Your message dated Thu, 29 Apr 2021 23:10:38 -0700
with message-id <YIufXjfeDp/CBH6D@localhost>
and subject line Re: aarch64 gcc defaults to -moutline-atomics but musl doesn't 
provide the corresponding functions
has caused the Debian Bug report #987591,
regarding aarch64 gcc defaults to -moutline-atomics but musl doesn't provide 
the corresponding functions
to be marked as done.

This means that you claim that the problem has been dealt with.
If this is not the case it is now your responsibility to reopen the
Bug report if necessary, and/or fix the problem forthwith.

(NB: If you are a system administrator and have no idea what this
message is talking about, this may indicate a serious mail system
misconfiguration somewhere. Please contact [email protected]
immediately.)


-- 
987591: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=987591
Debian Bug Tracking System
Contact [email protected] with problems
--- Begin Message ---
Package: musl
Version: 1.2.2-1
Severity: normal
X-Debbugs-Cc: [email protected]

aarch64 gcc appears to default to -moutline-atomics, which causes gcc to
emit references to function symbols for atomic operations rather than
inlining the corresponding instruction sequences.

However, musl doesn't appear to provide the corresponding function
symbols. This results in errors like these (from compiling libgit2):

attrcache.c:(.text.git_attr_cache__get+0x144): undefined reference to 
`__aarch64_swp8_acq_rel'
attrcache.c:(.text.git_attr_cache__get+0x158): undefined reference to 
`__aarch64_ldadd4_acq_rel'
attrcache.c:(.text.git_attr_cache__get+0x168): undefined reference to 
`__aarch64_swp8_acq_rel'
attrcache.c:(.text.git_attr_cache__get+0x180): undefined reference to 
`__aarch64_swp8_acq_rel'
attrcache.c:(.text.git_attr_cache__get+0x224): undefined reference to 
`__aarch64_cas8_acq_rel'
attrcache.c:(.text.git_attr_cache__get+0x248): undefined reference to 
`__aarch64_swp8_acq_rel'

Would it make sense for the aarch64 musl-gcc wrapper to pass
-mno-outline-atomics, which resolves this problem and makes gcc no
longer emit references to those symbols?

--- End Message ---
--- Begin Message ---
On Thu, 29 Apr 2021 19:33:42 -0700 Josh Triplett <[email protected]> wrote:
> On Mon, 26 Apr 2021 00:36:31 -0700 Josh Triplett <[email protected]> 
> wrote:
> > Package: musl
> > Version: 1.2.2-1
> > Severity: normal
> > X-Debbugs-Cc: [email protected]
> > 
> > aarch64 gcc appears to default to -moutline-atomics, which causes gcc to
> > emit references to function symbols for atomic operations rather than
> > inlining the corresponding instruction sequences.
> > 
> > However, musl doesn't appear to provide the corresponding function
> > symbols. This results in errors like these (from compiling libgit2):
> > 
> > attrcache.c:(.text.git_attr_cache__get+0x144): undefined reference to 
> > `__aarch64_swp8_acq_rel'
> > attrcache.c:(.text.git_attr_cache__get+0x158): undefined reference to 
> > `__aarch64_ldadd4_acq_rel'
> > attrcache.c:(.text.git_attr_cache__get+0x168): undefined reference to 
> > `__aarch64_swp8_acq_rel'
> > attrcache.c:(.text.git_attr_cache__get+0x180): undefined reference to 
> > `__aarch64_swp8_acq_rel'
> > attrcache.c:(.text.git_attr_cache__get+0x224): undefined reference to 
> > `__aarch64_cas8_acq_rel'
> > attrcache.c:(.text.git_attr_cache__get+0x248): undefined reference to 
> > `__aarch64_swp8_acq_rel'
> > 
> > Would it make sense for the aarch64 musl-gcc wrapper to pass
> > -mno-outline-atomics, which resolves this problem and makes gcc no
> > longer emit references to those symbols?
> 
> Following up on this: given the performance wins associated with
> -moutline-atomics, it may also make sense for musl to provide those
> symbols, or for the aarch64 musl-gcc to ensure that an appropriate
> libgcc with those symbols is linked in.

Based on some further analysis, it looks like this only applies when
passing -nodefaultlibs and manually linking in musl's libc; otherwise,
the aarch64 musl-gcc does seem to handle this by linking in libgcc.

Sorry for the noise; closing this bug.

--- End Message ---

Reply via email to