Re: interrupt cycles time

Tech-Archive recommends: Repair Windows Errors & Optimize Windows Performance

From: Mark Roddy (markr_at_hollistech.com)
Date: 03/04/05 

Date: Fri, 04 Mar 2005 08:52:10 -0500

Oliver Nuding wrote:
> Ok, I get a bit more in details.
>
> I wrote a firewire driver.

OK I'm officially confused. Do you mean that you wrote a driver for a
1394 controller on a pci bus? That would be a bit unusual. Generally you
  would want to use the existing 'in-box' driver for the 1394 controller
and write a function driver for the camera on the firewire bus
enumerated by the controller. This function driver would not have an isr
at all.

>
  The synchronisation of firewire is 125us. Within
> each of these intervals all cameras send one packet of data to the computer.
> When you use color cameras these packets are not RGB data but YUV data. I
> use to do this color conversion inside the interrupt. Each packet has its
> specific location inside an image. All packet headers except the fist of an
> image look excately the same. I may not mix them.
> In earlier versions I tested to do the work with DPCs but I got in
> synchronisation troubles because of unpredictable delays of DPCs. Even pci
> interrupts have a delay from the pci bus to the begin of the ISR (under
> windows 2000) between 4us and 22us. Tested with a 2GHz computer, measured
> with a scope.
> To be able to improve the speed of the color conversion I set an output
> (parallel port) directly in the interrupt and did time measurements with a
> scope. The latency of this output is a lot less than 1us. The result is that
> the conversion of one packet (equals the duration of an interrupt) takes
> about 50us. I do two of this conversions within 125us.
>
> My questions:
> - Why is it possible to do image processing during the same time an
> interrupt is working? This work needs a lot longer than the left time.
> - I get no real difference whether I work on safed images or captured images
> (interrupt <-> no interrupt).
> - This implies that an interrupt does not hold the cpu to be worked but has
> a dedicated area/ structure for his own.
> This would be in contrast to all that I learned. One pipeline for
> interrupts/one for the rest?
> - As conclusion? Can we save cycle time with excessive use of work in
> interrupts? (strange question!)
>
> Thanks
> Oliver Nuding
>
>
> "Stephan Wolf [MVP]" <stewo68@hotmail.com> wrote in message
> news:f1nd21hum6ngdr6rcasvegeklhcrvfkebm@4ax.com...
>
>>Sorry I don't get your point. What is your actual question?
>>
>>It seems you have two hardware interrupt requests (IRQ) within 125us.
>>Then your software interrupt service routine (ISR) probably writes to
>>some I/O port to disable the IRQ line.
>>
>>At this point one would usually leave the ISR and tell the system to
>>schedule a deferred procedure call (DPC), which will later run at a
>>lower priority level (IRQL).
>>
>>Note that there are no guarantees at all about interrupt delays in
>>Windows. Do not expect any minimum delay.
>>
>>The DPC should re-enable the hardware IRQ line right before it
>>returns. At this point, the next IRQ can be raised and the ISR can be
>>called again (note: even if the DPC has not yet returned).
>>
>>Stephan
>>---
>>On Wed, 2 Mar 2005 12:48:05 +0100, "Oliver Nuding"
>><olivernuding@filtec-europe.de> wrote:
>>
>>
>>>Dear all,
>>>
>>>Perhaps someone of you is able to explain me the following.
>>>
>>>I get two interrupts within 125us. At the beginning of the interrupt I
>
> set
>
>>>an output on one of the motherboard IOs. When I leave the interrupt I
>
> take
>
>>>it back. The scope tells me that the duration of one interrupt is about
>>>50us. This means that normaly a time of 25us would be left for other
>>>processing. But there is a lot more.
>>>
>>>This would mean that an interrupt is not worked completely in one part -
>
> or
>
>>>that it does not take all the resources of a processor. Like one pipeline
>>>for the interrupt and one for general stuff.
>>>
>>>Thanks
>>>Oliver Nuding
>
>
> 

-- 
=====================
Mark Roddy DDK MVP
Windows 2003/XP/2000 Consulting
Hollis Technology Solutions 603-321-1032
www.hollistech.com

Relevant Pages

  • Re: Assembling Visual Studio generated listing files
    ... card issues an interrupt each time it has a new position for any one ... An ISR is constrained to act in < 10us, and for most modern machines, 12.5us is a very ... Without MASM and in pure C or C++ and without any Windows API/MFC, ... This is a device driver. ...
    (microsoft.public.vc.mfc)
  • [RFC v2] HOWTO use NAPI to reduce TX interrupts
    ... A recent discussion about the spidernet driver resulted in the dicovery ... transmit paths, ... Many devices trigger an interrupt for each packet that has been ...
    (Linux-Kernel)
  • Re: bge(4) one packet wedge
    ... When the ISR starts the driver should record the status_tag value. ... If necessary the hardware will generate a new interrupt and start the ...
    (freebsd-net)
  • Re: Interrupt Problem in Our custom Serial Driver
    ... interrupt, your ISR is correctly activated and returns the right SYSINTR. ... you can focus on the IST and the driver ... required for the serial port. ...
    (microsoft.public.windowsce.platbuilder)
  • Re: Synchronization between ISR and IST - I2C Based device
    ... It sounds to me that your system design is a little flawed, your idea of a layered I2C driver is probably better then having the I2C driver in for each of your devices. ... We have implemented ISR that reads the Interrupt Status Register ... The problem we are facing one of the device holds the I2C bus longer ...
    (microsoft.public.windowsce.platbuilder)