lkml.org 
[lkml]   [2006]   [Mar]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: libata/sata_nv latency on NVIDIA CK804 [was Re: AMD64 X2 lost ticks on PM timer]

* Jeff Garzik <jeff@garzik.org> wrote:

> Ingo Molnar wrote:
> >* Bill Rugolsky Jr. <brugolsky@telemetry-investments.com> wrote:
> >
> >
> >> <...>-2913 0d.h. 8us : raise_softirq_irqoff (blk_complete_request)
> >> <...>-2913 0d.h. 8us : __ata_qc_complete (ata_qc_complete)
> >> <...>-2913 0d.h. 9us : ata_host_intr (nv_interrupt)
> >> <...>-2913 0d.h. 9us!: ata_bmdma_status (ata_host_intr)
> >> <...>-2913 0d.h. 16641us : nv_check_hotplug_ck804 (nv_interrupt)
> >> <...>-2913 0d.h. 16642us : _spin_unlock_irqrestore (nv_interrupt)
> >> <...>-2913 0d.h. 16642us : smp_apic_timer_interrupt
> >> (apic_timer_interrupt)
> >> <...>-2913 0d.h. 16642us : exit_idle (smp_apic_timer_interrupt)
> >
> >
> >ouch. The codepath in question (ata_host_intr()) doesnt seem to have any
> >loop that could take 16.6 msecs (!). This very much looks like some
> >hardware-triggered delay - some really screwed up DMA prioritization
> >perhaps, starving the host CPU for 16.6 msecs? But what DMA takes 16.6
> >msecs? That's enough time to transfer dozens of megabytes of data on a
> >midrange system.
>
> Yeah, I don't see anything offhand either.
>
> sata_nv's nv_interrupt() should be using spin_lock() rather than
> spin_lock_irqsave(), but I doubt that's a latency cause.

please see my previous codepath description. No spin_lock() calls in the
measured codepath. They would show up in the latency trace anyway.

> I would be surprised if the legacy PCI IDE registers, all PIO-based,
> would be implemented via BIOS SMM or some other similarly slow method.
> But its not impossible...

yeah. But then again - PIO ops easily take 16 usecs on almost arbitrary
PC hardware, so an SMM trap might not be all that noticeable -
especially if this register is considered legacy. (maybe Windows does
the MMIO variant)

Ingo
-
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/

\
 
 \ /
  Last update: 2006-03-15 23:40    [W:0.071 / U:1.388 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site