_queue_waiters == 0) {
+		unsigned long irq_flags;
+
+		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_ANY_FENCE);
+		spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags);
+	}
+	mutex_unlock(&dev_priv->hw_mutex);
+}
+
 int vmw_wait_seqno(struct vmw_private *dev_priv,
 		      bool lazy, uint32_t seqno,
 		      bool interruptible, unsigned long timeout)
 {
 	long ret;
-	unsigned long irq_flags;
 	struct vmw_fifo_state *fifo = &dev_priv->fifo;
 
 	if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP))
@@ -199,17 +230,7 @@ int vmw_wait_seqno(struct vmw_private *dev_priv,
 		return vmw_fallback_wait(dev_priv, lazy, false, seqno,
 					 interruptible, timeout);
 
-	mutex_lock(&dev_priv->hw_mutex);
-	if (atomic_add_return(1, &dev_priv->fence_queue_waiters) > 0) {
-		spin_lock_irqsave(&dev_priv->irq_lock, irq_flags);
-		outl(SVGA_IRQFLAG_ANY_FENCE,
-		     dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
-		vmw_write(dev_priv, SVGA_REG_IRQMASK,
-			  vmw_read(dev_priv, SVGA_REG_IRQMASK) |
-			  SVGA_IRQFLAG_ANY_FENCE);
-		spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags);
-	}
-	mutex_unlock(&dev_priv->hw_mutex);
+	vmw_seqno_waiter_add(dev_priv);
 
 	if (interruptible)
 		ret = wait_event_interruptible_timeout
@@ -222,21 +243,13 @@ int vmw_wait_seqno(struct vmw_private *dev_priv,
 		     vmw_seqno_passed(dev_priv, seqno),
 		     timeout);
 
+	vmw_seqno_waiter_remove(dev_priv);
+
 	if (unlikely(ret == 0))
 		ret = -EBUSY;
 	else if (likely(ret > 0))
 		ret = 0;
 
-	mutex_lock(&dev_priv->hw_mutex);
-	if (atomic_dec_and_test(&dev_priv->fence_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_ANY_FENCE);
-		spin_unlock_irqrestore(&dev_priv->irq_lock, irq_flags);
-	}
-	mutex_unlock(&dev_priv->hw_mutex);
-
 	return ret;
 }
 
-- 
1.7.4.4

[PATCH 7/8] vmwgfx: Implement fence objects

[Thomas Hellstrom]

Will be needed for queries and drm event-driven throttling.

As a benefit, they help avoid stale user-space fence handles.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
---
 drivers/gpu/drm/vmwgfx/Makefile          |    3 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c   |   26 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c      |   28 ++-
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h      |   16 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c  |  112 +++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_fence.c    |  619 ++++++++++++++++++++++++++++++
 drivers/gpu/drm/vmwgfx/vmwgfx_fence.h    |  105 +++++
 drivers/gpu/drm/vmwgfx/vmwgfx_irq.c      |   35 +--
 drivers/gpu/drm/vmwgfx/vmwgfx_resource.c |    4 -
 include/drm/vmwgfx_drm.h                 |  149 +++++++-
 10 files changed, 1010 insertions(+), 87 deletions(-)
 create mode 100644 drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
 create mode 100644 drivers/gpu/drm/vmwgfx/vmwgfx_fence.h

diff --git a/drivers/gpu/drm/vmwgfx/Makefile b/drivers/gpu/drm/vmwgfx/Makefile
index f41e8b4..7d8e9d5 100644
--- a/drivers/gpu/drm/vmwgfx/Makefile
+++ b/drivers/gpu/drm/vmwgfx/Makefile
@@ -4,6 +4,7 @@ ccflags-y := -Iinclude/drm
 vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \
 	    vmwgfx_fb.o vmwgfx_ioctl.o vmwgfx_resource.o vmwgfx_buffer.o \
 	    vmwgfx_fifo.o vmwgfx_irq.o vmwgfx_ldu.o vmwgfx_ttm_glue.o \
-	    vmwgfx_overlay.o vmwgfx_marker.o vmwgfx_gmrid_manager.o
+	    vmwgfx_overlay.o vmwgfx_marker.o vmwgfx_gmrid_manager.o \
+	    vmwgfx_fence.o
 
 obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index 72d9561..5d665ce 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -274,39 +274,39 @@ static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
 
 static void *vmw_sync_obj_ref(void *sync_obj)
 {
-	return sync_obj;
+
+	return (void *)
+		vmw_fence_obj_reference((struct vmw_fence_obj *) sync_obj);
 }
 
 static void vmw_sync_obj_unref(void **sync_obj)
 {
-	*sync_obj = NULL;
+	vmw_fence_obj_unreference((struct vmw_fence_obj **) sync_obj);
 }
 
 static int vmw_sync_obj_flush(void *sync_obj, void *sync_arg)
 {
-	struct vmw_private *dev_priv = (struct vmw_private *)sync_arg;
-
-	mutex_lock(&dev_priv->hw_mutex);
-	vmw_write(dev_priv, SVGA_REG_SYNC, SVGA_SYNC_GENERIC);
-	mutex_unlock(&dev_priv->hw_mutex);
+	vmw_fence_obj_flush((struct vmw_fence_obj *) sync_obj);
 	return 0;
 }
 
 static bool vmw_sync_obj_signaled(void *sync_obj, void *sync_arg)
 {
-	struct vmw_private *dev_priv = (struct vmw_private *)sync_arg;
-	uint32_t seqno = (unsigned long) sync_obj;
+	unsigned long flags = (unsigned long) sync_arg;
+	return	vmw_fence_obj_signaled((struct vmw_fence_obj *) sync_obj,
+				       (uint32_t) flags);
 
-	return vmw_seqno_passed(dev_priv, seqno);
 }
 
 static int vmw_sync_obj_wait(void *sync_obj, void *sync_arg,
 			     bool lazy, bool interruptible)
 {
-	struct vmw_private *dev_priv = (struct vmw_private *)sync_arg;
-	uint32_t seqno = (unsigned long) sync_obj;
+	unsigned long flags = (unsigned long) sync_arg;
 
-	return vmw_wait_seqno(dev_priv, false, seqno, false, 3*HZ);
+	return vmw_fence_obj_wait((struct vmw_fence_obj *) sync_obj,
+				  (uint32_t) flags,
+				  lazy, interruptible,
+				  VMW_FENCE_WAIT_TIMEOUT);
 }
 
 struct ttm_bo_driver vmw_bo_driver = {
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 4f65f1e..d4829cb 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -82,12 +82,18 @@
 #define DRM_IOCTL_VMW_EXECBUF					\
 	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_EXECBUF,		\
 		struct drm_vmw_execbuf_arg)
+#define DRM_IOCTL_VMW_GET_3D_CAP				\
+	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_GET_3D_CAP,		\
+		 struct drm_vmw_get_3d_cap_arg)
 #define DRM_IOCTL_VMW_FENCE_WAIT				\
 	DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FENCE_WAIT,		\
 		 struct drm_vmw_fence_wait_arg)
-#define DRM_IOCTL_VMW_GET_3D_CAP				\
-	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_GET_3D_CAP,		\
-		struct drm_vmw_get_3d_cap_arg)
+#define DRM_IOCTL_VMW_FENCE_SIGNALED				\
+	DRM_IOWR(DRM_COMMAND_BASE + DRM_VMW_FENCE_SIGNALED,	\
+		 struct drm_vmw_fence_signaled_arg)
+#define DRM_IOCTL_VMW_FENCE_UNREF				\
+	DRM_IOW(DRM_COMMAND_BASE + DRM_VMW_FENCE_UNREF,		\
+		 struct drm_vmw_fence_arg)
 
