Re: how to minimize interrupt latency using interrupt affinity in

Thanks Eliyas.

Xperf is pretty nice but it is missing lots of the DPC/ISR details that the
tracelog provides. Actually, it would be VERY useful if the CPU or ISR usage
in Xpref would include context switches due to interrupts and DPC (so that
one can see nested interrupts etc).

I'm looking forward to getting more information from you once your sources
get back to you.

Regards

Philip

"Eliyas Yakub [MSFT]" wrote:

1) I suggest using the new Xperf tool from Microsoft
(http://msdn.microsoft.com/en-us/library/cc305187.aspx) to analyze perf
issues.
2) You cannot configure IRQL value of your interrupt. You have to work with
whatever system assigns for your device.
3) I have forward this message to some folks I know. I will let you know if
they have any idea.

-Eliyas


"pgruebele" <pgruebele@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:C10C18D1-8B22-42E1-80A9-E564DF9A6280@xxxxxxxxxxxxxxxx
I have a PCI device for which I wrote a standard PnP WDM driver. This is
a
timer device and based on timer interrupts the ISR needs to do a very
short
amount of processing - no DPCs need queueing and only the PCI board itself
is
read/written in the ISR.

My requirement is that the ISR be called within .2ms of hardware
interrupt.
However, I am seeing jitter of up to at least 1ms (ISR gets called up to
1ms
after hardware interrupt). I understand that Vista (latest updates) is
not a
RTOS and that other ISRs can take up to .25ms (according to tracelog) and
that under load conditions, several of these other ISR calls may be
handled
before my driver's ISR. But I did the following in order to make sure
that
my ISR runs on a processor core which is not ever used by other ISRs:

1. I used the interrupt affinity policy tool to change the processor
affinity for all device drivers (except mine) to 0x3 (this is on a Q6600
4-core system).

2. I used the affinity policiy tool to set my driver's affinity to 0x4

3. Using tracelog, I verified that under system load, only cores 0 & 1 are
used by other drivers' interrupts and that only core 3 is used by my ISR.

In spite of this, I am getting large jitter in ISR latency for my driver
(up
to more than 1ms) under system load (using HD, Direct3D etc) - even though
my
device is the only device generating interrupts on core 3!


I have the following questions:

1. Does anyone have any idea why this might be the case?

2. On another note, even though I set affinity to 0x3 for all drivers
except
mine, they only generate interrupts on core 0 (instead of cores 0 & 1).
Why
is that?

3. How can one change Irql and SynchronizeIrql for the interrupt. The
documentation only states that these values are supposed to be taken from
CM_PARTIAL_RESOURCE_DESCRIPTOR. What if I want my IRQL to be very high so
that my driver ISR will preempt other ISR's even if they are running on
the
same cpu core? If this is possible then I would not have to prevent all
devices except for mine from generating interrupts on core 2.

Thanks

Philip

.

Relevant Pages

  • Re: how to minimize interrupt latency using interrupt affinity in
    ... driver raises interrupt on core 2). ... on core 0, then they can only disable interrupts for that core... ... nirlpk ISR calls in this part of the xperf table are 2.86ms apart. ...
    (microsoft.public.development.device.drivers)
  • Re: how to minimize interrupt latency using interrupt affinity in
    ... it would be VERY useful if the CPU or ISR ... one can see nested interrupts etc). ... my ISR runs on a processor core which is not ever used by other ISRs: ... even though I set affinity to 0x3 for all drivers ...
    (microsoft.public.development.device.drivers)
  • how to minimize interrupt latency using interrupt affinity in Vist
    ... I have a PCI device for which I wrote a standard PnP WDM driver. ... read/written in the ISR. ... used by other drivers' interrupts and that only core 3 is used by my ISR. ...
    (microsoft.public.development.device.drivers)
  • Re: Strange problem with semFlush/semTake
    ... As I stated in one of my previous messages, the cmpTime task takes only ... As for the use of semGive instead of semFlush, ... If the ISR fires, let's say, while your task is ... I am using binary semaphores to sync between interrupts and tasks and I ...
    (comp.os.vxworks)
  • Re: how to minimize interrupt latency using interrupt affinity in
    ... on core 0, then they can only disable interrupts for that core... ... Since it happens mainly with system load (not just CPU load since interrupts ... used in various parts of the kernel and in some exported APIs (the ... nirlpk ISR calls in this part of the xperf table are 2.86ms apart. ...
    (microsoft.public.development.device.drivers)
Quantcast