Messages in this thread | | | Date | Mon, 23 Nov 2015 09:59:07 -0500 (EST) | From | Nicolas Pitre <> | Subject | Re: [GIT PULL] optimize 64-by-32 ddivision for constant divisors on 32-bit machines |
| |
On Mon, 23 Nov 2015, Arnd Bergmann wrote:
> On Sunday 22 November 2015 17:28:33 Nicolas Pitre wrote: > > On Sun, 22 Nov 2015, Arnd Bergmann wrote: > > > On Monday 16 November 2015 20:20:38 you wrote: > > > I'm now getting a build regressing with the attached randconfig configuration, > > > when compiling drivers/net/wireless/iwlegacy/common.o: > > > > > > drivers/built-in.o: In function `il_send_rxon_timing': > > > :(.text+0xbbac80): undefined reference to `__aeabi_uldivmod' > > > :(.text+0xbbac9c): undefined reference to `__aeabi_uldivmod' > > > :(.text+0xbbacdc): undefined reference to `__aeabi_uldivmod' > > > :(.text+0xbbadc8): undefined reference to `__aeabi_uldivmod' > > > :(.text+0xbbadf8): undefined reference to `__aeabi_uldivmod' > > > :(.text+0xbbae3c): more undefined references to `__aeabi_uldivmod' follow > > > drivers/built-in.o: In function `il_send_rxon_timing': > > > :(.text+0xbbb11c): undefined reference to `____ilog2_NaN' > > > > This looks like some gcc bug from a few years ago. > > Yes, I clearly remember debugging something in this area before, > but I don't remember what we did about it. > > We already have the workaround for OABI (which you kept), and I have > another workaround to disable the optimized function for ARMv3 as > well, as newer gcc versions also get that wrong here (internal compiler > error IIRC).
Although we no longer support ARMv3 targets, the only reason -march=armv3 is used is to prevent gcc from using strh/ldrh instructions on RiscPC whose memory bus can't accommodate it. Still, this would be worth reporting a bug for it.
> We could add yet another such workaround for CONFIG_PROFILE_ALL_BRANCHES, > but I don't have a good feeling about doing this unless we understand > well enough why it breaks.
I'm able to reproduce it, and I am in the process of reintroducing the last patch piece by piece to see what breaks. The presence of ____ilog2_NaN is particularly intriguing as this would mean 0 is passed to ilog2() somewhere.
Nicolas
| |