Message ID | 20250216175545.35079-3-contact@artur-rojek.eu (mailing list archive) |
---|---|
State | New |
Headers | show |
Series | J2 Turtle Board fixes | expand |
On Sun, Feb 16, 2025 at 6:56 PM Artur Rojek <contact@artur-rojek.eu> wrote: > > The jcore-aic irqchip does not have separate interrupt numbers reserved > for cpu-local vs global interrupts. Instead, the task of selecting this > property is being delegated to the device drivers requesting the given > irq. > > This quirk has not been taken into account while migrating jcore-pit to > request_percpu_irq(), resulting in a failure to register PIT interrupts. > > Fix this behavior by making the following changes: > 1) Explicitly register irq_set_percpu_devid() in jcore-pit. > 2) Provide enable_percpu_irq()/disable_percpu_irq() calls in jcore-pit. > 3) Make jcore-aic pass the correct per-cpu cookie to the irq handler by > using handle_percpu_devid_irq() instead of handle_percpu_irq(). > > Fixes: 69a9dcbd2d65 ("clocksource/drivers/jcore: Use request_percpu_irq()") > > Signed-off-by: Artur Rojek <contact@artur-rojek.eu> I can confirm that this change compiles OK with strict percpu address space checks on x86 [1]. Acked-by: Uros Bizjak <ubizjak@gmail.com> [1] https://lore.kernel.org/lkml/20250127160709.80604-1-ubizjak@gmail.com/ Uros. > --- > drivers/clocksource/jcore-pit.c | 15 ++++++++++++++- > drivers/irqchip/irq-jcore-aic.c | 2 +- > 2 files changed, 15 insertions(+), 2 deletions(-) > > diff --git a/drivers/clocksource/jcore-pit.c b/drivers/clocksource/jcore-pit.c > index a3fe98cd3838..82815428f8f9 100644 > --- a/drivers/clocksource/jcore-pit.c > +++ b/drivers/clocksource/jcore-pit.c > @@ -114,6 +114,18 @@ static int jcore_pit_local_init(unsigned cpu) > pit->periodic_delta = DIV_ROUND_CLOSEST(NSEC_PER_SEC, HZ * buspd); > > clockevents_config_and_register(&pit->ced, freq, 1, ULONG_MAX); > + enable_percpu_irq(pit->ced.irq, IRQ_TYPE_NONE); > + > + return 0; > +} > + > +static int jcore_pit_local_teardown(unsigned cpu) > +{ > + struct jcore_pit *pit = this_cpu_ptr(jcore_pit_percpu); > + > + pr_info("Local J-Core PIT teardown on cpu %u\n", cpu); > + > + disable_percpu_irq(pit->ced.irq); > > return 0; > } > @@ -168,6 +180,7 @@ static int __init jcore_pit_init(struct device_node *node) > return -ENOMEM; > } > > + irq_set_percpu_devid(pit_irq); > err = request_percpu_irq(pit_irq, jcore_timer_interrupt, > "jcore_pit", jcore_pit_percpu); > if (err) { > @@ -237,7 +250,7 @@ static int __init jcore_pit_init(struct device_node *node) > > cpuhp_setup_state(CPUHP_AP_JCORE_TIMER_STARTING, > "clockevents/jcore:starting", > - jcore_pit_local_init, NULL); > + jcore_pit_local_init, jcore_pit_local_teardown); > > return 0; > } > diff --git a/drivers/irqchip/irq-jcore-aic.c b/drivers/irqchip/irq-jcore-aic.c > index b9dcc8e78c75..1f613eb7b7f0 100644 > --- a/drivers/irqchip/irq-jcore-aic.c > +++ b/drivers/irqchip/irq-jcore-aic.c > @@ -38,7 +38,7 @@ static struct irq_chip jcore_aic; > static void handle_jcore_irq(struct irq_desc *desc) > { > if (irqd_is_per_cpu(irq_desc_get_irq_data(desc))) > - handle_percpu_irq(desc); > + handle_percpu_devid_irq(desc); > else > handle_simple_irq(desc); > } > -- > 2.48.1 >
