Message ID | 20191229183153.3719869-5-chris@chris-wilson.co.uk (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [1/7] drm/i915/gt: Ensure that all new contexts clear STOP_RING | expand |
On Sun, 29 Dec 2019 at 18:32, Chris Wilson <chris@chris-wilson.co.uk> wrote: > > We leave the kernel_context on the HW as we suspend (and while idle). > There is no guarantee that is complete in memory, so we try to inhibit > restoration from the kernel_context. Reinforce the inhibition by > scrubbing the context. > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > --- > drivers/gpu/drm/i915/gt/intel_lrc.c | 17 +++++++++++++++-- > drivers/gpu/drm/i915/gt/intel_ring_submission.c | 2 +- > 2 files changed, 16 insertions(+), 3 deletions(-) > > diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c > index 14e7e179855f..b1508dbd1063 100644 > --- a/drivers/gpu/drm/i915/gt/intel_lrc.c > +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c > @@ -2494,6 +2494,11 @@ static int execlists_context_alloc(struct intel_context *ce) > > static void execlists_context_reset(struct intel_context *ce) > { > + u32 *regs; > + > + CE_TRACE(ce, "reset\n"); > + GEM_BUG_ON(!intel_context_is_pinned(ce)); > + > /* > * Because we emit WA_TAIL_DWORDS there may be a disparity > * between our bookkeeping in ce->ring->head and ce->ring->tail and > @@ -2510,8 +2515,17 @@ static void execlists_context_reset(struct intel_context *ce) > * So to avoid that we reset the context images upon resume. For > * simplicity, we just zero everything out. > */ > - intel_ring_reset(ce->ring, 0); > + intel_ring_reset(ce->ring, ce->ring->emit); > + > + regs = memset(ce->lrc_reg_state, 0, PAGE_SIZE); > + execlists_init_reg_state(regs, ce, ce->engine, ce->ring, true); > __execlists_update_reg_state(ce, ce->engine); > + > + /* Avoid trying to reload the garbage */ > + regs[CTX_CONTEXT_CONTROL] |= > + _MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT); > + > + ce->lrc_desc |= CTX_DESC_FORCE_RESTORE; > } > > static const struct intel_context_ops execlists_context_ops = { > @@ -3968,7 +3982,6 @@ static void init_common_reg_state(u32 * const regs, > CTX_CTRL_RS_CTX_ENABLE); > > regs[CTX_RING_CTL] = RING_CTL_SIZE(ring->size) | RING_VALID; > - regs[CTX_BB_STATE] = RING_BB_PPGTT; Zero clue what that does... Otherwise, Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Quoting Matthew Auld (2019-12-30 16:06:47) > On Sun, 29 Dec 2019 at 18:32, Chris Wilson <chris@chris-wilson.co.uk> wrote: > > static const struct intel_context_ops execlists_context_ops = { > > @@ -3968,7 +3982,6 @@ static void init_common_reg_state(u32 * const regs, > > CTX_CTRL_RS_CTX_ENABLE); > > > > regs[CTX_RING_CTL] = RING_CTL_SIZE(ring->size) | RING_VALID; > > - regs[CTX_BB_STATE] = RING_BB_PPGTT; > > Zero clue what that does... It's supposed to be a readonly bit that shows the state of the current batch buffer, and is supposed to be only set by MI_BB_START. Broadwell and Braswell disagree with the bspec. C'est la vie. I broke it out into a separate patch for clarity. -Chris
diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c index 14e7e179855f..b1508dbd1063 100644 --- a/drivers/gpu/drm/i915/gt/intel_lrc.c +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c @@ -2494,6 +2494,11 @@ static int execlists_context_alloc(struct intel_context *ce) static void execlists_context_reset(struct intel_context *ce) { + u32 *regs; + + CE_TRACE(ce, "reset\n"); + GEM_BUG_ON(!intel_context_is_pinned(ce)); + /* * Because we emit WA_TAIL_DWORDS there may be a disparity * between our bookkeeping in ce->ring->head and ce->ring->tail and @@ -2510,8 +2515,17 @@ static void execlists_context_reset(struct intel_context *ce) * So to avoid that we reset the context images upon resume. For * simplicity, we just zero everything out. */ - intel_ring_reset(ce->ring, 0); + intel_ring_reset(ce->ring, ce->ring->emit); + + regs = memset(ce->lrc_reg_state, 0, PAGE_SIZE); + execlists_init_reg_state(regs, ce, ce->engine, ce->ring, true); __execlists_update_reg_state(ce, ce->engine); + + /* Avoid trying to reload the garbage */ + regs[CTX_CONTEXT_CONTROL] |= + _MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT); + + ce->lrc_desc |= CTX_DESC_FORCE_RESTORE; } static const struct intel_context_ops execlists_context_ops = { @@ -3968,7 +3982,6 @@ static void init_common_reg_state(u32 * const regs, CTX_CTRL_RS_CTX_ENABLE); regs[CTX_RING_CTL] = RING_CTL_SIZE(ring->size) | RING_VALID; - regs[CTX_BB_STATE] = RING_BB_PPGTT; } static void init_wa_bb_reg_state(u32 * const regs, diff --git a/drivers/gpu/drm/i915/gt/intel_ring_submission.c b/drivers/gpu/drm/i915/gt/intel_ring_submission.c index 066c4eddf5d0..843111b7b015 100644 --- a/drivers/gpu/drm/i915/gt/intel_ring_submission.c +++ b/drivers/gpu/drm/i915/gt/intel_ring_submission.c @@ -1347,7 +1347,7 @@ static int ring_context_pin(struct intel_context *ce) static void ring_context_reset(struct intel_context *ce) { - intel_ring_reset(ce->ring, 0); + intel_ring_reset(ce->ring, ce->ring->emit); } static const struct intel_context_ops ring_context_ops = {
We leave the kernel_context on the HW as we suspend (and while idle). There is no guarantee that is complete in memory, so we try to inhibit restoration from the kernel_context. Reinforce the inhibition by scrubbing the context. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> --- drivers/gpu/drm/i915/gt/intel_lrc.c | 17 +++++++++++++++-- drivers/gpu/drm/i915/gt/intel_ring_submission.c | 2 +- 2 files changed, 16 insertions(+), 3 deletions(-)