mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Mark Rutland <mark.rutland@arm.com>
To: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: linux-arch@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org, linux-mips@vger.kernel.org,
	clang-built-linux@googlegroups.com, x86@kernel.org,
	Catalin Marinas <catalin.marinas@arm.com>,
	Will Deacon <will.deacon@arm.com>, Arnd Bergmann <arnd@arndb.de>,
	Russell King <linux@armlinux.org.uk>,
	Paul Burton <paul.burton@mips.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Andy Lutomirski <luto@kernel.org>, Ingo Molnar <mingo@redhat.com>,
	Borislav Petkov <bp@alien8.de>, Stephen Boyd <sboyd@kernel.org>,
	Mark Salyzyn <salyzyn@android.com>,
	Kees Cook <keescook@chromium.org>,
	Peter Collingbourne <pcc@google.com>,
	Dmitry Safonov <0x7f454c46@gmail.com>,
	Andrei Vagin <avagin@openvz.org>,
	Nick Desaulniers <ndesaulniers@google.com>,
	Marc Zyngier <maz@kernel.org>, Will Deacon <will@kernel.org>
Subject: Re: [PATCH v3 21/26] arm64: Introduce asm/vdso/arch_timer.h
Date: Mon, 16 Mar 2020 10:28:46 +0000	[thread overview]
Message-ID: <20200316102845.GB5746@lakrids.cambridge.arm.com> (raw)
In-Reply-To: <20200313154345.56760-22-vincenzo.frascino@arm.com>

Hi Vincenzo,

On Fri, Mar 13, 2020 at 03:43:40PM +0000, Vincenzo Frascino wrote:
> The vDSO library should only include the necessary headers required for
> a userspace library (UAPI and a minimal set of kernel headers). To make
> this possible it is necessary to isolate from the kernel headers the
> common parts that are strictly necessary to build the library.
> 
> Introduce asm/vdso/arch_timer.h to contain all the arm64 specific
> code. This allows to replace the second isb() in __arch_get_hw_counter()
> with a fake dependent stack read of the counter which improves the vdso
> library peformances of ~4.5%. Below the results of vdsotest [1] ran for
> 100 iterations.
> 
> Before the patch:
> =================
> clock-gettime-monotonic: syscall: 771 nsec/call
> clock-gettime-monotonic:    libc: 130 nsec/call
> clock-gettime-monotonic:    vdso: 111 nsec/call
> ...
> clock-gettime-realtime: syscall: 762 nsec/call
> clock-gettime-realtime:    libc: 130 nsec/call
> clock-gettime-realtime:    vdso: 111 nsec/call
> 
> After the patch:
> ================
> clock-gettime-monotonic: syscall: 792 nsec/call
> clock-gettime-monotonic:    libc: 124 nsec/call
> clock-gettime-monotonic:    vdso: 106 nsec/call
> ...
> clock-gettime-realtime: syscall: 776 nsec/call
> clock-gettime-realtime:    libc: 124 nsec/call
> clock-gettime-realtime:    vdso: 106 nsec/call
> 
> [1] https://github.com/nathanlynch/vdsotest
> 
> Cc: Catalin Marinas <catalin.marinas@arm.com>
> Cc: Will Deacon <will@kernel.org>
> Cc: Marc Zyngier <maz@kernel.org>
> Cc: Mark Rutland <Mark.Rutland@arm.com>
> Signed-off-by: Vincenzo Frascino <vincenzo.frascino@arm.com>
> ---
>  arch/arm64/include/asm/arch_timer.h        | 29 ++++---------------
>  arch/arm64/include/asm/vdso/arch_timer.h   | 33 ++++++++++++++++++++++
>  arch/arm64/include/asm/vdso/gettimeofday.h |  7 +++--
>  3 files changed, 42 insertions(+), 27 deletions(-)
>  create mode 100644 arch/arm64/include/asm/vdso/arch_timer.h
> 
> diff --git a/arch/arm64/include/asm/arch_timer.h b/arch/arm64/include/asm/arch_timer.h
> index 7ae54d7d333a..7f22cd00ad45 100644
> --- a/arch/arm64/include/asm/arch_timer.h
> +++ b/arch/arm64/include/asm/arch_timer.h
> @@ -164,24 +164,7 @@ static inline void arch_timer_set_cntkctl(u32 cntkctl)
>  	isb();
>  }
>  
> -/*
> - * Ensure that reads of the counter are treated the same as memory reads
> - * for the purposes of ordering by subsequent memory barriers.
> - *
> - * This insanity brought to you by speculative system register reads,
> - * out-of-order memory accesses, sequence locks and Thomas Gleixner.
> - *
> - * http://lists.infradead.org/pipermail/linux-arm-kernel/2019-February/631195.html
> - */
> -#define arch_counter_enforce_ordering(val) do {				\
> -	u64 tmp, _val = (val);						\
> -									\
> -	asm volatile(							\
> -	"	eor	%0, %1, %1\n"					\
> -	"	add	%0, sp, %0\n"					\
> -	"	ldr	xzr, [%0]"					\
> -	: "=r" (tmp) : "r" (_val));					\
> -} while (0)
> +#include <asm/vdso/arch_timer.h>
>  
>  static __always_inline u64 __arch_counter_get_cntpct_stable(void)
>  {
> @@ -189,7 +172,7 @@ static __always_inline u64 __arch_counter_get_cntpct_stable(void)
>  
>  	isb();
>  	cnt = arch_timer_reg_read_stable(cntpct_el0);
> -	arch_counter_enforce_ordering(cnt);
> +	cnt = arch_counter_enforce_ordering(cnt);
>  	return cnt;

