[Intel-gfx] [PATCH] i915: support page flipping
Chris Wilson
chris at chris-wilson.co.uk
Thu Feb 26 22:56:52 CET 2009
On Thu, 2009-02-26 at 13:28 -0800, Jesse Barnes wrote:
> Add a new page flip ioctl to the i915 driver. The new ioctl takes the new
> drm_i915_gem_page_flip arg, which contains a buffer object target for the
> flip, an offset into that buffer, and other buffer parameters to be used for
> the flip.
>
> If a flip is already pending on the object in question, the ioctl waits for
> it to finish first, then queues the flip. An optional wait flag causes the
> ioctl to wait for the flip to complete before the it returns.
>
> If an execbuffer containing an object with a pending flip comes in, it will
> stall until the flip completes, to preserve glXSwapBuffers semantics.
>
> Signed-off-by: Jesse Barnes <jbarnes at virtuousgeek.org>
Didn't do too bad in spotting the missing checks... ;-)
> diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c
> index 2d797ff..0d6a6d3 100644
> --- a/drivers/gpu/drm/i915/i915_dma.c
> +++ b/drivers/gpu/drm/i915/i915_dma.c
> @@ -829,6 +829,175 @@ static int i915_set_status_page(struct drm_device *dev, void *data,
> return 0;
> }
>
> +static int i915_pipe_to_plane(struct drm_device *dev, int pipe)
> +{
> + drm_i915_private_t *dev_priv = dev->dev_private;
> + u32 reg, pipe_mask;
> +
> + if (pipe == 0)
> + pipe_mask = DISPPLANE_SEL_PIPE_A;
> + else
> + pipe_mask = DISPPLANE_SEL_PIPE_B;
> +
> + pipe_mask |= DISPLAY_PLANE_ENABLE;
> +
> + reg = I915_READ(DSPACNTR);
> + if ((reg & (DISPLAY_PLANE_ENABLE | DISPPLANE_SEL_PIPE_MASK)) ==
> + pipe_mask)
> + return 0;
> +
> + reg = I915_READ(DSPBCNTR);
> + if ((reg & (DISPLAY_PLANE_ENABLE | DISPPLANE_SEL_PIPE_MASK)) ==
> + pipe_mask)
> + return 1;
> +
> + return -1;
> +}
> +
> +bool
> +i915_gem_flip_pending(struct drm_gem_object *obj)
> +{
> + struct drm_device *dev = obj->dev;
> + drm_i915_private_t *dev_priv = dev->dev_private;
> + struct drm_i915_gem_object *obj_priv = obj->driver_private;
> + unsigned long flags;
> + bool pending = false;
> +
> + spin_lock_irqsave(&dev_priv->vblank_lock, flags);
> + if (!list_empty(&obj_priv->vblank_head))
> + pending = true;
This has annoyed me every time (but I'm running out of things I can
complain about ;-), can we just say
pending = !list_empty();
and be done with the conditional.
> + spin_unlock_irqrestore(&dev_priv->vblank_lock, flags);
> +
> + return pending;
> +}
> +
> +static int i915_gem_page_flip(struct drm_device *dev, void *data,
> + struct drm_file *file_priv)
> +{
> + struct drm_i915_gem_page_flip *flip_data = data;
> + drm_i915_private_t *dev_priv = dev->dev_private;
> + struct drm_gem_object *obj;
> + struct drm_i915_gem_object *obj_priv;
> + unsigned long flags;
> + uint32_t offset;
> + unsigned int pitch, pipe, plane, tiled;
> + int ret = 0, vblank_ref_err = 0, reqno;
> + RING_LOCALS;
> +
> + if (!(dev->driver->driver_features & DRIVER_GEM))
> + return -ENODEV;
Guess I'll just have to comment this out to continue testing this
feature. :-)
> +
> + /*
> + * Reject unknown flags so future userspace knows what we (don't)
> + * support
> + */
> + if (flip_data->flags & (~I915_PAGE_FLIP_WAIT)) {
> + DRM_ERROR("bad page flip flags\n");
> + return -EINVAL;
> + }
> +
> + if ((pipe = flip_data->pipe) > 1) {
> + DRM_ERROR("bad pipe\n");
> + return -EINVAL;
> + }
> +
> + plane = i915_pipe_to_plane(dev, pipe);
> + if (plane < 0) {
> + DRM_ERROR("bad pipe (no planes enabled?)\n");
> + return -EINVAL;
> + }
> +
> + obj = drm_gem_object_lookup(dev, file_priv, flip_data->handle);
> + if (obj == NULL)
> + return -EBADF;
> +
> + /*
> + * Make sure the new buffer is in bounds
> + * FIXME: should probably check against current mode as well
> + */
> + if (flip_data->offset >= obj->size) {
> + DRM_ERROR("bad flip offset\n");
> + ret = -EINVAL;
> + goto out_unref_prelock;
> + }
> +
> + obj_priv = obj->driver_private;
Need to check obj_priv->stride&7 == 0. Hmm, that may be impossible
anyway.
