1538236 - >=firefox-66.0 fails to compile with lto enabled on armhf with error: undefined reference to '_PrepareAndDispatch'

Status: RESOLVED FIXED

(Core :: XPCOM, defect, P3)

Description

[tt_1]

Attachment: firefox-66.0-lto-armhf.log.zip

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Firefox/60.0

Steps to reproduce:

I tried to emerge firefox for armhf with lto enabled.

Actual results:

The initial compile without lto passed, but the lto phase fails with this beauty:

27:12.99 /usr/armv7a-unknown-linux-gnueabihf/tmp/portage/www-client/firefox-66.0-r1/temp/ccuBxeIi.ltrans0.ltrans.o(.text+0xc): error: undefined reference to '_PrepareAndDispatch'

compressed build.log is attached.

Expected results:

The expectation is that the compile passes.

Comment 1

[Kim Moir [:kmoir] ET]

:tt_1 Can you provide the exact steps you used to build? Your log just provides the output of the build process.

Comment 2

[Tom Tung [:tt, :ttung]]

redirect the ni to the right person :)

Comment 3

[tt_1]

I'm willing to provide more details, it's just that I'm not sure what you mean by exact steps used to build? I set up a cross-compiler, compiled all dependencies + rust-std for target. All configure and make steps are in the log.

Without lto it's passing.

Comment 4

[Nathan Froyd [:froydnj]]

Pretty sure this is an xpcom bug.

Comment 5

[Nathan Froyd [:froydnj]]

Attachment: Bug 1538236 - mark aarch64 unix's PrepareAndDispatch as used; r=mccr8

Otherwise, LTO will eliminate it because the compiler can't see the
calls to PrepareAndDispatch from the relevant assembly file.

Comment 6

[Nathan Froyd [:froydnj]]

Would you please see if the posted patch fixes your problem? I don't have an armhf system to test on. Thanks!

Comment 7

[tt_1]

hey there, this bug is against armv7a-unknown-linux-gnueabihf, it's essentially armv7 + hardfloat. It is not ARM64. Nontheless I tested your patch, but to no avail.

It might be a reincarnation of #336183

Comment 8

[Nathan Froyd [:froydnj]]

Attachment: fix for arm

(In reply to tt_1 from comment #7)

hey there, this bug is against armv7a-unknown-linux-gnueabihf, it's essentially armv7 + hardfloat. It is not ARM64. Nontheless I tested your patch, but to no avail.

It might be a reincarnation of #336183

Doh, sorry about that! I've had aarch64 on the brain for a while now.

I believe this new patch should work better, though I haven't tested it. Would you try it out?

Comment 9

[tt_1]

That patch fixes the compile, thank you!

Comment 10

[Nathan Froyd [:froydnj]]

Attachment: Bug 1538236 - fix LTO issues with arm xptcstubs; r=mccr8

ARM's xptcstubs use a slightly different setup for PrepareAndDispatch
than...well, all of our other stubs. This difference appears to be
causing problems with LTO builds. Change the setup to be more like our
other stubs, which additionally gets rid of some of the asm nonsense.

Comment 11

[tt_1]

Attachment: firefox-66.0.2-lto.png

runtime tested with --enable-lto=thin, works fine.

Comment 12

[tt_1]

An uplift to beta 67 would be highly appreciated.

Comment 13

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/576f9e652a20
fix LTO issues with arm xptcstubs; r=mccr8

Comment 14

[Bogdan Tara[:bogdan_tara | bogdant]]

https://hg.mozilla.org/mozilla-central/rev/576f9e652a20

Comment 15

[Nathan Froyd [:froydnj]]

Comment on attachment 9055989 [details]
Bug 1538236 - fix LTO issues with arm xptcstubs; r=mccr8

Beta/Release Uplift Approval Request

• Feature/Bug causing the regression: None
• User impact if declined: Third parties doing arm-specific builds with LTO will run into issues.
• Is this code covered by automated tests?: Yes
• Has the fix been verified in Nightly?: Yes
• Needs manual test from QE?: No
• If yes, steps to reproduce:
• List of other uplifts needed: None
• Risk to taking this patch: Low
• Why is the change risky/not risky? (and alternatives if risky): This change is in core code, but it's such core code that if it was broken, the entire browser would fall over in every test we run or builds would fail completely.
• String changes made/needed: None.

Comment 16

[Pascal Chevrel:pascalc (PTO until April 26)]

Comment on attachment 9055989 [details]
Bug 1538236 - fix LTO issues with arm xptcstubs; r=mccr8

Low risk, covered by tests and verified on nightly, approved for 67 beta 10, thanks.

Comment 17

[Andreea Pavel [:apavel]]

https://hg.mozilla.org/releases/mozilla-beta/rev/1cdd3ceab2bc

Comment 18

[tt_1]

excellent, thank you very much!