[PATCH] irqchip/gic-v3: use dsb(ishst) to synchronize data to smp before issuing ipi

[Barry Song]

On Mon, Feb 21, 2022 at 4:21 AM Marc Zyngier <maz at kernel.org> wrote:
>
> On 2022-02-20 15:04, Russell King (Oracle) wrote:
> > On Sat, Feb 19, 2022 at 05:55:49AM +0800, Barry Song wrote:
> >> dsb(ishst) should be enough here as we only need to guarantee the
> >> visibility of data to other CPUs in smp inner domain before we
> >> send the ipi.
> >>
> >> Signed-off-by: Barry Song <song.bao.hua at hisilicon.com>
> >> ---
> >>  drivers/irqchip/irq-gic-v3.c | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/irqchip/irq-gic-v3.c
> >> b/drivers/irqchip/irq-gic-v3.c
> >> index 5e935d97207d..0efe1a9a9f3b 100644
> >> --- a/drivers/irqchip/irq-gic-v3.c
> >> +++ b/drivers/irqchip/irq-gic-v3.c
> >> @@ -1211,7 +1211,7 @@ static void gic_ipi_send_mask(struct irq_data
> >> *d, const struct cpumask *mask)
> >>       * Ensure that stores to Normal memory are visible to the
> >>       * other CPUs before issuing the IPI.
> >>       */
> >> -    wmb();
> >> +    dsb(ishst);
> >
> > On ARM, wmb() is a dsb(st) followed by other operations which may
> > include a sync operation at the L2 cache, and SoC specific barriers
> > for the bus. Hopefully, nothing will break if the sledge hammer is
> > replaced by the tack hammer.
>
> The saving grace is that ARMv8 forbids (as per D4.4.11) these
> SW-visible,
> non architected caches (something like PL310 simply isn't allowed).
> Given
> that GICv3 requires ARMv8 the first place, we should be OK.
>
> As for SoC-specific bus barriers, I don't know of any that would affect
> an ARMv8 based SoC. But I'm always prepared to be badly surprised...2