> +
> + if (i915_gem_flip_pending(obj))
> + wait_for_completion(&obj_priv->vblank);
> +
> + mutex_lock(&dev->struct_mutex);
> + if (obj_priv->tiling_mode != I915_TILING_NONE &&
> + obj_priv->tiling_mode != I915_TILING_X) {
> + DRM_ERROR("can only flip non-tiled or X tiled pages\n");
> + ret = -EINVAL;
> + goto out_unref;
> + }
> +
> + vblank_ref_err = drm_vblank_get(dev, pipe);
> + if (!vblank_ref_err) {
Ok, this seems like a reasonable fallback. If vblank is not enabled then
you just queue the flip without waiting.
> + ret = i915_gem_object_pin(obj, 0);
> + if (ret) {
> + DRM_ERROR("failed to pin object for flip\n");
> + ret = -EBUSY;
What's the rationale for changing the reported error here? (And it might
be helpful to printk the original error value.)
> + goto out_unref;
> + }
> + }
> +
> + /*
> + * Put the object in the GTT domain before the flip,
> + * since there may be outstanding rendering
> + */
> + i915_gem_object_set_to_gtt_domain(obj, 0);
Need to check, report and handle errors here. And yes, these do occur in
the wild for some as of yet unknown reason.
> +
> + offset = obj_priv->gtt_offset + flip_data->offset;
> +
> + pitch = obj_priv->stride;
> + tiled = (obj_priv->tiling_mode == I915_TILING_X);
> +
> + BEGIN_LP_RING(4);
> + OUT_RING(CMD_OP_DISPLAYBUFFER_INFO | (plane << 20));
> + OUT_RING(pitch);
> + if (IS_I965G(dev)) {
> + OUT_RING(offset | tiled);
> + OUT_RING(((flip_data->x - 1) << 16) | (flip_data->y - 1));
> + } else {
> + OUT_RING(offset);
> + OUT_RING(MI_NOOP);
> + }
> + ADVANCE_LP_RING();
> +
> + reqno = i915_add_request(dev, 0);
Be consistent and call this seqno like the rest of the code, please.
> +
> + mutex_unlock(&dev->struct_mutex);
> +
> + /*
> + * This is a bit racy; the flip above may have already happened
> + * by the time we get here. If that happens, the new back buffer
> + * won't be available for rendering for one extra frame, since
> + * the vblank list won't have the object.
> + */
> + spin_lock_irqsave(&dev_priv->vblank_lock, flags);
> + if (!vblank_ref_err)
> + list_add_tail(&obj_priv->vblank_head,
> + &dev_priv->mm.vblank_list[pipe]);
> +
> + obj_priv->flip_seqno = reqno;
> + spin_unlock_irqrestore(&dev_priv->vblank_lock, flags);
> +
> + if (!vblank_ref_err && (flip_data->flags & I915_PAGE_FLIP_WAIT))
> + wait_for_completion(&obj_priv->vblank);
> +
> +out_unref_prelock:
> + /* Take lock again for unref */
> + mutex_lock(&dev->struct_mutex);
> +out_unref:
> + drm_gem_object_unreference(obj);
> + mutex_unlock(&dev->struct_mutex);
> +
> + return ret;
> +}
> +
> /**
> * i915_probe_agp - get AGP bootup configuration
> * @pdev: PCI device
> @@ -1307,6 +1476,7 @@ struct drm_ioctl_desc i915_ioctls[] = {
> DRM_IOCTL_DEF(DRM_I915_GEM_SET_TILING, i915_gem_set_tiling, 0),
> DRM_IOCTL_DEF(DRM_I915_GEM_GET_TILING, i915_gem_get_tiling, 0),
> DRM_IOCTL_DEF(DRM_I915_GEM_GET_APERTURE, i915_gem_get_aperture_ioctl, 0),
