| Message ID | 20190812162105.13428-1-benjamin.tissoires@redhat.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Jiri Kosina |
| Headers | show |
| Series | HID: uhid: actually use the err number from userspace | expand |
Hey On Mon, Aug 12, 2019 at 6:21 PM Benjamin Tissoires <benjamin.tissoires@redhat.com> wrote: > > This can help debugging the situation > > Signed-off-by: Benjamin Tissoires <benjamin.tissoires@redhat.com> > --- > > Hi, > > not entirely sure if we can use this in a such simple way. > > However, this is useful to mimic device behaviour from userspace. > > Cheers, > Benjamin > > drivers/hid/uhid.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/hid/uhid.c b/drivers/hid/uhid.c > index fa0cc0899827..2fa32e7fc733 100644 > --- a/drivers/hid/uhid.c > +++ b/drivers/hid/uhid.c > @@ -284,7 +284,7 @@ static int uhid_hid_set_report(struct hid_device *hid, unsigned char rnum, > goto unlock; > > if (uhid->report_buf.u.set_report_reply.err) > - ret = -EIO; > + ret = -uhid->report_buf.u.set_report_reply.err; I am generally in favor of this. But: 1) can you do this for both set_report *and* get_report? 2) I think you have to filter some of the error codes. For instance, if you return one of the -ERESTARTSYS codes, this might cause the syscall to restart (if auto-restart is enabled on this context). At the same time, this is not *that* bad. It might even be useful for the userspace driver to trigger an EINTR. At least we should be aware of this. So maybe filters are not necessary.. Mhhh. Comments? Thanks David > else > ret = count; > > -- > 2.19.2 >
On Tue, Aug 13, 2019 at 12:50 PM David Rheinsberg <david.rheinsberg@gmail.com> wrote: > > Hey > > On Mon, Aug 12, 2019 at 6:21 PM Benjamin Tissoires > <benjamin.tissoires@redhat.com> wrote: > > > > This can help debugging the situation > > > > Signed-off-by: Benjamin Tissoires <benjamin.tissoires@redhat.com> > > --- > > > > Hi, > > > > not entirely sure if we can use this in a such simple way. > > > > However, this is useful to mimic device behaviour from userspace. > > > > Cheers, > > Benjamin > > > > drivers/hid/uhid.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/drivers/hid/uhid.c b/drivers/hid/uhid.c > > index fa0cc0899827..2fa32e7fc733 100644 > > --- a/drivers/hid/uhid.c > > +++ b/drivers/hid/uhid.c > > @@ -284,7 +284,7 @@ static int uhid_hid_set_report(struct hid_device *hid, unsigned char rnum, > > goto unlock; > > > > if (uhid->report_buf.u.set_report_reply.err) > > - ret = -EIO; > > + ret = -uhid->report_buf.u.set_report_reply.err; > > I am generally in favor of this. But: > > 1) can you do this for both set_report *and* get_report? right :) > > 2) I think you have to filter some of the error codes. For instance, > if you return one of the -ERESTARTSYS codes, this might cause the > syscall to restart (if auto-restart is enabled on this context). At > the same time, this is not *that* bad. It might even be useful for the > userspace driver to trigger an EINTR. At least we should be aware of > this. So maybe filters are not necessary.. Mhhh. Comments? I haven't thought at all of the side effects of letting the user return a random error code. I have the impression that anything below EHWPOISON (133) is relatively safe. So maybe we should just make sure the error code is below 134? The ERESTARTSYS has a few warnings in the include file, so I guess the side effects might be too much for what we want to deal with. Cheers, Benjamin > > Thanks > David > > > else > > ret = count; > > > > -- > > 2.19.2 > >
Hey > > 2) I think you have to filter some of the error codes. For instance, > > if you return one of the -ERESTARTSYS codes, this might cause the > > syscall to restart (if auto-restart is enabled on this context). At > > the same time, this is not *that* bad. It might even be useful for the > > userspace driver to trigger an EINTR. At least we should be aware of > > this. So maybe filters are not necessary.. Mhhh. Comments? > > I haven't thought at all of the side effects of letting the user > return a random error code. > I have the impression that anything below EHWPOISON (133) is > relatively safe. So maybe we should just make sure the error code is > below 134? > > The ERESTARTSYS has a few warnings in the include file, so I guess the > side effects might be too much for what we want to deal with. How about `err < ERESTARTSYS`? That is, we grant user-space the entire range [1-511]. This seems to be the range reserved for uapi. I think the ERESTART* codes would be fine as well, but I also don't believe there to be any actual use-case for them. Anyway, I am fine with either range. Thanks David
diff --git a/drivers/hid/uhid.c b/drivers/hid/uhid.c index fa0cc0899827..2fa32e7fc733 100644 --- a/drivers/hid/uhid.c +++ b/drivers/hid/uhid.c @@ -284,7 +284,7 @@ static int uhid_hid_set_report(struct hid_device *hid, unsigned char rnum, goto unlock; if (uhid->report_buf.u.set_report_reply.err) - ret = -EIO; + ret = -uhid->report_buf.u.set_report_reply.err; else ret = count;
This can help debugging the situation Signed-off-by: Benjamin Tissoires <benjamin.tissoires@redhat.com> --- Hi, not entirely sure if we can use this in a such simple way. However, this is useful to mimic device behaviour from userspace. Cheers, Benjamin drivers/hid/uhid.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)