Moore, Robert wrote:
Here is an explanation from one of the ACPI experts:
In most cases, it cannot be cleared until after the method is
executed.
The method usually talks to the hardware that is holding the input
signal low to tell it to "let go". If you try to clear before the
method
is executed, it won't work because the GPE will still be held low.
I still don't see how the background method can execute with an interrupt
pending. I suggest that the handler should disable the pending event,
schedule the background method to deal with the hardware, and clear the
event status. The event/interrupt wont be refired because it is disabled.
The background task will have to deal with the hardware to remove the event
assertion. This would simplify the suggested fix. To that end, I have
created another patch which is attached.
I notice that in all cases, it should be that the interrupt status should be
cleared or a storm may be possible. I suggest moving the acpi_hw_clear_gpe
call in the dispatch handler to the end of the acpi_ev_gpe_dispatch routine.
This gives the code more strength (is cleared in any case) and provides
good bracketing - although it is not clear why the status should be cleared
later rather than sooner. It also sort of resembles the spec.
This does alter the hw status presented in the ctrl when the background
hander runs. But since that is asynchronous, if that data is important
it should be captured in the isr and forwarded via the handler parameter
interface.
morrow
-----Original Message-----
From: linux-acpi-owner@xxxxxxxxxxxxxxx [mailto:linux-acpi-
owner@xxxxxxxxxxxxxxx] On Behalf Of William Morrow
Sent: Friday, August 18, 2006 11:15 AM
To: Moore, Robert
Cc: akpm@xxxxxxxx; Brown, Len; linux-acpi@xxxxxxxxxxxxxxx;
jordan.crouse@xxxxxxx; Yu, Luming
Subject: Re: [patch 12/14] ACPI: Clear GPE before disabling it
Moore, Robert wrote:
Here is the required software behavior, from the ACPI spec version
3.0A,
page 138, section 5.6.2.2:
When OSPM receives a general-purpose event (the event is from a
GPEx_BLK
STS bit), OSPM does the following:
1. Disables the interrupt source (GPEx_BLK EN bit).
2. If an edge event, clears the status bit.
3. Performs one of the following:
* Dispatches to an ACPI-aware device driver.
* Queues the matching control method for execution.
* Manages a wake event using device _PRW objects.
4. If a level event, clears the status bit.
5. Enables the interrupt source.
I have to admit, I dont see how step 4 is supposed to be occurring.
If any unclearable interrupt source is active here, then the storm
will
ensue.
The background execution of a method will never happen to clear it.
I have to plead ignorance when it comes to the full understanding of
the acpi system. On re-reading your email, I think I may have missed
your point on clearing the events in other cases. If you are
advocating
that the fix should be conditioned to only clear the bit in the level
case
then I have to agree. Like I said, I frequently seem to miss the
point.
morrow
This is essentially what is implemented today in ACPICA. (It appears
that 1 and 2 are transposed in the code, this may or may not be a
problem.)
Note that edge GPEs are cleared before dispatch, but level GPEs are
cleared after dispatch.
Bob
-----Original Message-----
From: William Morrow [mailto:William.Morrow@xxxxxxx]
Sent: Thursday, August 17, 2006 9:39 AM
To: Moore, Robert
Cc: akpm@xxxxxxxx; Brown, Len; linux-acpi@xxxxxxxxxxxxxxx;
jordan.crouse@xxxxxxx; Yu, Luming
Subject: Re: [patch 12/14] ACPI: Clear GPE before disabling it
Moore, Robert wrote:
Worse, the GPE is already cleared in the edge case, at the start of
the
GPE dispatch function:
/*
* If edge-triggered, clear the GPE status bit now. Note that
* level-triggered events are cleared after the GPE is serviced.
*/
if ((GpeEventInfo->Flags & ACPI_GPE_XRUPT_TYPE_MASK) ==
ACPI_GPE_EDGE_TRIGGERED)
{
Status = AcpiHwClearGpe (GpeEventInfo);
if (ACPI_FAILURE (Status))
{
ACPI_EXCEPTION ((AE_INFO, Status,
"Unable to clear GPE[%2X]", GpeNumber));
return_UINT32 (ACPI_INTERRUPT_NOT_HANDLED);
}
}
Sorry for the late reply, but I have not looked at this for some
time.
I had to re-examine this issue.
The interrupt type is tagged as level triggered in this case.
The dispatch handler is ACPI_GPE_DISPATCH_METHOD and so the
method is queued, but the interrupt is not cleared (it is disabled
instead).
In this hw, that is all that happens. It is disabled but not
cleared.
The original comment would lead me to think that the prevailing
thought
was that disabling the gpe implied clearing the interrupt status.
That is
not the case in this hw environment, and the result is the interrupt
storm.
Although this handler is artfully done, the architecture of this
handler is unusual in that most interrupt handlers field status
and clear the offending event at once, then process the new input.
This handler is coded in a way that seems to indicate that interrupt
status is possible when events are not enabled, but is not willing
to
clear them - again indicating that the hw design is required to
block
interrupts if the event has been disabled, but the interrupt status
is
not clear. If that is true - then this code has no effect, or is
another
case which can result in an interrupt storm.
/* Check if there is anything active at all in this register */
enabled_status_byte = (u8) (status_reg & enable_reg);
if (!enabled_status_byte) {
/* No active GPEs in this register, move on */
continue;
}
Additionally, any error return which prevents the interrupt
source from being cleared will result in an interrupt storm.
This occurs several times.
The comment:
/*
* Execute the method associated with the GPE
* NOTE: Level-triggered GPEs are cleared after the method
completes.
