[PATCH v6 08/32] drm/i915: Initialize all contexts

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



From: Ben Widawsky <benjamin.widawsky@xxxxxxxxx>

The problem is we're going to switch to a new context, which could be
the default context. The plan was to use restore inhibit, which would be
fine, except if we are using dynamic page tables (which we will). If we
use dynamic page tables and we don't load new page tables, the previous
page tables might go away, and future operations will fault.

CTXA runs.
switch to default, restore inhibit
CTXA dies and has its address space taken away.
Run CTXB, tries to save using the context A's address space - this
fails.

The general solution is to make sure every context has it's own state,
and its own address space. For cases when we must restore inhibit, first
thing we do is load a valid address space. I thought this would be
enough, but apparently there are references within the context itself
which will refer to the old address space - therefore, we also must
reinitialize.

It was tricky to track this down as we don't have much insight into what
happens in a context save.

This is required for the next patch which enables dynamic page tables.

v2: to->ppgtt is only valid in full ppgtt.

Signed-off-by: Ben Widawsky <ben@xxxxxxxxxxxx>
Signed-off-by: Michel Thierry <michel.thierry@xxxxxxxxx> (v2)
---
 drivers/gpu/drm/i915/i915_gem_context.c | 25 +++++++++++--------------
 1 file changed, 11 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index 437cdcc..6a583c3 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -596,13 +596,6 @@ needs_pd_load_pre(struct intel_engine_cs *ring, struct intel_context *to)
 			(ring != &dev_priv->ring[RCS])) && to->ppgtt;
 }
 
-static bool
-needs_pd_load_post(struct intel_engine_cs *ring, struct intel_context *to)
-{
-	return IS_GEN8(ring->dev) &&
-			(to->ppgtt || &to->ppgtt->pd_dirty_rings);
-}
-
 static int do_switch(struct intel_engine_cs *ring,
 		     struct intel_context *to)
 {
@@ -683,20 +676,24 @@ static int do_switch(struct intel_engine_cs *ring,
 
 	/* GEN8 does *not* require an explicit reload if the PDPs have been
 	 * setup, and we do not wish to move them.
-	 *
-	 * XXX: If we implemented page directory eviction code, this
-	 * optimization needs to be removed.
 	 */
-	if (!to->legacy_hw_ctx.initialized || i915_gem_context_is_default(to))
+	if (!to->legacy_hw_ctx.initialized) {
 		hw_flags |= MI_RESTORE_INHIBIT;
-	else if (to->ppgtt && test_and_clear_bit(ring->id, &to->ppgtt->pd_dirty_rings))
+		/* NB: If we inhibit the restore, the context is not allowed to
+		 * die because future work may end up depending on valid address
+		 * space. This means we must enforce that a page table load
+		 * occur when this occurs. */
+	} else if (to->ppgtt && test_and_clear_bit(ring->id, &to->ppgtt->pd_dirty_rings))
 		hw_flags |= MI_FORCE_RESTORE;
 
 	ret = mi_set_context(ring, to, hw_flags);
 	if (ret)
 		goto unpin_out;
 
-	if (needs_pd_load_post(ring, to)) {
+	if (IS_GEN8(ring->dev) && to->ppgtt && (hw_flags & MI_RESTORE_INHIBIT)) {
+		/* We have a valid page directory (scratch) to switch to. This
+		 * allows the old VM to be freed. Note that if anything occurs
+		 * between the set context, and here, we are f*cked */
 		ret = to->ppgtt->switch_mm(to->ppgtt, ring);
 		/* The hardware context switch is emitted, but we haven't
 		 * actually changed the state - so it's probably safe to bail
@@ -746,7 +743,7 @@ static int do_switch(struct intel_engine_cs *ring,
 		i915_gem_context_unreference(from);
 	}
 
-	uninitialized = !to->legacy_hw_ctx.initialized && from == NULL;
+	uninitialized = !to->legacy_hw_ctx.initialized;
 	to->legacy_hw_ctx.initialized = true;
 
 done:
-- 
2.1.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/intel-gfx





[Index of Archives]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]
  Powered by Linux