On Wed, Sep 17, 2014 at 01:57:33PM +0200, Frans Klaver wrote: > On Wed, Sep 17, 2014 at 12:34 PM, Henrique de Moraes Holschuh > <hmh@xxxxxxxxxx> wrote: > > On Tue, 16 Sep 2014, Darren Hart wrote: > >> - When reading and writing sysfs device attribute files, avoid dependency > >> on specific error codes wherever possible. This minimizes coupling to > >> the error handling implemementation within the kernel. > >> > >> In general, failures to read or write sysfs device attributes shall > >> propogate errors wherever possible. Common errors include, but are not > >> limited to: > >> > >> -EIO: The read or store operation is not supported, typically returned by > >> the sysfs system itself if the read or store pointer is NULL. > >> > >> -ENXIO: The read or store operation failed > > > > from errno(3): > > EIO Input/output error (POSIX.1) > > ENXIO No such device or address (POSIX.1) > > > > It makes sense to retry EIO. ENXIO means there's nobody listening at the > > time, and isn't usually retried. > > > > The device-based interfaces get it right. A typical example is the > > cpu-based devices, where ENXIO means "no such processor", while EIO means > > "whatever you're trying to do failed", so a MSR read would return ENXIO if > > the processor core is offline/doesn't exist, and EIO if the processor core > > is there, but raised a #GP when the MSR read was attempted. > > Here's something I don't quite understand. How should one then > distinguish between sysfs's use of EIO "can't (read from|write to) > this file", and this example's EIO "something went wrong, you might > want to try again"? Why not use EAGAIN "Resource temporarily > unavailable" in the case where trying again makes sense? I wouldn't > normally retry the last operation if I was just told something > actually went wrong. > > I've only been at it for a couple of weeks, but I get the impression > that sysfs has never really been guided regarding error codes, or has > gone to live its own life now kept in check with "don't change the > errors, it may break userspace". Does that make sense? Right, this was the distinction I was trying to make with the above description. Henrique's points are valid, but based on the sysfs subsystem already using EIO in the way that it does, I felt the above made sense. That said, I'm not personally tied to them, it's just what I have derived from recent discussions on the subject and what I observed in existing usage. -- Darren Hart Intel Open Source Technology Center -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
