On Tue 21 Jul 12:16 PDT 2020, Siddharth Gupta wrote:
> On 7/15/2020 2:51 PM, Mathieu Poirier wrote:
> > On Wed, Jul 15, 2020 at 02:18:39PM -0600, Mathieu Poirier wrote:
> > > On Tue, Jul 07, 2020 at 12:07:49PM -0700, Siddharth Gupta wrote:
[..]
> > > > diff --git a/drivers/remoteproc/remoteproc_cdev.c b/drivers/remoteproc/remoteproc_cdev.c
[..]
> > > > +int rproc_char_device_add(struct rproc *rproc)
> > > > +{
> > > > + int ret;
> > > > + dev_t cdevt;
> > > > +
> > > > + cdev_init(&rproc->char_dev, &rproc_fops);
> > > > + rproc->char_dev.owner = THIS_MODULE;
> > > > +
> > > > + cdevt = MKDEV(rproc_major, rproc->index);
> > > > + ret = cdev_add(&rproc->char_dev, cdevt, 1);
> > Trying this patchset on my side gave me the following splat[1]. After finding
> > the root case I can't understand how you haven't see it on your side when you
> > tested the feature.
> >
> > [1]. https://pastebin.com/aYTUUCdQ
Mathieu, I've looked at this back and forth. Afaict this implies that
rproc_major is still 0. Could it be that either alloc_chrdev_region()
failed or somehow has yet to be called when you hit this point?
> Hey Mathieu,
>
> We aren't able to reproduce the error that you are seeing, the splat is
> coming
> from the check for whiteout device[1] - which shouldn't happen because of
> the
> find_dynamic_major call[2], right?
>
> We are successfully seeing all our character device files and able to
> successfully boot remoteprocs. From what I read and understood about
> whiteout
> devices they will be hidden in the fs.
>
> Could you provide more details about your configuration and testing?
>
> [1]: https://github.com/torvalds/linux/blob/master/fs/char_dev.c#L486
> <https://github.com/torvalds/linux/blob/master/fs/char_dev.c#L123>
> [2]: https://github.com/torvalds/linux/blob/master/fs/char_dev.c#L123
>
> <https://github.com/torvalds/linux/blob/master/fs/char_dev.c#L486>
> > > > + if (ret < 0)
> > > > + goto out;
> > > > +
> > > > + rproc->dev.devt = cdevt;
> > > > +out:
> > > > + return ret;
> > > > +}
> > > > +
> > > > +void rproc_char_device_remove(struct rproc *rproc)
> > > > +{
> > > > + __unregister_chrdev(rproc_major, rproc->index, 1, "remoteproc");
> > > > +}
> > > > +
> > > > +void __init rproc_init_cdev(void)
> > > > +{
> > > > + int ret;
> > > > +
> > > > + ret = alloc_chrdev_region(&rproc_major, 0, NUM_RPROC_DEVICES, "remoteproc");
> > > > + if (ret < 0)
> > > > + pr_err("Failed to alloc rproc_cdev region, err %d\n", ret);
> > > > +}
> > > > +
> > > > +void __exit rproc_exit_cdev(void)
> > > > +{
> > > > + unregister_chrdev_region(MKDEV(rproc_major, 0), NUM_RPROC_DEVICES);
> > > Please go back to the comment I made on this during my last review and respin.
> > After digging in the code while debugging the above problem, I don't see how
> > unregistering the chrdev region the way it is done here would have worked.
> Since this is compiled statically and not built as a module, we will never
> exercise the code path, so I will remove it in the next patchset.
>
You're right Siddharth, since we changed CONFIG_REMOTEPROC to bool it's no longer
possible to hit remoteproc_exit(), so you can omit this function
entirely. (And we should clean up the rest of that as well)
[..]
> > > > diff --git a/include/linux/remoteproc.h b/include/linux/remoteproc.h
[..]
> > > > @@ -488,6 +489,8 @@ struct rproc_dump_segment {
> > > > * @auto_boot: flag to indicate if remote processor should be auto-started
> > > > * @dump_segments: list of segments in the firmware
> > > > * @nb_vdev: number of vdev currently handled by rproc
> > > > + * @char_dev: character device of the rproc
> > > > + * @cdev_put_on_release: flag to indicate if remoteproc should be shutdown on @char_dev release
> > > > */
> > > > struct rproc {
> > > > struct list_head node;
> > > > @@ -523,6 +526,8 @@ struct rproc {
> > > > int nb_vdev;
> > > > u8 elf_class;
> > > > u16 elf_machine;
> > > > + struct cdev char_dev;
As stated privately, I assumed based on this name that this is a struct
device related to that character device. So please rename this cdev to
save me from doing this mistake again.
Thanks,
Bjorn
