Re: [Intel-gfx] [PATCH] drm/i915/dg2: Catch and log more unexpected values in DG1_MSTR_TILE_INTR
On 06/06/2022 16:21, Matt Roper wrote:
On Mon, Jun 06, 2022 at 12:55:20PM +0100, Tvrtko Ursulin wrote:
On 27/05/2022 19:42, Matt Roper wrote:
On Thu, May 26, 2022 at 11:18:17AM +0100, Tvrtko Ursulin wrote:
On 25/05/2022 19:05, Matt Roper wrote:
On Wed, May 25, 2022 at 05:03:13PM +0100, Tvrtko Ursulin wrote:
On 24/05/2022 18:51, Matt Roper wrote:
On Tue, May 24, 2022 at 10:43:39AM +0100, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin
Catch and log any garbage in the register, including no tiles marked, or
multiple tiles marked.
Signed-off-by: Tvrtko Ursulin
Cc: Matt Roper
---
We caught garbage in DG1_MSTR_TILE_INTR with DG2 (actual value 0xF9D2C008)
during glmark and more badness. So I thought lets log all possible failure
modes from here and also use per device logging.
---
drivers/gpu/drm/i915/i915_irq.c | 33 ++---
drivers/gpu/drm/i915/i915_reg.h | 1 +
2 files changed, 23 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 73cebc6aa650..79853d3fc1ed 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -2778,24 +2778,30 @@ static irqreturn_t dg1_irq_handler(int irq, void *arg)
u32 gu_misc_iir;
if (!intel_irqs_enabled(i915))
- return IRQ_NONE;
+ goto none;
master_tile_ctl = dg1_master_intr_disable(regs);
- if (!master_tile_ctl) {
- dg1_master_intr_enable(regs);
- return IRQ_NONE;
+ if (!master_tile_ctl)
+ goto enable_none;
+
+ if (master_tile_ctl & ~(DG1_MSTR_IRQ | DG1_MSTR_TILE_MASK)) {
+ drm_warn(&i915->drm, "Garbage in master_tile_ctl: 0x%08x!\n",
+master_tile_ctl);
I know we have a bunch of them already, but shouldn't we be avoiding
printk-based stuff like this inside interrupt handlers? Should we be
migrating all these error messages over to trace_printk or something
similar that's safer to use?
Not sure - I kind of think some really unexpected and worrying situations
should be loud and on by default. Risk is then spam if not ratelimited.
Maybe we should instead ratelimit most errors/warnings coming for irq
handlers?
It's not the risk of spam that's the problem, but rather that
printk-based stuff eventually calls into the console code to flush its
buffers. That's way more overhead than you want in an interrupt handler
so it's bad on its own, but if you're using something slow like a serial
console, it becomes even more of a problem.
Is it a problem for messages which we never expect to see?
Kind of. While not as catastrophic, it's the same argument for why we
don't use BUG() anymore...when the impossible does manage to happen
there's unnecessary collateral damage on things outside of graphics. If
we're adding huge delays inside an interrupt handler (while other
interrupts are disabled) that impacts the system-wide usability, not
just our own driver.
I'd also argue that these messages actually are semi-expected. Random
bits being set shouldn't happen, but in the world of dgpu's, we do
occasionally see cases where the PCI link itself goes down for reasons
outside our control and then all registers read back as 0x,
which will probably trigger error messages here (as well as a bunch of
other places).
Could you expand a bit on what is semi-expected and when? I mean the
circumstances of PCI link going down. We certainly don't have any code to
survive that.
Yeah, I'm referring to the "Lost access to MMIO BAR" errors; in the past
most of them have ultimately been tracked down to bugs in early
firmware, so flashing an updated IFWI/BIOS onto the device usually
solved the problems. Generally those buggy firmwares are an internal
problem that never make it into the wild, but I think we have also seen
cases where they get triggered by physical/electrical problems on a
specific part; that can potentially happen to anyone who's unlucky
enough to get a defective/damaged unit.
Basically "hardware returns all F's" happens because the CPU initiates
an MMIO transaction with the hardware, the hardware fails to produce any
response (possibly due to failing hardware, possibly due to
firmware/BIOS bugs), so 0x gets returned as an autocompletion to
prevent the CPU core from hanging.
It looks like we still have a few open here:
https://gitlab.freedesktop.org/search?search=%22Lost+access+to+MMIO+BAR%22&group_id=2642&project_id=4519&scope=issues&search_code=false&snippets=false&repository_ref=&nav_source=navbar
and there are some features on specific platforms we haven't turned on
yet because they also trigger these failures (which is still under
debug).
We don't/can't really do much to handle these problems in i915 today
except printing the 'lost access' error so that we know to ignore
whatever kinds of bogus errors we get after that point (usually lots of
messages about forcewake failing
Re: [Intel-gfx] [PATCH] drm/i915/dg2: Catch and log more unexpected values in DG1_MSTR_TILE_INTR
On Mon, Jun 06, 2022 at 12:55:20PM +0100, Tvrtko Ursulin wrote:
>
> On 27/05/2022 19:42, Matt Roper wrote:
> > On Thu, May 26, 2022 at 11:18:17AM +0100, Tvrtko Ursulin wrote:
> > > On 25/05/2022 19:05, Matt Roper wrote:
> > > > On Wed, May 25, 2022 at 05:03:13PM +0100, Tvrtko Ursulin wrote:
> > > > >
> > > > > On 24/05/2022 18:51, Matt Roper wrote:
> > > > > > On Tue, May 24, 2022 at 10:43:39AM +0100, Tvrtko Ursulin wrote:
> > > > > > > From: Tvrtko Ursulin
> > > > > > >
> > > > > > > Catch and log any garbage in the register, including no tiles
> > > > > > > marked, or
> > > > > > > multiple tiles marked.
> > > > > > >
> > > > > > > Signed-off-by: Tvrtko Ursulin
> > > > > > > Cc: Matt Roper
> > > > > > > ---
> > > > > > > We caught garbage in DG1_MSTR_TILE_INTR with DG2 (actual value
> > > > > > > 0xF9D2C008)
> > > > > > > during glmark and more badness. So I thought lets log all
> > > > > > > possible failure
> > > > > > > modes from here and also use per device logging.
