Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
IronPort-SDR: fRVUJcIsQ37FAWndKJTDp/Lf7/Dv8ELFxw3JaJPj3qAGbKbinoRnFiWS8cd5lGJryfI/CXzPQs
 sRdJjQlLuNqw==
IronPort-SDR: Qz6muSZFi2ahvoQYCouIM8vwxtXgOzHC7ph/awKQc8NhnCWD3LEcZa1By6dLhZGIsxPtciCSfC
 upSLURDM66xQ==
Subject: Re: [PATCH v2 6/8] perf auxtrace: Drop legacy __sync functions
To:     Leo Yan <leo.yan@linaro.org>,
        Arnaldo Carvalho de Melo <acme@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@redhat.com>,
        Namhyung Kim <namhyung@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>, x86@kernel.org,
        "H. Peter Anvin" <hpa@zytor.com>,
        Mathieu Poirier <mathieu.poirier@linaro.org>,
        Suzuki K Poulose <suzuki.poulose@arm.com>,
        Mike Leach <mike.leach@linaro.org>,
        Andi Kleen <ak@linux.intel.com>,
        linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org,
        coresight@lists.linaro.org, linux-arm-kernel@lists.infradead.org
References: <20210602103007.184993-1-leo.yan@linaro.org>
 <20210602103007.184993-7-leo.yan@linaro.org>
From:   Adrian Hunter <adrian.hunter@intel.com>
Organization: Intel Finland Oy, Registered Address: PL 281, 00181 Helsinki,
 Business Identity Code: 0357606 - 4, Domiciled in Helsinki
Message-ID: <fc7bfa25-d96e-ea6d-2756-f94653dc4771@intel.com>
Date:   Wed, 2 Jun 2021 13:47:42 +0300
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.10.1
MIME-Version: 1.0
In-Reply-To: <20210602103007.184993-7-leo.yan@linaro.org>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2/06/21 1:30 pm, Leo Yan wrote:
> The main purpose for using __sync built-in functions is to support
> compat mode for 32-bit perf with 64-bit kernel.  But using these
> built-in functions might cause couple potential issues.
> 
> Firstly, __sync functions originally support Intel Itanium processoer [1]
> but it cannot promise to support all 32-bit archs.  Now these
> functions have become the legacy functions.
> 
> As Peter also pointed out the logic issue in the function
> auxtrace_mmap__write_tail(), it does a cmpxchg with 0 values to load
> old_tail, and then executes a further cmpxchg with old_tail to write
> the new tail.  If consider the aux_tail might be assigned to '0' in the
> middle of loops, this can introduce mess for AUX buffer if the kernel
> fetches the temporary value '0'.
> 
> Considering __sync functions cannot really fix the 64-bit value
> atomicity on 32-bit archs, thus this patch drops __sync functions.
> 
> Credits to Peter for detailed analysis.
> 
> [1] https://gcc.gnu.org/onlinedocs/gcc/_005f_005fsync-Builtins.html#g_t_005f_005fsync-Builtins
> 
> Suggested-by: Peter Zijlstra <peterz@infradead.org>
> Signed-off-by: Leo Yan <leo.yan@linaro.org>
> ---
>  tools/perf/util/auxtrace.h | 19 -------------------
>  1 file changed, 19 deletions(-)
> 
> diff --git a/tools/perf/util/auxtrace.h b/tools/perf/util/auxtrace.h
> index 42b7ef811bde..e625bc76cdde 100644
> --- a/tools/perf/util/auxtrace.h
> +++ b/tools/perf/util/auxtrace.h
> @@ -432,12 +432,6 @@ struct auxtrace_cache;
>  
>  #ifdef HAVE_AUXTRACE_SUPPORT
>  
> -/*
> - * In snapshot mode the mmapped page is read-only which makes using
> - * __sync_val_compare_and_swap() problematic.  However, snapshot mode expects
> - * the buffer is not updated while the snapshot is made (e.g. Intel PT disables
> - * the event) so there is not a race anyway.
> - */
>  static inline u64 auxtrace_mmap__read_snapshot_head(struct auxtrace_mmap *mm)
>  {
>  	struct perf_event_mmap_page *pc = mm->userpg;
> @@ -451,11 +445,7 @@ static inline u64 auxtrace_mmap__read_snapshot_head(struct auxtrace_mmap *mm)
>  static inline u64 auxtrace_mmap__read_head(struct auxtrace_mmap *mm)
>  {
>  	struct perf_event_mmap_page *pc = mm->userpg;
> -#if BITS_PER_LONG == 64 || !defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT)

The test and setup for HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT is not needed anymore either.

>  	u64 head = READ_ONCE(pc->aux_head);
> -#else
> -	u64 head = __sync_val_compare_and_swap(&pc->aux_head, 0, 0);
> -#endif
>  
>  	/* Ensure all reads are done after we read the head */
>  	smp_rmb();
> @@ -465,19 +455,10 @@ static inline u64 auxtrace_mmap__read_head(struct auxtrace_mmap *mm)
>  static inline void auxtrace_mmap__write_tail(struct auxtrace_mmap *mm, u64 tail)
>  {
>  	struct perf_event_mmap_page *pc = mm->userpg;
> -#if BITS_PER_LONG != 64 && defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT)
> -	u64 old_tail;
> -#endif
>  
>  	/* Ensure all reads are done before we write the tail out */
>  	smp_mb();
> -#if BITS_PER_LONG == 64 || !defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT)
>  	pc->aux_tail = tail;
> -#else
> -	do {
> -		old_tail = __sync_val_compare_and_swap(&pc->aux_tail, 0, 0);
> -	} while (!__sync_bool_compare_and_swap(&pc->aux_tail, old_tail, tail));
> -#endif
>  }
>  
>  int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
>