[PATCH][0/7] perfctr-2.7.2 for 2.6.6-mm2: summary

From: Mikael Pettersson (mikpe_at_csd.uu.se)
Date: 05/14/04

  • Next message: Mikael Pettersson: "[PATCH][1/7] perfctr-2.7.2 for 2.6.6-mm2: core" 
    Date:	Fri, 14 May 2004 16:08:49 +0200 (MEST)
To: akpm@osdl.org

    This set of patches add perfctr-2.7.2, the performance-monitoring
    counters driver, to kernel 2.6.6-mm2.

    Based on comments from Andrew Morton, this version:
    - switches to a new syscall instead of ioctl()s
    - is split in several logically distinct parts

    Summary: perfctr drives the performance counters in i386,
    x86_64, and PowerPC processors. It supports virtualised
    per-process counters with low-overhead user-space sampling,
    and global-mode counters for system-wide measurements.

    Perfctr has been in use for several years, at major HPC centres
    and among various researchers, for application-level performance
    measurements.

    Documentation is limited at the moment. We're working on it.

    Invariably, kernel hackers ask how perfctr differs from oprofile.
    The short answer is that oprofile and perfctr are based on
    different approaches with different goals. Perfctr is primarily
    about allowing application developers to count performance-
    related events in application code. Overflow interrupts can
    be caught for profiling purposes, when the hardware and kernel
    allow this, but the counts themselves are the primary objects
    of interest. Oprofile primarily does interrupt-driven profiling
    using various clock-like interrupt sources, one of which may
    be the CPU's performance counters.

    The set of patches that follow are:

    1/7: core driver files and kernel changes
    2/7: i386 driver and arch changes
    3/7: x86_64 driver and arch changes
    4/7: PowerPC driver and arch changes
    5/7: driver for virtualised (per-process) performance counters
    6/7: driver for global-mode (system-wide) performance counters
    7/7: remaining small changes

    /Mikael Pettersson
    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

  • Next message: Mikael Pettersson: "[PATCH][1/7] perfctr-2.7.2 for 2.6.6-mm2: core"

    Relevant Pages

    • [PATCH][2.6.7-mm5] perfctr low-level documentation
      ... Control data may be per-counter, global, or both. ... The counters are not assumed to be interchangeable. ... +"struct perfctr_sum_ctrs" ... +some low-level driver operations. ...
      (Linux-Kernel)
    • [PATCH][2.6.7-mm7] perfctr documentation update
      ... For detailed hardware control register layouts, ... accesses the perfctr state from a forbidden thread. ... driver to form the value for the CESR. ... interrupt-mode counters was last suspended ...
      (Linux-Kernel)
    • [PATCH][0/6] perfctr-2.7.3 for 2.6.7-rc1-mm1: summary
      ... counters driver, to kernel 2.6.7-rc1-mm1. ... and 5 for per-process perfctrs. ... Moved detailed CPU type detection on x86 from the driver ... perfctr drives the performance counters in i386, ...
      (Linux-Kernel)
    • Re: Question re out-stabling or whatever its called
      ... what does the driver do with the ticket ... Loudwater Village) used a GS outstationed at the village Harry Cross, ... Storys of "electric drills" and "winding counters back" are probably not quite true. ...
      (uk.transport.buses)
    • Re: How to detect disk activity?
      ... counters from user mode. ... it is better to write a driver from the ... Your project of monitoring disk activity is really good for someone who ...
      (microsoft.public.development.device.drivers)