[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-id: <4B135417.1080302@majjas.com>
Date: Mon, 30 Nov 2009 00:11:51 -0500
From: Michael Breuer <mbreuer@...jas.com>
To: Arjan van de Ven <arjan@...radead.org>
Cc: Ingo Molnar <mingo@...e.hu>,
Peter Zijlstra <a.p.zijlstra@...llo.nl>,
Thomas Gleixner <tglx@...utronix.de>,
Len Brown <lenb@...nel.org>,
Arnaldo Carvalho de Melo <acme@...hat.com>,
linux-kernel@...r.kernel.org
Subject: Re: Problem? intel_iommu=off; perf top shows acpi_os_read_port as
extremely busy
Ok - one more rather odd (to me) data point...
I started playing around with various settings, and traced the calls to
acpi_os_read_port.
To summarize:
With intel_iommu=off, I see a large percentage of calls to
acpi_os_read_port resulting from user apps (portsentry is #1).
With intel_iommu=on, NONE of trace points to any user apps - all derive
from the idle loop.
To make things more interesting, when I enable intel_iommu and disable
vt-d in bios, the system performs much better (20% improvement in
glxgears, for example), perf top looks like this:
------------------------------------------------------------------------------
PerfTop: 4863 irqs/sec kernel:62.7% [100000 cycles], (all, 8 CPUs)
------------------------------------------------------------------------------
samples pcnt kernel function
_______ _____ _______________
2213.00 - 5.5% : acpi_idle_enter_bm
2001.00 - 5.0% : acpi_os_read_port
1544.00 - 3.9% : _spin_lock_irqsave
1075.00 - 2.7% : ioread32
928.00 - 2.3% : find_busiest_group
851.00 - 2.1% : _spin_unlock_irqrestore
823.00 - 2.1% : hpet_next_event
810.00 - 2.0% : tg_shares_up
655.00 - 1.6% : fget_light
641.00 - 1.6% : schedule
639.00 - 1.6% : tick_nohz_stop_sched_tick
638.00 - 1.6% : sub_preempt_count
634.00 - 1.6% : add_preempt_count
548.00 - 1.4% : do_sys_poll
446.00 - 1.1% : trace_hardirqs_off
And additionally, one recurring boot warning I've seen since I first
booted this box has disappeared - first boot message of IRQ16 disabled.
I'm thinking that bad VT-D bios is causing trouble even when intel_iommu
is disabled.
On 11/29/2009 03:47 PM, Arjan van de Ven wrote:
> On Sat, 28 Nov 2009 13:10:21 -0500
> Michael Breuer<mbreuer@...jas.com> wrote:
>
>
>> Ok - my only question then is why things appear so different with
>> intel_iommu enabled.
>>
> something else is even more expensive then :0
>
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists