[Lucid SRU 0/5] Enable Sandybridge AVX

Tim Gardner tcanonical at tpi.com
Wed Jul 21 12:15:44 UTC 2010


On 07/20/2010 12:51 PM, Steve Conklin wrote:
> BugLink: http://bugs.launchpad.net/bugs/607653
>
> These patches are from upstream but have not been queued for stable.
>
> Intel Advanced Vector Extensions (AVX) introduce 256-bit vector processing capability. it's the next generation of the SSE instructions, and useful for many things, especially multi-media codecs. More about AVX at http://software.intel.com/sites/avx.
>
> AVX enablement has taken place as follows:
> - Kernel xsave/xrestore (done in 2.6.28)
> - Kernel patch enabling AVX (done in 2.6.30)
> - gcc( done in 4.4.2) ; binutils (done in 2.20.51.0.2 )
>
> glibc 2.11 has AVX support, so we need to enable debugging support in the kernel to support glibc.
>
> Suresh Siddha (5):
>    x86, ptrace: regset extensions to support xstate
>    ptrace: Add support for generic PTRACE_GETREGSET/PTRACE_SETREGSET
>    ptrace: Fix ptrace_regset() comments and diagnose errors specifically
>    x86, ptrace: Simplify xstateregs_get()
>    x86, ptrace: Remove set_stopped_child_used_math() in [x]fpregs_set
>
>   arch/x86/include/asm/i387.h  |   12 +++++-
>   arch/x86/include/asm/user.h  |   58 +++++++++++++++++++++++++++
>   arch/x86/include/asm/xsave.h |    2 +
>   arch/x86/kernel/i387.c       |   71 ++++++++++++++++++++++++++++++++--
>   arch/x86/kernel/ptrace.c     |   34 +++++++++++++++-
>   arch/x86/kernel/xsave.c      |    1 +
>   include/linux/elf.h          |    8 +++-
>   include/linux/ptrace.h       |   20 +++++++++
>   kernel/ptrace.c              |   88 ++++++++++++++++++++++++++++++++++++++++++
>   9 files changed, 284 insertions(+), 10 deletions(-)
>
>

I'm not really in favor of this patch set. Lucid binutils is not likely 
going to change versions from 2.20.1 to the required version of 2.20.51. 
Without userspace support for the instruction, then I can see no reason 
to support saving the extra context.

rtg
-- 
Tim Gardner tim.gardner at canonical.com




More information about the kernel-team mailing list