 /**
  * The core DRM version of this macro doesn't account for
@@ -131,7 +137,12 @@ static struct drm_ioctl_desc vmw_ioctls[] = {
 		      DRM_AUTH | DRM_UNLOCKED),
 	VMW_IOCTL_DEF(VMW_EXECBUF, vmw_execbuf_ioctl,
 		      DRM_AUTH | DRM_UNLOCKED),
-	VMW_IOCTL_DEF(VMW_FENCE_WAIT, vmw_fence_wait_ioctl,
+	VMW_IOCTL_DEF(VMW_FENCE_WAIT, vmw_fence_obj_wait_ioctl,
+		      DRM_AUTH | DRM_UNLOCKED),
+	VMW_IOCTL_DEF(VMW_FENCE_SIGNALED,
+		      vmw_fence_obj_signaled_ioctl,
+		      DRM_AUTH | DRM_UNLOCKED),
+	VMW_IOCTL_DEF(VMW_FENCE_UNREF, vmw_fence_obj_unref_ioctl,
 		      DRM_AUTH | DRM_UNLOCKED),
 	VMW_IOCTL_DEF(VMW_GET_3D_CAP, vmw_get_cap_3d_ioctl,
 		      DRM_AUTH | DRM_UNLOCKED),
@@ -198,12 +209,14 @@ static int vmw_request_device(struct vmw_private *dev_priv)
 		DRM_ERROR("Unable to initialize FIFO.\n");
 		return ret;
 	}
+	vmw_fence_fifo_up(dev_priv->fman);
 
 	return 0;
 }
 
 static void vmw_release_device(struct vmw_private *dev_priv)
 {
+	vmw_fence_fifo_down(dev_priv->fman);
 	vmw_fifo_release(dev_priv, &dev_priv->fifo);
 }
 
@@ -434,6 +447,10 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 			goto out_no_device;
 		}
 	}
+
+	dev_priv->fman = vmw_fence_manager_init(dev_priv);
+	if (unlikely(dev_priv->fman == NULL))
+		goto out_no_fman;
 	ret = vmw_kms_init(dev_priv);
 	if (unlikely(ret != 0))
 		goto out_no_kms;
@@ -475,6 +492,8 @@ out_no_fifo:
 	vmw_overlay_close(dev_priv);
 	vmw_kms_close(dev_priv);
 out_no_kms:
+	vmw_fence_manager_takedown(dev_priv->fman);
+out_no_fman:
 	if (dev_priv->stealth)
 		pci_release_region(dev->pdev, 2);
 	else
@@ -518,6 +537,7 @@ static int vmw_driver_unload(struct drm_device *dev)
 	}
 	vmw_kms_close(dev_priv);
 	vmw_overlay_close(dev_priv);
+	vmw_fence_manager_takedown(dev_priv->fman);
 	if (dev_priv->stealth)
 		pci_release_region(dev->pdev, 2);
 	else
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 3018871..770f063 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -38,6 +38,7 @@
 #include "ttm/ttm_lock.h"
 #include "ttm/ttm_execbuf_util.h"
 #include "ttm/ttm_module.h"
+#include "vmwgfx_fence.h"
 
 #define VMWGFX_DRIVER_DATE "20100927"
 #define VMWGFX_DRIVER_MAJOR 1
@@ -53,6 +54,11 @@
 #define VMW_PL_GMR TTM_PL_PRIV0
 #define VMW_PL_FLAG_GMR TTM_PL_FLAG_PRIV0
 
+#define VMW_RES_CONTEXT ttm_driver_type0
+#define VMW_RES_SURFACE ttm_driver_type1
+#define VMW_RES_STREAM ttm_driver_type2
+#define VMW_RES_FENCE ttm_driver_type3
+
 struct vmw_fpriv {
 	struct drm_master *locked_master;
 	struct ttm_object_file *tfile;
@@ -245,6 +251,7 @@ struct vmw_private {
 	atomic_t fifo_queue_waiters;
 	uint32_t last_read_seqno;
 	spinlock_t irq_lock;
+	struct vmw_fence_manager *fman;
 
 	/*
 	 * Device state
@@ -456,8 +463,6 @@ extern int vmw_irq_postinstall(struct drm_device *dev);
 extern void vmw_irq_uninstall(struct drm_device *dev);
 extern bool vmw_seqno_passed(struct vmw_private *dev_priv,
 				uint32_t seqno);
-extern int vmw_fence_wait_ioctl(struct drm_device *dev, void *data,
-				struct drm_file *file_priv);
 extern int vmw_fallback_wait(struct vmw_private *dev_priv,
 			     bool lazy,
 			     bool fifo_idle,
@@ -466,7 +471,8 @@ extern int vmw_fallback_wait(struct vmw_private *dev_priv,
 			     unsigned long timeout);
 extern void vmw_update_seqno(struct vmw_private *dev_priv,
 				struct vmw_fifo_state *fifo_state);
-
+extern void vmw_seqno_waiter_add(struct vmw_private *dev_priv);
+extern void vmw_seqno_waiter_remove(struct vmw_private *dev_priv);
 
 /**
  * Rudimentary fence-like objects currently used only for throttling -
@@ -572,4 +578,8 @@ static inline struct vmw_dma_buffer *vmw_dmabuf_reference(struct vmw_dma_buffer
 	return NULL;
 }
 
+static inline struct ttm_mem_global *vmw_mem_glob(struct vmw_private *dev_priv)
+{
+	return (struct ttm_mem_global *) dev_priv->mem_global_ref.object;
+}
 #endif
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 4811c6b..0255abb 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -255,7 +255,7 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
 	if (unlikely(cur_validate_node == sw_context->cur_val_buf)) {
 		val_buf = &sw_context->val_bufs[cur_validate_node];
 		val_buf->bo = ttm_bo_reference(bo);
-		val_buf->new_sync_obj_arg = (void *) dev_priv;
+		val_buf->new_sync_obj_arg = (void *) DRM_VMW_FENCE_FLAG_EXEC;
 		list_add_tail(&val_buf->head, &sw_context->validate_nodes);
 		++sw_context->cur_val_buf;
 	}
@@ -320,7 +320,6 @@ static int vmw_cmd_wait_query(struct vmw_private *dev_priv,
 	return 0;
 }
 
-
 static int vmw_cmd_dma(struct vmw_private *dev_priv,
 		       struct vmw_sw_context *sw_context,
 		       SVGA3dCmdHeader *header)
@@ -675,6 +674,50 @@ static int vmw_resize_cmd_bounce(struct vmw_sw_context *sw_context,
 	return 0;
 }
 
