Re: [PATCH v3 1/2] ACPI: IPMI: Add helper to wait for when SMI is selected

[Corey Minyard <minyard@acm.org>]

On Thu, Jan 04, 2024 at 06:11:29PM +0100, Rafael J. Wysocki wrote:
> On Thu, Jan 4, 2024 at 4:42 PM Corey Minyard <minyard@acm.org> wrote:
> >
> > On Thu, Jan 04, 2024 at 02:34:52PM +0100, Rafael J. Wysocki wrote:
> > > On Thu, Jan 4, 2024 at 3:48 AM Kai-Heng Feng
> > > <kai.heng.feng@canonical.com> wrote:
> > > >
> > > > The function of acpi_power_meter module on Dell system requires IPMI
> > > > handler is installed and SMI is selected.
> > >
> > > Does the firmware use _DEP to let the OS know about this dependency?
> > >
> > > > So add a helper to let acpi_power_meter know when IPMI handler and SMI
> > > > are ready.
> > > >
> > > > Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
> > > > ---
> > > > v3:
> > > >  - New patch.
> > > >
> > > >  drivers/acpi/acpi_ipmi.c | 17 ++++++++++++++++-
> > > >  include/acpi/acpi_bus.h  |  5 +++++
> > > >  2 files changed, 21 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/acpi/acpi_ipmi.c b/drivers/acpi/acpi_ipmi.c
> > > > index 0555f68c2dfd..54862cab7171 100644
> > > > --- a/drivers/acpi/acpi_ipmi.c
> > > > +++ b/drivers/acpi/acpi_ipmi.c
> > > > @@ -23,6 +23,8 @@ MODULE_LICENSE("GPL");
> > > >  #define IPMI_TIMEOUT                   (5000)
> > > >  #define ACPI_IPMI_MAX_MSG_LENGTH       64
> > > >
> > > > +static struct completion smi_selected;
> > > > +
> > > >  struct acpi_ipmi_device {
> > > >         /* the device list attached to driver_data.ipmi_devices */
> > > >         struct list_head head;
> > > > @@ -463,8 +465,10 @@ static void ipmi_register_bmc(int iface, struct device *dev)
> > > >                 if (temp->handle == handle)
> > > >                         goto err_lock;
> > > >         }
> > > > -       if (!driver_data.selected_smi)
> > > > +       if (!driver_data.selected_smi) {
> > > >                 driver_data.selected_smi = ipmi_device;
> > > > +               complete(&smi_selected);
> > > > +       }
> > > >         list_add_tail(&ipmi_device->head, &driver_data.ipmi_devices);
> > > >         mutex_unlock(&driver_data.ipmi_lock);
> > > >
> > > > @@ -578,10 +582,21 @@ acpi_ipmi_space_handler(u32 function, acpi_physical_address address,
> > > >         return status;
> > > >  }
> > > >
> > > > +int acpi_wait_for_acpi_ipmi(void)
> > > > +{
> > > > +       long ret;
> > > > +
> > > > +       ret = wait_for_completion_interruptible_timeout(&smi_selected, 2 * HZ);
> > > > +
> > > > +       return ret > 0 ? 0 : -ETIMEDOUT;
> > >
> > > What will happen if the IPMI driver is unloaded after this has returned 0?
> >
> > The IPMI driver can't be unloaded if it has a user.
> 
> Because of the use of the exported symbol, right?

I wasn't thinking that, but yes, that would do it.  It also keep track
of users and doesn't allow the driver to be unloaded if anything has
registered as a user, either.

-corey

> 
> > I've been following this, but I know little about ACPI.  Beyond this
> > solution, the only other solution I could come up with was to start the
> > IPMI driver earlier.  But then you are in a chicken-and-egg situation
> > (https://dictionary.cambridge.org/dictionary/english/chicken-and-egg-situation).
> > Which was the reason for the SPMI table, but that's really kind of
> > useless for this, even if the SPMI table existed.
> 
> Fine.
> 
> Let me reply to the patch with some more comments then.