| Message ID | 20220430030025.58405-1-samuel@sholland.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | riscv: Fix irq_work when SMP is disabled | expand |
Am Samstag, 30. April 2022, 05:00:23 CEST schrieb Samuel Holland: > irq_work is triggered via an IPI, but the IPI infrastructure is not > included in uniprocessor kernels. As a result, irq_work never runs. > Fall back to the tick-based irq_work implementation on uniprocessor > configurations. > > Fixes: 298447928bb1 ("riscv: Support irq_work via self IPIs") > Signed-off-by: Samuel Holland <samuel@sholland.org> That uniprocessor part seems a tiny bit neglected - as I saw previously with alternatives not getting applied as well, so Reviewed-by: Heiko Stuebner <heiko@sntech.de> Though somehow I find the arm32 style a tad nicer by defining an is_smp() function [0] that holds the necessary checks. But I guess that is a style preference. Heiko [0] https://elixir.bootlin.com/linux/latest/source/arch/arm/include/asm/smp_plat.h#L18 > --- > This was found while bringing up cpufreq on D1. Switching cpufreq > governors was hanging on irq_work_sync(). > > arch/riscv/include/asm/irq_work.h | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/arch/riscv/include/asm/irq_work.h b/arch/riscv/include/asm/irq_work.h > index d6c277992f76..b53891964ae0 100644 > --- a/arch/riscv/include/asm/irq_work.h > +++ b/arch/riscv/include/asm/irq_work.h > @@ -4,7 +4,7 @@ > > static inline bool arch_irq_work_has_interrupt(void) > { > - return true; > + return IS_ENABLED(CONFIG_SMP); > } > extern void arch_irq_work_raise(void); > #endif /* _ASM_RISCV_IRQ_WORK_H */ >
On Tue, 03 May 2022 16:16:26 PDT (-0700), heiko@sntech.de wrote: > Am Samstag, 30. April 2022, 05:00:23 CEST schrieb Samuel Holland: >> irq_work is triggered via an IPI, but the IPI infrastructure is not >> included in uniprocessor kernels. As a result, irq_work never runs. >> Fall back to the tick-based irq_work implementation on uniprocessor >> configurations. >> >> Fixes: 298447928bb1 ("riscv: Support irq_work via self IPIs") >> Signed-off-by: Samuel Holland <samuel@sholland.org> > > That uniprocessor part seems a tiny bit neglected - as I saw previously > with alternatives not getting applied as well, so Ya, it definately is -- and not just me missing this fix, I'd also dropped the CONFIG_SMP=n test from my list by accident. Luckily it still boots in QEMU, so at least it's not as brokn as it could be. > > Reviewed-by: Heiko Stuebner <heiko@sntech.de> > > Though somehow I find the arm32 style a tad nicer by defining > an is_smp() function [0] that holds the necessary checks. > > But I guess that is a style preference. IIUC that's slightly different: this is a compile-time issue related to the IPI framework not coming up at all, not a runtime "are we on a uniprocessor" issue. I put this on for-next (no fixes right now, I'm still collecting 5.19 merge window material). Thanks! > > > Heiko > > > [0] https://elixir.bootlin.com/linux/latest/source/arch/arm/include/asm/smp_plat.h#L18 >> --- >> This was found while bringing up cpufreq on D1. Switching cpufreq >> governors was hanging on irq_work_sync(). >> >> arch/riscv/include/asm/irq_work.h | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/arch/riscv/include/asm/irq_work.h b/arch/riscv/include/asm/irq_work.h >> index d6c277992f76..b53891964ae0 100644 >> --- a/arch/riscv/include/asm/irq_work.h >> +++ b/arch/riscv/include/asm/irq_work.h >> @@ -4,7 +4,7 @@ >> >> static inline bool arch_irq_work_has_interrupt(void) >> { >> - return true; >> + return IS_ENABLED(CONFIG_SMP); >> } >> extern void arch_irq_work_raise(void); >> #endif /* _ASM_RISCV_IRQ_WORK_H */ >>
diff --git a/arch/riscv/include/asm/irq_work.h b/arch/riscv/include/asm/irq_work.h index d6c277992f76..b53891964ae0 100644 --- a/arch/riscv/include/asm/irq_work.h +++ b/arch/riscv/include/asm/irq_work.h @@ -4,7 +4,7 @@ static inline bool arch_irq_work_has_interrupt(void) { - return true; + return IS_ENABLED(CONFIG_SMP); } extern void arch_irq_work_raise(void); #endif /* _ASM_RISCV_IRQ_WORK_H */
irq_work is triggered via an IPI, but the IPI infrastructure is not included in uniprocessor kernels. As a result, irq_work never runs. Fall back to the tick-based irq_work implementation on uniprocessor configurations. Fixes: 298447928bb1 ("riscv: Support irq_work via self IPIs") Signed-off-by: Samuel Holland <samuel@sholland.org> --- This was found while bringing up cpufreq on D1. Switching cpufreq governors was hanging on irq_work_sync(). arch/riscv/include/asm/irq_work.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)