[Intel-gfx] [PATCH 06/13] drm/i915/bdw: implement semaphore signal
Ville Syrjälä
ville.syrjala at linux.intel.com
Wed Feb 12 10:29:20 CET 2014
On Tue, Feb 11, 2014 at 03:01:31PM -0800, Ben Widawsky wrote:
> On Tue, Feb 11, 2014 at 02:22:37PM -0800, Ben Widawsky wrote:
> > On Tue, Feb 11, 2014 at 02:11:04PM -0800, Ben Widawsky wrote:
> > > On Thu, Jan 30, 2014 at 02:38:17PM +0200, Ville Syrjälä wrote:
> > > > On Wed, Jan 29, 2014 at 11:55:26AM -0800, Ben Widawsky wrote:
> > > > > Semaphore signalling works similarly to previous GENs with the exception
> > > > > that the per ring mailboxes no longer exist. Instead you must define
> > > > > your own space, somewhere in the GTT.
> > > > >
> > > > > The comments in the code define the layout I've opted for, which should
> > > > > be fairly future proof. Ie. I tried to define offsets in abstract terms
> > > > > (NUM_RINGS, seqno size, etc).
> > > > >
> > > > > NOTE: If one wanted to move this to the HWSP they could. I've decided
> > > > > one 4k object would be easier to deal with, and provide potential wins
> > > > > with cache locality, but that's all speculative.
> > > > >
> > > > > v2: Update the macro to not need the other ring's ring->id (Chris)
> > > > > Update the comment to use the correct formula (Chris)
> > > > >
> > > > > Signed-off-by: Ben Widawsky <ben at bwidawsk.net>
> > > > > ---
> > > > > drivers/gpu/drm/i915/i915_drv.h | 1 +
> > > > > drivers/gpu/drm/i915/i915_reg.h | 5 +-
> > > > > drivers/gpu/drm/i915/intel_ringbuffer.c | 199 +++++++++++++++++++++++++-------
> > > > > drivers/gpu/drm/i915/intel_ringbuffer.h | 38 +++++-
> > > > > 4 files changed, 197 insertions(+), 46 deletions(-)
> > > > >
> > > > > diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> > > > > index 3673ba1..f521059 100644
> > > > > --- a/drivers/gpu/drm/i915/i915_drv.h
> > > > > +++ b/drivers/gpu/drm/i915/i915_drv.h
> > > > > @@ -1380,6 +1380,7 @@ typedef struct drm_i915_private {
> > > > >
> > > > > struct pci_dev *bridge_dev;
> > > > > struct intel_ring_buffer ring[I915_NUM_RINGS];
> > > > > + struct drm_i915_gem_object *semaphore_obj;
> > > > > uint32_t last_seqno, next_seqno;
> > > > >
> > > > > drm_dma_handle_t *status_page_dmah;
> > > > > diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
> > > > > index cbbaf26..8b745dc 100644
> > > > > --- a/drivers/gpu/drm/i915/i915_reg.h
> > > > > +++ b/drivers/gpu/drm/i915/i915_reg.h
> > > > > @@ -216,7 +216,7 @@
> > > > > #define MI_DISPLAY_FLIP_IVB_SPRITE_B (3 << 19)
> > > > > #define MI_DISPLAY_FLIP_IVB_PLANE_C (4 << 19)
> > > > > #define MI_DISPLAY_FLIP_IVB_SPRITE_C (5 << 19)
> > > > > -#define MI_SEMAPHORE_MBOX MI_INSTR(0x16, 1) /* gen6+ */
> > > > > +#define MI_SEMAPHORE_MBOX MI_INSTR(0x16, 1) /* gen6, gen7 */
> > > > > #define MI_SEMAPHORE_GLOBAL_GTT (1<<22)
> > > > > #define MI_SEMAPHORE_UPDATE (1<<21)
> > > > > #define MI_SEMAPHORE_COMPARE (1<<20)
> > > > > @@ -241,6 +241,8 @@
> > > > > #define MI_RESTORE_EXT_STATE_EN (1<<2)
> > > > > #define MI_FORCE_RESTORE (1<<1)
> > > > > #define MI_RESTORE_INHIBIT (1<<0)
> > > > > +#define MI_SEMAPHORE_SIGNAL MI_INSTR(0x1b, 0) /* GEN8+ */
> > > > > +#define MI_SEMAPHORE_TARGET(engine) ((engine)<<15)
> > > > > #define MI_STORE_DWORD_IMM MI_INSTR(0x20, 1)
> > > > > #define MI_MEM_VIRTUAL (1 << 22) /* 965+ only */