+/**
+ * vmw_execbuf_fence_commands - create and submit a command stream fence
+ *
+ * Creates a fence object and submits a command stream marker.
+ * If this fails for some reason, We sync the fifo and return NULL.
+ * It is then safe to fence buffers with a NULL pointer.
+ */
+
+int vmw_execbuf_fence_commands(struct drm_file *file_priv,
+			       struct vmw_private *dev_priv,
+			       struct vmw_fence_obj **p_fence,
+			       uint32_t *p_handle)
+{
+	uint32_t sequence;
+	int ret;
+	bool synced = false;
+
+
+	ret = vmw_fifo_send_fence(dev_priv, &sequence);
+	if (unlikely(ret != 0)) {
+		DRM_ERROR("Fence submission error. Syncing.\n");
+		synced = true;
+	}
+
+	if (p_handle != NULL)
+		ret = vmw_user_fence_create(file_priv, dev_priv->fman,
+					    sequence,
+					    DRM_VMW_FENCE_FLAG_EXEC,
+					    p_fence, p_handle);
+	else
+		ret = vmw_fence_create(dev_priv->fman, sequence,
+				       DRM_VMW_FENCE_FLAG_EXEC,
+				       p_fence);
+
+	if (unlikely(ret != 0 && !synced)) {
+		(void) vmw_fallback_wait(dev_priv, false, false,
+					 sequence, false,
+					 VMW_FENCE_WAIT_TIMEOUT);
+		*p_fence = NULL;
+	}
+
+	return 0;
+}
+
 int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
 		      struct drm_file *file_priv)
 {
@@ -685,9 +728,10 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
 	int ret;
 	void *user_cmd;
 	void *cmd;
-	uint32_t seqno;
 	struct vmw_sw_context *sw_context = &dev_priv->ctx;
 	struct vmw_master *vmaster = vmw_master(file_priv->master);
+	struct vmw_fence_obj *fence;
+	uint32_t handle;
 
 	ret = ttm_read_lock(&vmaster->lock, true);
 	if (unlikely(ret != 0))
@@ -754,34 +798,60 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
 	memcpy(cmd, sw_context->cmd_bounce, arg->command_size);
 	vmw_fifo_commit(dev_priv, arg->command_size);
 
-	ret = vmw_fifo_send_fence(dev_priv, &seqno);
-
-	ttm_eu_fence_buffer_objects(&sw_context->validate_nodes,
-				    (void *)(unsigned long) seqno);
-	vmw_clear_validations(sw_context);
-	mutex_unlock(&dev_priv->cmdbuf_mutex);
-
+	user_fence_rep = (struct drm_vmw_fence_rep __user *)
+		(unsigned long)arg->fence_rep;
+	ret = vmw_execbuf_fence_commands(file_priv, dev_priv,
+					 &fence,
+					 (user_fence_rep) ? &handle : NULL);
 	/*
 	 * This error is harmless, because if fence submission fails,
-	 * vmw_fifo_send_fence will sync.
+	 * vmw_fifo_send_fence will sync. The error will be propagated to
+	 * user-space in @fence_rep
 	 */
 
 	if (ret != 0)
 		DRM_ERROR("Fence submission error. Syncing.\n");
 
-	fence_rep.error = ret;
-	fence_rep.fence_seq = (uint64_t) seqno;
-	fence_rep.pad64 = 0;
+	ttm_eu_fence_buffer_objects(&sw_context->validate_nodes,
+				    (void *) fence);
 
-	user_fence_rep = (struct drm_vmw_fence_rep __user *)
-	    (unsigned long)arg->fence_rep;
+	vmw_clear_validations(sw_context);
+	mutex_unlock(&dev_priv->cmdbuf_mutex);
 
-	/*
-	 * copy_to_user errors will be detected by user space not
-	 * seeing fence_rep::error filled in.
-	 */
+	if (user_fence_rep) {
+		fence_rep.error = ret;
+		fence_rep.handle = handle;
+		fence_rep.seqno = fence->seqno;
+		vmw_update_seqno(dev_priv, &dev_priv->fifo);
+		fence_rep.passed_seqno = dev_priv->last_read_seqno;
+
+		/*
+		 * copy_to_user errors will be detected by user space not
+		 * seeing fence_rep::error filled in. Typically
+		 * user-space would have pre-set that member to -EFAULT.
+		 */
+		ret = copy_to_user(user_fence_rep, &fence_rep,
+				   sizeof(fence_rep));
+
+		/*
+		 * User-space lost the fence object. We need to sync
+		 * and unreference the handle.
+		 */
+		if (unlikely(ret != 0) && (fence_rep.error == 0)) {
+			BUG_ON(fence == NULL);
+
+			ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile,
+						  handle, TTM_REF_USAGE);
+			DRM_ERROR("Fence copy error. Syncing.\n");
+			(void) vmw_fence_obj_wait(fence,
+						  fence->signal_mask,
+						  false, false,
+						  VMW_FENCE_WAIT_TIMEOUT);
+		}
+	}
 
-	ret = copy_to_user(user_fence_rep, &fence_rep, sizeof(fence_rep));
+	if (likely(fence != NULL))
+		vmw_fence_obj_unreference(&fence);
 
