Bug#853122: [Regression] Changes to i2c-piix4.c initialisation prevent loading of sp5100_tco watchdog driver on AMD SB800 chipset

[Boszormenyi Zoltan <zboszor@pr.hu>]

2017-03-31 14:49 keltezéssel, Guenter Roeck írta:
> Hi Paul,
>
> On 03/31/2017 12:17 AM, Paul Menzel wrote:
> > Dear Wolfram,
> >
> >
> > Thank you for the reply, which we talked about briefly at the
> > Chemnitzer LinuxTage.
> >
> >
> > Am Freitag, den 03.03.2017, 11:17 +0100 schrieb Wolfram Sang:
> > > > Unfortunately, commit 2fee61d22e (i2c: piix4: Add support for
> > > > multiplexed main adapter in SB800) [1] caused a regression. Tim
> > > > reported that to the Linux Kernel Bugtracker as bug #170741 last
> > > > September [2], but it looks like the affected subsystems don’t use it.
> > >
> > > Jean Delvare pointed out this issue amongst others[1] last year already.
> > > Let me quote:
> > >
> > > ===
> > >
> > > 5* The I/O ports used for SMBus configuration and port switching are
> > > also needed by a watchdog driver, sp5100_tco. Both drivers request the
> > > region, so the first one wins, and the other driver can't be loaded.
> > > sp5100_tco was there first, so the changes done to the i2c-piix4 driver
> > > recently will cause a regression for some users by preventing them
> > > from using the sp5100_tco and i2c-piix4 drivers at the same time. In
> > > the long run I guess we will need a helper module to handle this shared
> > > resource. Unless IORESOURCE_MUXED can be used for that. Either way,
> > > that's more work than I can put into this before kernel v4.5 is
> > > released. For the time being, I think we should simply make it
> > > non-fatal if the I/O ports can't be requested, and continue without
> > > multiplexing (as before.)
> > >
> > > ===
> > >
> > > Seems nobody had the resources, so far.
> >
> > I still don’t understand, why Jean then not immediately reverted the
> > commit to adhere to the Linux Kernel’s no-regression-policy.
> >
> > > I don't have the HW and not much experience with non-embedded
> > > platforms. I wonder, though, if we really need to convert the drivers
> > > to MFD ones, or if we could use the simpler MFD_SYSCON mechanism
> > > which helps in exactly such cases for embedded platforms. But I am
> > > really lacking details here and am afraid this is probably all the
> > > input I can give currently.
> >
> > Zoltan stepped up, and uploaded a patch for review to the Kernel.org
> > Bugzilla [2], also attached to this message.
>
> Please don't send patches as attachments.
>
> request_muxed_region() can fail, and literally every other driver
> using it checks for that failure. Please do the same.

In what circumstances can request_muxed_region() fail? As far as
I can see, only if two drivers use the same I/O port base and the
already present region did not use IORESOURCE_MUXED which is
not the case here. When request_muxed_region() is used consistently,
subsequent requests are put on a wait queue and the first one is
woken up when the region is released. So, it's basically a mutex.
Am I missing something here?

Alternatively, the original "piix4_mutex_sb800" mutex can be moved out
to its own file and used by both drivers. This way, request_muxed_region()
would not be needed at all among these two drivers.

The third option is to remove the request_*region() calls and the mutex
completely.

After all, drivers/usb/host/pci-quirks.c::usb_amd_quirk_pll() also uses
this I/O port pair and this is done without request_*region() or locking.
It is for some AMD devices, SB800 included, for which the function
accesses the same I/O ports used by both i2c-piix4 and sp5100_tco.

/*
 * The hardware normally enables the A-link power management feature, which
 * lets the system lower the power consumption in idle states.
 *
 * This USB quirk prevents the link going into that lower power state
 * during isochronous transfers.
 *
 * Without this quirk, isochronous stream on OHCI/EHCI/xHCI controllers of
 * some AMD platforms may stutter or have breaks occasionally.
 */
static void usb_amd_quirk_pll(int disable)

This function is hidden behind two wrappers: usb_amd_quirk_pll_disable()
and usb_amd_quirk_pll_enable() and these are used from:

drivers/usb/host/ohci-q.c
drivers/usb/host/xhci-ring.c
drivers/usb/host/ehci-sched.c

> The sp5100_tco_dev_name change in the watchdog driver is unnecessary.

request_muxed_region() requires a const char * name to be passed.
sp5100_tco uses a different DEVNAME for the two device generations.
I wanted to preserve this distinction but dev_name is local to
sp5100_tco_setupdevice() and it would have been an overkill to call
tco_has_sp5100_reg_layout() every time the code executes
request_muxed_region().

Are you saying that this distinction is unnecessary, too?
I would prefer a single DEVNAME that just contains the driver name.
Less code, less confusion.

> There are some unnecessary { } in the watchdog driver after the patch
> is applied.

True, there are two places.

> Please split the patch into two patches so they can be reviewed and
> applied separately.

Likely I will, but I would like to hear the others' opinion, too.

1. a single patch with a single goal: protect I/O port access
   for SB800 across the whole kernel
2. patch series for same goal for the two drivers separately
   (or three if pci-quirks.c needs to change as well)

and

A) a common mutex across the two (three) drivers in the kernel, or
B) request_muxed_region() with retrying in the usual manner if it fails:

   #define enter_sb800() \
      do { \
      } while (!request_muxed_region(...))

Best regards,
Zoltán Böszörményi