> + DRM_IOCTL_DEF(DRM_I915_GEM_PAGE_FLIP, i915_gem_page_flip, DRM_AUTH|DRM_MASTER),
> };
>
> int i915_max_ioctl = DRM_ARRAY_SIZE(i915_ioctls);
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 3951a12..148e80a 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -159,6 +159,10 @@ typedef struct drm_i915_private {
> u32 irq_mask_reg;
> u32 pipestat[2];
>
> + /** Protects vblank list */
> + spinlock_t vblank_lock;
> + struct work_struct vblank_work;
> +
> int tex_lru_log_granularity;
> int allow_batchbuffer;
> struct mem_block *agp_heap;
> @@ -338,6 +342,11 @@ typedef struct drm_i915_private {
> */
> struct delayed_work retire_work;
>
> + /**
> + * List of objects waiting on vblank events (one per pipe)
> + */
> + struct list_head vblank_list[2];
> +
> uint32_t next_gem_seqno;
>
> /**
> @@ -389,6 +398,13 @@ struct drm_i915_gem_object {
> /** This object's place on the active/flushing/inactive lists */
> struct list_head list;
>
> + /** Object's place on the vblank list (protected by vblank_lock)*/
> + struct list_head vblank_head;
> + /** sequence number for flip (when it passes the flip is done),
> + * protected by vblank lock
> + */
> + int flip_seqno;
> +
> /**
> * This is set if the object is on the active or flushing lists
> * (has pending rendering), and is not set if it's on inactive (ready
> @@ -457,6 +473,9 @@ struct drm_i915_gem_object {
>
> /** for phy allocated objects */
> struct drm_i915_gem_phys_object *phys_obj;
> +
> + /** for page flips and other vblank related blocks */
> + struct completion vblank;
> };
>
> /**
> @@ -516,6 +535,7 @@ extern long i915_compat_ioctl(struct file *filp, unsigned int cmd,
> extern int i915_emit_box(struct drm_device *dev,
> struct drm_clip_rect __user *boxes,
> int i, int DR1, int DR4);
> +extern bool i915_gem_flip_pending(struct drm_gem_object *obj);
>
> /* i915_irq.c */
> extern int i915_irq_emit(struct drm_device *dev, void *data,
> @@ -625,6 +645,9 @@ int i915_gem_attach_phys_object(struct drm_device *dev,
> void i915_gem_detach_phys_object(struct drm_device *dev,
> struct drm_gem_object *obj);
> void i915_gem_free_all_phys_object(struct drm_device *dev);
> +uint32_t i915_add_request(struct drm_device *dev, uint32_t flush_domains);
> +int i915_seqno_passed(uint32_t seq1, uint32_t seq2);
> +uint32_t i915_get_gem_seqno(struct drm_device *dev);
>
> /* i915_gem_tiling.c */
> void i915_gem_detect_bit_6_swizzle(struct drm_device *dev);
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 3e025d5..5270d25 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -900,7 +900,7 @@ i915_gem_object_move_to_inactive(struct drm_gem_object *obj)
> *
> * Returned sequence numbers are nonzero on success.
> */
> -static uint32_t
> +uint32_t
> i915_add_request(struct drm_device *dev, uint32_t flush_domains)
> {
> drm_i915_private_t *dev_priv = dev->dev_private;
> @@ -1028,7 +1028,7 @@ i915_gem_retire_request(struct drm_device *dev,
> /**
> * Returns true if seq1 is later than seq2.
