[PATCH] riscv: Fix IPI not initialized in riscv_cpu_setup()

[PATCH] riscv: Fix IPI not initialized in riscv_cpu_setup()

[Nikita Shubin]

If riscv_cpu_setup() is skipped we end up with riscv_init_ipi() not
being called.

But we still use riscv_send_ipi() in spl_invoke_opensbi(), so even if
CONFIG_PHASE_CPU not set we still need init IPI and clear pending IPIs
on harts.

Fixes: d8810e1d4f82 ("riscv: Skip riscv_cpu_setup() when CPU driver is disabled")
Signed-off-by: Nikita Shubin <nikita.shubin@maquefel.me>
---
 arch/riscv/cpu/cpu.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/riscv/cpu/cpu.c b/arch/riscv/cpu/cpu.c
index 3bec7c7cb6d..a65e6951384 100644
--- a/arch/riscv/cpu/cpu.c
+++ b/arch/riscv/cpu/cpu.c
@@ -639,7 +639,7 @@ int riscv_cpu_setup(void)
 	struct udevice *dev;
 
 	if (!CONFIG_IS_ENABLED(CPU))
-		return 0;
+		goto init_riscv_ipi;
 
 	uclass_find_first_device(UCLASS_CPU, &dev);
 	if (!dev) {
@@ -687,6 +687,7 @@ int riscv_cpu_setup(void)
 #endif
 	}
 
+init_riscv_ipi:
 #if CONFIG_IS_ENABLED(SMP)
 	ret = riscv_init_ipi();
 	if (ret)

---
base-commit: cc557af4553382f6f50e3ed62b9577054e7bc54f
change-id: 20260904-fix_skipped_ipi_init-aa5575f55bc4

Best regards,
-- 
Nikita Shubin

Re: [PATCH] riscv: Fix IPI not initialized in riscv_cpu_setup()

[Michal Simek]

On 9/4/26 08:07, Nikita Shubin wrote:
> If riscv_cpu_setup() is skipped we end up with riscv_init_ipi() not
> being called.
> 
> But we still use riscv_send_ipi() in spl_invoke_opensbi(), so even if
> CONFIG_PHASE_CPU not set we still need init IPI and clear pending IPIs
> on harts.
> 
> Fixes: d8810e1d4f82 ("riscv: Skip riscv_cpu_setup() when CPU driver is disabled")
> Signed-off-by: Nikita Shubin <nikita.shubin@maquefel.me>
> ---
>   arch/riscv/cpu/cpu.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/riscv/cpu/cpu.c b/arch/riscv/cpu/cpu.c
> index 3bec7c7cb6d..a65e6951384 100644
> --- a/arch/riscv/cpu/cpu.c
> +++ b/arch/riscv/cpu/cpu.c
> @@ -639,7 +639,7 @@ int riscv_cpu_setup(void)
>   	struct udevice *dev;
>   
>   	if (!CONFIG_IS_ENABLED(CPU))
> -		return 0;
> +		goto init_riscv_ipi;
>   
>   	uclass_find_first_device(UCLASS_CPU, &dev);
>   	if (!dev) {
> @@ -687,6 +687,7 @@ int riscv_cpu_setup(void)
>   #endif
>   	}
>   
> +init_riscv_ipi:
>   #if CONFIG_IS_ENABLED(SMP)
>   	ret = riscv_init_ipi();
>   	if (ret)
> 
> ---
> base-commit: cc557af4553382f6f50e3ed62b9577054e7bc54f
> change-id: 20260904-fix_skipped_ipi_init-aa5575f55bc4
> 
> Best regards,

There are 2 other approaches how to do it in the code but it is up to maintainer 
to decide. No issue from my side to solve it via goto. Obviously one big if 
(CONFIG_IS_ENABLED(CPU)) is another one. Or move it to own function.

Anyway
Acked-by: Michal Simek <michal.simek@amd.com>

Thanks,
Michal

Re: [PATCH] riscv: Fix IPI not initialized in riscv_cpu_setup()

[Nikita Shubin]

Hi Yao Zi!

A shy ping on this patch.

Thanks !

