| Message ID | 20250402-arm-vdso-v1-1-2e7a12d75107@debian.org (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [RFC] arm64: vdso: Use __arch_counter_get_cntvct() | expand |
On Wed, 02 Apr 2025 20:22:47 +0100, Breno Leitao <leitao@debian.org> wrote: > > While reading how `cntvct_el0` was read in the kernel, I found that > __arch_get_hw_counter() is doing something very similar to what > __arch_counter_get_cntvct() is already doing. > > Use the existing __arch_counter_get_cntvct() function instead of > duplicating similar inline assembly code in __arch_get_hw_counter(). > > Both functions were performing nearly identical operations to read the > cntvct_el0 register. The only difference was that > __arch_get_hw_counter() included a memory clobber in its inline > assembly, which appears unnecessary in this context. > > This change simplifies the code by eliminating duplicate functionality > and improves maintainability by centralizing the counter access logic in > a single implementation. > > Signed-off-by: Breno Leitao <leitao@debian.org> > --- > I'm sharing this code as an RFC since I'm not intimately familiar with > different arm platforms, and I want to double-check that I haven't > missed anything subtle. > --- > arch/arm64/include/asm/vdso/gettimeofday.h | 22 ++-------------------- > 1 file changed, 2 insertions(+), 20 deletions(-) > > diff --git a/arch/arm64/include/asm/vdso/gettimeofday.h b/arch/arm64/include/asm/vdso/gettimeofday.h > index 92a2b59a9f3df..417b5b41b877d 100644 > --- a/arch/arm64/include/asm/vdso/gettimeofday.h > +++ b/arch/arm64/include/asm/vdso/gettimeofday.h > @@ -11,6 +11,7 @@ > #include <asm/barrier.h> > #include <asm/unistd.h> > #include <asm/sysreg.h> > +#include <asm/arch_timer.h> > > #define VDSO_HAS_CLOCK_GETRES 1 > > @@ -69,8 +70,6 @@ int clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts) > static __always_inline u64 __arch_get_hw_counter(s32 clock_mode, > const struct vdso_time_data *vd) > { > - u64 res; > - > /* > * Core checks for mode already, so this raced against a concurrent > * update. Return something. Core will do another round and then > @@ -79,24 +78,7 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode, > if (clock_mode == VDSO_CLOCKMODE_NONE) > return 0; > > - /* > - * If FEAT_ECV is available, use the self-synchronizing counter. > - * Otherwise the isb is required to prevent that the counter value > - * is speculated. > - */ > - asm volatile( > - ALTERNATIVE("isb\n" > - "mrs %0, cntvct_el0", > - "nop\n" > - __mrs_s("%0", SYS_CNTVCTSS_EL0), > - ARM64_HAS_ECV) > - : "=r" (res) > - : > - : "memory"); > - > - arch_counter_enforce_ordering(res); > - > - return res; > + return __arch_counter_get_cntvct(); I won't pretend I understand it all, but you really want to have a look at the link just above the arch_counter_enforce_ordering() definition, pasted below for your convenience: https://lore.kernel.org/r/alpine.DEB.2.21.1902081950260.1662@nanos.tec.linutronix.de/ Dropping this ordering enforcement seems pretty adventurous unless you have very strong guarantees about the context this executes in. M.
On Wed, 02 Apr 2025 23:04:41 +0100, Marc Zyngier <maz@kernel.org> wrote: > > On Wed, 02 Apr 2025 20:22:47 +0100, > Breno Leitao <leitao@debian.org> wrote: > > > > While reading how `cntvct_el0` was read in the kernel, I found that > > __arch_get_hw_counter() is doing something very similar to what > > __arch_counter_get_cntvct() is already doing. > > > > Use the existing __arch_counter_get_cntvct() function instead of > > duplicating similar inline assembly code in __arch_get_hw_counter(). > > > > Both functions were performing nearly identical operations to read the > > cntvct_el0 register. The only difference was that > > __arch_get_hw_counter() included a memory clobber in its inline > > assembly, which appears unnecessary in this context. > > > > This change simplifies the code by eliminating duplicate functionality > > and improves maintainability by centralizing the counter access logic in > > a single implementation. > > > > Signed-off-by: Breno Leitao <leitao@debian.org> > > --- > > I'm sharing this code as an RFC since I'm not intimately familiar with > > different arm platforms, and I want to double-check that I haven't > > missed anything subtle. > > --- > > arch/arm64/include/asm/vdso/gettimeofday.h | 22 ++-------------------- > > 1 file changed, 2 insertions(+), 20 deletions(-) > > > > diff --git