> > > > > > > ---
> > > > > > > drivers/gpu/drm/i915/i915_irq.c | 33
> > > > > > > ++---
> > > > > > > drivers/gpu/drm/i915/i915_reg.h | 1 +
> > > > > > > 2 files changed, 23 insertions(+), 11 deletions(-)
> > > > > > >
> > > > > > > diff --git a/drivers/gpu/drm/i915/i915_irq.c
> > > > > > > b/drivers/gpu/drm/i915/i915_irq.c
> > > > > > > index 73cebc6aa650..79853d3fc1ed 100644
> > > > > > > --- a/drivers/gpu/drm/i915/i915_irq.c
> > > > > > > +++ b/drivers/gpu/drm/i915/i915_irq.c
> > > > > > > @@ -2778,24 +2778,30 @@ static irqreturn_t dg1_irq_handler(int
> > > > > > > irq, void *arg)
> > > > > > > u32 gu_misc_iir;
> > > > > > > if (!intel_irqs_enabled(i915))
> > > > > > > - return IRQ_NONE;
> > > > > > > + goto none;
> > > > > > > master_tile_ctl = dg1_master_intr_disable(regs);
> > > > > > > - if (!master_tile_ctl) {
> > > > > > > - dg1_master_intr_enable(regs);
> > > > > > > - return IRQ_NONE;
> > > > > > > + if (!master_tile_ctl)
> > > > > > > + goto enable_none;
> > > > > > > +
> > > > > > > + if (master_tile_ctl & ~(DG1_MSTR_IRQ | DG1_MSTR_TILE_MASK)) {
> > > > > > > + drm_warn(&i915->drm, "Garbage in master_tile_ctl:
> > > > > > > 0x%08x!\n",
> > > > > > > + master_tile_ctl);
> > > > > >
> > > > > > I know we have a bunch of them already, but shouldn't we be avoiding
> > > > > > printk-based stuff like this inside interrupt handlers? Should we
> > > > > > be
> > > > > > migrating all these error messages over to trace_printk or something
> > > > > > similar that's safer to use?
> > > > >
> > > > > Not sure - I kind of think some really unexpected and worrying
> > > > > situations
> > > > > should be loud and on by default. Risk is then spam if not
> > > > > ratelimited.
> > > > > Maybe we should instead ratelimit most errors/warnings coming for irq
> > > > > handlers?
> > > >
> > > > It's not the risk of spam that's the problem, but rather that
> > > > printk-based stuff eventually calls into the console code to flush its
> > > > buffers. That's way more overhead than you want in an interrupt handler
> > > > so it's bad on its own, but if you're using something slow like a serial
> > > > console, it becomes even more of a problem.
> > >
> > > Is it a problem for messages which we never expect to see?
> >
> > Kind of. While not as catastrophic, it's the same argument for why we
> > don't use BUG() anymore...when the impossible does manage to happen
> > there's unnecessary collateral damage on things outside of graphics. If
> > we're adding huge delays inside an interrupt handler (while other
> > interrupts are disabled) that impacts the system-wide usability, not
> > just our own driver.
> >
> > I'd also argue that these messages actually are semi-expected. Random
> > bits being set shouldn't happen, but in the world of dgpu's, we do
> > occasionally see cases where the PCI link itself goes down for reasons
> > outside our control and then all registers read back as 0x,
> > which will probably trigger error messages here (as well as a bunch of
> > other places).
>
> Could you expand a bit on what is semi-expected and when? I mean the
> circumstances of PCI link going down. We certainly don't have any code to
> survive that.
Yeah, I'm referring to the "Lost access to MMIO BAR" errors; in the past
most of them have ultimately been tracked down to bugs in early
firmware, so flashing an updated IFWI/BIOS onto the device usually
solved the problems. Generally those buggy firmwares are an internal
problem that never make it into the wild, but I think we have also seen
cases where they get triggered by physical/electrical problems on a
specific part; that can potentially happen to anyone who's unlucky
enough to get a defective/damaged unit.
Basically "hardware returns all F's" happens because the CPU initiates
an MMIO transaction with the hardware, the hardware fai
Re: [Intel-gfx] [PATCH] drm/i915/dg2: Catch and log more unexpected values in DG1_MSTR_TILE_INTR
On 27/05/2022 19:42, Matt Roper wrote:
On Thu, May 26, 2022 at 11:18:17AM +0100, Tvrtko Ursulin wrote:
On 25/05/2022 19:05, Matt Roper wrote:
On Wed, May 25, 2022 at 05:03:13PM +0100, Tvrtko Ursulin wrote:
On 24/05/2022 18:51, Matt Roper wrote:
On Tue, May 24, 2022 at 10:43:39AM +0100, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin
Catch and log any garbage in the register, including no tiles marked, or
multiple tiles marked.
Signed-off-by: Tvrtko Ursulin
Cc: Matt Roper
---
We caught garbage in DG1_MSTR_TILE_INTR with DG2 (actual value 0xF9D2C008)
during glmark and more badness. So I thought lets log all possible failure
modes from here and also use per device logging.
---
drivers/gpu/drm/i915/i915_irq.c | 33 ++---
drivers/gpu/drm/i915/i915_reg.h | 1 +
2 files changed, 23 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 73cebc6aa650..79853d3fc1ed 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -2778,24 +2778,30 @@ static irqreturn_t dg1_irq_handler(int irq, void *arg)
u32 gu_misc_iir;
if (!intel_irqs_enabled(i915))
- return IRQ_NONE;
+ goto none;
master_tile_ctl = dg1_master_intr_disable(regs);
- if (!master_tile_ctl) {
- dg1_master_intr_enable(regs);
- return IRQ_NONE;
+ if (!master_tile_ctl)
+ goto enable_none;
+
+ if (master_tile_ctl & ~(DG1_MSTR_IRQ | DG1_MSTR_TILE_MASK)) {
+ drm_warn(&i915->drm, "Garbage in master_tile_ctl: 0x%08x!\n",
+master_tile_ctl);
I know we have a bunch of them already, but shouldn't we be avoiding
printk-based stuff like this inside interrupt handlers? Should we be
migrating all these error messages over to trace_printk or something
similar that's safer to use?
