| Message ID | d0f8dfb26c025d3e3eee1b5f610161ca19b942df.1577111367.git.christophe.leroy@c-s.fr (mailing list archive) |
|---|---|
| State | RFC |
| Delegated to: | Paul Burton |
| Headers | show |
| Series | powerpc/32: switch VDSO to C implementation. | expand |
On Mon, Dec 23, 2019 at 6:31 AM Christophe Leroy <christophe.leroy@c-s.fr> wrote: > > do_hres() is called from several places, so GCC doesn't inline > it at first. > > do_hres() takes a struct __kernel_timespec * parameter for > passing the result. In the 32 bits case, this parameter corresponds > to a local var in the caller. In order to provide a pointer > to this structure, the caller has to put it in its stack and > do_hres() has to write the result in the stack. This is suboptimal, > especially on RISC processor like powerpc. > > By making GCC inline the function, the struct __kernel_timespec > remains a local var using registers, avoiding the need to write and > read stack. > > The improvement is significant on powerpc. I'm okay with it, mainly because I don't expect many workloads to have more than one copy of the code hot at the same time.
On Mon, Dec 23, 2019 at 3:31 PM Christophe Leroy <christophe.leroy@c-s.fr> wrote: > > do_hres() is called from several places, so GCC doesn't inline > it at first. > > do_hres() takes a struct __kernel_timespec * parameter for > passing the result. In the 32 bits case, this parameter corresponds > to a local var in the caller. In order to provide a pointer > to this structure, the caller has to put it in its stack and > do_hres() has to write the result in the stack. This is suboptimal, > especially on RISC processor like powerpc. > > By making GCC inline the function, the struct __kernel_timespec > remains a local var using registers, avoiding the need to write and > read stack. > > The improvement is significant on powerpc. > > Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr> Good idea, I can see how this ends up being an improvement for most of the callers. Acked-by: Arnd Bergmann <arnd@arndb.de>
Arnd Bergmann <arnd@arndb.de> writes: > On Mon, Dec 23, 2019 at 3:31 PM Christophe Leroy > <christophe.leroy@c-s.fr> wrote: >> >> do_hres() is called from several places, so GCC doesn't inline >> it at first. >> >> do_hres() takes a struct __kernel_timespec * parameter for >> passing the result. In the 32 bits case, this parameter corresponds >> to a local var in the caller. In order to provide a pointer >> to this structure, the caller has to put it in its stack and >> do_hres() has to write the result in the stack. This is suboptimal, >> especially on RISC processor like powerpc. >> >> By making GCC inline the function, the struct __kernel_timespec >> remains a local var using registers, avoiding the need to write and >> read stack. >> >> The improvement is significant on powerpc. >> >> Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr> > > Good idea, I can see how this ends up being an improvement > for most of the callers. > > Acked-by: Arnd Bergmann <arnd@arndb.de> https://lore.kernel.org/r/20191112012724.250792-3-dima@arista.com On the way to be applied. Thanks, tglx
On 01/10/2020 09:07 PM, Thomas Gleixner wrote: > Arnd Bergmann <arnd@arndb.de> writes: >> On Mon, Dec 23, 2019 at 3:31 PM Christophe Leroy >> <christophe.leroy@c-s.fr> wrote: >>> >>> do_hres() is called from several places, so GCC doesn't inline >>> it at first. >>> >>> do_hres() takes a struct __kernel_timespec * parameter for >>> passing the result. In the 32 bits case, this parameter corresponds >>> to a local var in the caller. In order to provide a pointer >>> to this structure, the caller has to put it in its stack and >>> do_hres() has to write the result in the stack. This is suboptimal, >>> especially on RISC processor like powerpc. >>> >>> By making GCC inline the function, the struct __kernel_timespec >>> remains a local var using registers, avoiding the need to write and >>> read stack. >>> >>> The improvement is significant on powerpc. >>> >>> Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr> >> >> Good