Re: [PATCH] ipmi:si: Add async init to ipmi_si

[Michał Cłapiński]

On Fri, Jun 26, 2026 at 6:35 AM Corey Minyard <corey@xxxxxxxxxxx> wrote:
>
> On Thu, Jun 25, 2026 at 05:59:54PM +0200, Michal Clapinski wrote:
> > Added a new config option to allow offloading slow part of
> > init_ipmi_si. Saves 100ms on my system.
>
> Are you loading as a module or building IPMI into the kernel?

Building in. I don't know much about IPMI but my system won't boot
without it because it can't find the disk. I think some storage
controllers need IPMI?

> I'm thinking this is a good idea, but not quite done this way.
>
> I have another long-standing issue that if a BMC is not operational when
> the system comes up, it will not continue to try to bring it up, so
> you have to reboot or hotmod the device in when it becomes available.
>
> I'm thinking that instead of pushing off the whole process, push off
> just the individual calls to try_smi_init().  I'm assuming that's where
> all the time is spent at init.

I can implement it this way in v2.

> So with that it would be possible to periodically retry a BMC until
> it eventually comes up.
>
> Plus, that way the "unload_when_empty" function won't be broken with
> this feature.
>
> I am also not quite sure what will happen if you try to unload the
> module if things are pushed off in a startup state like this.
>
> I'm also not quite sure how this will affect the ACPI IPMI functions in
> the kernel.  I would guess it's ok, since it registers to know when the
> interface becomes available, but it might be delayed a bit which might
> confuse things.  Also, it might delay the driver being available til
> later at startup, which may confuse userland users.
>
> I'm also wondering if making this an option makes sense, or if this
> should be the way it always works.  An option might be nice if it broke
> things, I guess.  But almost everyone uses modules, and that will be
> delayed from boot, anyway.  I guess that means ACPI is not an issue,
> either.

Since my system won't boot without IPMI, it would probably also fail
the same way if IPMI was still initializing so I didn't want to break
unsuspecting users. In reality it probably wouldn't happen often since
IPMI initialization is fast enough to finish before the rest of the
kernel.

> Just kind of pondering this right now.
>
> -corey
>
> >
> > Signed-off-by: Michal Clapinski <mclapinski@xxxxxxxxxx>
> > ---
> >  drivers/char/ipmi/Kconfig        |  8 ++++++
> >  drivers/char/ipmi/ipmi_si_intf.c | 48 +++++++++++++++++++++-----------
> >  2 files changed, 40 insertions(+), 16 deletions(-)
> >
> > diff --git a/drivers/char/ipmi/Kconfig b/drivers/char/ipmi/Kconfig
> > index 669f76000197..8419409dc3b9 100644
> > --- a/drivers/char/ipmi/Kconfig
> > +++ b/drivers/char/ipmi/Kconfig
> > @@ -67,6 +67,14 @@ config IPMI_SI
> >         Currently, only KCS and SMIC are supported.  If
> >         you are using IPMI, you should probably say "y" here.
> >
> > +config IPMI_SI_ASYNC_INIT
> > +     bool 'Asynchronous initialization of IPMI System Interface'
> > +     depends on IPMI_SI
> > +     default n
> > +     help
> > +       Enables asynchronous init of the IPMI System Interface.
> > +       It speeds up the boot time.
> > +
> >  config IPMI_SSIF
> >       tristate 'IPMI SMBus handler (SSIF)'
> >       depends on I2C
> > diff --git a/drivers/char/ipmi/ipmi_si_intf.c b/drivers/char/ipmi/ipmi_si_intf.c
> > index 9a9d12be9bf7..3e422c7df60a 100644
> > --- a/drivers/char/ipmi/ipmi_si_intf.c
> > +++ b/drivers/char/ipmi/ipmi_si_intf.c
> > @@ -39,6 +39,7 @@
> >  #include <linux/rcupdate.h>
> >  #include <linux/ipmi.h>
> >  #include <linux/ipmi_smi.h>
> > +#include <linux/async.h>
> >  #include "ipmi_si.h"
> >  #include "ipmi_si_sm.h"
> >  #include <linux/string.h>
> > @@ -2174,25 +2175,10 @@ static bool __init ipmi_smi_info_same(struct smi_info *e1, struct smi_info *e2)
> >               e1->io.addr_data == e2->io.addr_data);
> >  }
> >
> > -static int __init init_ipmi_si(void)
> > +static int __init smi_init_scan(void)
> >  {
> >       struct smi_info *e, *e2;
> >
> > -     if (initialized)
> > -             return 0;
> > -
> > -     ipmi_hardcode_init();
> > -
> > -     pr_info("IPMI System Interface driver\n");
> > -
> > -     ipmi_si_platform_init();
> > -
> > -     ipmi_si_pci_init();
> > -
> > -     ipmi_si_ls2k_init();
> > -
> > -     ipmi_si_parisc_init();
> > -
> >       mutex_lock(&smi_infos_lock);
> >
> >       /*
> > @@ -2271,6 +2257,36 @@ static int __init init_ipmi_si(void)
> >               return 0;
> >       }
> >  }
> > +
> > +static void __init async_smi_init(void *data, async_cookie_t cookie)
> > +{
> > +     smi_init_scan();
> > +}
> > +
> > +static int __init init_ipmi_si(void)
> > +{
> > +     if (initialized)
> > +             return 0;
> > +
> > +     ipmi_hardcode_init();
> > +
> > +     pr_info("IPMI System Interface driver\n");
> > +
> > +     ipmi_si_platform_init();
> > +
> > +     ipmi_si_pci_init();
> > +
> > +     ipmi_si_ls2k_init();
> > +
> > +     ipmi_si_parisc_init();
> > +
> > +     if (IS_ENABLED(CONFIG_IPMI_SI_ASYNC_INIT)) {
> > +             async_schedule(async_smi_init, NULL);
> > +             return 0;
> > +     }
> > +
> > +     return smi_init_scan();
> > +}
> >  module_init(init_ipmi_si);
> >
> >  static void wait_msg_processed(struct smi_info *smi_info)
> > --
> > 2.55.0.rc0.799.gd6f94ed593-goog
> >