On Wed, 7 Mar 2018, Matthias Kaehlcke wrote:
> El Tue, Feb 20, 2018 at 10:59:49PM +0100 Arnd Bergmann ha dit:
>
> > Trying to build an LTO-Enabled kernel with Thumb2 instructions failed
> > horribly for me, with an endless output of things like
> >
> > ccVnNycO.s:2665: Error: thumb conditional instruction should be in IT block
> > -- `bxne lr'
> > ccVnNycO.s:7128: Error: thumb conditional instruction should be in IT block
> > -- `strexeq r5,r2,[r3]'
> > ccVnNycO.s:7258: Error: thumb conditional instruction should be in IT block
> > -- `strexeq lr,r0,[r3]'
> > ccVnNycO.s:17380: Error: thumb conditional instruction should be in IT
> > block -- `strexeq r1,r2,[r6]'
> > ccVnNycO.s:19163: Error: thumb conditional instruction should be in IT
> > block -- `strexeq r8,r6,[r3]'
> > ccVnNycO.s:22722: Error: thumb conditional instruction should be in IT
> > block -- `strexeq r7,r1,[r0]'
> > ccVnNycO.s:24105: conditional infixes are deprecated in unified syntax
> > ccVnNycO.s:24105: Error: thumb conditional instruction should be in IT
> > block -- `sbcccs r1,r1,r3'
> > ccVnNycO.s:24105: Error: thumb conditional instruction should be in IT
> > block -- `movcc r3,#0'
> > ccVnNycO.s:24210: conditional infixes are deprecated in unified syntax
> > ccVnNycO.s:24210: Error: thumb conditional instruction should be in IT
> > block -- `sbcccs r2,r2,r3'
> > ccVnNycO.s:24210: Error: thumb conditional instruction should be in IT
> > block -- `movcc r3,#0'
>
> For the record:
>
> The errors about sbcccs and movcc probably stem from here:
>
> /* We use 33-bit arithmetic here... */
> #define __range_ok(addr, size) ({ \
> unsigned long flag, roksum; \
> __chk_user_ptr(addr); \
> __asm__("adds %1, %2, %3; sbcccs %1, %1, %0; movcc %0, #0" \
> : "=&r" (flag), "=&r" (roksum) \
> : "r" (addr), "Ir" (size), "0"
> (current_thread_info()->addr_limit) \
> : "cc"); \
> flag; })
>
> arch/arm/include/asm/uaccess.h
>
> I stumbled across this when trying to build a 32-bit ARM kernel with
> clang.
You have to tell clang to pass -mno-warn-deprecated to gas. That's what
the gcc build currently does.
In this particular case with LTO, the same trick as done for
-mimplicit-it=always would do it to work around
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78353.
Nicolas