Not sure - I kind of think some really unexpected and worrying situations
should be loud and on by default. Risk is then spam if not ratelimited.
Maybe we should instead ratelimit most errors/warnings coming for irq
handlers?
It's not the risk of spam that's the problem, but rather that
printk-based stuff eventually calls into the console code to flush its
buffers. That's way more overhead than you want in an interrupt handler
so it's bad on its own, but if you're using something slow like a serial
console, it becomes even more of a problem.
Is it a problem for messages which we never expect to see?
Kind of. While not as catastrophic, it's the same argument for why we
don't use BUG() anymore...when the impossible does manage to happen
there's unnecessary collateral damage on things outside of graphics. If
we're adding huge delays inside an interrupt handler (while other
interrupts are disabled) that impacts the system-wide usability, not
just our own driver.
I'd also argue that these messages actually are semi-expected. Random
bits being set shouldn't happen, but in the world of dgpu's, we do
occasionally see cases where the PCI link itself goes down for reasons
outside our control and then all registers read back as 0x,
which will probably trigger error messages here (as well as a bunch of
other places).
Could you expand a bit on what is semi-expected and when? I mean the
circumstances of PCI link going down. We certainly don't have any code
to survive that.
While the unexpected bits in the master tile register are strange and
may point to a bigger problem somewhere else, they're also harmless on
their own since we should just ignore those bits and only process the
valid tiles.
Yes, I was expecting that a patch belonging to multi-tile enablement would
be incoming soon, which would be changing:
+ if (REG_FIELD_GET(DG1_MSTR_TILE_MASK, master_tile_ctl) !=
+ DG1_MSTR_TILE(0)) {
+ drm_warn(&i915->drm, "Unexpected irq from tile %u!\n",
+ilog2(REG_FIELD_GET(DG1_MSTR_TILE_MASK,
+master_tile_ctl)));
+ goto enable_none;
}
From this patch, into something completely different like walking bit by
bit, handling the present tiles, and warning on unexpected ones. What should
remain though is warning on no tiles signaled (which what we saw, together
with garbage in reserved bits).
Yeah. Although I still feel the interrupt handler should really just be
flagging the errors so that the actual prints themselves can happen
outside the interrupt.
In this particular case at least DRM_ERROR with no device info is the odd
one out in the entire file so I'd suggest changing at least that, if the
rest of my changes is of questionable benefit.
Changing DRM_ERROR -> drm_err would probably be fine in the short term
since it doesn't really make us any worse off. Changing to drm_warn
might not be great since we're generating a lot more lines of output and
Sorry I don't follow - why doe
Re: [Intel-gfx] [PATCH] drm/i915/dg2: Catch and log more unexpected values in DG1_MSTR_TILE_INTR
On Thu, May 26, 2022 at 11:18:17AM +0100, Tvrtko Ursulin wrote:
>
> On 25/05/2022 19:05, Matt Roper wrote:
> > On Wed, May 25, 2022 at 05:03:13PM +0100, Tvrtko Ursulin wrote:
> > >
> > > On 24/05/2022 18:51, Matt Roper wrote:
> > > > On Tue, May 24, 2022 at 10:43:39AM +0100, Tvrtko Ursulin wrote:
> > > > > From: Tvrtko Ursulin
> > > > >
> > > > > Catch and log any garbage in the register, including no tiles marked,
> > > > > or
> > > > > multiple tiles marked.
> > > > >
> > > > > Signed-off-by: Tvrtko Ursulin
> > > > > Cc: Matt Roper
> > > > > ---
> > > > > We caught garbage in DG1_MSTR_TILE_INTR with DG2 (actual value
> > > > > 0xF9D2C008)
> > > > > during glmark and more badness. So I thought lets log all possible
> > > > > failure
> > > > > modes from here and also use per device logging.
> > > > > ---
> > > > >drivers/gpu/drm/i915/i915_irq.c | 33
> > > > > ++---
> > > > >drivers/gpu/drm/i915/i915_reg.h | 1 +
> > > > >2 files changed, 23 insertions(+), 11 deletions(-)
> > > > >
> > > > > diff --git a/drivers/gpu/drm/i915/i915_irq.c
> > > > > b/drivers/gpu/drm/i915/i915_irq.c
> > > > > index 73cebc6aa650..79853d3fc1ed 100644
> > > > > --- a/drivers/gpu/drm/i915/i915_irq.c
> > > > > +++ b/drivers/gpu/drm/i915/i915_irq.c
> > > > > @@ -2778,24 +2778,30 @@ static irqreturn_t dg1_irq_handler(int irq,
> > > > > void *arg)
> > > > > u32 gu_misc_iir;
> > > > > if (!intel_irqs_enabled(i915))
> > > > > - return IRQ_NONE;
> > > > > + goto none;
> > > > > master_tile_ctl = dg1_master_intr_disable(regs);
> > > > > - if (!master_tile_ctl) {
> > > > > - dg1_master_intr_enable(regs);
> > > > > - return IRQ_NONE;
> > > > > + if (!master_tile_ctl)
> > > > > + goto enable_none;
> > > > > +
> > > > > + if (master_tile_ctl & ~(DG1_MSTR_IRQ | DG1_MSTR_TILE_MASK)) {
> > > > > + drm_warn(&i915->drm, "Garbage in master_tile_ctl:
> > > > > 0x%08x!\n",
> > > > > + master_tile_ctl);
> > > >
> > > > I know we have a bunch of them already, but shouldn't we be avoiding
> > > > printk-based stuff like this inside interrupt handlers? Should we be
> > > > migrating all these error messages over to trace_printk or something
> > > > similar that's safer to use?
> > >
> > > Not sure - I kind of think some really unexpected and worrying situations
> > > should be loud and on by default. Risk is then spam if not ratelimited.
> > > Maybe we should instead ratelimit most errors/warnings coming for irq
> > > handlers?