On Fri, 2026-09-04 at 08:20 +0200, Michal Simek wrote:
> 
> 
> On 9/4/26 08:07, Nikita Shubin wrote:
> > If riscv_cpu_setup() is skipped we end up with riscv_init_ipi() not
> > being called.
> > 
> > But we still use riscv_send_ipi() in spl_invoke_opensbi(), so even
> > if
> > CONFIG_PHASE_CPU not set we still need init IPI and clear pending
> > IPIs
> > on harts.
> > 
> > Fixes: d8810e1d4f82 ("riscv: Skip riscv_cpu_setup() when CPU driver
> > is disabled")
> > Signed-off-by: Nikita Shubin <nikita.shubin@maquefel.me>
> > ---
> >   arch/riscv/cpu/cpu.c | 3 ++-
> >   1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/arch/riscv/cpu/cpu.c b/arch/riscv/cpu/cpu.c
> > index 3bec7c7cb6d..a65e6951384 100644
> > --- a/arch/riscv/cpu/cpu.c
> > +++ b/arch/riscv/cpu/cpu.c
> > @@ -639,7 +639,7 @@ int riscv_cpu_setup(void)
> >   	struct udevice *dev;
> >   
> >   	if (!CONFIG_IS_ENABLED(CPU))
> > -		return 0;
> > +		goto init_riscv_ipi;
> >   
> >   	uclass_find_first_device(UCLASS_CPU, &dev);
> >   	if (!dev) {
> > @@ -687,6 +687,7 @@ int riscv_cpu_setup(void)
> >   #endif
> >   	}
> >   
> > +init_riscv_ipi:
> >   #if CONFIG_IS_ENABLED(SMP)
> >   	ret = riscv_init_ipi();
> >   	if (ret)
> > 
> > ---
> > base-commit: cc557af4553382f6f50e3ed62b9577054e7bc54f
> > change-id: 20260904-fix_skipped_ipi_init-aa5575f55bc4
> > 
> > Best regards,
> 
> There are 2 other approaches how to do it in the code but it is up to
> maintainer 
> to decide. No issue from my side to solve it via goto. Obviously one
> big if 
> (CONFIG_IS_ENABLED(CPU)) is another one. Or move it to own function.
> 
> Anyway
> Acked-by: Michal Simek <michal.simek@amd.com>
> 
> Thanks,
> Michal

Re: [PATCH] riscv: Fix IPI not initialized in riscv_cpu_setup()

[Yao Zi]

On Fri, Sep 25, 2026 at 01:19:15PM +0300, Nikita Shubin wrote:
> Hi Yao Zi!
> 
> A shy ping on this patch.
> 
> Thanks !

Sorry for the delay.

> On Fri, 2026-09-04 at 08:20 +0200, Michal Simek wrote:
> > 
> > 
> > On 9/4/26 08:07, Nikita Shubin wrote:
> > > If riscv_cpu_setup() is skipped we end up with riscv_init_ipi() not
> > > being called.
> > > 
> > > But we still use riscv_send_ipi() in spl_invoke_opensbi(), so even
> > > if
> > > CONFIG_PHASE_CPU not set we still need init IPI and clear pending
> > > IPIs
> > > on harts.
> > > 
> > > Fixes: d8810e1d4f82 ("riscv: Skip riscv_cpu_setup() when CPU driver
> > > is disabled")
> > > Signed-off-by: Nikita Shubin <nikita.shubin@maquefel.me>
> > > ---
> > >   arch/riscv/cpu/cpu.c | 3 ++-
> > >   1 file changed, 2 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/arch/riscv/cpu/cpu.c b/arch/riscv/cpu/cpu.c
> > > index 3bec7c7cb6d..a65e6951384 100644
> > > --- a/arch/riscv/cpu/cpu.c
> > > +++ b/arch/riscv/cpu/cpu.c
> > > @@ -639,7 +639,7 @@ int riscv_cpu_setup(void)
> > >   	struct udevice *dev;
> > >   
> > >   	if (!CONFIG_IS_ENABLED(CPU))
> > > -		return 0;
> > > +		goto init_riscv_ipi;
> > >   
> > >   	uclass_find_first_device(UCLASS_CPU, &dev);
> > >   	if (!dev) {
> > > @@ -687,6 +687,7 @@ int riscv_cpu_setup(void)
> > >   #endif
> > >   	}
> > >   
> > > +init_riscv_ipi:
> > >   #if CONFIG_IS_ENABLED(SMP)
> > >   	ret = riscv_init_ipi();
> > >   	if (ret)
> > > 
> > > ---
> > > base-commit: cc557af4553382f6f50e3ed62b9577054e7bc54f
> > > change-id: 20260904-fix_skipped_ipi_init-aa5575f55bc4
> > > 
> > > Best regards,
> > 
> > There are 2 other approaches how to do it in the code but it is up to
> > maintainer 
> > to decide. No issue from my side to solve it via goto. Obviously one
> > big if 
> > (CONFIG_IS_ENABLED(CPU)) is another one. Or move it to own function.
> > 
> > Anyway
> > Acked-by: Michal Simek <michal.simek@amd.com>

I think it's likely a bad idea to have IPI stuff in riscv_cpu_setup(),
but this fix itself is okay. I should have noticed it when reviewing.