i assume we have been safe since dsb(ish) has been widely used for
smp data visibility in arm64:
arch/arm64/include/asm/assembler.h: dsb ish
arch/arm64/include/asm/cacheflush.h: dsb(ish);
arch/arm64/include/asm/pgtable.h: dsb(ishst);
arch/arm64/include/asm/pgtable.h: dsb(ishst);
arch/arm64/include/asm/pgtable.h: dsb(ishst);
arch/arm64/include/asm/pgtable.h: dsb(ishst);
arch/arm64/include/asm/pgtable.h: * is doing a dsb(ishst), but that
penalises the fastpath.
arch/arm64/include/asm/smp.h: dsb(ishst);
arch/arm64/include/asm/tlbflush.h:        "dsb ish\n tlbi " #op,        \
arch/arm64/include/asm/tlbflush.h:        "dsb ish\n tlbi " #op ", %0",     \
arch/arm64/include/asm/tlbflush.h: dsb(ishst);
arch/arm64/include/asm/tlbflush.h: dsb(ish);
arch/arm64/include/asm/tlbflush.h: dsb(ishst);
arch/arm64/include/asm/tlbflush.h: dsb(ish);
arch/arm64/include/asm/tlbflush.h: dsb(ishst);
arch/arm64/include/asm/tlbflush.h: dsb(ish);
arch/arm64/include/asm/tlbflush.h: dsb(ishst);
arch/arm64/include/asm/tlbflush.h: dsb(ish);
arch/arm64/include/asm/tlbflush.h: dsb(ishst);
arch/arm64/include/asm/tlbflush.h: dsb(ish);
arch/arm64/include/asm/tlbflush.h: dsb(ishst);
arch/arm64/include/asm/tlbflush.h: dsb(ish);
arch/arm64/kernel/alternative.c: dsb(ish);
arch/arm64/kernel/entry.S: dsb ish
arch/arm64/kernel/head.S: dsb ishst // Make zero page visible to PTW
arch/arm64/kernel/hibernate-asm.S: dsb ish /* wait for PoU cleaning to finish */
arch/arm64/kernel/hibernate-asm.S: dsb ish
arch/arm64/kernel/mte.c: dsb(ish);
arch/arm64/kernel/process.c: dsb(ish);
arch/arm64/kernel/sys_compat.c: dsb(ish);
arch/arm64/kvm/hyp/nvhe/tlb.c: dsb(ishst);
arch/arm64/kvm/hyp/nvhe/tlb.c: dsb(ish);
arch/arm64/kvm/hyp/nvhe/tlb.c: dsb(ish);
arch/arm64/kvm/hyp/nvhe/tlb.c: dsb(ishst);
arch/arm64/kvm/hyp/nvhe/tlb.c: dsb(ish);
arch/arm64/kvm/hyp/nvhe/tlb.c: dsb(ishst);
arch/arm64/kvm/hyp/nvhe/tlb.c: dsb(ish);
arch/arm64/kvm/hyp/pgtable.c: dsb(ishst);
arch/arm64/kvm/hyp/pgtable.c: dsb(ishst);
arch/arm64/kvm/hyp/pgtable.c: dsb(ishst);
arch/arm64/kvm/hyp/pgtable.c: dsb(ish);
arch/arm64/kvm/hyp/pgtable.c: dsb(ishst);
arch/arm64/kvm/hyp/pgtable.c: dsb(ishst);
arch/arm64/kvm/hyp/pgtable.c: dsb(ishst);
arch/arm64/kvm/hyp/vhe/tlb.c: dsb(ishst);
arch/arm64/kvm/hyp/vhe/tlb.c: dsb(ish);
arch/arm64/kvm/hyp/vhe/tlb.c: dsb(ish);
arch/arm64/kvm/hyp/vhe/tlb.c: dsb(ishst);
arch/arm64/kvm/hyp/vhe/tlb.c: dsb(ish);
arch/arm64/kvm/hyp/vhe/tlb.c: dsb(ishst);
arch/arm64/kvm/hyp/vhe/tlb.c: dsb(ish);
arch/arm64/mm/cache.S: dsb     ishst
arch/arm64/mm/cache.S: dsb ishst
arch/arm64/mm/kasan_init.c: dsb(ishst);
arch/arm64/mm/mmu.c: * We need dsb(ishst) here to ensure the
page-table-walker sees
arch/arm64/mm/mmu.c: dsb(ishst);
arch/arm64/mm/proc.S: dsb ish
drivers/irqchip/irq-gic-v3-its.c: dsb(ishst);
drivers/irqchip/irq-gic-v3-its.c: dsb(ishst);


Plus, is it even a problem to arm since arm only requires soc-level barrier
for system-level memory barrier rather than smp-level barrier?

#if defined(CONFIG_ARM_DMA_MEM_BUFFERABLE) || defined(CONFIG_SMP)
#define mb()            __arm_heavy_mb()
#define rmb()           dsb()
#define wmb()           __arm_heavy_mb(st)
#define dma_rmb()       dmb(osh)
#define dma_wmb()       dmb(oshst)
#else
#define mb()            barrier()
#define rmb()           barrier()
#define wmb()           barrier()
#define dma_rmb()       barrier()
#define dma_wmb()       barrier()
#endif

#define __smp_mb()      dmb(ish)
#define __smp_rmb()     __smp_mb()
#define __smp_wmb()     dmb(ishst)

I am not seeing arm requires soc-level mb for smp-level barrier though
the mandatory
barriers are using heavy_mb which has soc-level mb.

In this particular case, we are asking the data visibility for
smp-level not system-level. I am
not quite sure Russell's concern is relevant.

>
>          M.
> --
> Jazz is not dead. It just smells funny...

Thanks
Barry