On Tuesday, June 11, 2013 3:53 AM, Ian Abbott wrote:
> On 2013-06-11 01:58, H Hartley Sweeten wrote:
>> Introduce a new subdevice flag, SDF_FREE_SPRIV, to allow the comedi
>> core to automatically free the subdevice private data during the
>> cleanup_device() stage of the detach.
>
> This is an implementation detail that should really be hidden from
> userspace. With this patch it is exposed through the 'subd_flags'
> member of 'struct comedi_subdinfo'. That can easily be fixed by masking
> out the 'SDF_FREE_SPRIV' flag when setting 'us->subd_flags' in
> 'do_subdinfo_ioctl()'. Ideally, 'SDF_FREE_SPRIV' shouldn't appear in
> "comedi.h" at all as should be of no interest to userspace.
I agree, but I couldn't find a clean place to put the flag.
> As we're running out of SDF_ values, we might want to reassign the
> current 'SDF_FREE_SPRIV' value to some other user-visible meaning at a
> later date.
>
> We could put the internal flags in a different member of 'struct
> comedi_subdevice'; either define a new member or abuse the existing
> 'runflags' member. (A new member could be placed in a 32-bit hole in
> the structure for 64-bit kernels, so would only use extra space for
> 32-bit kernels.) (There are various members of 'struct
> comedi_subdevice' that could be stripped out to save space if necessary,
> such as 'settling_time_0', 'flags' and 'flaglist'. These were intended
> to pass various details from drivers to userspace but none of the
> existing drivers actually set them.)
I'll move the new flag to the 'runflags' as:
SRF_FREE_SPRIV = 0x80000000,
Note, the 'runflags' are normally accessed using the static helper functions
comedi_{set,get}_subdevice_runflags() in comedi_fops.c and protected
by the s->spin_lock. Since this flag is only used during the attach/detach
I don't think the spinlock is needed. This avoids needing to expose the
Helpers.
I'll post an update shortly.
Thanks,
Hartley
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/devel