Reviewed-by: Yao Zi <me@ziyao.cc>

However, I'm RISC-V's reviewer instead of maintainer, so I couldn't pick
the patch for you though.

> > Thanks,
> > Michal

Best regards,
Yao Zi

Re: [PATCH] riscv: Fix IPI not initialized in riscv_cpu_setup()

[Nikita Shubin]

Hi Tim Ouyang and Leo Liang !

Any comments on this ?

Thanks !

On Fri, 2026-09-25 at 11:38 +0000, Yao Zi wrote:
> On Fri, Sep 25, 2026 at 01:19:15PM +0300, Nikita Shubin wrote:
> > Hi Yao Zi!
> > 
> > A shy ping on this patch.
> > 
> > Thanks !
> 
> Sorry for the delay.
> 
> > On Fri, 2026-09-04 at 08:20 +0200, Michal Simek wrote:
> > > 
> > > 
> > > On 9/4/26 08:07, Nikita Shubin wrote:
> > > > If riscv_cpu_setup() is skipped we end up with riscv_init_ipi()
> > > > not
> > > > being called.
> > > > 
> > > > But we still use riscv_send_ipi() in spl_invoke_opensbi(), so
> > > > even
> > > > if
> > > > CONFIG_PHASE_CPU not set we still need init IPI and clear
> > > > pending
> > > > IPIs
> > > > on harts.
> > > > 
> > > > Fixes: d8810e1d4f82 ("riscv: Skip riscv_cpu_setup() when CPU
> > > > driver
> > > > is disabled")
> > > > Signed-off-by: Nikita Shubin <nikita.shubin@maquefel.me>
> > > > ---
> > > >   arch/riscv/cpu/cpu.c | 3 ++-
> > > >   1 file changed, 2 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/arch/riscv/cpu/cpu.c b/arch/riscv/cpu/cpu.c
> > > > index 3bec7c7cb6d..a65e6951384 100644
> > > > --- a/arch/riscv/cpu/cpu.c
> > > > +++ b/arch/riscv/cpu/cpu.c
> > > > @@ -639,7 +639,7 @@ int riscv_cpu_setup(void)
> > > >   	struct udevice *dev;
> > > >   
> > > >   	if (!CONFIG_IS_ENABLED(CPU))
> > > > -		return 0;
> > > > +		goto init_riscv_ipi;
> > > >   
> > > >   	uclass_find_first_device(UCLASS_CPU, &dev);
> > > >   	if (!dev) {
> > > > @@ -687,6 +687,7 @@ int riscv_cpu_setup(void)
> > > >   #endif
> > > >   	}
> > > >   
> > > > +init_riscv_ipi:
> > > >   #if CONFIG_IS_ENABLED(SMP)
> > > >   	ret = riscv_init_ipi();
> > > >   	if (ret)
> > > > 
> > > > ---
> > > > base-commit: cc557af4553382f6f50e3ed62b9577054e7bc54f
> > > > change-id: 20260904-fix_skipped_ipi_init-aa5575f55bc4
> > > > 
> > > > Best regards,
> > > 
> > > There are 2 other approaches how to do it in the code but it is
> > > up to
> > > maintainer 
> > > to decide. No issue from my side to solve it via goto. Obviously
> > > one
> > > big if 
> > > (CONFIG_IS_ENABLED(CPU)) is another one. Or move it to own
> > > function.
> > > 
> > > Anyway
> > > Acked-by: Michal Simek <michal.simek@amd.com>
> 
> I think it's likely a bad idea to have IPI stuff in
> riscv_cpu_setup(),
> but this fix itself is okay. I should have noticed it when reviewing.
> 
> Reviewed-by: Yao Zi <me@ziyao.cc>
> 
> However, I'm RISC-V's reviewer instead of maintainer, so I couldn't
> pick
> the patch for you though.
> 
> > > Thanks,
> > > Michal
> 
> Best regards,
> Yao Zi

Re: [PATCH] riscv: Fix IPI not initialized in riscv_cpu_setup()

[Leo Liang]

Hi Nikita,

On Tue, Sep 29, 2026 at 12:58 PM Nikita Shubin
<nikita.shubin@maquefel.me> wrote:
>
> Hi Tim Ouyang and Leo Liang !
>
> Any comments on this ?
>
> Thanks !