> >
> > It's not the risk of spam that's the problem, but rather that
> > printk-based stuff eventually calls into the console code to flush its
> > buffers. That's way more overhead than you want in an interrupt handler
> > so it's bad on its own, but if you're using something slow like a serial
> > console, it becomes even more of a problem.
>
> Is it a problem for messages which we never expect to see?
Kind of. While not as catastrophic, it's the same argument for why we
don't use BUG() anymore...when the impossible does manage to happen
there's unnecessary collateral damage on things outside of graphics. If
we're adding huge delays inside an interrupt handler (while other
interrupts are disabled) that impacts the system-wide usability, not
just our own driver.
I'd also argue that these messages actually are semi-expected. Random
bits being set shouldn't happen, but in the world of dgpu's, we do
occasionally see cases where the PCI link itself goes down for reasons
outside our control and then all registers read back as 0x,
which will probably trigger error messages here (as well as a bunch of
other places).
>
> > While the unexpected bits in the master tile register are strange and
> > may point to a bigger problem somewhere else, they're also harmless on
> > their own since we should just ignore those bits and only process the
> > valid tiles.
>
> Yes, I was expecting that a patch belonging to multi-tile enablement would
> be incoming soon, which would be changing:
>
> + if (REG_FIELD_GET(DG1_MSTR_TILE_MASK, master_tile_ctl) !=
> + DG1_MSTR_TILE(0)) {
> + drm_warn(&i915->drm, "Unexpected irq from tile %u!\n",
> + ilog2(REG_FIELD_GET(DG1_MSTR_TILE_MASK,
> + master_tile_ctl)));
> + goto enable_none;
> }
>
> From this patch, into something completely different like walking bit by
> bit, handling the present tiles, and warning on unexpected ones. What should
> remain though is warning on no tiles signaled (which what we saw, together
> with garbage in reserved bits).
Yeah. Although I still feel the interrupt handler should really just be
flagging the errors so that the actual prints themselves can happen
outside the interrupt.
>
> > > In this particular case at least DRM_ERROR with no device info is the odd
> > > one out in t
Re: [Intel-gfx] [PATCH] drm/i915/dg2: Catch and log more unexpected values in DG1_MSTR_TILE_INTR
On 25/05/2022 19:14, Lucas De Marchi wrote:
On Wed, May 25, 2022 at 05:03:13PM +0100, Tvrtko Ursulin wrote:
On 24/05/2022 18:51, Matt Roper wrote:
On Tue, May 24, 2022 at 10:43:39AM +0100, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin
Catch and log any garbage in the register, including no tiles
marked, or
multiple tiles marked.
Signed-off-by: Tvrtko Ursulin
Cc: Matt Roper
---
We caught garbage in DG1_MSTR_TILE_INTR with DG2 (actual value
0xF9D2C008)
during glmark and more badness. So I thought lets log all possible
failure
modes from here and also use per device logging.
---
drivers/gpu/drm/i915/i915_irq.c | 33 ++---
drivers/gpu/drm/i915/i915_reg.h | 1 +
2 files changed, 23 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_irq.c
b/drivers/gpu/drm/i915/i915_irq.c
index 73cebc6aa650..79853d3fc1ed 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -2778,24 +2778,30 @@ static irqreturn_t dg1_irq_handler(int irq,
void *arg)
u32 gu_misc_iir;
if (!intel_irqs_enabled(i915))
- return IRQ_NONE;
+ goto none;
master_tile_ctl = dg1_master_intr_disable(regs);
- if (!master_tile_ctl) {
- dg1_master_intr_enable(regs);
- return IRQ_NONE;
+ if (!master_tile_ctl)
+ goto enable_none;
+
+ if (master_tile_ctl & ~(DG1_MSTR_IRQ | DG1_MSTR_TILE_MASK)) {
+ drm_warn(&i915->drm, "Garbage in master_tile_ctl: 0x%08x!\n",
+ master_tile_ctl);
I know we have a bunch of them already, but shouldn't we be avoiding
printk-based stuff like this inside interrupt handlers? Should we be
migrating all these error messages over to trace_printk or something
similar that's safer to use?
Not sure - I kind of think some really unexpected and worrying
situations should be loud and on by default. Risk is then spam if not
ratelimited. Maybe we should instead ratelimit most errors/warnings
coming for irq handlers?
In this particular case at least DRM_ERROR with no device info is the
odd one out in the entire file so I'd suggest changing at least that,
if the rest of my changes is of questionable benefit.
I'd rather remove the printk's from irq rather than adding more. At the
very
least, they should be the _once variant or ratelimited. One of the few
cases to even deserve a unlikely(), even to document this shouldn't
really be happening.
I would support ratelimited for all the unexpected bits set, no bits
set, or similar conditions. I wouldn't remove such printks to
micro-optimize things. That would potentially lose important feedback in
cases when we hit truly unexpected situations.
But annotating them as unlikely would be a good thing.
Our irq handlers (particularly on dgfx and multi-gt) are already too
long running... I don't like making them any onger or slower.
How come? I mean which interrupts are a problem there? Isn't GuC
supposed to be taking on that load on itself, isn't that one of the main
selling points?
Regards,
Tvrtko
Re: [Intel-gfx] [PATCH] drm/i915/dg2: Catch and log more unexpected values in DG1_MSTR_TILE_INTR
On 25/05/2022 19:05, Matt Roper wrote:
On Wed, May 25, 2022 at 05:03:13PM +0100, Tvrtko Ursulin wrote:
On 24/05/2022 18:51, Matt Roper wrote:
On Tue, May 24, 2022 at 10:43:39AM +0100, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin
Catch and log any garbage in the register, including no tiles marked, or
multiple tiles marked.
Signed-off-by: Tvrtko Ursulin
Cc: Matt Roper
---
We caught garbage in DG1_MSTR_TILE_INTR with DG2 (actual value 0xF9D2C008)
during glmark and more badness. So I thought lets log all possible failure
modes from here and also use per device logging.
