Message-ID: <4244f22a-a208-321f-ec32-150ffde35c8e@gmail.com>
Date: Sat, 29 Jun 2024 15:00:00 +0300
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.4.2
Subject: Re: LogwrtResult contended spinlock
Content-Language: en-US
To: Alvaro Herrera <alvherre@alvh.no-ip.org>
Cc: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>,
 Tom Lane <tgl@sss.pgh.pa.us>, Andres Freund <andres@anarazel.de>,
 Jeff Davis <pgsql@j-davis.com>, pgsql-hackers@lists.postgresql.org,
 Jaime Casanova <jcasanov@systemguards.com.ec>
References: <202406291023.6plsxeg2zho7@alvherre.pgsql>
From: Alexander Lakhin <exclusion@gmail.com>
In-Reply-To: <202406291023.6plsxeg2zho7@alvherre.pgsql>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Archived-At: <https://www.postgresql.org/message-id/4244f22a-a208-321f-ec32-150ffde35c8e%40gmail.com>
Precedence: bulk

Hi Alvaro,

Thank you for looking at this!

29.06.2024 13:23, Alvaro Herrera wrote:
>> TRAP: failed Assert("TYPEALIGN(8, (uintptr_t)(&currval)) ==
>> (uintptr_t)(&currval)"), File: "...\src\include\port/atomics.h", Line: 597,
>> PID: 7556
>> child process was terminated by exception 0xC0000409
> Oh.  This is the new assertion in pg_atomic_monotonic_advance_u64() and
> therefore the only possible culprit could be logInsertResult in
> XLogCtlData.  Does it work if you do this?
>
> diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
> index 8dcdf5a764..e581488d57 100644
> --- a/src/backend/access/transam/xlog.c
> +++ b/src/backend/access/transam/xlog.c
> @@ -468,7 +468,7 @@ typedef struct XLogCtlData
>   	XLogRecPtr	lastSegSwitchLSN;
>   
>   	/* These are accessed using atomics -- info_lck not needed */
> -	pg_atomic_uint64 logInsertResult;	/* last byte + 1 inserted to buffers */
> +	pg_atomic_uint64 logInsertResult pg_attribute_aligned(8);	/* last byte + 1 inserted to buffers */
>   	pg_atomic_uint64 logWriteResult;	/* last byte + 1 written out */
>   	pg_atomic_uint64 logFlushResult;	/* last byte + 1 flushed */

It doesn't, but the following works for me:
  static inline uint64
  pg_atomic_monotonic_advance_u64(volatile pg_atomic_uint64 *ptr, uint64 target_)
  {
-       uint64          currval;
+       pg_attribute_aligned(8) uint64          currval;

because the failed assertion is:
#ifndef PG_HAVE_ATOMIC_U64_SIMULATION
     AssertPointerAlignment(&currval, 8);
#endif

Best regards,
Alexander