idea, I can see how this ends up being an improvement >> for most of the callers. >> >> Acked-by: Arnd Bergmann <arnd@arndb.de> > > https://lore.kernel.org/r/20191112012724.250792-3-dima@arista.com > > On the way to be applied. > Oh nice, I get even better result with the way it is done by Dmitry compared to my own first patch. On an mpc8xx at 132Mhz (32bits powerpc), before the patch I have gettimeofday: vdso: 1256 nsec/call clock-gettime-monotonic-raw: vdso: 1449 nsec/call clock-gettime-monotonic-coarse: vdso: 768 nsec/call clock-gettime-monotonic: vdso: 1390 nsec/call With the patch I have: gettimeofday: vdso: 947 nsec/call clock-gettime-monotonic-raw: vdso: 1156 nsec/call clock-gettime-monotonic-coarse: vdso: 638 nsec/call clock-gettime-monotonic: vdso: 1094 nsec/call So that's a 20-25% improvement. I modified it slightly as follows: diff --git a/lib/vdso/gettimeofday.c b/lib/vdso/gettimeofday.c index 9e474d54814f..b793f211bca8 100644 --- a/lib/vdso/gettimeofday.c +++ b/lib/vdso/gettimeofday.c @@ -37,8 +37,8 @@ u64 vdso_calc_delta(u64 cycles, u64 last, u64 mask, u32 mult) } #endif -static int do_hres(const struct vdso_data *vd, clockid_t clk, - struct __kernel_timespec *ts) +static __always_inline int do_hres(const struct vdso_data *vd, clockid_t clk, + struct __kernel_timespec *ts) { const struct vdso_timestamp *vdso_ts = &vd->basetime[clk]; u64 cycles, last, sec, ns; @@ -67,8 +67,8 @@ static int do_hres(const struct vdso_data *vd, clockid_t clk, return 0; } -static void do_coarse(const struct vdso_data *vd, clockid_t clk, - struct __kernel_timespec *ts) +static __always_inline int do_coarse(const struct vdso_data *vd, clockid_t clk, + struct __kernel_timespec *ts) { const struct vdso_timestamp *vdso_ts = &vd->basetime[clk]; u32 seq; @@ -78,6 +78,8 @@ static void do_coarse(const struct vdso_data *vd, clockid_t clk, ts->tv_sec = vdso_ts->sec; ts->tv_nsec = vdso_ts->nsec; } while (unlikely(vdso_read_retry(vd, seq))); + + return 0; } static __maybe_unused int @@ -95,15 +97,16 @@ __cvdso_clock_gettime_common(const struct vdso_data *vd, clockid_t clock, * clocks are handled in the VDSO directly. */ msk = 1U << clock; - if (likely(msk & VDSO_HRES)) { - return do_hres(&vd[CS_HRES_COARSE], clock, ts); - } else if (msk & VDSO_COARSE) { - do_coarse(&vd[CS_HRES_COARSE], clock, ts); - return 0; - } else if (msk & VDSO_RAW) { - return do_hres(&vd[CS_RAW], clock, ts); - } - return -1; + if (likely(msk & VDSO_HRES)) + vd += CS_HRES_COARSE; + else if (msk & VDSO_COARSE) + return do_coarse(&vd[CS_HRES_COARSE], clock, ts); + else if (msk & VDSO_RAW) + vd += CS_RAW; + else + return -1; + + return do_hres(vd, clock, ts); } static __maybe_unused int --- Christophe
diff --git a/lib/vdso/gettimeofday.c b/lib/vdso/gettimeofday.c index 24e1ba838260..86d5b1c8796b 100644 --- a/lib/vdso/gettimeofday.c +++ b/lib/vdso/gettimeofday.c @@ -34,8 +34,8 @@ u64 vdso_calc_delta(u64 cycles, u64 last, u64 mask, u32 mult) } #endif -static int do_hres(const struct vdso_data *vd, clockid_t clk, - struct __kernel_timespec *ts) +static inline int do_hres(const struct vdso_data *vd, clockid_t clk, + struct __kernel_timespec *ts) { const struct vdso_timestamp *vdso_ts = &vd->basetime[clk]; u64 cycles, last, sec, ns;
do_hres() is called from several places, so GCC doesn't inline it at first. do_hres() takes a struct __kernel_timespec * parameter for passing the result. In the 32 bits case, this parameter corresponds to a local var in the caller. In order to provide a pointer to this structure, the caller has to put it in its stack and do_hres() has to write the result in the stack. This is suboptimal, especially on RISC processor like powerpc. By making GCC inline the function, the struct __kernel_timespec remains a local var using registers, avoiding the need to write and read stack. The improvement is significant on powerpc. Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr> --- lib/vdso/gettimeofday.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)