---
drivers/gpu/drm/i915/i915_irq.c | 33 ++---
drivers/gpu/drm/i915/i915_reg.h | 1 +
2 files changed, 23 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 73cebc6aa650..79853d3fc1ed 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -2778,24 +2778,30 @@ static irqreturn_t dg1_irq_handler(int irq, void *arg)
u32 gu_misc_iir;
if (!intel_irqs_enabled(i915))
- return IRQ_NONE;
+ goto none;
master_tile_ctl = dg1_master_intr_disable(regs);
- if (!master_tile_ctl) {
- dg1_master_intr_enable(regs);
- return IRQ_NONE;
+ if (!master_tile_ctl)
+ goto enable_none;
+
+ if (master_tile_ctl & ~(DG1_MSTR_IRQ | DG1_MSTR_TILE_MASK)) {
+ drm_warn(&i915->drm, "Garbage in master_tile_ctl: 0x%08x!\n",
+master_tile_ctl);
I know we have a bunch of them already, but shouldn't we be avoiding
printk-based stuff like this inside interrupt handlers? Should we be
migrating all these error messages over to trace_printk or something
similar that's safer to use?
Not sure - I kind of think some really unexpected and worrying situations
should be loud and on by default. Risk is then spam if not ratelimited.
Maybe we should instead ratelimit most errors/warnings coming for irq
handlers?
It's not the risk of spam that's the problem, but rather that
printk-based stuff eventually calls into the console code to flush its
buffers. That's way more overhead than you want in an interrupt handler
so it's bad on its own, but if you're using something slow like a serial
console, it becomes even more of a problem.
Is it a problem for messages which we never expect to see?
While the unexpected bits in the master tile register are strange and
may point to a bigger problem somewhere else, they're also harmless on
their own since we should just ignore those bits and only process the
valid tiles.
Yes, I was expecting that a patch belonging to multi-tile enablement
would be incoming soon, which would be changing:
+ if (REG_FIELD_GET(DG1_MSTR_TILE_MASK, master_tile_ctl) !=
+ DG1_MSTR_TILE(0)) {
+ drm_warn(&i915->drm, "Unexpected irq from tile %u!\n",
+ilog2(REG_FIELD_GET(DG1_MSTR_TILE_MASK,
+master_tile_ctl)));
+ goto enable_none;
}
From this patch, into something completely different like walking bit
by bit, handling the present tiles, and warning on unexpected ones. What
should remain though is warning on no tiles signaled (which what we saw,
together with garbage in reserved bits).
In this particular case at least DRM_ERROR with no device info is the odd
one out in the entire file so I'd suggest changing at least that, if the
rest of my changes is of questionable benefit.
Changing DRM_ERROR -> drm_err would probably be fine in the short term
since it doesn't really make us any worse off. Changing to drm_warn
might not be great since we're generating a lot more lines of output and
Sorry I don't follow - why does replacing drm_err with drm_warn generate
(a lot) more lines of output?
But it can be drm_err for all I care, I don't think we really have
consistent story between errors and warnings in this area.
probably multiplying the already bad overhead that shouldn't be
happening in an interrupt handler. But if we could update the interrupt
handler to just save away the details and do the actual drm_warn later,
outside the interrupt handler code, that would be okay. We should
probably work toward something like that for all of our interrupt
handler warning/error messages.
Not sure I agree - for messages which we don't expect to see it doesn't
really matter that there will be overhead when they are hit. Presumably
bad things are already happening there so spending effort to optimise
those path is questionable.
Regards,
Tvrtko
Re: [Intel-gfx] [PATCH] drm/i915/dg2: Catch and log more unexpected values in DG1_MSTR_TILE_INTR
On Wed, May 25, 2022 at 05:03:13PM +0100, Tvrtko Ursulin wrote:
On 24/05/2022 18:51, Matt Roper wrote:
On Tue, May 24, 2022 at 10:43:39AM +0100, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin
Catch and log any garbage in the register, including no tiles marked, or
multiple tiles marked.
Signed-off-by: Tvrtko Ursulin
Cc: Matt Roper
---
We caught garbage in DG1_MSTR_TILE_INTR with DG2 (actual value 0xF9D2C008)
during glmark and more badness. So I thought lets log all possible failure
modes from here and also use per device logging.
---
drivers/gpu/drm/i915/i915_irq.c | 33 ++---
drivers/gpu/drm/i915/i915_reg.h | 1 +
2 files changed, 23 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 73cebc6aa650..79853d3fc1ed 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -2778,24 +2778,30 @@ static irqreturn_t dg1_irq_handler(int irq, void *arg)
u32 gu_misc_iir;
if (!intel_irqs_enabled(i915))
- return IRQ_NONE;
+ goto none;
master_tile_ctl = dg1_master_intr_disable(regs);
- if (!master_tile_ctl) {
- dg1_master_intr_enable(regs);
- return IRQ_NONE;
+ if (!master_tile_ctl)
+ goto enable_none;
+
+ if (master_tile_ctl & ~(DG1_MSTR_IRQ | DG1_MSTR_TILE_MASK)) {
+ drm_warn(&i915->drm, "Garbage in master_tile_ctl: 0x%08x!\n",
+master_tile_ctl);
I know we have a bunch of them already, but shouldn't we be avoiding
printk-based stuff like this inside interrupt handlers? Should we be
migrating all these error messages over to trace_printk or something
similar that's safer to use?
Not sure - I kind of think some really unexpected and worrying
situations should be loud and on by default. Risk is then spam if not
ratelimited. Maybe we should instead ratelimit most errors/warnings
coming for irq handlers?
In this particular case at least DRM_ERROR with no device info is the
odd one out in the entire file so I'd suggest changing at least that,
if the rest of my changes is of questionable benefit.
I'd rather remove the printk's from irq rather than adding more. At the very
least, they should be the _once variant or ratelimited. One of the few
cases to even deserve a unlikely(), even to document this shouldn't
really be happening.
Our irq handlers (particularly on dgfx and multi-gt) are already too
long running... I don't like making them any onger or slower.
