search for: nvc0_push_vbo

When nvc0_push_vbo calls nouveau_scratch_done it does not mean scratch buffers can be freed immediately. It means "when hardware advances to this place in the command stream the scratch buffers can be freed". The bug existed for a very long time. Nobody noticed, because "scratch runout" code path...

Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu> --- src/gallium/drivers/nouveau/nv50/nv50_context.c | 22 ++++++++++++++++++- src/gallium/drivers/nouveau/nv50/nv50_context.h | 1 + src/gallium/drivers/nouveau/nv50/nv50_vbo.c | 29 ++++++++++++++++++++++++- 3 files changed, 50 insertions(+), 2 deletions(-) diff --git a/src/gallium/drivers/nouveau/nv50/nv50_context.c

The main patches are from Christoph. Unfortunately they're a little beyond my understanding of all the vertex-related details, but they generally seemed fine. I'm just going to push these unless someone steps up to review them. Christoph Bumiller (2): nvc0: add support for indirect drawing nvc0: fix translate path for PRIM_RESTART_WITH_DRAW_ARRAYS Ilia Mirkin (1): nouveau: check if

...must ensure that (min_index + index_bias) is >= 0 */ nvc0->vb_elt_first = info->min_index + info->index_bias; nvc0->vb_elt_limit = info->max_index - info->min_index; @@ -1031,6 +1033,7 @@ nvc0_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info) nvc0_push_vbo(nvc0, info); push->kick_notify = nvc0_default_kick_notify; nouveau_pushbuf_bufctx(push, NULL); + pipe_mutex_unlock(screen->base.push_mutex); return; } @@ -1083,4 +1086,5 @@ nvc0_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info) nvc0_rel...

As it turns out, with the proprietary firmware, the VP3 and VP4 interfaces are identical. Furthermore, this is all already implemented for nvc0. So these patches (a) move the easily sharable bits of the nvc0 implementation into the nouveau directory, and then (b) implement the other parts in nv50. The non-shared parts are still largely copies, but there are some differences, not the least of which