gcc rodata regression
Hi, I have been fighting with a strange problem on ARM where gcc puts all the .rodata from a number of files lumped into a single .rodata section even when -fdata-sections is used. I searched and found a bug report here: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54303 which exactly describes the problem I see. According to the bug report this does not happen on x86_64, and there is a test case provided. Does anyone have any clues as to how to fix this bug? I am willing to give it a try with a few pointers. I currently have to add loads of #ifdefs in the C code to work around this bug and remove the unwanted and unreferenced data. Thanks, Simon
Re: gcc rodata regression
Hi Andreas, On 1 February 2015 at 01:55, Andreas Schwab wrote: > Simon Glass writes: > >> According to the bug report this does not happen on x86_64, > > That is not true, see comment #6. OK, so it is a widespread problem then. Do you have any hints on how I could begin to figure out how to fix this? It is blocking me from making progress so I'm willing to invest some time. Regards, Simon
Re: gcc rodata regression
Hi Segher, On 2 February 2015 at 01:16, Segher Boessenkool wrote: > On Sat, Jan 31, 2015 at 05:09:29PM -0700, Simon Glass wrote: >> I have been fighting with a strange problem on ARM where gcc puts all >> the .rodata from a number of files lumped into a single .rodata >> section even when -fdata-sections is used. >> >> I searched and found a bug report here: >> >> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54303 >> >> which exactly describes the problem I see. > > You could try this patch. I think there were still some problems with > it, there must have been some reason I didn't submit it. Cannot have > been just laziness :-P Thanks, I will give it a try and let you know what I find. [snip] Regards, Simon
Fwd: Re: gcc rodata regression
Hi Segher, On 2 February 2015 at 01:16, Segher Boessenkool wrote: > On Sat, Jan 31, 2015 at 05:09:29PM -0700, Simon Glass wrote: >> I have been fighting with a strange problem on ARM where gcc puts all >> the .rodata from a number of files lumped into a single .rodata >> section even when -fdata-sections is used. >> >> I searched and found a bug report here: >> >> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54303 >> >> which exactly describes the problem I see. > > You could try this patch. I think there were still some problems with > it, there must have been some reason I didn't submit it. Cannot have > been just laziness :-P I tried this patch and it solved my problem - the rodata section is now split up and most of it is discarded in the link phase. I hope some form of this can make it into gcc one day soon and a suitable regression test case can be added! Thanks, Simon [snip]