> */
> -static int
> +int
> i915_seqno_passed(uint32_t seq1, uint32_t seq2)
> {
> return (int32_t)(seq1 - seq2) >= 0;
> @@ -2498,6 +2498,25 @@ i915_gem_execbuffer(struct drm_device *dev, void *data,
> goto pre_mutex_err;
> }
>
> + /* Look up object handles */
> + for (i = 0; i < args->buffer_count; i++) {
> + object_list[i] = drm_gem_object_lookup(dev, file_priv,
> + exec_list[i].handle);
> + if (object_list[i] == NULL) {
> + DRM_ERROR("Invalid object handle %d at index %d\n",
> + exec_list[i].handle, i);
> + ret = -EBADF;
> + goto pre_mutex_err;
Hmm, can't jump straight to pre_mutex_err as that means you leak the
references on object already looked up. Take the mutex and goto err;
> + }
> +
> + if (i915_gem_flip_pending(object_list[i])) {
> + struct drm_i915_gem_object *obj_priv;
> +
> + obj_priv = object_list[i]->driver_private;
> + wait_for_completion(&obj_priv->vblank);
> + }
> + }
> +
> mutex_lock(&dev->struct_mutex);
>
> i915_verify_inactive(dev, __FILE__, __LINE__);
> @@ -2516,18 +2535,6 @@ i915_gem_execbuffer(struct drm_device *dev, void *data,
> goto pre_mutex_err;
> }
>
> - /* Look up object handles */
> - for (i = 0; i < args->buffer_count; i++) {
> - object_list[i] = drm_gem_object_lookup(dev, file_priv,
> - exec_list[i].handle);
> - if (object_list[i] == NULL) {
> - DRM_ERROR("Invalid object handle %d at index %d\n",
> - exec_list[i].handle, i);
> - ret = -EBADF;
> - goto err;
> - }
> - }
> -
> /* Pin and relocate */
> for (pin_tries = 0; ; pin_tries++) {
> ret = 0;
> @@ -2920,6 +2927,9 @@ int i915_gem_init_object(struct drm_gem_object *obj)
> obj_priv->obj = obj;
> obj_priv->fence_reg = I915_FENCE_REG_NONE;
> INIT_LIST_HEAD(&obj_priv->list);
> + INIT_LIST_HEAD(&obj_priv->vblank_head);
> +
> + init_completion(&obj_priv->vblank);
>
> return 0;
> }
> @@ -2980,8 +2990,10 @@ int
> i915_gem_idle(struct drm_device *dev)
> {
> drm_i915_private_t *dev_priv = dev->dev_private;
> + struct drm_i915_gem_object *obj_priv, *tmp;
> + unsigned long irqflags;
> uint32_t seqno, cur_seqno, last_seqno;
> - int stuck, ret;
> + int stuck, ret, pipe;
>
> mutex_lock(&dev->struct_mutex);
>
> @@ -2995,10 +3007,25 @@ i915_gem_idle(struct drm_device *dev)
> */
> dev_priv->mm.suspended = 1;
>
> + mutex_unlock(&dev->struct_mutex);
> +
> + /* Wait for any outstanding flips */
> + spin_lock_irqsave(&dev_priv->vblank_lock, irqflags);
> + for (pipe = 0; pipe < 2; pipe++) {
> + list_for_each_entry_safe(obj_priv, tmp,
> + &dev_priv->mm.vblank_list[pipe],
> + vblank_head) {
> + spin_unlock_irqrestore(&dev_priv->vblank_lock, irqflags);
> + wait_for_completion(&obj_priv->vblank);
> + spin_lock_irqsave(&dev_priv->vblank_lock, irqflags);
> + }
> + }
> + spin_unlock_irqrestore(&dev_priv->vblank_lock, irqflags);
> +
> /* Cancel the retire work handler, wait for it to finish if running
> */
> - mutex_unlock(&dev->struct_mutex);
> cancel_delayed_work_sync(&dev_priv->mm.retire_work);
> +
> mutex_lock(&dev->struct_mutex);
>
> i915_kernel_lost_context(dev);
> @@ -3358,9 +3385,12 @@ i915_gem_load(struct drm_device *dev)
> INIT_LIST_HEAD(&dev_priv->mm.flushing_list);
> INIT_LIST_HEAD(&dev_priv->mm.inactive_list);
> INIT_LIST_HEAD(&dev_priv->mm.request_list);
> + INIT_LIST_HEAD(&dev_priv->mm.vblank_list[0]);
> + INIT_LIST_HEAD(&dev_priv->mm.vblank_list[1]);
> INIT_DELAYED_WORK(&dev_priv->mm.retire_work,
> i915_gem_retire_work_handler);
> dev_priv->mm.next_gem_seqno = 1;
> + spin_lock_init(&dev_priv->vblank_lock);
>
> /* Old X drivers will take 0-2 for front, back, depth buffers */
> dev_priv->fence_reg_start = 3;
> diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
> index 548ff2c..f50df88 100644
> --- a/drivers/gpu/drm/i915/i915_irq.c
> +++ b/drivers/gpu/drm/i915/i915_irq.c
