Re: [PATCH 5/5] lib: utils/suspend: add Andes ATCSMU suspend driver

[Ben Zong-You Xie <ben717@andestech.com>]

> >
> > -     ae350_non_ret_save(sbi_scratch_thishart_ptr());
> >       ae350_disable_coherency();
> >       wfi();
> > -     return 0;
> > +
> > +     /* Light sleep resumes here */
> > +     ae350_enable_coherency();
> > +
> > +     return SBI_ENOTSUPP;
>
> I just skimmed this patch really fast, so I probably missed something,
> but it seems odd to return SBI_ENOTSUPP here.
>
> Thanks,
> drew

Hi Drew,

The reason for returning SBI_ENOTSUPP here is to satisfy the logic in
sbi_hsm_exit(). In sbi_hsm_exit(), there is a specific check:

  if (hsm_device_has_hart_hotplug()) {
    if (hsm_device_hart_stop() != SBI_ENOTSUPP)
      goto fail_exit;
  }

  /**
    * As platform is lacking support for hotplug, directly jump to warmboot
    * and wait for interrupts in warmboot. We do it preemptively in order
    * preserve the hart states and reuse the code path for hotplug.
    */
  jump_warmboot();

If we return 0 (SBI_OK), the condition (0 != SBI_ENOTSUPP) evaluates to
true, and the code jumps to fail_exit, hanging the hart.

Returning SBI_ENOTSUPP allows the execution to fall through to the
jump_warmboot() path, which is the intended behavior for 'light sleep'
on AE350 platform.

-- 
opensbi mailing list
opensbi@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/opensbi