drm/vmwgfx: Fix fencing on SVGAv3

Port of the vmwgfx to SVGAv3 lacked support for fencing. SVGAv3 removed
FIFO's and replaced them with command buffers and extra registers.
The initial version of SVGAv3 lacked support for most advanced features
(e.g. 3D) which made fences unnecessary. That is no longer the case,
especially as 3D support is being turned on.

Switch from FIFO commands and capabilities to command buffers and extra
registers to enable fences on SVGAv3.

Fixes: 2cd80db ("drm/vmwgfx: Add basic support for SVGA3")
Signed-off-by: Zack Rusin <[email protected]>
Reviewed-by: Martin Krastev <[email protected]>
Reviewed-by: Maaz Mombasawala <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

drivers/gpu/drm/vmwgfx/vmwgfx_cmd.c

@@ -528,7 +528,7 @@ int vmw_cmd_send_fence(struct vmw_private *dev_priv, uint32_t *seqno)
  *seqno = atomic_add_return(1, &dev_priv->marker_seq);
  } while (*seqno == 0);
 
- if (!(vmw_fifo_caps(dev_priv) & SVGA_FIFO_CAP_FENCE)) {
+ if (!vmw_has_fences(dev_priv)) {
 
  /*
  * Don't request hardware to send a fence. The

drivers/gpu/drm/vmwgfx/vmwgfx_drv.h

@@ -1679,4 +1679,12 @@ static inline void vmw_irq_status_write(struct vmw_private *vmw,
  outl(status, vmw->io_start + SVGA_IRQSTATUS_PORT);
 }
 
+static inline bool vmw_has_fences(struct vmw_private *vmw)
+{
+ if ((vmw->capabilities & (SVGA_CAP_COMMAND_BUFFERS |
+ SVGA_CAP_CMD_BUFFERS_2)) != 0)
+ return true;
+ return (vmw_fifo_caps(vmw) & SVGA_FIFO_CAP_FENCE) != 0;
+}
+
 #endif

drivers/gpu/drm/vmwgfx/vmwgfx_fence.c

@@ -82,6 +82,22 @@ fman_from_fence(struct vmw_fence_obj *fence)
  return container_of(fence->base.lock, struct vmw_fence_manager, lock);
 }
 
+static u32 vmw_fence_goal_read(struct vmw_private *vmw)
+{
+ if ((vmw->capabilities2 & SVGA_CAP2_EXTRA_REGS) != 0)
+ return vmw_read(vmw, SVGA_REG_FENCE_GOAL);
+ else
+ return vmw_fifo_mem_read(vmw, SVGA_FIFO_FENCE_GOAL);
+}
+
+static void vmw_fence_goal_write(struct vmw_private *vmw, u32 value)
+{
+ if ((vmw->capabilities2 & SVGA_CAP2_EXTRA_REGS) != 0)
+ vmw_write(vmw, SVGA_REG_FENCE_GOAL, value);
+ else
+ vmw_fifo_mem_write(vmw, SVGA_FIFO_FENCE_GOAL, value);
+}
+
 /*
  * Note on fencing subsystem usage of irqs:
  * Typically the vmw_fences_update function is called
@@ -392,17 +408,16 @@ static bool vmw_fence_goal_new_locked(struct vmw_fence_manager *fman,
  if (likely(!fman->seqno_valid))
  return false;
 
- goal_seqno = vmw_fifo_mem_read(fman->dev_priv, SVGA_FIFO_FENCE_GOAL);
+ goal_seqno = vmw_fence_goal_read(fman->dev_priv);
  if (likely(passed_seqno - goal_seqno >= VMW_FENCE_WRAP))
  return false;
 
  fman->seqno_valid = false;
  list_for_each_entry(fence, &fman->fence_list, head) {
  if (!list_empty(&fence->seq_passed_actions)) {
  fman->seqno_valid = true;
- vmw_fifo_mem_write(fman->dev_priv,
- SVGA_FIFO_FENCE_GOAL,
- fence->base.seqno);
+ vmw_fence_goal_write(fman->dev_priv,
+ fence->base.seqno);
  break;
  }
  }
@@ -434,13 +449,12 @@ static bool vmw_fence_goal_check_locked(struct vmw_fence_obj *fence)
  if (dma_fence_is_signaled_locked(&fence->base))
  return false;
 
- goal_seqno = vmw_fifo_mem_read(fman->dev_priv, SVGA_FIFO_FENCE_GOAL);
+ goal_seqno = vmw_fence_goal_read(fman->dev_priv);
  if (likely(fman->seqno_valid &&
  goal_seqno - fence->base.seqno < VMW_FENCE_WRAP))
  return false;
 
- vmw_fifo_mem_write(fman->dev_priv, SVGA_FIFO_FENCE_GOAL,
- fence->base.seqno);
+ vmw_fence_goal_write(fman->dev_priv, fence->base.seqno);
  fman->seqno_valid = true;
 
  return true;

drivers/gpu/drm/vmwgfx/vmwgfx_irq.c

@@ -32,6 +32,14 @@
 
 #define VMW_FENCE_WRAP (1 << 24)
 
