On 17/11/14 11:42, Thomas Leonard wrote: > On 26 October 2014 10:25, Thomas Leonard <tal...@gmail.com> wrote: >> On 26 October 2014 09:55, Samuel Thibault <samuel.thiba...@ens-lyon.org> >> wrote: >>> Thomas Leonard, le Sun 26 Oct 2014 09:46:09 +0000, a écrit : >>>> Could you say a bit more about the linker problems you had? >>> Really digging in the archives this time :) >>> >>> ia64-specific: >>> http://lists.xen.org/archives/html/xen-ia64-devel/2008-03/msg00126.html >>> http://lists.xen.org/archives/html/xen-ia64-devel/2008-12/msg00070.html >>> x86_64-specific (missing red zone support) >>> http://lists.xen.org/archives/html/xen-ia64-devel/2008-02/msg00251.html >>> >>> So I guess it could be OK on arm, but you have to make sure that Mini-OS >>> implements the whole ABI that gcc will use. Testing is not enough, I got >>> hit by the red zone for instance. >> On ARM, we have a separate stack for the IRQ handler, so the red zone >> at least shouldn't be a problem. > Incidentally, why doesn't Mini-OS/x86 use a red zone? I assume there's > a worthwhile performance benefit to it, and it would prevent subtle > bugs when software is accidentally compiled for the normal ABI. > >
Kernel code (intended for ring0) cannot use red zones as a stack switch doesn't occur when an exception/interrupt occurs. Compiling any kernel code with red zone support leads to subtle clobbering of state. ~Andrew _______________________________________________ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