 	vmw_kms_cursor_post_execbuf(dev_priv);
 	ttm_read_unlock(&vmaster->lock);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
new file mode 100644
index 0000000..8df679f
--- /dev/null
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
@@ -0,0 +1,619 @@
+/**************************************************************************
+ *
+ * Copyright © 2011 VMware, Inc., Palo Alto, CA., USA
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS 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 "drmP.h"
+#include "vmwgfx_drv.h"
+
+#define VMW_FENCE_WRAP (1 << 31)
+
+struct vmw_fence_manager {
+	int num_fence_objects;
+	struct vmw_private *dev_priv;
+	spinlock_t lock;
+	u32 next_seqno;
+	struct list_head fence_list;
+	struct work_struct work;
+	u32 user_fence_size;
+	u32 fence_size;
+	bool fifo_down;
+	struct list_head cleanup_list;
+};
+
+struct vmw_user_fence {
+	struct ttm_base_object base;
+	struct vmw_fence_obj fence;
+};
+
+/**
+ * vmw_fence_destroy_locked
+ *
+ */
+
+static void vmw_fence_obj_destroy_locked(struct kref *kref)
+{
+	struct vmw_fence_obj *fence =
+		container_of(kref, struct vmw_fence_obj, kref);
+
+	struct vmw_fence_manager *fman = fence->fman;
+	unsigned int num_fences;
+
+	list_del_init(&fence->head);
+	num_fences = --fman->num_fence_objects;
+	spin_unlock_irq(&fman->lock);
+	if (fence->destroy)
+		fence->destroy(fence);
+	else
+		kfree(fence);
+
+	spin_lock_irq(&fman->lock);
+}
+
+
+/**
+ * Execute signal actions on fences recently signaled.
+ * This is done from a workqueue so we don't have to execute
+ * signal actions from atomic context.
+ */
+
+static void vmw_fence_work_func(struct work_struct *work)
+{
+	struct vmw_fence_manager *fman =
+		container_of(work, struct vmw_fence_manager, work);
+	struct list_head list;
+	struct vmw_fence_action *action, *next_action;
+
+	do {
+		INIT_LIST_HEAD(&list);
+		spin_lock_irq(&fman->lock);
+		list_splice_init(&fman->cleanup_list, &list);
+		spin_unlock_irq(&fman->lock);
+
+		if (list_empty(&list))
+			return;
+
+		/*
+		 * At this point, only we should be able to manipulate the
+		 * list heads of the actions we have on the private list.
+		 */
+
+		list_for_each_entry_safe(action, next_action, &list, head) {
+			list_del_init(&action->head);
+			action->cleanup(action);
+		}
+	} while (1);
+}
+
+struct vmw_fence_manager *vmw_fence_manager_init(struct vmw_private *dev_priv)
+{
+	struct vmw_fence_manager *fman = kzalloc(sizeof(*fman), GFP_KERNEL);
+
+	if (unlikely(fman == NULL))
+		return NULL;
+
+	fman->dev_priv = dev_priv;
+	spin_lock_init(&fman->lock);
+	INIT_LIST_HEAD(&fman->fence_list);
+	INIT_LIST_HEAD(&fman->cleanup_list);
+	INIT_WORK(&fman->work, &vmw_fence_work_func);
+	fman->fifo_down = true;
+	fman->user_fence_size = ttm_round_pot(sizeof(struct vmw_user_fence));
+	fman->fence_size = ttm_round_pot(sizeof(struct vmw_fence_obj));
+
+	return fman;
+}
+
+void vmw_fence_manager_takedown(struct vmw_fence_manager *fman)
+{
+	unsigned long irq_flags;
+	bool lists_empty;
+
+	(void) cancel_work_sync(&fman->work);
+
+	spin_lock_irqsave(&fman->lock, irq_flags);
+	lists_empty = list_empty(&fman->fence_list) &&
+		list_empty(&fman->cleanup_list);
+	spin_unlock_irqrestore(&fman->lock, irq_flags);
+
+	BUG_ON(!lists_empty);
+	kfree(fman);
+}
+
+static int vmw_fence_obj_init(struct vmw_fence_manager *fman,
+			      struct vmw_fence_obj *fence,
+			      u32 seqno,
+			      uint32_t mask,
+			      void (*destroy) (struct vmw_fence_obj *fence))
+{
+	unsigned long irq_flags;
+	unsigned int num_fences;
+	int ret = 0;
+
+	fence->seqno = seqno;
+	INIT_LIST_HEAD(&fence->seq_passed_actions);
+	fence->fman = fman;
+	fence->signaled = 0;
+	fence->signal_mask = mask;
+	kref_init(&fence->kref);
+	fence->destroy = destroy;
+	init_waitqueue_head(&fence->queue);
+
+	spin_lock_irqsave(&fman->lock, irq_flags);
+	if (unlikely(fman->fifo_down)) {
+		ret = -EBUSY;
+		goto out_unlock;
+	}
+	list_add_tail(&fence->head, &fman->fence_list);
+	num_fences = ++fman->num_fence_objects;
+
+out_unlock:
+	spin_unlock_irqrestore(&fman->lock, irq_flags);
+	return ret;
+
+}
+
+struct vmw_fence_obj *vmw_fence_obj_reference(struct vmw_fence_obj *fence)
+{
+	kref_get(&fence->kref);
+	return fence;
+}
+
+/**
+ * vmw_fence_obj_unreference
+ *
+ * Note that this function may not be entered with disabled irqs since
+ * it may re-enable them in the destroy function.
+ *
+ */
+void vmw_fence_obj_unreference(struct vmw_fence_obj **fence_p)
+{
+	struct vmw_fence_obj *fence = *fence_p;
+	struct vmw_fence_manager *fman = fence->fman;
+
+	*fence_p = NULL;
+	spin_lock_irq(&fman->lock);
+	BUG_ON(atomic_read(&fence->kref.refcount) == 0);
+	kref_put(&fence->kref, vmw_fence_obj_destroy_locked);
+	spin_unlock_irq(&fman->lock);
+}
+
+void vmw_fences_perform_actions(struct vmw_fence_manager *fman,
+				struct list_head *list)
+{
+	struct vmw_fence_action *action, *next_action;
+
+	list_for_each_entry_safe(action, next_action, list, head) {
+		list_del_init(&action->head);
+		if (action->seq_passed != NULL)
+			action->seq_passed(action);
+
+		/*
+		 * Add the cleanup action to the cleanup list so that
+		 * it will be performed by a worker task.
+		 */
+
+		if (action->cleanup != NULL)
+			list_add_tail(&action->head, &fman->cleanup_list);
+	}
+}
+
+void vmw_fences_update(struct vmw_fence_manager *fman, u32 seqno)
+{
+	unsigned long flags;
+	struct vmw_fence_obj *fence, *next_fence;
+	struct list_head action_list;
+
+	spin_lock_irqsave(&fman->lock, flags);
+	list_for_each_entry_safe(fence, next_fence, &fman->fence_list, head) {
+		if (seqno - fence->seqno < VMW_FENCE_WRAP) {
+			list_del_init(&fence->head);
+			fence->signaled |= DRM_VMW_FENCE_FLAG_EXEC;
+			INIT_LIST_HEAD(&action_list);
+			list_splice_init(&fence->seq_passed_actions,
+					 &action_list);
+			vmw_fences_perform_actions(fman, &action_list);
+			wake_up_all(&fence->queue);
+		}
+
+	}
+	if (!list_empty(&fman->cleanup_list))
+		(void) schedule_work(&fman->work);
+	spin_unlock_irqrestore(&fman->lock, flags);
+}
+
+
+bool vmw_fence_obj_signaled(struct vmw_fence_obj *fence,
+			    uint32_t flags)
+{
+	struct vmw_fence_manager *fman = fence->fman;
+	unsigned long irq_flags;
