On Fri, Oct 28, 2011 at 09:26:23PM +0800, Hillf Danton wrote: When we were chasing the TLB crash recently for a while I was suspecting the default get_user_fast implementation so I ported gup.c from x86 but I never really finished it, so thanks for ridding me of one item from my to do list. This probably wants some more testing in particular on 32-bit systems with 64-bit pagetables such as some Alchemy configurations or 32-bit kernels on Sibyte systems. I fixed up a reject in arch/mips/mm/Makefile due to whitespace differences and queued the patch for 3.3. Have you made any benchmarks of the new gup.c? Thanks, Ralf