> > > > > #define MI_STORE_DWORD_INDEX MI_INSTR(0x21, 1)
> > > > > @@ -329,6 +331,7 @@
> > > > > #define PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE (1<<10) /* GM45+ only */
> > > > > #define PIPE_CONTROL_INDIRECT_STATE_DISABLE (1<<9)
> > > > > #define PIPE_CONTROL_NOTIFY (1<<8)
> > > > > +#define PIPE_CONTROL_FLUSH_ENABLE (1<<7) /* gen7+ */
> > > > > #define PIPE_CONTROL_VF_CACHE_INVALIDATE (1<<4)
> > > > > #define PIPE_CONTROL_CONST_CACHE_INVALIDATE (1<<3)
> > > > > #define PIPE_CONTROL_STATE_CACHE_INVALIDATE (1<<2)
> > > > > diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
> > > > > index 37ae2b1..b750835 100644
> > > > > --- a/drivers/gpu/drm/i915/intel_ringbuffer.c
> > > > > +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
> > > > > @@ -619,6 +619,13 @@ static int init_render_ring(struct intel_ring_buffer *ring)
> > > > > static void render_ring_cleanup(struct intel_ring_buffer *ring)
> > > > > {
> > > > > struct drm_device *dev = ring->dev;
> > > > > + struct drm_i915_private *dev_priv = dev->dev_private;
> > > > > +
> > > > > + if (dev_priv->semaphore_obj) {
> > > > > + i915_gem_object_ggtt_unpin(dev_priv->semaphore_obj);
> > > > > + drm_gem_object_unreference(&dev_priv->semaphore_obj->base);
> > > > > + dev_priv->semaphore_obj = NULL;
> > > > > + }
> > > > >
> > > > > if (ring->scratch.obj == NULL)
> > > > > return;
> > > > > @@ -632,6 +639,86 @@ static void render_ring_cleanup(struct intel_ring_buffer *ring)
> > > > > ring->scratch.obj = NULL;
> > > > > }
> > > > >
> > > > > +static int gen8_rcs_signal(struct intel_ring_buffer *signaller,
> > > > > + unsigned int num_dwords)
> > > > > +{
> > > > > +#define MBOX_UPDATE_DWORDS 8
> > > > > + struct drm_device *dev = signaller->dev;
> > > > > + struct drm_i915_private *dev_priv = dev->dev_private;
> > > > > + struct intel_ring_buffer *waiter;
> > > > > + int i, ret, num_rings;
> > > > > +
> > > > > + num_rings = hweight_long(INTEL_INFO(dev)->ring_mask);
> > > > > + num_dwords = (num_rings-1) * MBOX_UPDATE_DWORDS;
> > > >
> > > > Again num_dwords +=
> > > >
> > > > > +#undef MBOX_UPDATE_DWORDS
> > > > > +
> > > > > + /* XXX: + 4 for the caller */
> > > > > + ret = intel_ring_begin(signaller, num_dwords + 4);
> > > >
> > > > and the +4 goes away.
> > > >
> > > > > + if (ret)
> > > > > + return ret;
> > > > > +
> > > > > + for_each_ring(waiter, dev_priv, i) {
> > > > > + u64 gtt_offset = signaller->semaphore.signal_ggtt[i];
> > > > > + if (gtt_offset == MI_SEMAPHORE_SYNC_INVALID)
> > > > > + continue;
> > > > > +
> > > > > + intel_ring_emit(signaller, GFX_OP_PIPE_CONTROL(6));
> > > > > + intel_ring_emit(signaller, PIPE_CONTROL_GLOBAL_GTT_IVB |
> > > > > + PIPE_CONTROL_QW_WRITE |
> > > > > + PIPE_CONTROL_FLUSH_ENABLE);
> > > > > + intel_ring_emit(signaller, lower_32_bits(gtt_offset));
> > > > > + intel_ring_emit(signaller, upper_32_bits(gtt_offset));
> > > > > + intel_ring_emit(signaller, signaller->outstanding_lazy_seqno);
> > > > > + intel_ring_emit(signaller, 0);
> > > > > + intel_ring_emit(signaller, MI_SEMAPHORE_SIGNAL |
> > > > > + MI_SEMAPHORE_TARGET(waiter->id));
> > > > > + intel_ring_emit(signaller, 0);
> > > > > + }
> > > > > +
> > > > > + WARN_ON(i != num_rings);
> > > > > +
> > > > > + return 0;
> > > > > +}
> > > >
> > > > <snip>
> > >
> > > Got those, thanks.