Thanks for the patch.
Reviewed-by: Leo Yu-Chi Liang <leo.liang@sifive.com>

>
> On Fri, 2026-09-25 at 11:38 +0000, Yao Zi wrote:
> > On Fri, Sep 25, 2026 at 01:19:15PM +0300, Nikita Shubin wrote:
> > > Hi Yao Zi!
> > >
> > > A shy ping on this patch.
> > >
> > > Thanks !
> >
> > Sorry for the delay.
> >
> > > On Fri, 2026-09-04 at 08:20 +0200, Michal Simek wrote:
> > > >
> > > >
> > > > On 9/4/26 08:07, Nikita Shubin wrote:
> > > > > If riscv_cpu_setup() is skipped we end up with riscv_init_ipi()
> > > > > not
> > > > > being called.
> > > > >
> > > > > But we still use riscv_send_ipi() in spl_invoke_opensbi(), so
> > > > > even
> > > > > if
> > > > > CONFIG_PHASE_CPU not set we still need init IPI and clear
> > > > > pending
> > > > > IPIs
> > > > > on harts.
> > > > >
> > > > > Fixes: d8810e1d4f82 ("riscv: Skip riscv_cpu_setup() when CPU
> > > > > driver
> > > > > is disabled")
> > > > > Signed-off-by: Nikita Shubin <nikita.shubin@maquefel.me>
> > > > > ---
> > > > >   arch/riscv/cpu/cpu.c | 3 ++-
> > > > >   1 file changed, 2 insertions(+), 1 deletion(-)
> > > > >
> > > > > diff --git a/arch/riscv/cpu/cpu.c b/arch/riscv/cpu/cpu.c
> > > > > index 3bec7c7cb6d..a65e6951384 100644
> > > > > --- a/arch/riscv/cpu/cpu.c
> > > > > +++ b/arch/riscv/cpu/cpu.c
> > > > > @@ -639,7 +639,7 @@ int riscv_cpu_setup(void)
> > > > >         struct udevice *dev;
> > > > >
> > > > >         if (!CONFIG_IS_ENABLED(CPU))
> > > > > -               return 0;
> > > > > +               goto init_riscv_ipi;
> > > > >
> > > > >         uclass_find_first_device(UCLASS_CPU, &dev);
> > > > >         if (!dev) {
> > > > > @@ -687,6 +687,7 @@ int riscv_cpu_setup(void)
> > > > >   #endif
> > > > >         }
> > > > >
> > > > > +init_riscv_ipi:
> > > > >   #if CONFIG_IS_ENABLED(SMP)
> > > > >         ret = riscv_init_ipi();
> > > > >         if (ret)
> > > > >
> > > > > ---
> > > > > base-commit: cc557af4553382f6f50e3ed62b9577054e7bc54f
> > > > > change-id: 20260904-fix_skipped_ipi_init-aa5575f55bc4
> > > > >
> > > > > Best regards,
> > > >
> > > > There are 2 other approaches how to do it in the code but it is
> > > > up to
> > > > maintainer
> > > > to decide. No issue from my side to solve it via goto. Obviously
> > > > one
> > > > big if
> > > > (CONFIG_IS_ENABLED(CPU)) is another one. Or move it to own
> > > > function.
> > > >
> > > > Anyway
> > > > Acked-by: Michal Simek <michal.simek@amd.com>
> >
> > I think it's likely a bad idea to have IPI stuff in
> > riscv_cpu_setup(),
> > but this fix itself is okay. I should have noticed it when reviewing.
> >
> > Reviewed-by: Yao Zi <me@ziyao.cc>
> >
> > However, I'm RISC-V's reviewer instead of maintainer, so I couldn't
> > pick
> > the patch for you though.
> >
> > > > Thanks,
> > > > Michal
> >
> > Best regards,
> > Yao Zi