+	uint32_t signaled;
+
+	spin_lock_irqsave(&fman->lock, irq_flags);
+	signaled = fence->signaled;
+	spin_unlock_irqrestore(&fman->lock, irq_flags);
+
+	flags &= fence->signal_mask;
+	if ((signaled & flags) == flags)
+		return 1;
+
+	if ((signaled & DRM_VMW_FENCE_FLAG_EXEC) == 0) {
+		struct vmw_private *dev_priv = fman->dev_priv;
+		__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+		u32 seqno;
+
+		seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
+		vmw_fences_update(fman, seqno);
+	}
+
+	spin_lock_irqsave(&fman->lock, irq_flags);
+	signaled = fence->signaled;
+	spin_unlock_irqrestore(&fman->lock, irq_flags);
+
+	return ((signaled & flags) == flags);
+}
+
+int vmw_fence_obj_wait(struct vmw_fence_obj *fence,
+		       uint32_t flags, bool lazy,
+		       bool interruptible, unsigned long timeout)
+{
+	struct vmw_private *dev_priv = fence->fman->dev_priv;
+	long ret;
+
+	if (likely(vmw_fence_obj_signaled(fence, flags)))
+		return 0;
+
+	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
+	vmw_seqno_waiter_add(dev_priv);
+
+	if (interruptible)
+		ret = wait_event_interruptible_timeout
+			(fence->queue,
+			 vmw_fence_obj_signaled(fence, flags),
+			 timeout);
+	else
+		ret = wait_event_timeout
+			(fence->queue,
+			 vmw_fence_obj_signaled(fence, flags),
+			 timeout);
+
+	vmw_seqno_waiter_remove(dev_priv);
+
+	if (unlikely(ret == 0))
+		ret = -EBUSY;
+	else if (likely(ret > 0))
+		ret = 0;
+
+	return ret;
+}
+
+void vmw_fence_obj_flush(struct vmw_fence_obj *fence)
+{
+	struct vmw_private *dev_priv = fence->fman->dev_priv;
+
+	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
+}
+
+static void vmw_fence_destroy(struct vmw_fence_obj *fence)
+{
+	struct vmw_fence_manager *fman = fence->fman;
+
+	kfree(fence);
+	/*
+	 * Free kernel space accounting.
+	 */
+	ttm_mem_global_free(vmw_mem_glob(fman->dev_priv),
+			    fman->fence_size);
+}
+
+int vmw_fence_create(struct vmw_fence_manager *fman,
+		     uint32_t seqno,
+		     uint32_t mask,
+		     struct vmw_fence_obj **p_fence)
+{
+	struct ttm_mem_global *mem_glob = vmw_mem_glob(fman->dev_priv);
+	struct vmw_fence_obj *fence;
+	int ret;
+
+	ret = ttm_mem_global_alloc(mem_glob, fman->fence_size,
+				   false, false);
+	if (unlikely(ret != 0))
+		return ret;
+
+	fence = kzalloc(sizeof(*fence), GFP_KERNEL);
+	if (unlikely(fence == NULL)) {
+		ret = -ENOMEM;
+		goto out_no_object;
+	}
+
+	ret = vmw_fence_obj_init(fman, fence, seqno, mask,
+				 vmw_fence_destroy);
+	if (unlikely(ret != 0))
+		goto out_err_init;
+
+	*p_fence = fence;
+	return 0;
+
+out_err_init:
+	kfree(fence);
+out_no_object:
+	ttm_mem_global_free(mem_glob, fman->fence_size);
+	return ret;
+}
+
+
+static void vmw_user_fence_destroy(struct vmw_fence_obj *fence)
+{
+	struct vmw_user_fence *ufence =
+		container_of(fence, struct vmw_user_fence, fence);
+	struct vmw_fence_manager *fman = fence->fman;
+
+	kfree(ufence);
+	/*
+	 * Free kernel space accounting.
+	 */
+	ttm_mem_global_free(vmw_mem_glob(fman->dev_priv),
+			    fman->user_fence_size);
+}
+
+static void vmw_user_fence_base_release(struct ttm_base_object **p_base)
+{
+	struct ttm_base_object *base = *p_base;
+	struct vmw_user_fence *ufence =
+		container_of(base, struct vmw_user_fence, base);
+	struct vmw_fence_obj *fence = &ufence->fence;
+
+	*p_base = NULL;
+	vmw_fence_obj_unreference(&fence);
+}
+
+int vmw_user_fence_create(struct drm_file *file_priv,
+			  struct vmw_fence_manager *fman,
+			  uint32_t seqno,
+			  uint32_t mask,
+			  struct vmw_fence_obj **p_fence,
+			  uint32_t *p_handle)
+{
+	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+	struct vmw_user_fence *ufence;
+	struct vmw_fence_obj *tmp;
+	struct ttm_mem_global *mem_glob = vmw_mem_glob(fman->dev_priv);
+	int ret;
+
+	/*
+	 * Kernel memory space accounting, since this object may
+	 * be created by a user-space request.
+	 */
+
+	ret = ttm_mem_global_alloc(mem_glob, fman->user_fence_size,
+				   false, false);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ufence = kzalloc(sizeof(*ufence), GFP_KERNEL);
+	if (unlikely(ufence == NULL)) {
+		ret = -ENOMEM;
+		goto out_no_object;
+	}
+
+	ret = vmw_fence_obj_init(fman, &ufence->fence, seqno,
+				 mask, vmw_user_fence_destroy);
+	if (unlikely(ret != 0)) {
+		kfree(ufence);
+		goto out_no_object;
+	}
+
+	/*
+	 * The base object holds a reference which is freed in
+	 * vmw_user_fence_base_release.
+	 */
+	tmp = vmw_fence_obj_reference(&ufence->fence);
+	ret = ttm_base_object_init(tfile, &ufence->base, false,
+				   VMW_RES_FENCE,
+				   &vmw_user_fence_base_release, NULL);
+
+
+	if (unlikely(ret != 0)) {
+		/*
+		 * Free the base object's reference
+		 */
+		vmw_fence_obj_unreference(&tmp);
+		goto out_err;
+	}
+
+	*p_fence = &ufence->fence;
+	*p_handle = ufence->base.hash.key;
+
+	return 0;
+out_err:
+	tmp = &ufence->fence;
+	vmw_fence_obj_unreference(&tmp);
+out_no_object:
+	ttm_mem_global_free(mem_glob, fman->user_fence_size);
+	return ret;
+}
+
+
+/**
+ * vmw_fence_fifo_down - signal all unsignaled fence objects.
+ */
+
+void vmw_fence_fifo_down(struct vmw_fence_manager *fman)
+{
+	unsigned long irq_flags;
+	struct list_head action_list;
+	int ret;
+
+	/*
+	 * The list may be altered while we traverse it, so always
+	 * restart when we've released the fman->lock.
+	 */
+
+	spin_lock_irqsave(&fman->lock, irq_flags);
+	fman->fifo_down = true;
+	while (!list_empty(&fman->fence_list)) {
+		struct vmw_fence_obj *fence =
+			list_entry(fman->fence_list.prev, struct vmw_fence_obj,
+				   head);
+		kref_get(&fence->kref);
+		spin_unlock_irq(&fman->lock);
+
+		ret = vmw_fence_obj_wait(fence, fence->signal_mask,
+					 false, false,
+					 VMW_FENCE_WAIT_TIMEOUT);
+
+		if (unlikely(ret != 0)) {
+			list_del_init(&fence->head);
+			fence->signaled |= DRM_VMW_FENCE_FLAG_EXEC;
+			INIT_LIST_HEAD(&action_list);
+			list_splice_init(&fence->seq_passed_actions,
+					 &action_list);
+			vmw_fences_perform_actions(fman, &action_list);
+			wake_up_all(&fence->queue);
+		}
+
+		spin_lock_irq(&fman->lock);
+
+		BUG_ON(!list_empty(&fence->head));
+		kref_put(&fence->kref, vmw_fence_obj_destroy_locked);
+	}
+	spin_unlock_irqrestore(&fman->lock, irq_flags);
+}
+
