Bug 93811 - __builtin___clear_cache() is a noop on powerpc (which is incorrect)
Summary: __builtin___clear_cache() is a noop on powerpc (which is incorrect)
Status: NEW
Alias: None
Product: gcc
Classification: Unclassified
Component: target (show other bugs)
Version: 7.5.0
: P3 normal
Target Milestone: ---
Assignee: Not yet assigned to anyone
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2020-02-18 17:17 UTC by anton
Modified: 2021-10-30 05:33 UTC (History)
1 user (show)

See Also:
Host: powerpc64-unknown-linux-gnu
Target: powerpc64-unknown-linux-gnu
Build: powerpc64-unknown-linux-gnu
Known to work:
Known to fail:
Last reconfirmed: 2020-05-01 00:00:00

Attachments
source code for demonstrating the bug (77 bytes, text/plain)
2020-02-18 17:17 UTC, anton 		Details
Disassembly output for demonstration file (160 bytes, text/plain)
2020-02-18 17:20 UTC, anton 		Details
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description anton 2020-02-18 17:17:52 UTC 
Created attachment 47869 [details]
source code for demonstrating the bug

A call to __builtin___clear_cache() results in no code being generated.  This can lead to stale I-cache contents on at least the PPC970 and the PPC7447A (I have seen breakage from this bug on these CPUs).  This bug is present in at least gcc-4.3.2, gcc-5.5.0, and gcc-7.5.0.  Given that it has not been fixed for such a long time, it probably is still present in gcc-9 (which I could not test for other reasons).

You can find a working implementation of a similar function at <http://git.savannah.gnu.org/cgit/gforth.git/tree/arch/power/_sync_cache_range.c>
Comment 1 anton 2020-02-18 17:20:12 UTC 
Created attachment 47870 [details]
Disassembly output for demonstration file
Comment 2 Andrew Pinski 2020-02-20 03:37:25 UTC 
Note ___clear_cache was originally designed for trampolines. PoewrPC64v1 ABI does not use trampolines for nested functions.
Comment 3 Segher Boessenkool 2020-05-01 02:06:31 UTC 
Confirmed.  Apparently this builtin isn't used much at all :-/
Comment 4 Andrew Pinski 2021-10-30 05:33:58 UTC 
Mono's implementation seems like the best so far:
https://github.com/mono/mono/blob/main/mono/mini/mini-ppc.c#L759-L795