+static u32 vmw_irqflag_fence_goal(struct vmw_private *vmw)
+{
+ if ((vmw->capabilities2 & SVGA_CAP2_EXTRA_REGS) != 0)
+ return SVGA_IRQFLAG_REG_FENCE_GOAL;
+ else
+ return SVGA_IRQFLAG_FENCE_GOAL;
+}
+
 /**
  * vmw_thread_fn - Deferred (process context) irq handler
  *
@@ -96,7 +104,7 @@ static irqreturn_t vmw_irq_handler(int irq, void *arg)
  wake_up_all(&dev_priv->fifo_queue);
 
  if ((masked_status & (SVGA_IRQFLAG_ANY_FENCE |
- SVGA_IRQFLAG_FENCE_GOAL)) &&
+ vmw_irqflag_fence_goal(dev_priv))) &&
  !test_and_set_bit(VMW_IRQTHREAD_FENCE, dev_priv->irqthread_pending))
  ret = IRQ_WAKE_THREAD;
 
@@ -137,8 +145,7 @@ bool vmw_seqno_passed(struct vmw_private *dev_priv,
  if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP))
  return true;
 
- if (!(vmw_fifo_caps(dev_priv) & SVGA_FIFO_CAP_FENCE) &&
- vmw_fifo_idle(dev_priv, seqno))
+ if (!vmw_has_fences(dev_priv) && vmw_fifo_idle(dev_priv, seqno))
  return true;
 
  /**
@@ -160,6 +167,7 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
  unsigned long timeout)
 {
  struct vmw_fifo_state *fifo_state = dev_priv->fifo;
+ bool fifo_down = false;
 
  uint32_t count = 0;
  uint32_t signal_seq;
@@ -176,12 +184,14 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
  */
 
  if (fifo_idle) {
- down_read(&fifo_state->rwsem);
  if (dev_priv->cman) {
  ret = vmw_cmdbuf_idle(dev_priv->cman, interruptible,
  10*HZ);
  if (ret)
  goto out_err;
+ } else if (fifo_state) {
+ down_read(&fifo_state->rwsem);
+ fifo_down = true;
  }
  }
 
@@ -218,12 +228,12 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
  }
  }
  finish_wait(&dev_priv->fence_queue, &__wait);
- if (ret == 0 && fifo_idle)
+ if (ret == 0 && fifo_idle && fifo_state)
  vmw_fence_write(dev_priv, signal_seq);
 
  wake_up_all(&dev_priv->fence_queue);
 out_err:
- if (fifo_idle)
+ if (fifo_down)
  up_read(&fifo_state->rwsem);
 
  return ret;
@@ -266,13 +276,13 @@ void vmw_seqno_waiter_remove(struct vmw_private *dev_priv)
 
 void vmw_goal_waiter_add(struct vmw_private *dev_priv)
 {
- vmw_generic_waiter_add(dev_priv, SVGA_IRQFLAG_FENCE_GOAL,
+ vmw_generic_waiter_add(dev_priv, vmw_irqflag_fence_goal(dev_priv),
  &dev_priv->goal_queue_waiters);
 }
 
 void vmw_goal_waiter_remove(struct vmw_private *dev_priv)
 {
- vmw_generic_waiter_remove(dev_priv, SVGA_IRQFLAG_FENCE_GOAL,
+ vmw_generic_waiter_remove(dev_priv, vmw_irqflag_fence_goal(dev_priv),
  &dev_priv->goal_queue_waiters);
 }
 

drivers/gpu/drm/vmwgfx/vmwgfx_kms.c

@@ -1344,7 +1344,6 @@ vmw_kms_new_framebuffer(struct vmw_private *dev_priv,
  ret = vmw_kms_new_framebuffer_surface(dev_priv, surface, &vfb,
  mode_cmd,
  is_bo_proxy);
-
  /*
  * vmw_create_bo_proxy() adds a reference that is no longer
  * needed
@@ -1385,13 +1384,16 @@ static struct drm_framebuffer *vmw_kms_fb_create(struct drm_device *dev,
  ret = vmw_user_lookup_handle(dev_priv, file_priv,
  mode_cmd->handles[0],
  &surface, &bo);
- if (ret)
+ if (ret) {
+ DRM_ERROR("Invalid buffer object handle %u (0x%x).\n",
+ mode_cmd->handles[0], mode_cmd->handles[0]);
  goto err_out;
+ }
 
 
  if (!bo &&
  !vmw_kms_srf_ok(dev_priv, mode_cmd->width, mode_cmd->height)) {
- DRM_ERROR("Surface size cannot exceed %dx%d",
+ DRM_ERROR("Surface size cannot exceed %dx%d\n",
  dev_priv->texture_max_width,
  dev_priv->texture_max_height);
  goto err_out;