> @@ -187,6 +187,36 @@ u32 gm45_get_vblank_counter(struct drm_device *dev, int pipe)
> return I915_READ(reg);
> }
>
> +static void i915_vblank_work_func(struct work_struct *work)
> +{
> + drm_i915_private_t *dev_priv = container_of(work, drm_i915_private_t,
> + vblank_work);
> + struct drm_device *dev = dev_priv->dev;
> + struct drm_i915_gem_object *obj_priv, *tmp;
> + unsigned long irqflags;
> + int pipe, cur_seqno;
> +
> + mutex_lock(&dev->struct_mutex);
> +
> + spin_lock_irqsave(&dev_priv->vblank_lock, irqflags);
> + for (pipe = 0; pipe < 2; pipe++) {
> + list_for_each_entry_safe(obj_priv, tmp,
> + &dev_priv->mm.vblank_list[pipe],
> + vblank_head) {
> + cur_seqno = i915_get_gem_seqno(dev);
> + if (i915_seqno_passed(cur_seqno,
> + obj_priv->flip_seqno)) {
> + list_del_init(&obj_priv->vblank_head);
> + drm_vblank_put(dev, pipe);
> + i915_gem_object_unpin(obj_priv->obj);
> + complete(&obj_priv->vblank);
> + }
> + }
> + }
> + spin_unlock_irqrestore(&dev_priv->vblank_lock, irqflags);
> + mutex_unlock(&dev->struct_mutex);
> +}
> +
> irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS)
> {
> struct drm_device *dev = (struct drm_device *) arg;
> @@ -269,6 +299,9 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS)
> drm_handle_vblank(dev, 1);
> }
>
> + if (vblank)
> + schedule_work(&dev_priv->vblank_work);
> +
> if ((pipeb_stats & I915_LEGACY_BLC_EVENT_STATUS) ||
> (iir & I915_ASLE_INTERRUPT))
> opregion_asle_intr(dev);
> @@ -533,6 +566,7 @@ void i915_driver_irq_preinstall(struct drm_device * dev)
> I915_WRITE(IMR, 0xffffffff);
> I915_WRITE(IER, 0x0);
> (void) I915_READ(IER);
> + INIT_WORK(&dev_priv->vblank_work, i915_vblank_work_func);
> }
>
> int i915_driver_irq_postinstall(struct drm_device *dev)
> diff --git a/include/drm/i915_drm.h b/include/drm/i915_drm.h
> index 912cd52..1ac4ded 100644
> --- a/include/drm/i915_drm.h
> +++ b/include/drm/i915_drm.h
> @@ -184,6 +184,7 @@ typedef struct _drm_i915_sarea {
> #define DRM_I915_GEM_GET_TILING 0x22
> #define DRM_I915_GEM_GET_APERTURE 0x23
> #define DRM_I915_GEM_MMAP_GTT 0x24
> +#define DRM_I915_GEM_PAGE_FLIP 0x25
>
> #define DRM_IOCTL_I915_INIT DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT, drm_i915_init_t)
> #define DRM_IOCTL_I915_FLUSH DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLUSH)
> @@ -219,6 +220,7 @@ typedef struct _drm_i915_sarea {
> #define DRM_IOCTL_I915_GEM_SET_TILING DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_SET_TILING, struct
> drm_i915_gem_set_tiling)
> #define DRM_IOCTL_I915_GEM_GET_TILING DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_GET_TILING, struct
> drm_i915_gem_get_tiling)
> #define DRM_IOCTL_I915_GEM_GET_APERTURE DRM_IOR (DRM_COMMAND_BASE + DRM_I915_GEM_GET_APERTURE, struct
> drm_i915_gem_get_aperture)
> +#define DRM_IOCTL_I915_GEM_PAGE_FLIP DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_PAGE_FLIP, struct
> drm_i915_gem_page_flip)
>
> /* Allow drivers to submit batchbuffers directly to hardware, relying
> * on the security mechanisms provided by hardware.
> @@ -654,4 +656,30 @@ struct drm_i915_gem_get_aperture {
> uint64_t aper_available_size;
> };
>
> +#define I915_PAGE_FLIP_WAIT (1<<0) /* block on page flip completion */
> +
> +struct drm_i915_gem_page_flip {
> + /** Handle of new front buffer */
> + uint32_t handle;
> +
> + /** Offset into the object to use */
> + uint64_t offset;
> +
> + /**
> + * page flip flags (wait on flip only for now)
> + */
> + uint32_t flags;
> +
> + /**
> + * pipe to flip
> + */
> + uint32_t pipe;
> +
> + /**
> + * screen dimensions for flip
> + */
> + uint32_t x;
> + uint32_t y;
> +};
> +
> #endif /* _I915_DRM_H_ */
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx
More information about the Intel-gfx
mailing list