a/arch/arm64/include/asm/vdso/gettimeofday.h b/arch/arm64/include/asm/vdso/gettimeofday.h > > index 92a2b59a9f3df..417b5b41b877d 100644 > > --- a/arch/arm64/include/asm/vdso/gettimeofday.h > > +++ b/arch/arm64/include/asm/vdso/gettimeofday.h > > @@ -11,6 +11,7 @@ > > #include <asm/barrier.h> > > #include <asm/unistd.h> > > #include <asm/sysreg.h> > > +#include <asm/arch_timer.h> > > > > #define VDSO_HAS_CLOCK_GETRES 1 > > > > @@ -69,8 +70,6 @@ int clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts) > > static __always_inline u64 __arch_get_hw_counter(s32 clock_mode, > > const struct vdso_time_data *vd) > > { > > - u64 res; > > - > > /* > > * Core checks for mode already, so this raced against a concurrent > > * update. Return something. Core will do another round and then > > @@ -79,24 +78,7 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode, > > if (clock_mode == VDSO_CLOCKMODE_NONE) > > return 0; > > > > - /* > > - * If FEAT_ECV is available, use the self-synchronizing counter. > > - * Otherwise the isb is required to prevent that the counter value > > - * is speculated. > > - */ > > - asm volatile( > > - ALTERNATIVE("isb\n" > > - "mrs %0, cntvct_el0", > > - "nop\n" > > - __mrs_s("%0", SYS_CNTVCTSS_EL0), > > - ARM64_HAS_ECV) > > - : "=r" (res) > > - : > > - : "memory"); > > - > > - arch_counter_enforce_ordering(res); > > - > > - return res; > > + return __arch_counter_get_cntvct(); > > I won't pretend I understand it all, but you really want to have a > look at the link just above the arch_counter_enforce_ordering() > definition, pasted below for your convenience: > > https://lore.kernel.org/r/alpine.DEB.2.21.1902081950260.1662@nanos.tec.linutronix.de/ > > Dropping this ordering enforcement seems pretty adventurous unless you > have very strong guarantees about the context this executes in. Ah, I appear to have misread this patch, and __arch_counter_get_cntvct() does have the same ordering requirements. Apologies for the noise. M.
diff --git a/arch/arm64/include/asm/vdso/gettimeofday.h b/arch/arm64/include/asm/vdso/gettimeofday.h index 92a2b59a9f3df..417b5b41b877d 100644 --- a/arch/arm64/include/asm/vdso/gettimeofday.h +++ b/arch/arm64/include/asm/vdso/gettimeofday.h @@ -11,6 +11,7 @@ #include <asm/barrier.h> #include <asm/unistd.h> #include <asm/sysreg.h> +#include <asm/arch_timer.h> #define VDSO_HAS_CLOCK_GETRES 1 @@ -69,8 +70,6 @@ int clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts) static __always_inline u64 __arch_get_hw_counter(s32 clock_mode, const struct vdso_time_data *vd) { - u64 res; - /* * Core checks for mode already, so this raced against a concurrent * update. Return something. Core will do another round and then @@ -79,24 +78,7 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode, if (clock_mode == VDSO_CLOCKMODE_NONE) return 0; - /* - * If FEAT_ECV is available, use the self-synchronizing counter. - * Otherwise the isb is required to prevent that the counter value - * is speculated. - */ - asm volatile( - ALTERNATIVE("isb\n" - "mrs %0, cntvct_el0", - "nop\n" - __mrs_s("%0", SYS_CNTVCTSS_EL0), - ARM64_HAS_ECV) - : "=r" (res) - : - : "memory"); - - arch_counter_enforce_ordering(res); - - return res; + return __arch_counter_get_cntvct(); } #endif /* !__ASSEMBLY__ */
While reading how `cntvct_el0` was read in the kernel, I found that __arch_get_hw_counter() is doing something very similar to what __arch_counter_get_cntvct() is already doing. Use the existing __arch_counter_get_cntvct() function instead of duplicating similar inline assembly code in __arch_get_hw_counter(). Both functions were performing nearly identical operations to read the cntvct_el0 register. The only difference was that __arch_get_hw_counter() included a memory clobber in its inline assembly, which appears unnecessary in this context. This change simplifies the code by eliminating duplicate functionality and improves maintainability by centralizing the counter access logic in a single implementation. Signed-off-by: Breno Leitao <leitao@debian.org> --- I'm sharing this code as an RFC since I'm not intimately familiar with different arm platforms, and I want to double-check that I haven't missed anything subtle. --- arch/arm64/include/asm/vdso/gettimeofday.h | 22 ++-------------------- 1 file changed, 2 insertions(+), 20 deletions(-) --- base-commit: acc4d5ff0b61eb1715c498b6536c38c1feb7f3c1 change-id: 20250402-arm-vdso-c4b0c9f35300 Best regards,