Re: SYSFS "errors"

[Mauro Carvalho Chehab]

Em Mon, 18 Feb 2013 20:46:33 +0200
Felipe Balbi <balbi@xxxxxx> escreveu:

> Hi, On Mon, Feb 18, 2013 at 09:49:16AM -0800, Greg KH wrote:
> > > Input/output error - /sys/devices/cpu/power/autosuspend_delay_ms
> > 
> > The issue with this file is, if the power.use_autosuspend flag is not
> > set for the device, then it can't be read or written to.  This flag
> > changes dynamically with the system state
> > (__pm_runtime_use_autosuspend() can change it), so we can't just not
> > show the file if the flag is not set properly, sorry.
> > 
> > So the "error" is correct here, as is the 0644 file value.
> 
> hmm... we could create the file at pm_runtime_enable() time and remove
> it on pm_runtime_disable() time, no ? Addin Rafael to Cc

...


> > 
> > > No such device - /sys/devices/system/edac/mc/mc0/sdram_scrub_rate
> > 
> > Odd, go ask the edac developers
> 
> will do ;-)

Well, the question is missing ;) /me assumes that you want to talk about
suspending/resume and EDAC, right?

In general, memory controllers don't supports suspend, as far as I can tell. 
Still, I've seen a few ones that support, but the current drivers and/or the
EDAC core currently doesn't offer any support to it, as such setup is done
by the BIOS, when it detects the used DIMM banks.

I suspect that, when the OS puts the machine on a suspend state, the BIOS may
also suspend also the memory controller or put it into a low power consumption
mode, but it does it without any help from the EDAC drivers.

For most of what's there at EDAC, I don't think it is worth to add any PM support
inside it. There are, however, two cases were we may need to add some support:

1) if user changed the SDRAM scrub rate before suspending, it makes sense to
restore it after resume, on the memory controller drivers that support such
feature (not all supports it);

2) hot-pluggable DIMMs. EDAC currently doesn't support. This could be needed
on some future. In this case, the core may need to re-scan the memory controller,
changing the memory properties. I've no idea if is there any real case needing
it, nor what event would trigger the memory-controller re-scan. Resume is likely
one of the candidates for it, on machines that support hot-pluggable memories.

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