Re: [PATCH] ACPI / LPSS: Work around wrong sdio _ADR 0 entry on some byt/cht devices

[Mika Westerberg <mika.westerberg@linux.intel.com>]

On Wed, Dec 28, 2016 at 11:30:35PM +0100, Rafael J. Wysocki wrote:
> On Wed, Dec 28, 2016 at 10:14 AM, Mika Westerberg
> <mika.westerberg@linux.intel.com> wrote:
> > On Mon, Dec 26, 2016 at 12:25:19AM +0100, Rafael J. Wysocki wrote:
> >> CC Mika and Andy.
> >>
> >> Plus I don't think -stable is going to take your patches directly.
> >>
> >> On Sun, Dec 25, 2016 at 11:21 AM, Hans de Goede <hdegoede@redhat.com> wrote:
> >> > The firmware on some cherrytrail devices wrongly adds _ADR 0 to their
> >> > entry describing the 80860F14 uid "2" sd-controller.
> >> >
> >> > I believe the firmware writers intended this as a sdio function address,
> >> > but it is in the wrong place for this, so it gets interpreted as a pci
> >> > address, causing the node describing the sd-controller used for the
> >> > sdio-wifi to get seen as a firmware_node for the pci host bridge, rather
> >> > then being stand-alone device.
> >> >
> >> > This commit adds a byt_sdio_setup function which detects this scenario
> >> > and removes the wrong firmware_node link from the pci host bridge, which
> >> > fixes acpi_create_platform_device returning NULL, leading to non-working
> >> > sdio-wifi.
> >> >
> >> > BugLink: https://github.com/hadess/rtl8723bs/issues/80
> >> > Cc: stable@vger.kernel.org
> >> > Signed-off-by: Hans de Goede <hdegoede@redhat.com>
> >> > ---
> >> >  drivers/acpi/acpi_lpss.c | 30 ++++++++++++++++++++++++++++++
> >> >  1 file changed, 30 insertions(+)
> >> >
> >> > diff --git a/drivers/acpi/acpi_lpss.c b/drivers/acpi/acpi_lpss.c
> >> > index 373657f..df9cc66 100644
> >> > --- a/drivers/acpi/acpi_lpss.c
> >> > +++ b/drivers/acpi/acpi_lpss.c
> >> > @@ -84,6 +84,7 @@ static const struct lpss_device_desc lpss_dma_desc = {
> >> >  };
> >> >
> >> >  struct lpss_private_data {
> >> > +       struct acpi_device *adev;
> >> >         void __iomem *mmio_base;
> >> >         resource_size_t mmio_size;
> >> >         unsigned int fixed_clk_rate;
> >> > @@ -154,6 +155,33 @@ static void byt_i2c_setup(struct lpss_private_data *pdata)
> >> >         writel(0, pdata->mmio_base + LPSS_I2C_ENABLE);
> >> >  }
> >> >
> >> > +static void byt_sdio_setup(struct lpss_private_data *pdata)
> >> > +{
> >> > +       unsigned long long adr;
> >> > +       acpi_status status;
> >> > +       struct device *dev;
> >> > +
> >> > +       /*
> >> > +        * Some firmware has a broken _ADR 0 enter for the 80860F14:2
> >> > +        * device, which causes it to get seen as the firmware_node
> >> > +        * for the pci host bridge, rather then a stand alone device.
> >> > +        *
> >> > +        * Check if this is the case, and if it is remove the link.
> >> > +        */
> >> > +       if (strcmp(acpi_device_uid(pdata->adev), "2") != 0)
> >> > +               return;
> >> > +
> >> > +       status = acpi_evaluate_integer(pdata->adev->handle, "_ADR", NULL, &adr);
> >> > +       if (ACPI_FAILURE(status) || adr != 0)
> >> > +               return;
> >> > +
> >> > +       dev = acpi_get_first_physical_node(pdata->adev);
> >> > +       if (!dev)
> >> > +               return;
> >> > +
> >> > +       acpi_unbind_one(dev);
> >> > +}
> >
> > IIRC the _ADR 0 problem is not only limited to SDIO devices. I think we
> > should just fix the match heuristics in acpi_find_child_device() to
> > cover all other devices as well.
> >
> > Something like below might do the job.
> >
> > diff --git a/drivers/acpi/glue.c b/drivers/acpi/glue.c
> > index 5ea5dc219f56..7cfd48f55b6f 100644
> > --- a/drivers/acpi/glue.c
> > +++ b/drivers/acpi/glue.c
> > @@ -85,7 +85,7 @@ static struct acpi_bus_type *acpi_get_bus_type(struct device *dev)
> >
> >  static int find_child_checks(struct acpi_device *adev, bool check_children)
> >  {
> > -       bool sta_present = true;
> > +       bool hid_present, sta_present = true;
> >         unsigned long long sta;
> >         acpi_status status;
> >
> > @@ -98,7 +98,22 @@ static int find_child_checks(struct acpi_device *adev, bool check_children)
> >         if (check_children && list_empty(&adev->children))
> >                 return -ENODEV;
> >
> > -       return sta_present ? FIND_CHILD_MAX_SCORE : FIND_CHILD_MIN_SCORE;
> > +       /*
> > +        * If there is a _HID or _CID present on a bus that is supposed to
> > +        * be matched using _ADR, we prioritize those devices without ID
> > +        * higher.
> > +        *
> > +        * This is because there are many BIOSes out there having ACPI
> > +        * enumerated devices and the _ADR is set to 0. This will match the
> > +        * root PCI bridge to the first found device with _ADR 0 which is
> > +        * not always the right device.
> > +        */
> > +       hid_present = !list_empty(&adev->pnp.ids);
> > +
> > +       if (sta_present)
> > +               return hid_present ? FIND_CHILD_MIN_SCORE : FIND_CHILD_MAX_SCORE;
> > +
> > +       return FIND_CHILD_MIN_SCORE;
> >  }
> >
> >  struct acpi_device *acpi_find_child_device(struct acpi_device *parent,
> >
> 
> I'd rather only change the return statement, like this:
> 
> return sta_present && list_empty(&adev->pnp.ids) ?
> FIND_CHILD_MAX_SCORE : FIND_CHILD_MIN_SCORE;
> 
> I'll send a patch tomorrow unless there's something wrong with this.

Looks good to me, thanks.

> BTW, the comment is not correct too, because the host bridge has a
> _HID.  What is matched against _ADR(0) is the first function on bus 0
> IIRC.

Yes, you are right.