Doug White wrote: > On Tue, 4 May 2004, David O'Brien wrote: > > >>On Sat, May 01, 2004 at 08:42:46AM -0700, Doug White wrote: >> >>>2. Changing COPTFLAGS to '-O -frename-registers -pipe' from the default >>>'-O2 -frename-registers -pipe' stops the crash. So it may be an >>>O2-specific optimization bug. >> >>... >> >>>So in the end, at -O2, (nswbuf*MAXPHYS) != nswbuf*MAXPHYS. Creepy, eh? >>>I'll see if I can come up with a standalone test case to submit to the gcc >>>folks. >> >>Thanks! >>This is the first report of an -O2 problem, and I'm very happy you're >>willing to try to narrow this down so we don't have to return to -O[1]. > > > I caught kan's attention with this last night on irc, and he's going to > try to isolate the bad code generation. We could really use a standalone > case to test with & submit, but a naieve test doesn't come up with any > differences. > > In order to keep my amd64 stable I'm having to leave COPTFLAGS set, so if > someone commits a fix to gcc or imports a new version, try to get my > attention so I can test it with -O2 again. > > Also if we get into the 5.3 release timeframe we'll have to make a call to > ship with a -O2 compiled kernel or not. > Are you proposing dropping -O2 for 5.3 because you are afraid that there are other, not-yet-detected broken cases? I thought that we found a source code work-around that forced gcc to generate the correct assembly with -O2. Maybe that should be committed for the time being? ScottReceived on Fri May 07 2004 - 11:32:09 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:53 UTC