Re: unparseable, undocumented /sys/class/drm/.../pstate

[Greg KH <greg@kroah.com>]

On Mon, Jun 23, 2014 at 12:40:43PM -0400, Ilia Mirkin wrote:
> On Mon, Jun 23, 2014 at 12:36 PM, Greg KH <greg@kroah.com> wrote:
> > On Mon, Jun 23, 2014 at 12:18:51PM -0400, Ilia Mirkin wrote:
> >> On Mon, Jun 23, 2014 at 12:07 PM, Greg KH <greg@kroah.com> wrote:
> >> > On Sun, Jun 22, 2014 at 10:12:14PM -0400, Ilia Mirkin wrote:
> >> >> On Sat, Jun 21, 2014 at 3:45 PM, Greg KH <greg@kroah.com> wrote:
> >> >> > On Sat, Jun 21, 2014 at 02:22:59PM -0400, Ilia Mirkin wrote:
> >> >> >> On Sat, Jun 21, 2014 at 2:02 PM, Pavel Machek <pavel@ucw.cz> wrote:
> >> >> >> > Hi!
> >> >> >> >
> >> >> >> > AFAICT, pstate file will contain something like
> >> >> >> >
> >> >> >> > 07: core 100 MHz memory 123 MHz *
> >> >> >> > 08: core 100-200 MHz memory 123 MHz
> >> >> >> >
> >> >> >> > ...which does not look exactly like one-value-per-file, and I'm pretty
> >> >> >> > sure userspace will get it wrong if it tries to parse it. Plus, I
> >> >> >> > don't see required documentation in Documentation/ABI.
> >> >> >> >
> >> >> >> > Should we disable it for now, so that userspace does not start
> >> >> >> > depending on it and we'll not have to maintain it forever?
> >> >> >> >
> >> >> >> > I guess better interface would be something like
> >> >> >> >
> >> >> >> > pstate/07/core_clock_min
> >> >> >> >           core_clock_max
> >> >> >> >           memory_clock_min
> >> >> >> >           memory_clock_max
> >> >> >> >
> >> >> >> > and then pstate/active containing just the number of active state?
> >> >> >> >
> >> >> >> > Thanks,
> >> >> >> >                                                                 Pavel
> >> >> >> >
> >> >> >> > PS: I have no nvidia, got the news at
> >> >> >> >
> >> >> >> > http://www.phoronix.com/scan.php?page=article&item=nouveau_try_linux316&num=2
> >> >> >>
> >> >> >> FTR, this file has been in place since 3.13, and there was a different
> >> >> >> file before it (performance_levels), with a comparable format since
> >> >> >> much earlier (definitely 3.8, probably earlier). I think it's meant a
> >> >> >> lot more for people looking at it and echo'ing stuff to it to modify
> >> >> >> the levels (where supported), than for programs parsing it. Perhaps
> >> >> >> sysfs is the wrong place for this -- what is the right place? debugfs?
> >> >> >
> >> >> > Yes, please move it to debugfs.
> >> >>
> >> >> Could we just say that the format of this file is one-per-line of
> >> >>
> >> >> level: information-for-the-user
> >> >>
> >> >> And you can echo a level into it to switch to that level? That seems
> >> >> like a reasonable ABI to have... would be happy to throw it into a
> >> >> file somewhere... not sure where though.
> >> >
> >> > sysfs files are "one value per file", that's it.  Do anything other than
> >> > that, and it can not be in sysfs, sorry.
> >>
> >> I think that's a little inconsistent. There are *tons* of files in
> >> sysfs with multiple values. For example "local_cpulist" which contains
> >> the string "0-3" for me, the per-connector "modes" file which has a
> >> list of modes supported, and a "resource" file which has a list of hex
> >> values (which probably have something to do with PCI resources?). [I
> >> purposely picked values coming from different parts of the kernel not
> >> to focus on one subsystem...]
> >
> > A list of valid "values" that a file can be in is fine if you just then
> > write one value back to that file.  That's the one exception, but a
> > minor one given the huge number of sysfs files.  Other than that, if you
> 
> Which is pretty much what the pstate file is. Would it make things
> better if we removed the descriptive info while leaving the pstate
> file in place?
> 
> > know of exceptions to that rule, please point them out and I will be
> > glad to yell at the developers.
> >
> > PCI device resources are binary sysfs files, which are just pass-through
> > files from the firmware/device to userspace, with no parsing done in the
> > kernel.  So that's just a single 'value' as well.
> 
> $ cat /sys/class/drm/card0/device/resource
> 0x00000000f0000000 0x00000000f03fffff 0x0000000000140204
> 0x0000000000000000 0x0000000000000000 0x0000000000000000
> 
> Doesn't seem like "binary" in the true sense, but perhaps that's close enough.

It's "close enough" :)