On 2/16/25 11:55, Artur Rojek wrote: > The jcore-aic irqchip does not have separate interrupt numbers reserved > for cpu-local vs global interrupts. Instead, the task of selecting this > property is being delegated to the device drivers requesting the given > irq. > > This quirk has not been taken into account while migrating jcore-pit to > request_percpu_irq(), resulting in a failure to register PIT interrupts. > > Fix this behavior by making the following changes: > 1) Explicitly register irq_set_percpu_devid() in jcore-pit. > 2) Provide enable_percpu_irq()/disable_percpu_irq() calls in jcore-pit. > 3) Make jcore-aic pass the correct per-cpu cookie to the irq handler by > using handle_percpu_devid_irq() instead of handle_percpu_irq(). > > Fixes: 69a9dcbd2d65 ("clocksource/drivers/jcore: Use request_percpu_irq()") > > Signed-off-by: Artur Rojek <contact@artur-rojek.eu> Tested-by: Rob Landley <rob@landley.net> Rob
On 16/02/2025 18:55, Artur Rojek wrote: > The jcore-aic irqchip does not have separate interrupt numbers reserved > for cpu-local vs global interrupts. Instead, the task of selecting this > property is being delegated to the device drivers requesting the given > irq. > > This quirk has not been taken into account while migrating jcore-pit to > request_percpu_irq(), resulting in a failure to register PIT interrupts. > > Fix this behavior by making the following changes: > 1) Explicitly register irq_set_percpu_devid() in jcore-pit. > 2) Provide enable_percpu_irq()/disable_percpu_irq() calls in jcore-pit. > 3) Make jcore-aic pass the correct per-cpu cookie to the irq handler by > using handle_percpu_devid_irq() instead of handle_percpu_irq(). > > Fixes: 69a9dcbd2d65 ("clocksource/drivers/jcore: Use request_percpu_irq()") > > Signed-off-by: Artur Rojek <contact@artur-rojek.eu> > --- How this patch should be merged ? It is touching irqchip and clocksource at the same time. May I pick it in the clocksource tree ?
Hi Daniel, On Wed, 19 Feb 2025 at 15:43, Daniel Lezcano <daniel.lezcano@linaro.org> wrote: > On 16/02/2025 18:55, Artur Rojek wrote: > > The jcore-aic irqchip does not have separate interrupt numbers reserved > > for cpu-local vs global interrupts. Instead, the task of selecting this > > property is being delegated to the device drivers requesting the given > > irq. > > > > This quirk has not been taken into account while migrating jcore-pit to > > request_percpu_irq(), resulting in a failure to register PIT interrupts. > > > > Fix this behavior by making the following changes: > > 1) Explicitly register irq_set_percpu_devid() in jcore-pit. > > 2) Provide enable_percpu_irq()/disable_percpu_irq() calls in jcore-pit. > > 3) Make jcore-aic pass the correct per-cpu cookie to the irq handler by > > using handle_percpu_devid_irq() instead of handle_percpu_irq(). > > > > Fixes: 69a9dcbd2d65 ("clocksource/drivers/jcore: Use request_percpu_irq()") > > > > Signed-off-by: Artur Rojek <contact@artur-rojek.eu> > > --- > > How this patch should be merged ? > > It is touching irqchip and clocksource at the same time. > > May I pick it in the clocksource tree ? Thomas already took it, cfr. commit d7e3fd658248f257 ("irqchip/jcore-aic, clocksource/drivers/jcore: Fix jcore-pit interrupt request") in next-20250219. Gr{oetje,eeting}s, Geert
