Messages in this thread Patch in this message | | | Date | Fri, 12 Feb 2021 17:21:28 +0000 | From | Catalin Marinas <> | Subject | Re: [PATCH v13 4/7] arm64: mte: Enable TCO in functions that can read beyond buffer limits |
| |
On Thu, Feb 11, 2021 at 03:33:50PM +0000, Vincenzo Frascino wrote: > diff --git a/arch/arm64/kernel/mte.c b/arch/arm64/kernel/mte.c > index 706b7ab75f31..65ecb86dd886 100644 > --- a/arch/arm64/kernel/mte.c > +++ b/arch/arm64/kernel/mte.c > @@ -26,6 +26,10 @@ u64 gcr_kernel_excl __ro_after_init; > > static bool report_fault_once = true; > > +/* Whether the MTE asynchronous mode is enabled. */ > +DEFINE_STATIC_KEY_FALSE(mte_async_mode); > +EXPORT_SYMBOL_GPL(mte_async_mode); > + > static void mte_sync_page_tags(struct page *page, pte_t *ptep, bool check_swap) > { > pte_t old_pte = READ_ONCE(*ptep); > @@ -119,12 +123,24 @@ static inline void __mte_enable_kernel(const char *mode, unsigned long tcf) > void mte_enable_kernel_sync(void) > { > __mte_enable_kernel("synchronous", SCTLR_ELx_TCF_SYNC); > + > + /* > + * This function is called on each active smp core at boot > + * time, hence we do not need to take cpu_hotplug_lock again. > + */ > + static_branch_disable_cpuslocked(&mte_async_mode); > } > EXPORT_SYMBOL_GPL(mte_enable_kernel_sync); > > void mte_enable_kernel_async(void) > { > __mte_enable_kernel("asynchronous", SCTLR_ELx_TCF_ASYNC); > + > + /* > + * This function is called on each active smp core at boot > + * time, hence we do not need to take cpu_hotplug_lock again. > + */ > + static_branch_enable_cpuslocked(&mte_async_mode); > }
Sorry, I missed the cpuslocked aspect before. Is there any reason you need to use this API here? I suggested to add it to the mte_enable_kernel_sync() because kasan may at some point do this dynamically at run-time, so the boot-time argument doesn't hold. But it's also incorrect as this function will be called for hot-plugged CPUs as well after boot.
The only reason for static_branch_*_cpuslocked() is if it's called from a region that already invoked cpus_read_lock() which I don't think is the case here.
-- Catalin
| |