Re: [PATCH 1/2] remoteproc: Introduce deny_sysfs_ops flag

[Bjorn Andersson <bjorn.andersson@linaro.org>]

On Thu 03 Feb 10:37 PST 2022, Mathieu Poirier wrote:

> Hi Puranjay,
> 
> On Mon, Jan 31, 2022 at 07:41:51PM +0530, Puranjay Mohan wrote:
> > The remoteproc framework provides sysfs interfaces for changing
> > the firmware name and for starting/stopping a remote processor
> > through the sysfs files 'state' and 'firmware'. The 'recovery'
> > sysfs file can also be used similarly to control the error recovery
> > state machine of a remoteproc. These interfaces are currently
> > allowed irrespective of how the remoteprocs were booted (like
> > remoteproc self auto-boot, remoteproc client-driven boot etc).
> > These interfaces can adversely affect a remoteproc and its clients
> > especially when a remoteproc is being controlled by a remoteproc
> > client driver(s). Also, not all remoteproc drivers may want to
> > support the sysfs interfaces by default.
> > 
> > Add support to deny the sysfs state/firmware/recovery change by
> > introducing a state flag 'deny_sysfs_ops' that the individual
> > remoteproc drivers can set based on their usage needs. The default
> > behavior is to allow the sysfs operations as before.
> > 
> > Implement attribute_group->is_visible() to hide the sysfs
> > state/firmware/recovery entries when deny_sysfs_ops flag is set.
> > 
> 
> The address in the "To:" field of this email doesn't match the one in the SoB,
> Something that makes checkpatch angry.
> 
> > Signed-off-by: Puranjay Mohan <p-mohan@ti.com>
> > ---
> >  drivers/remoteproc/remoteproc_sysfs.c | 18 +++++++++++++++++-
> >  include/linux/remoteproc.h            |  2 ++
> >  2 files changed, 19 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/remoteproc/remoteproc_sysfs.c b/drivers/remoteproc/remoteproc_sysfs.c
> > index ea8b89f97d7b..4a41abdd1f7b 100644
> > --- a/drivers/remoteproc/remoteproc_sysfs.c
> > +++ b/drivers/remoteproc/remoteproc_sysfs.c
> > @@ -230,6 +230,21 @@ static ssize_t name_show(struct device *dev, struct device_attribute *attr,
> >  }
> >  static DEVICE_ATTR_RO(name);
> >  
> > +static umode_t rproc_is_visible(struct kobject *kobj, struct attribute *attr,
> > +			       int n)
> > +{
> > +	struct device *dev = kobj_to_dev(kobj);
> > +	struct rproc *rproc = to_rproc(dev);
> > +	umode_t mode = attr->mode;
> > +
> > +	if (rproc->deny_sysfs_ops && (attr == &dev_attr_recovery.attr ||
> > +				      attr == &dev_attr_firmware.attr ||
> > +				      attr == &dev_attr_state.attr))
> 
> I toyed with this solution for a little while.  I think the use case is valid
> but hiding the above options will also result in a system that is difficult to
> use (and debug) because they convey important information.
> 
> I suggest introducing a new kernel configuration options to make the attributes
> of the rproc_devgroup return -EINVAL when it is set.  So in remoteproc_sysfs.c
> do something like:
> 
> #if CONFIG_REMOTEPROC_SYSFS_RO

We should assume that people runs the common multi_v7_defconfig for arm,
or defconfig for arm64, and still want this support.

Based on Puranjay's proposed patch I think it's also reasonable to
assume that we might have platforms with a mixture of "read-only" and
"normal" remoteprocs (e.g. some power-management uC + some DSP).

> static bool option_is_read_only()
> {
>         return true;
> }
> #else
> static bool option_is_read_only()
> {
>         return false;
> }
> #endif
> 
> [...]
> 
> static ssize_t recovery_store(struct device *dev,
> 			      struct device_attribute *attr,
> 			      const char *buf, size_t count)
> {
> 	struct rproc *rproc = to_rproc(dev);
> 
>         if (option_is_read_only())

struct attribute_group has a .is_visible callback which I think is
better to base this on.

Regards,
Bjorn

>                 return -EINVAL;
> 
> 	if (sysfs_streq(buf, "enabled")) {
> 		/* change the flag and begin the recovery process if needed */
> 		rproc->recovery_disabled = false;
> 		rproc_trigger_recovery(rproc);
> 	} else if (sysfs_streq(buf, "disabled")) {
> 		rproc->recovery_disabled = true;
> 	} else if (sysfs_streq(buf, "recover")) {
> 		/* begin the recovery process without changing the flag */
> 		rproc_trigger_recovery(rproc);
> 	} else {
> 		return -EINVAL;
> 	}
> 
> 	return count;
> }
> 
> Thanks,
> Mathieu
> 
> > +		mode = 0;
> > +
> > +	return mode;
> > +}
> > +
> >  static struct attribute *rproc_attrs[] = {
> >  	&dev_attr_coredump.attr,
> >  	&dev_attr_recovery.attr,
> > @@ -240,7 +255,8 @@ static struct attribute *rproc_attrs[] = {
> >  };
> >  
> >  static const struct attribute_group rproc_devgroup = {
> > -	.attrs = rproc_attrs
> > +	.attrs = rproc_attrs,
> > +	.is_visible = rproc_is_visible,
> >  };
> >  
> >  static const struct attribute_group *rproc_devgroups[] = {
> > diff --git a/include/linux/remoteproc.h b/include/linux/remoteproc.h
> > index e0600e1e5c17..3849c66ce38f 100644
> > --- a/include/linux/remoteproc.h
> > +++ b/include/linux/remoteproc.h
> > @@ -523,6 +523,7 @@ struct rproc_dump_segment {
> >   * @table_sz: size of @cached_table
> >   * @has_iommu: flag to indicate if remote processor is behind an MMU
> >   * @auto_boot: flag to indicate if remote processor should be auto-started
> > + * @deny_sysfs_ops: flag to not permit sysfs operations on state, firmware and recovery
> >   * @dump_segments: list of segments in the firmware
> >   * @nb_vdev: number of vdev currently handled by rproc
> >   * @elf_class: firmware ELF class
> > @@ -562,6 +563,7 @@ struct rproc {
> >  	size_t table_sz;
> >  	bool has_iommu;
> >  	bool auto_boot;
> > +	bool deny_sysfs_ops;
> >  	struct list_head dump_segments;
> >  	int nb_vdev;
> >  	u8 elf_class;
> > -- 
> > 2.24.3
> >