Lucas De Marchi
Re: [Intel-gfx] [PATCH] drm/i915/dg2: Catch and log more unexpected values in DG1_MSTR_TILE_INTR
On Wed, May 25, 2022 at 05:03:13PM +0100, Tvrtko Ursulin wrote:
>
> On 24/05/2022 18:51, Matt Roper wrote:
> > On Tue, May 24, 2022 at 10:43:39AM +0100, Tvrtko Ursulin wrote:
> > > From: Tvrtko Ursulin
> > >
> > > Catch and log any garbage in the register, including no tiles marked, or
> > > multiple tiles marked.
> > >
> > > Signed-off-by: Tvrtko Ursulin
> > > Cc: Matt Roper
> > > ---
> > > We caught garbage in DG1_MSTR_TILE_INTR with DG2 (actual value 0xF9D2C008)
> > > during glmark and more badness. So I thought lets log all possible failure
> > > modes from here and also use per device logging.
> > > ---
> > > drivers/gpu/drm/i915/i915_irq.c | 33 ++---
> > > drivers/gpu/drm/i915/i915_reg.h | 1 +
> > > 2 files changed, 23 insertions(+), 11 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/i915/i915_irq.c
> > > b/drivers/gpu/drm/i915/i915_irq.c
> > > index 73cebc6aa650..79853d3fc1ed 100644
> > > --- a/drivers/gpu/drm/i915/i915_irq.c
> > > +++ b/drivers/gpu/drm/i915/i915_irq.c
> > > @@ -2778,24 +2778,30 @@ static irqreturn_t dg1_irq_handler(int irq, void
> > > *arg)
> > > u32 gu_misc_iir;
> > > if (!intel_irqs_enabled(i915))
> > > - return IRQ_NONE;
> > > + goto none;
> > > master_tile_ctl = dg1_master_intr_disable(regs);
> > > - if (!master_tile_ctl) {
> > > - dg1_master_intr_enable(regs);
> > > - return IRQ_NONE;
> > > + if (!master_tile_ctl)
> > > + goto enable_none;
> > > +
> > > + if (master_tile_ctl & ~(DG1_MSTR_IRQ | DG1_MSTR_TILE_MASK)) {
> > > + drm_warn(&i915->drm, "Garbage in master_tile_ctl: 0x%08x!\n",
> > > + master_tile_ctl);
> >
> > I know we have a bunch of them already, but shouldn't we be avoiding
> > printk-based stuff like this inside interrupt handlers? Should we be
> > migrating all these error messages over to trace_printk or something
> > similar that's safer to use?
>
> Not sure - I kind of think some really unexpected and worrying situations
> should be loud and on by default. Risk is then spam if not ratelimited.
> Maybe we should instead ratelimit most errors/warnings coming for irq
> handlers?
It's not the risk of spam that's the problem, but rather that
printk-based stuff eventually calls into the console code to flush its
buffers. That's way more overhead than you want in an interrupt handler
so it's bad on its own, but if you're using something slow like a serial
console, it becomes even more of a problem.
While the unexpected bits in the master tile register are strange and
may point to a bigger problem somewhere else, they're also harmless on
their own since we should just ignore those bits and only process the
valid tiles.
>
> In this particular case at least DRM_ERROR with no device info is the odd
> one out in the entire file so I'd suggest changing at least that, if the
> rest of my changes is of questionable benefit.
Changing DRM_ERROR -> drm_err would probably be fine in the short term
since it doesn't really make us any worse off. Changing to drm_warn
might not be great since we're generating a lot more lines of output and
probably multiplying the already bad overhead that shouldn't be
happening in an interrupt handler. But if we could update the interrupt
handler to just save away the details and do the actual drm_warn later,
outside the interrupt handler code, that would be okay. We should
probably work toward something like that for all of our interrupt
handler warning/error messages.
Matt
>
> Regards,
>
> Tvrtko
>
> >
> >
> > Matt
> >
> > > + goto enable_none;
> > > }
> > > /* FIXME: we only support tile 0 for now. */
> > > - if (master_tile_ctl & DG1_MSTR_TILE(0)) {
> > > - master_ctl = raw_reg_read(regs, GEN11_GFX_MSTR_IRQ);
> > > - raw_reg_write(regs, GEN11_GFX_MSTR_IRQ, master_ctl);
> > > - } else {
> > > - DRM_ERROR("Tile not supported: 0x%08x\n", master_tile_ctl);
> > > - dg1_master_intr_enable(regs);
> > > - return IRQ_NONE;
> > > + if (REG_FIELD_GET(DG1_MSTR_TILE_MASK, master_tile_ctl) !=
> > > + DG1_MSTR_TILE(0)) {
> > > + drm_warn(&i915->drm, "Unexpected irq from tile %u!\n",
> > > + ilog2(REG_FIELD_GET(DG1_MSTR_TILE_MASK,
> > > + master_tile_ctl)));
> > > + goto enable_none;
> > > }
> > > + master_ctl = raw_reg_read(regs, GEN11_GFX_MSTR_IRQ);
> > > + raw_reg_write(regs, GEN11_GFX_MSTR_IRQ, master_ctl);
> > > +
> > > gen11_gt_irq_handler(gt, master_ctl);
> > > if (master_ctl & GEN11_DISPLAY_IRQ)
> > > @@ -2810,6 +2816,11 @@ static irqreturn_t dg1_irq_handler(int irq, void
> > > *arg)
> > > pmu_irq_stats(i915, IRQ_HANDLED);
> > > return IRQ_HANDLED;
> > > +
> > > +enable_none:
> > > + dg1_master_intr_enable(regs);
> > > +none:
> > > + return IRQ_NONE;
> > > }
> > > /* Called from drm generic code,
Re: [Intel-gfx] [PATCH] drm/i915/dg2: Catch and log more unexpected values in DG1_MSTR_TILE_INTR
On 24/05/2022 18:51, Matt Roper wrote:
On Tue, May 24, 2022 at 10:43:39AM +0100, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin
Catch and log any garbage in the register, including no tiles marked, or
multiple tiles marked.
