[ltp] Request some more detailed data on recent thinkpads, T and R Series

Henrique de Moraes Holschuh linux-thinkpad@linux-thinkpad.org
Sat, 1 Aug 2009 11:02:09 -0300


On Thu, 30 Jul 2009, hollunder@gmx.at wrote:
> On Wed, 29 Jul 2009 19:55:33 -0300
> Henrique de Moraes Holschuh <hmh@hmh.eng.br> wrote:
> > On Wed, 29 Jul 2009, hollunder@gmx.at wrote:
> > > Also important for me is whether the usb buses/ports share
> > > interrupts with something else, like graphics, and whether there's
> > > a bios option to change that.
> > 
> > Why do you need that information?  What are the constraints in which
> > interrupt sharing would make a difference?
> 
> Well, I'm into audio, which often requires quite low latencies. When
> the audio interface, or the bus it uses, shares an interrupt with say
> graphics or networking you have a problem.

Are you already used to ThinkPads for low-latency work, and you're sure it
works?  If the IBM ones did, chances are very good the Lenovo ones will also
work fine.

SMIs are system management interrupts that place the processor in SMM
(system management mode), where the OS is halted and BIOS code runs to do
some "management" function (usually, work around design defects in the CPU
or other hardware, but on thinkpads, it is used for a lot more).

So, ThinkPads have this annoying and aggravating habit of doing a lot of
SMIs and the OS has little knowledge, let alone control, about time spent in
SMM.  This can cause unpredictable latencies that the OS is often not even
aware of (unless you are using the HPET or another external counter as the
timebase, in which case, time seems to just "jump forward").

I don't know if the time spent in SMM is large enough to matter for audio,
but there is also the syndrome it causes (jitter that only goes in one way,
so it has a cumulative effect instead of an average zero effect) which might
matter.

As I said, I never used it for audio, maybe the above doesn't cause enough
of a bother to matter.  But desktops are a LOT more predictable than laptops
if you care about latency.

> > Also, go read about MSI/MSI-X, if you care about this stuff, you need
> > to know about message-signaled interrupts.

Devices operating in MSI/MSI-X mode do not share, or use, any interrupt or
IOAPIC resources.  But you only have a good chance of it working well when
using good northbridges (i.e. AMD and Intel, forget substandard stuff like
nVidia, SiS, VIA...) and good devices.  ThinkPads are _good_ on this, since
only good stuff from Intel and Broadcom are used (and nVidia GPUs do MSI
just fine AFAIK, unlike some of their crappy northbridges).

However, closed source kernel drivers are a major source of worry for
low-latency AFAIK (they cannot be fixed, and all "gaming" benchmarks are
about throughput), so I'd suggest that you don't get anything that will
require closed source, proprietary drivers (don't confuse it with closed
source firmware which doesn't matter much for this) to run.

-- 
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh