[Intel-gfx] [PATCH 1/2] drm/i915: Keep the ctx workarounds tightly packed
Chris Wilson
chris at chris-wilson.co.uk
Fri Jun 15 11:34:51 UTC 2018
Quoting Mika Kuoppala (2018-06-15 12:29:23)
> Chris Wilson <chris at chris-wilson.co.uk> writes:
>
> > For each platform, we have a few registers that rewritten with multiple
> > values -- they are not part of a sequence, just different parts of a
> > masked register set at different times (e.g. platform and gen
> > workarounds). Consolidate these into a single register write to keep the
> > table compact.
> >
> > While adjusting the construction of the wa table, make it non fatal so
> > that the driver still loads but keeping the warning and extra details
> > for inspection.
> >
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > Cc: Oscar Mateo <oscar.mateo at intel.com>
> > Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> > Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
> > ---
> > drivers/gpu/drm/i915/i915_debugfs.c | 25 ++--------
> > drivers/gpu/drm/i915/i915_drv.h | 2 +-
> > drivers/gpu/drm/i915/intel_workarounds.c | 63 +++++++++++++++++-------
> > 3 files changed, 52 insertions(+), 38 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
> > index c600279d3db5..f78895ffab9b 100644
> > --- a/drivers/gpu/drm/i915/i915_debugfs.c
> > +++ b/drivers/gpu/drm/i915/i915_debugfs.c
> > @@ -3378,28 +3378,13 @@ static int i915_shared_dplls_info(struct seq_file *m, void *unused)
> >
> > static int i915_wa_registers(struct seq_file *m, void *unused)
> > {
> > - struct drm_i915_private *dev_priv = node_to_i915(m->private);
> > - struct i915_workarounds *workarounds = &dev_priv->workarounds;
> > + struct i915_workarounds *wa = &node_to_i915(m->private)->workarounds;
> > int i;
> >
> > - intel_runtime_pm_get(dev_priv);
> > -
> > - seq_printf(m, "Workarounds applied: %d\n", workarounds->count);
> > - for (i = 0; i < workarounds->count; ++i) {
> > - i915_reg_t addr;
> > - u32 mask, value, read;
> > - bool ok;
> > -
> > - addr = workarounds->reg[i].addr;
> > - mask = workarounds->reg[i].mask;
> > - value = workarounds->reg[i].value;
> > - read = I915_READ(addr);
> > - ok = (value & mask) == (read & mask);
> > - seq_printf(m, "0x%X: 0x%08X, mask: 0x%08X, read: 0x%08x, status: %s\n",
> > - i915_mmio_reg_offset(addr), value, mask, read, ok ? "OK" : "FAIL");
> > - }
> > -
> > - intel_runtime_pm_put(dev_priv);
> > + seq_printf(m, "Workarounds applied: %d\n", wa->count);
> > + for (i = 0; i < wa->count; ++i)
> > + seq_printf(m, "0x%X: 0x%08X, mask: 0x%08X\n",
> > + wa->reg[i].addr, wa->reg[i].value, wa->reg[i].mask);
>
> It seems that gem_workarounds is already smart enough to not
> parse/trust the driver provided read value.
>
> Perhaps the only value in here was that we saw what
> were context and non context saved ones. But we
> have other tools for that.
We couldn't even rely on it doing that, since what it read would be
random (we may have had a context, the power context may or may not have
the same values, we may never had loaded the values)...
Fortunately, we stopped using them :)
> > return 0;
> > }
> > diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> > index 2c12de678e32..91c389622217 100644
> > --- a/drivers/gpu/drm/i915/i915_drv.h
> > +++ b/drivers/gpu/drm/i915/i915_drv.h
> > @@ -1308,7 +1308,7 @@ struct i915_frontbuffer_tracking {
> > };
> >
> > struct i915_wa_reg {
> > - i915_reg_t addr;
> > + u32 addr;
> > u32 value;
> > /* bitmask representing WA bits */
> > u32 mask;
> > diff --git a/drivers/gpu/drm/i915/intel_workarounds.c b/drivers/gpu/drm/i915/intel_workarounds.c
> > index 24b929ce3341..f8bb32e974f6 100644
> > --- a/drivers/gpu/drm/i915/intel_workarounds.c
> > +++ b/drivers/gpu/drm/i915/intel_workarounds.c
> > @@ -48,29 +48,58 @@
> > * - Public functions to init or apply the given workaround type.
> > */
> >
> > -static int wa_add(struct drm_i915_private *dev_priv,
> > - i915_reg_t addr,
> > - const u32 mask, const u32 val)
> > +static void wa_add(struct drm_i915_private *i915,
> > + i915_reg_t reg, const u32 mask, const u32 val)
> > {
> > - const unsigned int idx = dev_priv->workarounds.count;
> > + struct i915_workarounds *wa = &i915->workarounds;
> > + unsigned int start = 0, end = wa->count;
> > + unsigned int addr = i915_mmio_reg_offset(reg);
> > + struct i915_wa_reg *r;
> > +
> > + while (start < end) {
> > + unsigned int mid = start + (end - start) / 2;
> > +
> > + if (wa->reg[mid].addr < addr) {
> > + start = mid + 1;
> > + } else if (wa->reg[mid].addr > addr) {
> > + end = mid;
> > + } else {
> > + r = &wa->reg[mid];
> > +
> > + if ((mask & ~r->mask) == 0) {
> > + DRM_ERROR("Discarding overwritten w/a for reg %04x (mask: %08x, value: %08x)\n",
> > + addr, r->mask, r->value);
> > +
> > + r->value &= ~mask;
> > + }
> > +
> > + r->value |= val;
> > + r->mask |= mask;
> > + return;
> > + }
> > + }
> >
> > - if (WARN_ON(idx >= I915_MAX_WA_REGS))
> > - return -ENOSPC;
> > + if (WARN_ON_ONCE(wa->count >= I915_MAX_WA_REGS)) {
> > + DRM_ERROR("Dropping w/a for reg %04x (mask: %08x, value: %08x)\n",
> > + addr, mask, val);
> > + return;
> > + }
> >
> > - dev_priv->workarounds.reg[idx].addr = addr;
> > - dev_priv->workarounds.reg[idx].value = val;
> > - dev_priv->workarounds.reg[idx].mask = mask;
> > + r = &wa->reg[wa->count++];
> > + r->addr = addr;
> > + r->value = val;
> > + r->mask = mask;
> >
> > - dev_priv->workarounds.count++;
> > + while (r-- > wa->reg) {
> > + GEM_BUG_ON(r[0].addr == r[1].addr);
> > + if (r[1].addr > r[0].addr)
> > + break;
>
> Keeping the workarounds in a sorted order by addr,
> could be mentioned in the header. But not insisting
> on that as this is rather clear from the code.
>
> >
> > - return 0;
> > + swap(r[1], r[0]);
> > + }
> > }
> >
> > -#define WA_REG(addr, mask, val) do { \
> > - const int r = wa_add(dev_priv, (addr), (mask), (val)); \
> > - if (r) \
> > - return r; \
> > - } while (0)
> > +#define WA_REG(addr, mask, val) wa_add(dev_priv, (addr), (mask), (val))
>
> If you do this, then the whole chain of initing workarounds
> can be converted to omitting return value.
I was thinking the same for a followup. As well as not storing the
workarounds but computing them on the fly -- for the normal case, we
only need them once for the initial context image.
-Chris
More information about the Intel-gfx
mailing list