This is an archive of the discontinued LLVM Phabricator instance.

[test] Skip ThinLTO cache tests requiring atime setting on NetBSD
ClosedPublic

Authored by mgorny on Dec 4 2018, 8:07 AM.

Download Raw Diff

Details

Reviewers

krytarowski
jhenderson
inglorion
pcc
rnk

Commits

rG929ce27cc77d: [test] Skip ThinLTO cache tests requiring atime setting on NetBSD
rL348355: [test] Skip ThinLTO cache tests requiring atime setting on NetBSD

Summary

Skip the ThinLTO cache tests on NetBSD. They require 'touch' being
able to alter atime of files, while NetBSD inhibits atime updates
when filesystem is mounted noatime.

Diff Detail

Repository: rL LLVM

Event Timeline

mgorny created this revision.Dec 4 2018, 8:07 AM

Herald added subscribers: dexonsmith, steven_wu, eraman, mehdi_amini. · View Herald TranscriptDec 4 2018, 8:07 AM

mgorny added a parent revision: D55271: [test] Split strip-preserve-time.test, and skip atime test on NetBSD.Dec 4 2018, 8:07 AM

krytarowski added inline comments.Dec 4 2018, 8:08 AM

test/ThinLTO/X86/cache.ll
2 ↗	(On Diff #176641)	Please include an inline comment why unsupported.

Updated wrt comment.

What would happen if the thin LTO cache is used on such a system? Do we need to put something in the thin LTO code to emit some sort of error, if this is attempted?

I don't really know; I suppose the same thing as if it were used on any regular system with noatime. However, I believe NetBSD is only special here as we can't set up the test environment with past dates.

krytarowski accepted this revision.Dec 4 2018, 9:59 AM

This revision is now accepted and ready to land.Dec 4 2018, 9:59 AM

In D55273#1318680, @jhenderson wrote:

What would happen if the thin LTO cache is used on such a system? Do we need to put something in the thin LTO code to emit some sort of error, if this is attempted?

Is this only optimization (i.e. not about generating correct result)? If so we will ignore it and align the NetBSD kernel behavior to Linux/FreeBSD.

LGTM. Thinking about it a bit more, and I don't think using the cache will cause any actual problems if everything has the same access time, beyond potentially doing unnecessary work.

test/ThinLTO/X86/cache.ll
1 ↗	(On Diff #176645)	Please move this comment down to the first set of RUN commands, since that's what it's describing. The UNSUPPORTED directive and associated comment are best kept at the top of the file before anything else.

mgorny updated this revision to Diff 176784.Dec 5 2018, 3:17 AM

mgorny marked an inline comment as done.

Closed by commit rL348355: [test] Skip ThinLTO cache tests requiring atime setting on NetBSD (authored by mgorny). · Explain WhyDec 5 2018, 3:18 AM

This revision was automatically updated to reflect the committed changes.

Revision Contents

Path

Size

llvm/

trunk/

test/

ThinLTO/

X86/

cache.ll

4 lines

Diff 176785

llvm/trunk/test/ThinLTO/X86/cache.ll

	; Verify first that without hash, we don't use the cache.			; NetBSD: noatime mounts currently inhibit 'touch -a' updates
				; UNSUPPORTED: system-netbsd

				; Verify first that without hash, we don't use the cache.
	; RUN: opt -module-summary %s -o %t.bc			; RUN: opt -module-summary %s -o %t.bc
	; RUN: opt -module-summary %p/Inputs/cache.ll -o %t2.bc			; RUN: opt -module-summary %p/Inputs/cache.ll -o %t2.bc

	; Verify that enabling caching is ignoring module without hash			; Verify that enabling caching is ignoring module without hash
	; RUN: rm -Rf %t.cache && mkdir %t.cache			; RUN: rm -Rf %t.cache && mkdir %t.cache
	; RUN: llvm-lto -thinlto-action=run -exported-symbol=globalfunc %t2.bc %t.bc -thinlto-cache-dir %t.cache			; RUN: llvm-lto -thinlto-action=run -exported-symbol=globalfunc %t2.bc %t.bc -thinlto-cache-dir %t.cache
	; RUN: ls %t.cache/llvmcache.timestamp			; RUN: ls %t.cache/llvmcache.timestamp
	; RUN: ls %t.cache \| count 1			; RUN: ls %t.cache \| count 1
	▲ Show 20 Lines • Show All 157 Lines • Show Last 20 Lines