error returns of pm_runtime_get()

error returns of pm_runtime_get()

[Oliver Neukum]

Hi,

looking through multiple drivers it seems to me that a great majority
is returning the error returns of pm_runtime_get() to user space.
They come quite directly from rpm_resume(). So we are led to return
things like -EACCESS or -EINVAL from system calls like open()
where they have no place.

So it seems to me like we should filter more or less like this:

-EBUSY		-> -EBUSY
-ENOMEM 	-> -ENOMEM
-EINPROGRES	-> -EAGAIN
any other error	-> -EIO

And it would be best to retain the current name of pm_runtime_get()
and introduce a version with unfiltered error codes.
What do you think?

	Regards
		Oliver

Re: error returns of pm_runtime_get()

[Rafael J. Wysocki]

On Friday, September 05, 2014 02:43:36 PM Oliver Neukum wrote:
> 
> Hi,
> 
> looking through multiple drivers it seems to me that a great majority
> is returning the error returns of pm_runtime_get() to user space.
> They come quite directly from rpm_resume(). So we are led to return
> things like -EACCESS or -EINVAL from system calls like open()
> where they have no place.

Aren't those driver bugs?

> So it seems to me like we should filter more or less like this:
> 
> -EBUSY		-> -EBUSY
> -ENOMEM 	-> -ENOMEM
> -EINPROGRES	-> -EAGAIN
> any other error	-> -EIO
> 
> And it would be best to retain the current name of pm_runtime_get()
> and introduce a version with unfiltered error codes.
> What do you think?

It doesn't sound right to me.  I may be tired at the moment, but I'm not sure
what the point would be.

Rafael

Re: error returns of pm_runtime_get()

[Oliver Neukum]

On Sat, 2014-09-06 at 01:42 +0200, Rafael J. Wysocki wrote:
> On Friday, September 05, 2014 02:43:36 PM Oliver Neukum wrote:
> > 
> > Hi,
> > 
> > looking through multiple drivers it seems to me that a great majority
> > is returning the error returns of pm_runtime_get() to user space.
> > They come quite directly from rpm_resume(). So we are led to return
> > things like -EACCESS or -EINVAL from system calls like open()
> > where they have no place.
> 
> Aren't those driver bugs?

Sure, but almost all drivers using runtime PM are affected.

> > So it seems to me like we should filter more or less like this:
> > 
> > -EBUSY		-> -EBUSY
> > -ENOMEM 	-> -ENOMEM
> > -EINPROGRES	-> -EAGAIN
> > any other error	-> -EIO
> > 
> > And it would be best to retain the current name of pm_runtime_get()
> > and introduce a version with unfiltered error codes.
> > What do you think?
> 
> It doesn't sound right to me.  I may be tired at the moment, but I'm not sure
> what the point would be.

The alternative is to make a conversion helper and to go through all
drivers. Not nice, but if necessary I'll do it.
But we cannot just return things like -EACCESS for internal reasons
while the file permissions are good.

	Regards
		Oliver

Re: error returns of pm_runtime_get()

[Rafael J. Wysocki]

On Wednesday, September 10, 2014 03:04:30 PM Oliver Neukum wrote:
> On Sat, 2014-09-06 at 01:42 +0200, Rafael J. Wysocki wrote:
> > On Friday, September 05, 2014 02:43:36 PM Oliver Neukum wrote:
> > > 
> > > Hi,
> > > 
> > > looking through multiple drivers it seems to me that a great majority
> > > is returning the error returns of pm_runtime_get() to user space.
> > > They come quite directly from rpm_resume(). So we are led to return
> > > things like -EACCESS or -EINVAL from system calls like open()
> > > where they have no place.
> > 
> > Aren't those driver bugs?
> 
> Sure, but almost all drivers using runtime PM are affected.

Well, what can I say?

> > > So it seems to me like we should filter more or less like this:
> > > 
> > > -EBUSY		-> -EBUSY
> > > -ENOMEM 	-> -ENOMEM
> > > -EINPROGRES	-> -EAGAIN
> > > any other error	-> -EIO
> > > 
> > > And it would be best to retain the current name of pm_runtime_get()
> > > and introduce a version with unfiltered error codes.
> > > What do you think?
> > 
> > It doesn't sound right to me.  I may be tired at the moment, but I'm not sure
> > what the point would be.
> 
> The alternative is to make a conversion helper and to go through all
> drivers. Not nice, but if necessary I'll do it.

I honestly would prefer that.

> But we cannot just return things like -EACCESS for internal reasons
> while the file permissions are good.

Sure.

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.

Re: error returns of pm_runtime_get()

[Oliver Neukum]

On Thu, 2014-09-11 at 02:03 +0200, Rafael J. Wysocki wrote:
> > The alternative is to make a conversion helper and to go through all
> > drivers. Not nice, but if necessary I'll do it.
> 
> I honestly would prefer that.

Very well, then I shall do that.

	Regards
		Oliver