On Sat, Oct 12, 2019 at 7:18 PM Jonathan Cameron <jic23@xxxxxxxxxx> wrote: > > On Sat, 12 Oct 2019 17:28:41 +0200 > Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote: > > > On Sat, Oct 12, 2019 at 05:27:44PM +0200, Greg KH wrote: > > > On Sat, Oct 12, 2019 at 12:19:46PM +0100, Jonathan Cameron wrote: > > > > The following changes since commit b73b93a2af3392b9b7b8ba7e818ee767499f9655: > > > > > > > > iio: adc: ad7192: Add sysfs ABI documentation (2019-09-08 10:34:49 +0100) > > > > > > > > are available in the Git repository at: > > > > > > > > https://git.kernel.org/pub/scm/linux/kernel/git/jic23/iio.git tags/iio-for-5.5a-take2 > > > > > > Better, but I see this now: > > > > > > drivers/iio/imu/adis.c: In function ‘__adis_check_status’: > > > drivers/iio/imu/adis.c:295:9: warning: ‘status’ may be used uninitialized in this function [-Wmaybe-uninitialized] > > > 295 | status &= adis->data->status_error_mask; > > > | ^~ > > > > > > > > > I'll take this, can you just send a follow-on patch for this? > > > > Also I see: > > > > drivers/iio/imu/adis16480.c: In function ‘adis16480_enable_irq’: > > drivers/iio/imu/adis16480.c:950:6: warning: ‘val’ may be used uninitialized in this function [-Wmaybe-uninitialized] > > 950 | val &= ~ADIS16480_DRDY_EN_MSK; > > | ^~ > > CC [M] drivers/iio/magnetometer/hmc5843_i2c.o > > drivers/iio/imu/adis16480.c: In function ‘adis16480_write_raw’: > > drivers/iio/imu/adis16480.c:571:7: warning: ‘val’ may be used uninitialized in this function [-Wmaybe-uninitialized] > > 571 | val &= ~enable_mask; > > | ^~ > > drivers/iio/imu/adis16480.c:557:11: note: ‘val’ was declared here > > 557 | uint16_t val; > > | ^~~ > > > > > > So did you really fix anything here? > > > > I'll drop this pull again. > > > > What version of gcc are you using? Might I suggest a newer one (i.e. a > > modern one?) > > Ah. This is my mistake. I did see all of these, but still thought we were > in the category of tidying up some compiler version caused issues. > > The adis16400 came up in my local tests, so I previously pinged Alex on > basis it was something to do in a follow up. The other two showed up, but > again I still thought these were compiler version issues, particularly > as 0-day didn't highlight them (there were several other issues it > did highlight this week). Hence again I requested a follow up to tidy > it up. > > Anyhow, did some digging. The issue here was a 'fix' I put in to an initial > 0-day issue in the inline functions that Alex added. Note that one > appears to be compiler version dependent as it didn't turn up in my > local builds. There are now inline functions that check if (ret) > and don't set the value if ret is non 0. > > Out in the drivers, the check is the more specific (unnecessarily) > if (ret < 0) and hence the compiler is concluded that there might be a path to > val not being set. Previously it was giving up figuring this out. > So reality is they are a false positive (sort of as in reality ret > is never positive) but the compiler has made a reasonable point > that it can't see that. > > Never mind, I'll do a new pull request once fixes are in place. > Given there are two obvious ways of suppressing this and it's Alex's > driver I'll wait until he has time to take a look. > > Sorry for wasting your time. > If it helps, let's drop the ADIS patches for this round, and I can take a closer look as well. The cleanup does seem to have revealed a few gaps in our CI in relation to upstreaming things. We use Travis-CI for our stuff and stuff is public: https://travis-ci.org/analogdevicesinc/linux So, if anyone sees anything we should do better, I'm open to improvements/suggestions. I am in the process of adding sparse-builds, maybe some build hardening is next (stronger compile/build flags), adding our patches on top of a newer kernel (that's partially done). That should help us catch things a bit earlier. Sorry for the noise from my side as-well. Alex > Jonathan
