Re: [PATCH v1] drivers: make struct device_driver::remove return void

From: Greg Kroah-Hartman
Date: Fri Nov 13 2020 - 10:38:38 EST


On Thu, Nov 12, 2020 at 09:51:12AM +0100, Uwe Kleine-König wrote:
> Hello Greg,
>
> On Tue, Nov 10, 2020 at 09:31:06PM +0100, Greg Kroah-Hartman wrote:
> > On Tue, Nov 10, 2020 at 04:07:23PM +0100, Uwe Kleine-König wrote:
> > > The driver core doesn't check the return value of the remove callback
> > > because there is only little software can do when hardware disappears.
> > >
> > > So change the callback to not return a value at all and adapt all users.
> > > The motivation for this change is that some driver authors have a
> > > misconception about failures in the remove callback. Making it void
> > > makes it pretty obvious that there is no error handling to be expected.
> > >
> > > Most drivers were easy to convert as they returned 0 unconditionally, I
> > > added a warning to code paths that returned an error code (that was
> > > ignored already before).
> > >
> > > Signed-off-by: Uwe Kleine-König <u.kleine-koenig@xxxxxxxxxxxxxx>
> > > ---
> > > Hello,
> > >
> > > I expect that there are still a few driver that need adaption as I only
> > > build tested allmodconfig on ARM and amd64.
> > >
> > > Best regards
> > > Uwe
> > >
> > > drivers/acpi/processor_driver.c | 10 ++++------
> > > drivers/amba/bus.c | 7 ++++---
> > > drivers/base/platform.c | 13 +++++++------
> > > drivers/bus/fsl-mc/fsl-mc-bus.c | 7 ++-----
> > > drivers/bus/mhi/core/init.c | 6 ++----
> > > drivers/char/hw_random/optee-rng.c | 4 +---
> > > drivers/char/tpm/tpm_ftpm_tee.c | 8 ++++----
> > > drivers/firmware/broadcom/tee_bnxt_fw.c | 4 +---
> > > drivers/fsi/fsi-master-hub.c | 4 +---
> > > drivers/fsi/fsi-sbefifo.c | 4 +---
> > > drivers/fsi/fsi-scom.c | 4 +---
> > > drivers/gpu/drm/drm_mipi_dsi.c | 7 +++++--
> > > drivers/gpu/host1x/bus.c | 11 +++++++----
> > > drivers/greybus/core.c | 4 +---
> > > drivers/hsi/clients/cmt_speech.c | 4 +---
> > > drivers/hsi/clients/hsi_char.c | 4 +---
> > > drivers/hsi/clients/nokia-modem.c | 4 +---
> > > drivers/hsi/clients/ssi_protocol.c | 4 +---
> > > drivers/i2c/busses/i2c-fsi.c | 4 +---
> > > drivers/input/rmi4/rmi_bus.c | 4 +---
> > > drivers/input/rmi4/rmi_driver.c | 4 +---
> > > drivers/input/touchscreen/wm97xx-core.c | 7 +++----
> > > drivers/mfd/ucb1400_core.c | 3 +--
> > > drivers/net/ethernet/3com/3c509.c | 5 ++---
> > > drivers/net/ethernet/3com/3c59x.c | 3 +--
> > > drivers/net/ethernet/dec/tulip/de4x5.c | 4 +---
> > > drivers/net/fddi/defxx.c | 5 ++---
> > > drivers/net/phy/mdio_device.c | 4 +---
> > > drivers/net/phy/phy_device.c | 4 +---
> > > drivers/pci/pcie/portdrv_core.c | 5 ++---
> > > drivers/scsi/advansys.c | 3 +--
> > > drivers/scsi/aha1740.c | 4 +---
> > > drivers/scsi/aic7xxx/aic7770_osm.c | 3 +--
> > > drivers/scsi/ch.c | 3 +--
> > > drivers/scsi/sd.c | 6 ++----
> > > drivers/scsi/ses.c | 3 +--
> > > drivers/scsi/sim710.c | 3 +--
> > > drivers/scsi/sr.c | 6 ++----
> > > drivers/scsi/st.c | 5 ++---
> > > drivers/siox/siox-core.c | 6 ++++--
> > > drivers/soundwire/bus_type.c | 13 +++++++------
> > > drivers/spi/spi.c | 8 +++++---
> > > drivers/usb/core/driver.c | 7 ++-----
> > > drivers/visorbus/visorbus_main.c | 5 +----
> > > include/linux/device/driver.h | 2 +-
> > > sound/core/seq/oss/seq_oss_synth.c | 6 ++----
> > > sound/core/seq/oss/seq_oss_synth.h | 2 +-
> > > sound/core/seq/seq_midi.c | 6 +++---
> > > sound/drivers/opl3/opl3_seq.c | 10 ++++++----
> > > sound/drivers/opl4/opl4_seq.c | 9 +++++----
> > > sound/hda/ext/hdac_ext_bus.c | 9 +++++++--
> > > sound/isa/sb/emu8000_synth.c | 5 ++---
> > > sound/pci/emu10k1/emu10k1_synth.c | 5 ++---
> > > sound/pci/hda/hda_bind.c | 11 +++++++----
> > > 54 files changed, 129 insertions(+), 172 deletions(-)
> >
> > First off, that's a lot of drivers with a "raw" remove function, why is
> > anyone doing that? It should all be wrapped up in the bus that the
> > drivers are on.
> >
> > In digging, ugh, looks like there's some sound driver abuse here that
> > should be fixed up first, which will cause you to get these "for free",
> > and some busses should be fixed up as well.
> >
> > This type of patch should only have to bus code, if things are right,
> > not individual drivers. It's not your fault, but fixing that up will
> > make this patch easier as it will be in bisectable pieces :)
>
> I don't understand how this gets more bisectable. The desired cleanups
> won't look considerably different, will they? Also irrespective of their
> order the intermediate steps will build and run just fine?!

But you should clean up individual drivers separately from the busses.

Those drivers should not be using that interface today, somehow that has
slipped by without people realizing it.

> I agree that there are quite some strange drivers, but given my limited
> time to work on this now (and expecting to have to rework this patch if
> I pick it up again for the next or even after-next merge window) I would
> like to see this breed in next already now.

I'm not going to apply anything to my trees that I know I will not take,
sorry.

thanks,

greg k-h