On 19/02/2025 15:50, Geert Uytterhoeven wrote: > Hi Daniel, > > On Wed, 19 Feb 2025 at 15:43, Daniel Lezcano <daniel.lezcano@linaro.org> wrote: >> On 16/02/2025 18:55, Artur Rojek wrote: >>> The jcore-aic irqchip does not have separate interrupt numbers reserved >>> for cpu-local vs global interrupts. Instead, the task of selecting this >>> property is being delegated to the device drivers requesting the given >>> irq. >>> >>> This quirk has not been taken into account while migrating jcore-pit to >>> request_percpu_irq(), resulting in a failure to register PIT interrupts. >>> >>> Fix this behavior by making the following changes: >>> 1) Explicitly register irq_set_percpu_devid() in jcore-pit. >>> 2) Provide enable_percpu_irq()/disable_percpu_irq() calls in jcore-pit. >>> 3) Make jcore-aic pass the correct per-cpu cookie to the irq handler by >>> using handle_percpu_devid_irq() instead of handle_percpu_irq(). >>> >>> Fixes: 69a9dcbd2d65 ("clocksource/drivers/jcore: Use request_percpu_irq()") >>> >>> Signed-off-by: Artur Rojek <contact@artur-rojek.eu> >>> --- >> >> How this patch should be merged ? >> >> It is touching irqchip and clocksource at the same time. >> >> May I pick it in the clocksource tree ? > > Thomas already took it, cfr. commit d7e3fd658248f257 > ("irqchip/jcore-aic, clocksource/drivers/jcore: Fix jcore-pit interrupt > request") in next-20250219. Oh, ok, thanks for letting me know
diff --git a/drivers/clocksource/jcore-pit.c b/drivers/clocksource/jcore-pit.c index a3fe98cd3838..82815428f8f9 100644 --- a/drivers/clocksource/jcore-pit.c +++ b/drivers/clocksource/jcore-pit.c @@ -114,6 +114,18 @@ static int jcore_pit_local_init(unsigned cpu) pit->periodic_delta = DIV_ROUND_CLOSEST(NSEC_PER_SEC, HZ * buspd); clockevents_config_and_register(&pit->ced, freq, 1, ULONG_MAX); + enable_percpu_irq(pit->ced.irq, IRQ_TYPE_NONE); + + return 0; +} + +static int jcore_pit_local_teardown(unsigned cpu) +{ + struct jcore_pit *pit = this_cpu_ptr(jcore_pit_percpu); + + pr_info("Local J-Core PIT teardown on cpu %u\n", cpu); + + disable_percpu_irq(pit->ced.irq); return 0; } @@ -168,6 +180,7 @@ static int __init jcore_pit_init(struct device_node *node) return -ENOMEM; } + irq_set_percpu_devid(pit_irq); err = request_percpu_irq(pit_irq, jcore_timer_interrupt, "jcore_pit", jcore_pit_percpu); if (err) { @@ -237,7 +250,7 @@ static int __init jcore_pit_init(struct device_node *node) cpuhp_setup_state(CPUHP_AP_JCORE_TIMER_STARTING, "clockevents/jcore:starting", - jcore_pit_local_init, NULL); + jcore_pit_local_init, jcore_pit_local_teardown); return 0; } diff --git a/drivers/irqchip/irq-jcore-aic.c b/drivers/irqchip/irq-jcore-aic.c index b9dcc8e78c75..1f613eb7b7f0 100644 --- a/drivers/irqchip/irq-jcore-aic.c +++ b/drivers/irqchip/irq-jcore-aic.c @@ -38,7 +38,7 @@ static struct irq_chip jcore_aic; static void handle_jcore_irq(struct irq_desc *desc) { if (irqd_is_per_cpu(irq_desc_get_irq_data(desc))) - handle_percpu_irq(desc); + handle_percpu_devid_irq(desc); else handle_simple_irq(desc); }
The jcore-aic irqchip does not have separate interrupt numbers reserved for cpu-local vs global interrupts. Instead, the task of selecting this property is being delegated to the device drivers requesting the given irq. This quirk has not been taken into account while migrating jcore-pit to request_percpu_irq(), resulting in a failure to register PIT interrupts. Fix this behavior by making the following changes: 1) Explicitly register irq_set_percpu_devid() in jcore-pit. 2) Provide enable_percpu_irq()/disable_percpu_irq() calls in jcore-pit. 3) Make jcore-aic pass the correct per-cpu cookie to the irq handler by using handle_percpu_devid_irq() instead of handle_percpu_irq(). Fixes: 69a9dcbd2d65 ("clocksource/drivers/jcore: Use request_percpu_irq()") Signed-off-by: Artur Rojek <contact@artur-rojek.eu> --- drivers/clocksource/jcore-pit.c | 15 ++++++++++++++- drivers/irqchip/irq-jcore-aic.c | 2 +- 2 files changed, 15 insertions(+), 2 deletions(-)