+void vmw_fence_fifo_up(struct vmw_fence_manager *fman)
+{
+	unsigned long irq_flags;
+
+	spin_lock_irqsave(&fman->lock, irq_flags);
+	fman->fifo_down = false;
+	spin_unlock_irqrestore(&fman->lock, irq_flags);
+}
+
+
+int vmw_fence_obj_wait_ioctl(struct drm_device *dev, void *data,
+			     struct drm_file *file_priv)
+{
+	struct drm_vmw_fence_wait_arg *arg =
+	    (struct drm_vmw_fence_wait_arg *)data;
+	unsigned long timeout;
+	struct ttm_base_object *base;
+	struct vmw_fence_obj *fence;
+	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+	int ret;
+	uint64_t wait_timeout = ((uint64_t)arg->timeout_us * HZ);
+
+	/*
+	 * 64-bit division not present on 32-bit systems, so do an
+	 * approximation. (Divide by 1000000).
+	 */
+
+	wait_timeout = (wait_timeout >> 20) + (wait_timeout >> 24) -
+	  (wait_timeout >> 26);
+
+	if (!arg->cookie_valid) {
+		arg->cookie_valid = 1;
+		arg->kernel_cookie = jiffies + wait_timeout;
+	}
+
+	base = ttm_base_object_lookup(tfile, arg->handle);
+	if (unlikely(base == NULL)) {
+		printk(KERN_ERR "Wait invalid fence object handle "
+		       "0x%08lx.\n",
+		       (unsigned long)arg->handle);
+		return -EINVAL;
+	}
+
+	fence = &(container_of(base, struct vmw_user_fence, base)->fence);
+
+	timeout = jiffies;
+	if (time_after_eq(timeout, (unsigned long)arg->kernel_cookie)) {
+		ret = ((vmw_fence_obj_signaled(fence, arg->flags)) ?
+		       0 : -EBUSY);
+		goto out;
+	}
+
+	timeout = (unsigned long)arg->kernel_cookie - timeout;
+
+	ret = vmw_fence_obj_wait(fence, arg->flags, arg->lazy, true, timeout);
+
+out:
+	ttm_base_object_unref(&base);
+
+	/*
+	 * Optionally unref the fence object.
+	 */
+
+	if (ret == 0 && (arg->wait_options & DRM_VMW_WAIT_OPTION_UNREF))
+		return ttm_ref_object_base_unref(tfile, arg->handle,
+						 TTM_REF_USAGE);
+	return ret;
+}
+
+int vmw_fence_obj_signaled_ioctl(struct drm_device *dev, void *data,
+				 struct drm_file *file_priv)
+{
+	struct drm_vmw_fence_signaled_arg *arg =
+		(struct drm_vmw_fence_signaled_arg *) data;
+	struct ttm_base_object *base;
+	struct vmw_fence_obj *fence;
+	struct vmw_fence_manager *fman;
+	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+	struct vmw_private *dev_priv = vmw_priv(dev);
+
+	base = ttm_base_object_lookup(tfile, arg->handle);
+	if (unlikely(base == NULL)) {
+		printk(KERN_ERR "Fence signaled invalid fence object handle "
+		       "0x%08lx.\n",
+		       (unsigned long)arg->handle);
+		return -EINVAL;
+	}
+
+	fence = &(container_of(base, struct vmw_user_fence, base)->fence);
+	fman = fence->fman;
+
+	arg->signaled = vmw_fence_obj_signaled(fence, arg->flags);
+	spin_lock_irq(&fman->lock);
+
+	arg->signaled_flags = fence->signaled;
+	arg->passed_seqno = dev_priv->last_read_seqno;
+	spin_unlock_irq(&fman->lock);
+
+	ttm_base_object_unref(&base);
+
+	return 0;
+}
+
+
+int vmw_fence_obj_unref_ioctl(struct drm_device *dev, void *data,
+			      struct drm_file *file_priv)
+{
+	struct drm_vmw_fence_arg *arg =
+		(struct drm_vmw_fence_arg *) data;
+
+	return ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile,
+					 arg->handle,
+					 TTM_REF_USAGE);
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fence.h b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.h
new file mode 100644
index 0000000..9307406
--- /dev/null
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fence.h
@@ -0,0 +1,105 @@
+/**************************************************************************
+ *
+ * Copyright © 2011 VMware, Inc., Palo Alto, CA., USA
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS 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.
+ *
+ **************************************************************************/
+
+#ifndef _VMWGFX_FENCE_H_
+
+#define VMW_FENCE_WAIT_TIMEOUT (5*HZ)
+
+struct vmw_private;
+
+struct vmw_fence_manager;
+
+/**
+ *
+ *
+ */
+struct vmw_fence_action {
+	struct list_head head;
+	void (*seq_passed) (struct vmw_fence_action *action);
+	void (*cleanup) (struct vmw_fence_action *action);
+};
+
+struct vmw_fence_obj {
+	struct kref kref;
+	u32 seqno;
+
+	struct vmw_fence_manager *fman;
+	struct list_head head;
+	uint32_t signaled;
+	uint32_t signal_mask;
+	struct list_head seq_passed_actions;
+	void (*destroy)(struct vmw_fence_obj *fence);
+	wait_queue_head_t queue;
+};
+
+extern struct vmw_fence_manager *
+vmw_fence_manager_init(struct vmw_private *dev_priv);
+
+extern void vmw_fence_manager_takedown(struct vmw_fence_manager *fman);
+
+extern void vmw_fence_obj_unreference(struct vmw_fence_obj **fence_p);
+
+extern struct vmw_fence_obj *
+vmw_fence_obj_reference(struct vmw_fence_obj *fence);
+
+extern void vmw_fences_update(struct vmw_fence_manager *fman,
+			      u32 sequence);
+
+extern bool vmw_fence_obj_signaled(struct vmw_fence_obj *fence,
+				   uint32_t flags);
+
+extern int vmw_fence_obj_wait(struct vmw_fence_obj *fence, uint32_t flags,
+			      bool lazy,
+			      bool interruptible, unsigned long timeout);
+
+extern void vmw_fence_obj_flush(struct vmw_fence_obj *fence);
+
+extern int vmw_fence_create(struct vmw_fence_manager *fman,
+			    uint32_t seqno,
+			    uint32_t mask,
+			    struct vmw_fence_obj **p_fence);
+
+extern int vmw_user_fence_create(struct drm_file *file_priv,
+				 struct vmw_fence_manager *fman,
+				 uint32_t sequence,
+				 uint32_t mask,
+				 struct vmw_fence_obj **p_fence,
+				 uint32_t *p_handle);
+
+extern void vmw_fence_fifo_up(struct vmw_fence_manager *fman);
+
+extern void vmw_fence_fifo_down(struct vmw_fence_manager *fman);
+
+extern int vmw_fence_obj_wait_ioctl(struct drm_device *dev, void *data,
+				    struct drm_file *file_priv);
+
+extern int vmw_fence_obj_signaled_ioctl(struct drm_device *dev, void *data,
+					struct drm_file *file_priv);
+
+extern int vmw_fence_obj_unref_ioctl(struct drm_device *dev, void *data,
+				     struct drm_file *file_priv);
+#endif /* _VMWGFX_FENCE_H_ */
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
index 13dde06..a005292 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_irq.c
@@ -40,8 +40,13 @@ irqreturn_t vmw_irq_handler(DRM_IRQ_ARGS)
 	status = inl(dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
 	spin_unlock(&dev_priv->irq_lock);
 
