[Intel-gfx] [igt-dev] [PATCH i-g-t] i915/gem_exec_balancer: Disable pre-parser for rewritten batches

Mika Kuoppala mika.kuoppala at linux.intel.com
Tue Jun 2 08:57:20 UTC 2020


Chris Wilson <chris at chris-wilson.co.uk> writes:

> Quoting Mika Kuoppala (2020-06-01 15:56:55)
>> Chris Wilson <chris at chris-wilson.co.uk> writes:
>> 
>> > As we rewrite the batches on the fly to implement the non-preemptible
>> > lock, we need to tell Tigerlake to read the batch afresh each time.
>> > Amusingly, the disable is a part of an arb-check, so we have to be
>> > careful not to include the arbitration point inside our unpreemptible
>> > loop.
>> >
>> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>> > ---
>> >  tests/i915/gem_exec_balancer.c | 13 +++++++++----
>> >  1 file changed, 9 insertions(+), 4 deletions(-)
>> >
>> > diff --git a/tests/i915/gem_exec_balancer.c b/tests/i915/gem_exec_balancer.c
>> > index 026f8347e..0e3b52900 100644
>> > --- a/tests/i915/gem_exec_balancer.c
>> > +++ b/tests/i915/gem_exec_balancer.c
>> > @@ -1350,6 +1350,11 @@ static void __bonded_dual(int i915,
>> >       *out = cycles;
>> >  }
>> >  
>> > +static uint32_t preparser_disable(void)
>> > +{
>> > +     return 0x5 << 23 | 1 << 8 | 1; /* preparser masked disable */
>> 
>> there is MI_ARB_CHECK
>> 
>> > +}
>> > +
>> >  static uint32_t sync_from(int i915, uint32_t addr, uint32_t target)
>> >  {
>> >       uint32_t handle = gem_create(i915, 4096);
>> > @@ -1363,14 +1368,14 @@ static uint32_t sync_from(int i915, uint32_t addr, uint32_t target)
>> >       *cs++ = 0;
>> >       *cs++ = 0;
>> >  
>> > -     *cs++ = MI_NOOP;
>> > +     *cs++ = preparser_disable();
>> >       *cs++ = MI_NOOP;
>> >       *cs++ = MI_NOOP;
>> >       *cs++ = MI_NOOP;
>> >  
>> >       /* wait for them to cancel us */
>> >       *cs++ = MI_BATCH_BUFFER_START | 1 << 8 | 1;
>> > -     *cs++ = addr + 16;
>> > +     *cs++ = addr + 24;
>> 
>> I must be totally confused about the layout as I can't get
>> the +8. You take one nop out and put one arb check in
>> and everything moves with 8?
>
> It's just skipping over the MI_ARB_CHECK, +4, aligned to the next qword
> because some old habits die hard.

Well that explains it,
Reviewed-by: Mika Kuoppala <mika.kuoppala at linux.intel.com>

> -Chris


More information about the Intel-gfx mailing list