Why have you changed the structure of arch_counter_enforce_ordering() to
return a value? The commit message has no rationale for that.

If there is a reason to change that, I'd prefer the driver change as one
patch, before moving the definition.

[...]

> +/*
> + * Ensure that reads of the counter are treated the same as memory reads
> + * for the purposes of ordering by subsequent memory barriers.
> + *
> + * This insanity brought to you by speculative system register reads,
> + * out-of-order memory accesses, sequence locks and Thomas Gleixner.
> + *
> + * http://lists.infradead.org/pipermail/linux-arm-kernel/2019-February/631195.html
> + *
> + */
> +static u64 arch_counter_enforce_ordering(u64 val)
> +{
> +	u64 tmp, _val = (val);
> +
> +	asm volatile(
> +	"	eor	%0, %1, %1\n"
> +	"	add	%0, sp, %0\n"
> +	"	ldr	xzr, [%0]"
> +	: "=r" (tmp) : "r" (_val));
> +
> +	return _val;
> +}

This change has no functional effect. Since `_val` is only passed in as
an input parameter, the compiler can assume the assembly has no effect
on it.

As above, what is the rationale for changing this?

> @@ -82,10 +83,10 @@ static __always_inline u64 __arch_get_hw_counter(s32 clock_mode)
>  	isb();
>  	asm volatile("mrs %0, cntvct_el0" : "=r" (res) :: "memory");
>  	/*
> -	 * This isb() is required to prevent that the seq lock is
> -	 * speculated.#
> +	 * arch_counter_enforce_ordering() is required to prevent that
> +	 * the seq lock is speculated.
>  	 */
> -	isb();
> +	res = arch_counter_enforce_ordering(res);

Can we delete the comment entirely? We don't bother in <asm/arch_timer.h>.

Even better, can we factor out __arch_counter_get_cntvct(), and use
that?