-	if (status & SVGA_IRQFLAG_ANY_FENCE)
+	if (status & SVGA_IRQFLAG_ANY_FENCE) {
+		__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
+		uint32_t seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
+
+		vmw_fences_update(dev_priv->fman, seqno);
 		wake_up_all(&dev_priv->fence_queue);
+	}
 	if (status & SVGA_IRQFLAG_FIFO_PROGRESS)
 		wake_up_all(&dev_priv->fifo_queue);
 
@@ -68,12 +73,12 @@ void vmw_update_seqno(struct vmw_private *dev_priv,
 			 struct vmw_fifo_state *fifo_state)
 {
 	__le32 __iomem *fifo_mem = dev_priv->mmio_virt;
-
 	uint32_t seqno = ioread32(fifo_mem + SVGA_FIFO_FENCE);
 
 	if (dev_priv->last_read_seqno != seqno) {
 		dev_priv->last_read_seqno = seqno;
 		vmw_marker_pull(&fifo_state->marker_queue, seqno);
+		vmw_fences_update(dev_priv->fman, seqno);
 	}
 }
 
@@ -175,7 +180,7 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
 	return ret;
 }
 
-static void vmw_seqno_waiter_add(struct vmw_private *dev_priv)
+void vmw_seqno_waiter_add(struct vmw_private *dev_priv)
 {
 	mutex_lock(&dev_priv->hw_mutex);
 	if (dev_priv->fence_queue_waiters++ == 0) {
@@ -192,7 +197,7 @@ static void vmw_seqno_waiter_add(struct vmw_private *dev_priv)
 	mutex_unlock(&dev_priv->hw_mutex);
 }
 
-static void vmw_seqno_waiter_remove(struct vmw_private *dev_priv)
+void vmw_seqno_waiter_remove(struct vmw_private *dev_priv)
 {
 	mutex_lock(&dev_priv->hw_mutex);
 	if (--dev_priv->fence_queue_waiters == 0) {
@@ -286,25 +291,3 @@ void vmw_irq_uninstall(struct drm_device *dev)
 	status = inl(dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
 	outl(status, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
 }
-
-#define VMW_FENCE_WAIT_TIMEOUT 3*HZ;
-
-int vmw_fence_wait_ioctl(struct drm_device *dev, void *data,
-			 struct drm_file *file_priv)
-{
-	struct drm_vmw_fence_wait_arg *arg =
-	    (struct drm_vmw_fence_wait_arg *)data;
-	unsigned long timeout;
-
-	if (!arg->cookie_valid) {
-		arg->cookie_valid = 1;
-		arg->kernel_cookie = jiffies + VMW_FENCE_WAIT_TIMEOUT;
-	}
-
-	timeout = jiffies;
-	if (time_after_eq(timeout, (unsigned long)arg->kernel_cookie))
-		return -EBUSY;
-
-	timeout = (unsigned long)arg->kernel_cookie - timeout;
-	return vmw_wait_seqno(vmw_priv(dev), true, arg->seqno, true, timeout);
-}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index 4b53803..c1b6ffd 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
@@ -31,10 +31,6 @@
 #include "ttm/ttm_placement.h"
 #include "drmP.h"
 
-#define VMW_RES_CONTEXT ttm_driver_type0
-#define VMW_RES_SURFACE ttm_driver_type1
-#define VMW_RES_STREAM ttm_driver_type2
-
 struct vmw_user_context {
 	struct ttm_base_object base;
 	struct vmw_resource res;
diff --git a/include/drm/vmwgfx_drm.h b/include/drm/vmwgfx_drm.h
index c2b3909..763a7a3 100644
--- a/include/drm/vmwgfx_drm.h
+++ b/include/drm/vmwgfx_drm.h
@@ -48,8 +48,12 @@
 #define DRM_VMW_UNREF_SURFACE        10
 #define DRM_VMW_REF_SURFACE          11
 #define DRM_VMW_EXECBUF              12
-#define DRM_VMW_FENCE_WAIT           13
-#define DRM_VMW_GET_3D_CAP           14
+#define DRM_VMW_GET_3D_CAP           13
+#define DRM_VMW_FENCE_WAIT           14
+#define DRM_VMW_FENCE_SIGNALED       15
+#define DRM_VMW_FENCE_UNREF          16
+#define DRM_VMW_FENCE_EVENT          17
+
 
 /*************************************************************************/
 /**
@@ -318,14 +322,23 @@ struct drm_vmw_execbuf_arg {
 	uint32_t command_size;
 	uint32_t throttle_us;
 	uint64_t fence_rep;
-	 uint32_t version;
-	 uint32_t flags;
+	uint32_t version;
+	uint32_t flags;
 };
 
 /**
  * struct drm_vmw_fence_rep
  *
- * @fence_seq: Fence seqno associated with a command submission.
+ * @handle: Fence object handle for fence associated with a command submission.
+ * @mask: Fence flags relevant for this fence object.
+ * @seqno: Fence sequence number in fifo. A fence object with a lower
+ * seqno will signal the EXEC flag before a fence object with a higher
+ * seqno. This can be used by user-space to avoid kernel calls to determine
+ * whether a fence has signaled the EXEC flag. Note that @seqno will
+ * wrap at 32-bit.
+ * @passed_seqno: The highest seqno number processed by the hardware
+ * so far. This can be used to mark user-space fence objects as signaled, and
+ * to determine whether a fence seqno might be stale.
  * @error: This member should've been set to -EFAULT on submission.
  * The following actions should be take on completion:
  * error == -EFAULT: Fence communication failed. The host is synchronized.
@@ -339,9 +352,12 @@ struct drm_vmw_execbuf_arg {
  */
 
 struct drm_vmw_fence_rep {
-	uint64_t fence_seq;
-	int32_t error;
+	uint32_t handle;
+	uint32_t mask;
+	uint32_t seqno;
+	uint32_t passed_seqno;
 	uint32_t pad64;
+	int32_t error;
 };
 
 /*************************************************************************/
@@ -430,14 +446,6 @@ struct drm_vmw_unref_dmabuf_arg {
 	uint32_t pad64;
 };
 
-
-struct drm_vmw_fence_wait_arg {
-	uint64_t seqno;
-	uint64_t kernel_cookie;
-	int32_t cookie_valid;
-	int32_t pad64;
-};
-
 /*************************************************************************/
 /**
  * DRM_VMW_CONTROL_STREAM - Control overlays, aka streams.
@@ -559,6 +567,7 @@ struct drm_vmw_stream_arg {
  * Return a single stream that was claimed by this process. Also makes
  * sure that the stream has been stopped.
  */
+
 /*************************************************************************/
 /**
  * DRM_VMW_GET_3D_CAP
@@ -607,4 +616,114 @@ struct drm_vmw_update_layout_arg {
 	uint64_t rects;
 };
 
