Re: [PATCH 1/3] tools include: Add uapi mman.h for each architecture

[Wangnan (F)]

On 2016/9/14 18:46, Naveen N. Rao wrote:
> On 2016/09/14 06:23PM, Wang Nan wrote:
> > On 2016/9/14 18:00, Naveen N. Rao wrote:
> > > On 2016/09/14 05:36PM, Wang Nan wrote:
> > > > On 2016/9/14 17:28, Naveen N. Rao wrote:
> > > > > On 2016/09/12 06:15PM, Arnaldo Carvalho de Melo wrote:
> > > > > > Em Mon, Sep 12, 2016 at 04:07:42PM -0300, Arnaldo Carvalho de Melo escreveu:
> > > > > > > Em Mon, Sep 12, 2016 at 12:54:29PM +0000, Wang Nan escreveu:
> > > > > > > > Some mmap related macros have different values for different
> > > > > > > > architectures. This patch introduces uapi mman.h for each
> > > > > > > > architectures.
> > > > > > > >
> > > > > > > > Three headers are cloned from kernel include to tools/include:
> > > > > > > >
> > > > > > > >     tools/include/uapi/asm-generic/mman-common.h
> > > > > > > >     tools/include/uapi/asm-generic/mman.h
> > > > > > > >     tools/include/uapi/linux/mman.h
> > > > > > > Cool, the above was done as copies, why not the rest? IIRC you mentioned
> > > > > > > some reasoning behind that decision, but we need it spelled out here.
> > > > > > >
> > > > > > > For instance, I went on and look at arch/xtensa/include/uapi/asm/mman.h,
> > > > > > > and couldn't find why we shouldn't copy it just like the three files
> > > > > > > above.
> > > > > > >
> > > > > > > I'm looking now at why the build breaks in so many systems, first hunch
> > > > > > > is that bits/ part (the ones without the failure details probably have
> > > > > > > the same errors), alpine uses musl libc, but some that broke are glibc
> > > > > > > based.
> > > > > > So, please take a look at my perf/core branch, I applied 1/3 and 3/3,
> > > > > > but took a different path for 2/3, now it builds for all systems I have
> > > > > > containers for:
> > > > > This still fails for me on ppc64. Perhaps we should guard
> > > > > P_MMAP_FLAG(32BIT) and potentially others with a #ifdef, which was
> > > > > earlier reverted by commit 256763b0 ("perf trace beauty mmap: Add more
> > > > > conditional defines")?
> > > > Perhaps we should set all non-exist flag to 0 in each uapi mman.h?
> > > Will that work for MADV_* since the macro there is for a case statement?
> > Then fall back to include/uapi/asm-generic/mman-common.h. And I
> mman-common.h is already included on powerpc:
>
> # cat tools/arch/powerpc/include/uapi/asm/mman.h
> #ifndef TOOLS_ARCH_POWERPC_UAPI_ASM_MMAN_FIX_H
> #define TOOLS_ARCH_POWERPC_UAPI_ASM_MMAN_FIX_H
> #define MAP_DENYWRITE	0x0800
> #define MAP_EXECUTABLE	0x1000
> #define MAP_GROWSDOWN	0x0100
> #define MAP_HUGETLB	0x40000
> #define MAP_LOCKED	0x80
> #define MAP_NONBLOCK	0x10000
> #define MAP_NORESERVE   0x40
> #define MAP_POPULATE	0x8000
> #define MAP_STACK	0x20000
> #include <uapi/asm-generic/mman-common.h>
> #endif

So for powerpc is free from MADV_* problem.

Alpha, MIPS, parisc and xtensa misses some MADV_* macros,
we should define them in their uapi/mman.h.

Thank you.

> > realized the missing of MADV_FEEE in tools/perf/trace/beauty/mmap.c.
> > Is that intentionally?
> Not sure, Arnaldo should know.
>
> Thanks,
> Naveen