Horton

Re: Compile error on the aarch64 platform: Missing asm/hwcap.h

Tom Lane <tgl@sss.pgh.pa.us>

View thread
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Steven Niu <niushiji@gmail.com>
Cc: John Naylor <johncnaylorls@gmail.com>, 高增琦 <pgf00a@gmail.com>, "tmunro@postgresql.org" <tmunro@postgresql.org>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2025-11-17T19:52:39Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Fix pg_popcount_aarch64.c to build with ancient glibc releases.

  2. Fix pg_crc32c_armv8_choose.c to build with ancient glibc releases.

Steven Niu <niushiji@gmail.com> writes:
> IvorySQL team found the same build issue on our building machine when we built IvorySQL code which is based on PG 18.0.

> pg_crc32c_armv8_choose.c:58:32:error:'HWCAP CRC32' undeclared(first use in this function)
>     58 | return (getauxval(AT_HWCAP) & HWCAP_CRC32) != θ;
>           |                                                                     ^~~~~~~~~~~

Bleah ... I confess to having misread the initial message as being a
complaint about HWCAP2_CRC32 not HWCAP_CRC32.  -ENOCAFFEINE I guess.

So what we've got here is that somewhere along the line glibc decided
that sys/auxv.h should duplicate the HWCAPxxx macros from the kernel's
header.  On recent (and even not so recent) aarch64 Fedora,
/usr/include/bits/hwcap.h has

/* The following must match the kernel's <asm/hwcap.h> and update the
   list together with sysdeps/unix/sysv/linux/aarch64/dl-procinfo.c.  */
#define HWCAP_FP                (1 << 0)
...

but that is not the case if you go back as far as RHEL7.

A minimal fix might be to change

 #if defined(HAVE_ELF_AUX_INFO) || defined(HAVE_GETAUXVAL)
 #include <sys/auxv.h>
-#if defined(__linux__) && !defined(__aarch64__) && !defined(HWCAP2_CRC32)
+#if defined(__linux__)
 #include <asm/hwcap.h>
 #endif
 #endif

but that risks compiler warnings if there are any macro discrepancies
at all between bits/hwcap.h and asm/hwcap.h.  I'm inclined to think
it's better to do something like

+#if defined(__linux__) && (defined(__aarch64__) ? !defined(HWCAP_CRC32) : !defined(HWCAP2_CRC32))

or perhaps that's too unreadable and we should break it out into
multiple #if's.

			regards, tom lane