> > >
> > > >
> > > > > diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
> > > > > index c69ae10..f1e7a66 100644
> > > > > --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> > > > > +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> > > > > @@ -111,6 +111,39 @@ struct intel_ring_buffer {
> > > > > #define I915_DISPATCH_PINNED 0x2
> > > > > void (*cleanup)(struct intel_ring_buffer *ring);
> > > > >
> > > > > + /* GEN8 signal/wait table
> > > > > + * signal to signal to signal to signal to
> > > > > + * RCS VCS BCS VECS
> > > > > + * ------------------------------------------------------
> > > > > + * RCS | NOP (0x00) | BCS (0x08) | VCS (0x10) | VECS (0x18) |
> > > > > + * |-----------------------------------------------------
> > > > > + * VCS | RCS (0x20) | NOP (0x28) | BCS (0x30) | VECS (0x38) |
> > > > > + * |-----------------------------------------------------
> > > > > + * BCS | RCS (0x40) | VCS (0x48) | NOP (0x50) | VECS (0x58) |
> > > > > + * |-----------------------------------------------------
> > > > > + * VECS | RCS (0x60) | VCS (0x68) | BCS (0x70) | NOP (0x78) |
> > > > > + * |-----------------------------------------------------
> > > > > + *
> > > > > + * Generalization:
> > > > > + * f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id)
> > > > > + * ie. transpose of g(x, y)
> > > > > + *
> > > > > + * sync from sync from sync from sync from
> > > > > + * RCS VCS BCS VECS
> > > > > + * ------------------------------------------------------
> > > > > + * RCS | NOP (0x00) | BCS (0x20) | VCS (0x40) | VECS (0x60) |
> > > > > + * |-----------------------------------------------------
> > > > > + * VCS | RCS (0x08) | NOP (0x28) | BCS (0x48) | VECS (0x68) |
> > > > > + * |-----------------------------------------------------
> > > > > + * BCS | RCS (0x10) | VCS (0x30) | NOP (0x50) | VECS (0x60) |
> > > > > + * |-----------------------------------------------------
> > > > > + * VECS | RCS (0x18) | VCS (0x38) | BCS (0x58) | NOP (0x78) |
> > > > > + * |-----------------------------------------------------
> > > > > + *
> > > > > + * Generalization:
> > > > > + * g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id)
> > > > > + * ie. transpose of f(x, y)
> > > > > + */
> > > > > struct {
> > > > > u32 sync_seqno[I915_NUM_RINGS-1];
> > > > > /* AKA wait() */
> > > > > @@ -120,7 +153,10 @@ struct intel_ring_buffer {
> > > > > /* our mbox written by others */
> > > > > u32 mbox[I915_NUM_RINGS];
> > > >
> > > > mbox should also get a u64 friend, right?
> > >
> > > mbox should be gen6 only, given the change to using the gtt on gen8. In
> > > this point in the series, semaphores should be forcibly disabled on
> > > gen8, so the code looks wrong, but the path cannot [should not] be
> > > taken.
> > >
> > > I suppose I should kill the initialization of mbox for gen8, or somehow
> > > consolidate with a union to prevent confusion.
> > >
> >
> > Just to clarify it should be
> >
> > gen6:
> > signal uses signal_mbox for signal
> > wait uses mbox
> >
> > gen8:
> > signal uses signal_ggtt for signal
> > wait uses arithmetic to figure out the offset
> >
>
> Ok, I've fixed this up to make things clearer, but it ends up in the
> next patch. So look there when I repost.
I was confused by these:
+ ring->semaphore.mbox[RCS] = GEN8_WAIT_OFFSET(ring, RCS);
So you did store the wait offset into mbox, but then you didn't use the
precomputed values and instead recomputed on the spot in gen8
ring_sync().
>
> > > >
> > > > > /* mboxes this ring signals to */
> > > > > - u32 signal_mbox[I915_NUM_RINGS];
> > > > > + union {
> > > > > + u32 signal_mbox[I915_NUM_RINGS];
> > > > > + u64 signal_ggtt[I915_NUM_RINGS];
> > > > > + };
> > > > >
> > > > > /* num_dwords is space the caller will need for atomic update */
> > > > > int (*signal)(struct intel_ring_buffer *signaller,
> > > > > --
> > > > > 1.8.5.3
> > > > >
> > > > > _______________________________________________
> > > > > Intel-gfx mailing list
> > > > > Intel-gfx at lists.freedesktop.org
> > > > > http://lists.freedesktop.org/mailman/listinfo/intel-gfx
> > > >
> > > > --
> > > > Ville Syrjälä
> > > > Intel OTC
> > >
> > > --
> > > Ben Widawsky, Intel Open Source Technology Center
> >
> > --
> > Ben Widawsky, Intel Open Source Technology Center
>
> --
> Ben Widawsky, Intel Open Source Technology Center
--
Ville Syrjälä
Intel OTC
More information about the Intel-gfx
mailing list