Thanks,
Mark.

  parent reply	other threads:[~2020-03-16 10:28 UTC|newest]

Thread overview: 53+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-03-13 15:43 [PATCH v3 00/26] Introduce common headers for vDSO Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 01/26] linux/const.h: Extract common header " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 02/26] linux/bits.h: " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 03/26] linux/limits.h: " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 04/26] x86:Introduce asm/vdso/clocksource.h Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 05/26] arm: Introduce asm/vdso/clocksource.h Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 06/26] arm64: " Vincenzo Frascino
2020-03-15 18:29   ` Catalin Marinas
2020-03-13 15:43 ` [PATCH v3 07/26] mips: " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 08/26] linux/clocksource.h: Extract common header for vDSO Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 09/26] linux/math64.h: " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 10/26] linux/time.h: " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 11/26] linux/time32.h: " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 12/26] linux/time64.h: " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 13/26] linux/jiffies.h: " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 14/26] linux/ktime.h: " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 15/26] common: Introduce processor.h Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 16/26] scripts: Fix the inclusion order in modpost Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 17/26] linux/elfnote.h: Replace elf.h with UAPI equivalent Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 18/26] arm64: Introduce asm/vdso/processor.h Vincenzo Frascino
2020-03-15 18:30   ` Catalin Marinas
2020-03-16  9:42     ` Vincenzo Frascino
2020-03-16 10:22       ` Mark Rutland
2020-03-16 10:26         ` Catalin Marinas
2020-03-16 10:29           ` Mark Rutland
2020-03-16 10:30           ` Vincenzo Frascino
2020-03-16 10:29         ` Vincenzo Frascino
2020-03-16 10:34       ` Catalin Marinas
2020-03-16 10:55         ` Vincenzo Frascino
2020-03-16 11:22           ` Catalin Marinas
2020-03-16 13:35             ` Vincenzo Frascino
2020-03-16 14:43               ` Catalin Marinas
2020-03-16 15:33                 ` Vincenzo Frascino
2020-03-16 15:49                   ` Catalin Marinas
2020-03-16 16:05                     ` Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 19/26] arm64: vdso: Include common headers in the vdso library Vincenzo Frascino
2020-03-15 18:30   ` Catalin Marinas
2020-03-13 15:43 ` [PATCH v3 20/26] arm64: vdso32: " Vincenzo Frascino
2020-03-15 18:30   ` Catalin Marinas
2020-03-13 15:43 ` [PATCH v3 21/26] arm64: Introduce asm/vdso/arch_timer.h Vincenzo Frascino
2020-03-15 18:32   ` Catalin Marinas
2020-03-16 15:37     ` Vincenzo Frascino
2020-04-09 13:26       ` Will Deacon
2020-04-09 13:36         ` Vincenzo Frascino
2020-03-16 10:28   ` Mark Rutland [this message]
2020-03-13 15:43 ` [PATCH v3 22/26] mips: vdso: Enable mips to use common headers Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 23/26] x86: vdso: Enable x86 " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 24/26] arm: vdso: Enable arm " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 25/26] lib: vdso: Enable " Vincenzo Frascino
2020-03-13 15:43 ` [PATCH v3 26/26] arm64: vdso32: Enable Clang Compilation Vincenzo Frascino
2020-03-16  9:17   ` Catalin Marinas
2020-03-13 16:19 ` [PATCH v3 00/26] Introduce common headers for vDSO Dmitry Safonov
2020-03-15 10:02   ` Thomas Gleixner

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200316102845.GB5746@lakrids.cambridge.arm.com \
    --to=mark.rutland@arm.com \
    --cc=0x7f454c46@gmail.com \
    --cc=arnd@arndb.de \
    --cc=avagin@openvz.org \
    --cc=bp@alien8.de \
    --cc=catalin.marinas@arm.com \
    --cc=clang-built-linux@googlegroups.com \
    --cc=keescook@chromium.org \
    --cc=linux-arch@vger.kernel.org \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mips@vger.kernel.org \
    --cc=linux@armlinux.org.uk \
    --cc=luto@kernel.org \
    --cc=maz@kernel.org \
    --cc=mingo@redhat.com \
    --cc=ndesaulniers@google.com \
    --cc=paul.burton@mips.com \
    --cc=pcc@google.com \
    --cc=salyzyn@android.com \
    --cc=sboyd@kernel.org \
    --cc=tglx@linutronix.de \
    --cc=vincenzo.frascino@arm.com \
    --cc=will.deacon@arm.com \
    --cc=will@kernel.org \
    --cc=x86@kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®