2015-07-02 21:43 GMT+09:00 Albert ARIBAUD albert.u.boot@aribaud.net:
Hello Masahiro,
On Thu, 2 Jul 2015 21:21:12 +0900, Masahiro Yamada yamada.masahiro@socionext.com wrote:
Hi Albert,
2015-07-02 6:21 GMT+09:00 Albert ARIBAUD albert.u.boot@aribaud.net:
On Thu, 16 Apr 2015 11:21:44 +0200, Albert ARIBAUD albert.u.boot@aribaud.net wrote:
Hello Masahiro,
Your patch clashes with Pavel's already committed break-if-private-libgcc-and-thumb, causing many boards to fail building.
I am putting your patch in 'under review' state until I can have a look at what happens with private libgcc and thumb.
Hmm, even with the libgcc+thumb thing fixed, this patch still causes quite a few target to fail:
- without the patch, buildman yields 503 1 12 / 516.
- With the patch, buildman yields 488 1 27 / 516.
Of these, 10 are aarch64 boards which now try to build a private libgcc with some hardwired 32-bit mnemonics :
| arch/arm/lib/_ashldi3.S:21: Error: operand 1 should be an integer register -- `subs r3,r2,#32' | arch/arm/lib/_ashldi3.S:22: Error: unknown mnemonic `rsb' -- `rsb ip,r2,#32' | arch/arm/lib/_ashldi3.S:23: Error: unknown mnemonic `movmi' -- `movmi r1,r1,lsl r2' | arch/arm/lib/_ashldi3.S:24: Error: unknown mnemonic `movpl' -- `movpl r1,r0,lsl r3' | arch/arm/lib/_ashldi3.S:25: Error: unknown mnemonic `orrmi' -- `orrmi r1,r1,r0,lsr ip' | arch/arm/lib/_ashldi3.S:26: Error: operand 1 should be an integer register -- `mov r0,r0,lsl r2' | arch/arm/lib/_ashldi3.S:27: Error: operand 1 should be an integer register -- `mov pc,lr'
This should be fairly easy to fix by defaulting CONFIG_USE_PRIVATE_LIBGCC to
The other 5 failures are snow, smdk5250, peach-pi, smdk5420, and peach-pit, all of which show the same error:
arch/arm/lib/lib.a(div0.o): In function `__div0': arch/arm/lib/div0.c:13: undefined reference to `hang' make[2]: *** [spl/u-boot-spl] Error 1 make[1]: *** [spl/u-boot-spl] Error 2
That one is be a dependency on hang(). This function... hangs out... in lib/hang.c, which according to lib/Makefile should be included in any build, SPL or otherwise; but it appears lib/ is not built at all for SPL, at least for these five targets.
Masahiro, can you have a look at e.g. snow (arm) and ls2085ardb (aarch64) and see if you can update your patch to make sure these two targets build? The other 13 should follow then.
OK. I will take a look.
Ok -- don't rush, though. Seeing as there are contrary opinions on whether we should use a private libgcc or not, I won't decide on committing the patch right away.
OK.
If you think we should take our time for this discussion, will you send a pull-request with currently applied change sets? The 2015.07 will be released in 10 days.