[lkml]   [2005]   [Jul]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectCONFIG_ALPHA_GENERIC problem with gcc-4.1
    I've been doing builds of linux-2.6.11 as a sanity check
    for new versions of gcc, and a problem just popped up
    in arch/alpha/Makefile (see
    I think I can work around this myself by using CONFIG_ALPHA_EV6
    instead of CONFIG_ALPHA_GENERIC, but here's my analysis
    of the problem; maybe the alpha kernel maintainer can take it from here?

    Take the following with a grain of salt; I don't know much
    about alpha or gcc, I'm just doing a little QA.

    arch/alpha/Makefile says:
    36 # If GENERIC, make sure to turn off any instruction set extensions that
    37 # the host compiler might have on by default. Given that EV4 and EV5
    38 # have the same instruction set, prefer EV5 because an EV5 schedule is
    39 # more likely to keep an EV4 processor busy than vice-versa.
    40 ifeq ($(CONFIG_ALPHA_GENERIC),y)
    41 mcpu := ev5
    42 mcpu_done := y
    43 endif
    84 # For TSUNAMI, we must have the assembler not emulate our instructions.
    85 # The same is true for IRONGATE, POLARIS, PYXIS.
    86 # BWX is most important, but we don't really want any emulation ever.
    87 CFLAGS += $(cflags-y) -Wa,-mev6

    Thus when you pick CONFIG_ALPHA_GENERIC, gcc is invoked with
    the contradictory options -mcpu=ev5 -Wa,-mev6

    This probably means that even on ev5, some ev6 instructions are used.
    In particular, see include/asm-alpha/compiler.h:

    #if defined(__alpha_bwx__)
    #define __kernel_ldbu(mem) (mem)
    #define __kernel_ldwu(mem) (mem)
    #define __kernel_stb(val,mem) ((mem) = (val))
    #define __kernel_stw(val,mem) ((mem) = (val))
    #define __kernel_ldbu(mem) \
    ({ unsigned char __kir; \
    __asm__("ldbu %0,%1" : "=r"(__kir) : "m"(mem)); \
    __kir; })

    That inline assembly is fine on ev5, but only if the assembler
    is emulating the ldbu instruction with a macro -- exactly
    the kind of thing arch/alpha/Makefile is trying to
    inhibit when it says -Wa,-mev6.

    This is an issue now because building the kernel with CONFIG_ALPHA_GENERIC
    fails on the current gcc-4.1 snapshot with

    > {standard input}:496: Error: macro requires $at register while noat in effect
    > make[1]: *** [arch/alpha/kernel/core_cia.o] Error 1

    and it looks like a kernel problem, not a gcc problem:
    don't try to use ev6 instructions on ev5 or earlier processors.

    That probably means conditionalizing that -Wa,ev6 properly,
    but if that's hard, maybe it means dropping support for ev4 and ev5 processors,
    and mapping CONFIG_ALPHA_GENERIC to ev6. I wouldn't know...
    - Dan

    Trying to get a job as a c++ developer? See
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to
    More majordomo info at
    Please read the FAQ at

     \ /
      Last update: 2005-07-11 16:28    [W:0.021 / U:21.496 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site