Signed-off-by: Tvrtko Ursulin
Cc: Matt Roper
---
We caught garbage in DG1_MSTR_TILE_INTR with DG2 (actual value 0xF9D2C008)
during glmark and more badness. So I thought lets log all possible failure
modes from here and also use per device logging.
---
drivers/gpu/drm/i915/i915_irq.c | 33 ++---
drivers/gpu/drm/i915/i915_reg.h | 1 +
2 files changed, 23 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 73cebc6aa650..79853d3fc1ed 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -2778,24 +2778,30 @@ static irqreturn_t dg1_irq_handler(int irq, void *arg)
u32 gu_misc_iir;
if (!intel_irqs_enabled(i915))
- return IRQ_NONE;
+ goto none;
master_tile_ctl = dg1_master_intr_disable(regs);
- if (!master_tile_ctl) {
- dg1_master_intr_enable(regs);
- return IRQ_NONE;
+ if (!master_tile_ctl)
+ goto enable_none;
+
+ if (master_tile_ctl & ~(DG1_MSTR_IRQ | DG1_MSTR_TILE_MASK)) {
+ drm_warn(&i915->drm, "Garbage in master_tile_ctl: 0x%08x!\n",
+master_tile_ctl);
I know we have a bunch of them already, but shouldn't we be avoiding
printk-based stuff like this inside interrupt handlers? Should we be
migrating all these error messages over to trace_printk or something
similar that's safer to use?
Not sure - I kind of think some really unexpected and worrying
situations should be loud and on by default. Risk is then spam if not
ratelimited. Maybe we should instead ratelimit most errors/warnings
coming for irq handlers?
In this particular case at least DRM_ERROR with no device info is the
odd one out in the entire file so I'd suggest changing at least that, if
the rest of my changes is of questionable benefit.
Regards,
Tvrtko
Matt
+ goto enable_none;
}
/* FIXME: we only support tile 0 for now. */
- if (master_tile_ctl & DG1_MSTR_TILE(0)) {
- master_ctl = raw_reg_read(regs, GEN11_GFX_MSTR_IRQ);
- raw_reg_write(regs, GEN11_GFX_MSTR_IRQ, master_ctl);
- } else {
- DRM_ERROR("Tile not supported: 0x%08x\n", master_tile_ctl);
- dg1_master_intr_enable(regs);
- return IRQ_NONE;
+ if (REG_FIELD_GET(DG1_MSTR_TILE_MASK, master_tile_ctl) !=
+ DG1_MSTR_TILE(0)) {
+ drm_warn(&i915->drm, "Unexpected irq from tile %u!\n",
+ilog2(REG_FIELD_GET(DG1_MSTR_TILE_MASK,
+master_tile_ctl)));
+ goto enable_none;
}
+ master_ctl = raw_reg_read(regs, GEN11_GFX_MSTR_IRQ);
+ raw_reg_write(regs, GEN11_GFX_MSTR_IRQ, master_ctl);
+
gen11_gt_irq_handler(gt, master_ctl);
if (master_ctl & GEN11_DISPLAY_IRQ)
@@ -2810,6 +2816,11 @@ static irqreturn_t dg1_irq_handler(int irq, void *arg)
pmu_irq_stats(i915, IRQ_HANDLED);
return IRQ_HANDLED;
+
+enable_none:
+ dg1_master_intr_enable(regs);
+none:
+ return IRQ_NONE;
}
/* Called from drm generic code, passed 'crtc' which
diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
index d8579ab9384c..eefa301c6430 100644
--- a/drivers/gpu/drm/i915/i915_reg.h
+++ b/drivers/gpu/drm/i915/i915_reg.h
@@ -5774,6 +5774,7 @@
#define DG1_MSTR_TILE_INTR _MMIO(0x190008)
#define DG1_MSTR_IRQREG_BIT(31)
+#define DG1_MSTR_TILE_MASK REG_GENMASK(3, 0)
#define DG1_MSTR_TILE(t)REG_BIT(t)
#define GEN11_DISPLAY_INT_CTL _MMIO(0x44200)
--
2.32.0
Re: [Intel-gfx] [PATCH] drm/i915/dg2: Catch and log more unexpected values in DG1_MSTR_TILE_INTR
On Tue, May 24, 2022 at 10:43:39AM +0100, Tvrtko Ursulin wrote:
> From: Tvrtko Ursulin
>
> Catch and log any garbage in the register, including no tiles marked, or
> multiple tiles marked.
>
> Signed-off-by: Tvrtko Ursulin
> Cc: Matt Roper
> ---
> We caught garbage in DG1_MSTR_TILE_INTR with DG2 (actual value 0xF9D2C008)
> during glmark and more badness. So I thought lets log all possible failure
> modes from here and also use per device logging.
> ---
> drivers/gpu/drm/i915/i915_irq.c | 33 ++---
> drivers/gpu/drm/i915/i915_reg.h | 1 +
> 2 files changed, 23 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
> index 73cebc6aa650..79853d3fc1ed 100644
> --- a/drivers/gpu/drm/i915/i915_irq.c
> +++ b/drivers/gpu/drm/i915/i915_irq.c
> @@ -2778,24 +2778,30 @@ static irqreturn_t dg1_irq_handler(int irq, void *arg)
> u32 gu_misc_iir;
>
> if (!intel_irqs_enabled(i915))
> - return IRQ_NONE;
> + goto none;
>
> master_tile_ctl = dg1_master_intr_disable(regs);
> - if (!master_tile_ctl) {
> - dg1_master_intr_enable(regs);
> - return IRQ_NONE;
> + if (!master_tile_ctl)
> + goto enable_none;
> +
> + if (master_tile_ctl & ~(DG1_MSTR_IRQ | DG1_MSTR_TILE_MASK)) {
> + drm_warn(&i915->drm, "Garbage in master_tile_ctl: 0x%08x!\n",
> + master_tile_ctl);
I know we have a bunch of them already, but shouldn't we be avoiding
printk-based stuff like this inside interrupt handlers? Should we be
migrating all these error messages over to trace_printk or something
similar that's safer to use?