*/
Status = AcpiOsExecute (OSL_GPE_HANDLER,
AcpiEvAsynchExecuteGpeMethod, GpeEventInfo);
is alarming, since it documents that the interrupt is active and
expects the OS to be able to schedule with pending interrupts.
My personal preference is that the code be modified to work in
the more traditional way, but since that would be a large change
and this produced the desired result - I opted for the minimum
coding distance change.
If there are any other materials you need to evaluate this change,
let
me know.
I hope this addresses your point. I am sort of renowned for being a
little askew
when trying to explain myself.
Thanks for your attention!
morrow
-----Original Message-----
From: Moore, Robert
Sent: Wednesday, August 16, 2006 2:50 PM
To: 'akpm@xxxxxxxx'; Brown, Len
Cc: linux-acpi@xxxxxxxxxxxxxxx; william.morrow@xxxxxxx;
jordan.crouse@xxxxxxx; Yu, Luming
Subject: RE: [patch 12/14] ACPI: Clear GPE before disabling it
How does this patch relate to level-triggered GPEs, where we have
the
following comment in the code just after the patch:
/*
* Execute the method associated with the GPE
* NOTE: Level-triggered GPEs are cleared after the method
completes.
*/
Status = AcpiOsExecute (OSL_GPE_HANDLER,
AcpiEvAsynchExecuteGpeMethod, GpeEventInfo);
-----Original Message-----
From: akpm@xxxxxxxx [mailto:akpm@xxxxxxxx]
Sent: Monday, August 14, 2006 10:38 PM
To: Brown, Len
Cc: linux-acpi@xxxxxxxxxxxxxxx; akpm@xxxxxxxx;
william.morrow@xxxxxxx;
jordan.crouse@xxxxxxx; Yu, Luming; Moore, Robert
Subject: [patch 12/14] ACPI: Clear GPE before disabling it
From: William Morrow <william.morrow@xxxxxxx>
On some BIOSen, the GPE bit will remain set even if it is
disabled,
resulting in a interrupt storm. This patch clears the bit before
disabling
it.
Signed-off-by: William Morrow <william.morrow@xxxxxxx>
Signed-off-by: Jordan Crouse <jordan.crouse@xxxxxxx>
Cc: "Yu, Luming" <luming.yu@xxxxxxxxx>
Cc: "Brown, Len" <len.brown@xxxxxxxxx>
Cc: "Moore, Robert" <robert.moore@xxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxx>
---
drivers/acpi/events/evgpe.c | 14 +++++++++++++-
1 file changed, 13 insertions(+), 1 deletion(-)
diff -puN
drivers/acpi/events/evgpe.c~acpi-clear-gpe-before-disabling-it
drivers/acpi/events/evgpe.c
---
a/drivers/acpi/events/evgpe.c~acpi-clear-gpe-before-disabling-it
+++ a/drivers/acpi/events/evgpe.c
@@ -677,10 +677,22 @@ acpi_ev_gpe_dispatch(struct acpi_gpe_eve
case ACPI_GPE_DISPATCH_METHOD:
/*
- * Disable GPE, so it doesn't keep firing before
the
method
has a
+ * Clear GPE, so it doesn't keep firing before
the
method has
a
* chance to run.
*/
+ status = acpi_hw_clear_gpe(gpe_event_info);
+ if (ACPI_FAILURE(status)) {
+ ACPI_EXCEPTION((AE_INFO, status,
+ "Unable to clear
GPE[%2X]",
+ gpe_number));
+
return_UINT32(ACPI_INTERRUPT_NOT_HANDLED);
+ }
+ /*
+ * Disable GPE, so it doesn't keep happen again.
+ */
+
status = acpi_ev_disable_gpe(gpe_event_info);
+
if (ACPI_FAILURE(status)) {
ACPI_EXCEPTION((AE_INFO, status,
"Unable to disable GPE[%2X]",
_
-
To unsubscribe from this list: send the line "unsubscribe linux-acpi"
in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
--- drivers/acpi/events/evgpe.c.orig 2006-08-03 18:52:32.000000000 -0600
+++ drivers/acpi/events/evgpe.c 2006-08-18 15:17:03.000000000 -0600
@@ -637,18 +637,6 @@
handler->
context);
- /* It is now safe to clear level-triggered events. */
-
- if ((gpe_event_info->flags & ACPI_GPE_XRUPT_TYPE_MASK) ==
- ACPI_GPE_LEVEL_TRIGGERED) {
- status = acpi_hw_clear_gpe(gpe_event_info);
- if (ACPI_FAILURE(status)) {
- ACPI_EXCEPTION((AE_INFO, status,
- "Unable to clear GPE[%2X]",
- gpe_number));
- return_UINT32(ACPI_INTERRUPT_NOT_HANDLED);
- }
- }
break;
case ACPI_GPE_DISPATCH_METHOD:
@@ -701,6 +689,21 @@
break;
}
+ /* It is now safe to clear level-triggered events. */
+ if ((gpe_event_info->flags & ACPI_GPE_XRUPT_TYPE_MASK) ==
+ ACPI_GPE_LEVEL_TRIGGERED) {
+ /*
+ * Clear GPE, so interrupt doesn't storm
+ */
+ status = acpi_hw_clear_gpe(gpe_event_info);
+ if (ACPI_FAILURE(status)) {
+ ACPI_EXCEPTION((AE_INFO, status,
+ "Unable to clear GPE[%2X]",
+ gpe_number));
+ return_UINT32(ACPI_INTERRUPT_NOT_HANDLED);
+ }
+ }
+
return_UINT32(ACPI_INTERRUPT_HANDLED);
}