+
+/*************************************************************************/
+/**
+ * DRM_VMW_FENCE_WAIT
+ *
+ * Waits for a fence object to signal. The wait is interruptible, so that
+ * signals may be delivered during the interrupt. The wait may timeout,
+ * in which case the calls returns -EBUSY. If the wait is restarted,
+ * that is restarting without resetting @cookie_valid to zero,
+ * the timeout is computed from the first call.
+ *
+ * The flags argument to the DRM_VMW_FENCE_WAIT ioctl indicates what to wait
+ * on:
+ * DRM_VMW_FENCE_FLAG_EXEC: All commands ahead of the fence in the command
+ * stream
+ * have executed.
+ * DRM_VMW_FENCE_FLAG_QUERY: All query results resulting from query finish
+ * commands
+ * in the buffer given to the EXECBUF ioctl returning the fence object handle
+ * are available to user-space.
+ *
+ * DRM_VMW_WAIT_OPTION_UNREF: If this wait option is given, and the
+ * fenc wait ioctl returns 0, the fence object has been unreferenced after
+ * the wait.
+ */
+
+#define DRM_VMW_FENCE_FLAG_EXEC   (1 << 0)
+#define DRM_VMW_FENCE_FLAG_QUERY  (1 << 1)
+
+#define DRM_VMW_WAIT_OPTION_UNREF (1 << 0)
+
+/**
+ * struct drm_vmw_fence_wait_arg
+ *
+ * @handle: Fence object handle as returned by the DRM_VMW_EXECBUF ioctl.
+ * @cookie_valid: Must be reset to 0 on first call. Left alone on restart.
+ * @kernel_cookie: Set to 0 on first call. Left alone on restart.
+ * @timeout_us: Wait timeout in microseconds. 0 for indefinite timeout.
+ * @lazy: Set to 1 if timing is not critical. Allow more than a kernel tick
+ * before returning.
+ * @flags: Fence flags to wait on.
+ * @wait_options: Options that control the behaviour of the wait ioctl.
+ *
+ * Input argument to the DRM_VMW_FENCE_WAIT ioctl.
+ */
+
+struct drm_vmw_fence_wait_arg {
+	uint32_t handle;
+	int32_t  cookie_valid;
+	uint64_t kernel_cookie;
+	uint64_t timeout_us;
+	int32_t lazy;
+	int32_t flags;
+	int32_t wait_options;
+	int32_t pad64;
+};
+
+/*************************************************************************/
+/**
+ * DRM_VMW_FENCE_SIGNALED
+ *
+ * Checks if a fence object is signaled..
+ */
+
+/**
+ * struct drm_vmw_fence_signaled_arg
+ *
+ * @handle: Fence object handle as returned by the DRM_VMW_EXECBUF ioctl.
+ * @flags: Fence object flags input to DRM_VMW_FENCE_SIGNALED ioctl
+ * @signaled: Out: Flags signaled.
+ * @sequence: Out: Highest sequence passed so far. Can be used to signal the
+ * EXEC flag of user-space fence objects.
+ *
+ * Input/Output argument to the DRM_VMW_FENCE_SIGNALED and DRM_VMW_FENCE_UNREF
+ * ioctls.
+ */
+
+struct drm_vmw_fence_signaled_arg {
+	 uint32_t handle;
+	 uint32_t flags;
+	 int32_t signaled;
+	 uint32_t passed_seqno;
+	 uint32_t signaled_flags;
+	 uint32_t pad64;
+};
+
+/*************************************************************************/
+/**
+ * DRM_VMW_FENCE_UNREF
+ *
+ * Unreferences a fence object, and causes it to be destroyed if there are no
+ * other references to it.
+ *
+ */
+
+/**
+ * struct drm_vmw_fence_arg
+ *
+ * @handle: Fence object handle as returned by the DRM_VMW_EXECBUF ioctl.
+ *
+ * Input/Output argument to the DRM_VMW_FENCE_UNREF ioctl..
+ */
+
+struct drm_vmw_fence_arg {
+	 uint32_t handle;
+	 uint32_t pad64;
+};
+
+
+
 #endif
-- 
1.7.4.4

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 8/8] vmwgfx: Bump major

[Thomas Hellstrom]

This bumps driver major version as a result of previous incompatible
interface changes.

In addition, a leftover command definition is removed from the
vmwgfx_drm.h header.

Also a strict version check is enforced on the exebuf ioctl.

This is intended to be the last major bump before exiting staging.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h     |    6 ++--
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c |   14 +++++++++
 include/drm/vmwgfx_drm.h                |   46 +------------------------------
 3 files changed, 18 insertions(+), 48 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 770f063..564a815 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -40,9 +40,9 @@
 #include "ttm/ttm_module.h"
 #include "vmwgfx_fence.h"
 
-#define VMWGFX_DRIVER_DATE "20100927"
-#define VMWGFX_DRIVER_MAJOR 1
-#define VMWGFX_DRIVER_MINOR 4
+#define VMWGFX_DRIVER_DATE "20110901"
+#define VMWGFX_DRIVER_MAJOR 2
+#define VMWGFX_DRIVER_MINOR 0
 #define VMWGFX_DRIVER_PATCHLEVEL 0
 #define VMWGFX_FILE_PAGE_OFFSET 0x00100000
 #define VMWGFX_FIFO_STATIC_SIZE (1024*1024)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 0255abb..df1c7c5 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -733,6 +733,20 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
 	struct vmw_fence_obj *fence;
 	uint32_t handle;
 
+	/*
+	 * This will allow us to extend the ioctl argument while
+	 * maintaining backwards compatibility:
+	 * We take different code paths depending on the value of
+	 * arg->version.
+	 */
+
+	if (unlikely(arg->version != DRM_VMW_EXECBUF_VERSION)) {
+		DRM_ERROR("Incorrect execbuf version.\n");
+		DRM_ERROR("You're running outdated experimental "
+			  "vmwgfx user-space drivers.");
+		return -EINVAL;
+	}
+
 	ret = ttm_read_lock(&vmaster->lock, true);
 	if (unlikely(ret != 0))
 		return ret;
diff --git a/include/drm/vmwgfx_drm.h b/include/drm/vmwgfx_drm.h
index 763a7a3..29cd9cf 100644
--- a/include/drm/vmwgfx_drm.h
+++ b/include/drm/vmwgfx_drm.h
@@ -31,7 +31,6 @@
 #define DRM_VMW_MAX_SURFACE_FACES 6
 #define DRM_VMW_MAX_MIP_LEVELS 24
 
-#define DRM_VMW_EXT_NAME_LEN 128
 
 #define DRM_VMW_GET_PARAM            0
 #define DRM_VMW_ALLOC_DMABUF         1
@@ -92,49 +91,6 @@ struct drm_vmw_getparam_arg {
 
 /*************************************************************************/
 /**
- * DRM_VMW_EXTENSION - Query device extensions.
- */
-
-/**
- * struct drm_vmw_extension_rep
- *
- * @exists: The queried extension exists.
- * @driver_ioctl_offset: Ioctl number of the first ioctl in the extension.
- * @driver_sarea_offset: Offset to any space in the DRI SAREA
- * used by the extension.
- * @major: Major version number of the extension.
- * @minor: Minor version number of the extension.
- * @pl: Patch level version number of the extension.
- *
- * Output argument to the DRM_VMW_EXTENSION Ioctl.
- */
-
-struct drm_vmw_extension_rep {
-	int32_t exists;
-	uint32_t driver_ioctl_offset;
-	uint32_t driver_sarea_offset;
-	uint32_t major;
-	uint32_t minor;
-	uint32_t pl;
-	uint32_t pad64;
-};
-
-/**
- * union drm_vmw_extension_arg
- *
- * @extension - Ascii name of the extension to be queried. //In
- * @rep - Reply as defined above. //Out
- *
- * Argument to the DRM_VMW_EXTENSION Ioctl.
- */
-
-union drm_vmw_extension_arg {
-	char extension[DRM_VMW_EXT_NAME_LEN];
-	struct drm_vmw_extension_rep rep;
-};
-
-/*************************************************************************/
-/**
  * DRM_VMW_CREATE_CONTEXT - Create a host context.
  *
  * Allocates a device unique context id, and queues a create context command
@@ -315,7 +271,7 @@ union drm_vmw_surface_reference_arg {
  * Argument to the DRM_VMW_EXECBUF Ioctl.
  */
 
-#define DRM_VMW_EXECBUF_VERSION 0
+#define DRM_VMW_EXECBUF_VERSION 1
 
 struct drm_vmw_execbuf_arg {
 	uint64_t commands;
-- 
1.7.4.4