Matt
> + goto enable_none;
> }
>
> /* FIXME: we only support tile 0 for now. */
> - if (master_tile_ctl & DG1_MSTR_TILE(0)) {
> - master_ctl = raw_reg_read(regs, GEN11_GFX_MSTR_IRQ);
> - raw_reg_write(regs, GEN11_GFX_MSTR_IRQ, master_ctl);
> - } else {
> - DRM_ERROR("Tile not supported: 0x%08x\n", master_tile_ctl);
> - dg1_master_intr_enable(regs);
> - return IRQ_NONE;
> + if (REG_FIELD_GET(DG1_MSTR_TILE_MASK, master_tile_ctl) !=
> + DG1_MSTR_TILE(0)) {
> + drm_warn(&i915->drm, "Unexpected irq from tile %u!\n",
> + ilog2(REG_FIELD_GET(DG1_MSTR_TILE_MASK,
> + master_tile_ctl)));
> + goto enable_none;
> }
>
> + master_ctl = raw_reg_read(regs, GEN11_GFX_MSTR_IRQ);
> + raw_reg_write(regs, GEN11_GFX_MSTR_IRQ, master_ctl);
> +
> gen11_gt_irq_handler(gt, master_ctl);
>
> if (master_ctl & GEN11_DISPLAY_IRQ)
> @@ -2810,6 +2816,11 @@ static irqreturn_t dg1_irq_handler(int irq, void *arg)
> pmu_irq_stats(i915, IRQ_HANDLED);
>
> return IRQ_HANDLED;
> +
> +enable_none:
> + dg1_master_intr_enable(regs);
> +none:
> + return IRQ_NONE;
> }
>
> /* Called from drm generic code, passed 'crtc' which
> diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
> index d8579ab9384c..eefa301c6430 100644
> --- a/drivers/gpu/drm/i915/i915_reg.h
> +++ b/drivers/gpu/drm/i915/i915_reg.h
> @@ -5774,6 +5774,7 @@
>
> #define DG1_MSTR_TILE_INTR _MMIO(0x190008)
> #define DG1_MSTR_IRQ REG_BIT(31)
> +#define DG1_MSTR_TILE_MASK REG_GENMASK(3, 0)
> #define DG1_MSTR_TILE(t) REG_BIT(t)
>
> #define GEN11_DISPLAY_INT_CTL_MMIO(0x44200)
> --
> 2.32.0
>
--
Matt Roper
Graphics Software Engineer
VTT-OSGC Platform Enablement
Intel Corporation
[Intel-gfx] [PATCH] drm/i915/dg2: Catch and log more unexpected values in DG1_MSTR_TILE_INTR
From: Tvrtko Ursulin
Catch and log any garbage in the register, including no tiles marked, or
multiple tiles marked.
Signed-off-by: Tvrtko Ursulin
Cc: Matt Roper
---
We caught garbage in DG1_MSTR_TILE_INTR with DG2 (actual value 0xF9D2C008)
during glmark and more badness. So I thought lets log all possible failure
modes from here and also use per device logging.
---
drivers/gpu/drm/i915/i915_irq.c | 33 ++---
drivers/gpu/drm/i915/i915_reg.h | 1 +
2 files changed, 23 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 73cebc6aa650..79853d3fc1ed 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -2778,24 +2778,30 @@ static irqreturn_t dg1_irq_handler(int irq, void *arg)
u32 gu_misc_iir;
if (!intel_irqs_enabled(i915))
- return IRQ_NONE;
+ goto none;
master_tile_ctl = dg1_master_intr_disable(regs);
- if (!master_tile_ctl) {
- dg1_master_intr_enable(regs);
- return IRQ_NONE;
+ if (!master_tile_ctl)
+ goto enable_none;
+
+ if (master_tile_ctl & ~(DG1_MSTR_IRQ | DG1_MSTR_TILE_MASK)) {
+ drm_warn(&i915->drm, "Garbage in master_tile_ctl: 0x%08x!\n",
+master_tile_ctl);
+ goto enable_none;
}
/* FIXME: we only support tile 0 for now. */
- if (master_tile_ctl & DG1_MSTR_TILE(0)) {
- master_ctl = raw_reg_read(regs, GEN11_GFX_MSTR_IRQ);
- raw_reg_write(regs, GEN11_GFX_MSTR_IRQ, master_ctl);
- } else {
- DRM_ERROR("Tile not supported: 0x%08x\n", master_tile_ctl);
- dg1_master_intr_enable(regs);
- return IRQ_NONE;
+ if (REG_FIELD_GET(DG1_MSTR_TILE_MASK, master_tile_ctl) !=
+ DG1_MSTR_TILE(0)) {
+ drm_warn(&i915->drm, "Unexpected irq from tile %u!\n",
+ilog2(REG_FIELD_GET(DG1_MSTR_TILE_MASK,
+master_tile_ctl)));
+ goto enable_none;
}
+ master_ctl = raw_reg_read(regs, GEN11_GFX_MSTR_IRQ);
+ raw_reg_write(regs, GEN11_GFX_MSTR_IRQ, master_ctl);
+
gen11_gt_irq_handler(gt, master_ctl);
if (master_ctl & GEN11_DISPLAY_IRQ)
@@ -2810,6 +2816,11 @@ static irqreturn_t dg1_irq_handler(int irq, void *arg)
pmu_irq_stats(i915, IRQ_HANDLED);
return IRQ_HANDLED;
+
+enable_none:
+ dg1_master_intr_enable(regs);
+none:
+ return IRQ_NONE;
}
/* Called from drm generic code, passed 'crtc' which
diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
index d8579ab9384c..eefa301c6430 100644
--- a/drivers/gpu/drm/i915/i915_reg.h
+++ b/drivers/gpu/drm/i915/i915_reg.h
@@ -5774,6 +5774,7 @@
#define DG1_MSTR_TILE_INTR _MMIO(0x190008)
#define DG1_MSTR_IRQ REG_BIT(31)
+#define DG1_MSTR_TILE_MASK REG_GENMASK(3, 0)
#define DG1_MSTR_TILE(t) REG_BIT(t)
#define GEN11_DISPLAY_INT_CTL _MMIO(0x44200)
--
2.32.0
