Re: [PATCH v6 3/5] firmware: psci: Read and use vendor reset types

[Elliot Berman <quic_eberman@quicinc.com>]

On Fri, Nov 15, 2024 at 02:35:52PM +0100, Lorenzo Pieralisi wrote:
> On Wed, Oct 23, 2024 at 09:30:21AM -0700, Elliot Berman wrote:
> > On Fri, Oct 18, 2024 at 10:42:46PM -0700, Stephen Boyd wrote:
> > > Quoting Elliot Berman (2024-10-18 12:39:48)
> > > > diff --git a/drivers/firmware/psci/psci.c b/drivers/firmware/psci/psci.c
> > > > index 2328ca58bba6..60bc285622ce 100644
> > > > --- a/drivers/firmware/psci/psci.c
> > > > +++ b/drivers/firmware/psci/psci.c
> > > > @@ -29,6 +29,8 @@
> > > >  #include <asm/smp_plat.h>
> > > >  #include <asm/suspend.h>
> > > >
> > > > +#define REBOOT_PREFIX "mode-"
> > > 
> > > Maybe move this near the function that uses it.
> > > 
> > > > +
> > > >  /*
> > > >   * While a 64-bit OS can make calls with SMC32 calling conventions, for some
> > > >   * calls it is necessary to use SMC64 to pass or return 64-bit values.
> > > > @@ -305,9 +315,29 @@ static int get_set_conduit_method(const struct device_node *np)
> > > >         return 0;
> > > >  }
> > > >
> > > > +static void psci_vendor_sys_reset2(unsigned long action, void *data)
> > > > +{
> > > > +       const char *cmd = data;
> > > > +       unsigned long ret;
> > > > +       size_t i;
> > > > +
> > > > +       for (i = 0; i < num_psci_reset_params; i++) {
> > > > +               if (!strcmp(psci_reset_params[i].mode, cmd)) {
> > > > +                       ret = invoke_psci_fn(PSCI_FN_NATIVE(1_1, SYSTEM_RESET2),
> > > > +                                            psci_reset_params[i].reset_type,
> > > > +                                            psci_reset_params[i].cookie, 0);
> > > > +                       pr_err("failed to perform reset \"%s\": %ld\n",
> > > > +                               cmd, (long)ret);
> > > 
> > > Do this intentionally return? Should it be some other function that's
> > > __noreturn instead and a while (1) if the firmware returns back to the
> > > kernel?
> > > 
> > 
> > Yes, I think it's best to make sure we fall back to the architectural
> > reset (whether it's the SYSTEM_RESET or architectural SYSTEM_RESET2)
> > since device would reboot then.
> 
> Well, that's one of the doubts I have about enabling this code. From
> userspace we are requesting a reboot (I don't even think that user
> space knows which reboot modes are actually implemented (?)) and we may
> end up issuing one with completely different semantics ?

You're right here, userspace issue a "reboot bootloader" and if kernel
doesn't have the support to set up the right cookie, the device would do
a normal reboot and not stop at the bootloader. This problem exists
today and I think whether this is an issue to solve is out of scope here.

> 
> Are these "reset types" exported to user space ?
> 

No mechanism exists to do that. We could do something specific for PSCI
or do something generic for everybody. I don't think something specific
for PSCI is the right approach because it's a general problem. I don't
think there's enough interest to change reboot command plumbing to
advertise valid reset types to userspace.

- Elliot