Re: CoreSight framework and drivers

[Pratik Patel <pratikp@codeaurora.org>]

On Mon, Jan 07, 2013 at 11:58:36AM +0000, Will Deacon wrote:
> On Thu, Jan 03, 2013 at 06:06:43PM +0000, Pratik Patel wrote:
> > On Sun, Dec 23, 2012 at 11:32:39AM +0000, Will Deacon wrote:
> > > On Fri, Dec 21, 2012 at 10:18:28PM +0000, Pratik Patel wrote:
> > > > What user interface do you plan to provide for the CTI? Maybe
> > > > something consistent with other CoreSight components in sysfs to
> > > > allow users to enable, disable, map and unmap ???
> > > > 
> > > > Please let me know your thoughts.
> > > 
> > > Rather than have your current approach of dev nodes + sysfs config files for
> > > each coresight device, I think it might be better to follow something closer
> > > to ftrace and stick per-device directories under debugfs/coresight/. Then you
> > > can have a pipe file and some config files in the same directory for each
> > > component. You also don't need to do any mapping operations with this (just
> > > post-process the stream directly).
> > > 
> > 
> > Thanks for the suggestion. I had initially debated between debugfs
> > and sysfs but chose sysfs + dev nodes since using device attributes
> > relieves the drivers from manually managing directories and files,
> > its taken care of by the device and sysfs layers. Moreover, since
> > these are physical devices, device attributes made sense at the
> > time.
> > 
> > The map and unmap I was referring to was for the CTI trigger
> > mappings. Dev nodes are currently intended to provide the raw
> > data collected in the sinks.
> > 
> > Whats the advantage in using debugfs here?
> 
> The main things I like about debugfs are (a) it's a text-driven interface
> and easy to script with and (b) it matches what we do for ftrace.
> 
> Furthermore, it means that subtle differences between devices can be hidden
> in the driver and not require different vendor tools for parsing the trace
> data.
> 
Sorry for the delay and maybe I am missing something but it seems
we can take care of such protocol or parsing/decoding differences
even with device nodes since that seems independent of the
interface used - per device debugfs attributes or per device
device nodes.

CoreSight trace solution is typically a SoC wide solution and
hence can get used by non-Linux processors or hardware. Using
debugfs would imply compiling it and exposing all the debug
knobs even if the use case was to use the CoreSight solution for
something that didn't need all that.

-- 
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation