On Sun, Apr 5, 2015 at 9:33 PM, Renato Golin <[email protected]> wrote: > On 5 April 2015 at 19:18, Venkataramanan Kumar > <[email protected]> wrote: >> (1) Can I post the patch to LLVM by increasing >> SANITIZER_MMAP_RANGE_SIZE to 1<<42? > > Venkat, > > Let's wait and make sure we can cleanly apply Jakub's patch [1] to > LLVM without any license issues. > > >> For 39 bit VA, TwoLevelByteMap uses more dimensions now. >> If someone wants to support 48 bit VA later, we may need to >> increase it by 1<<48. > > I don't like increasing the memory footprint on the most common > AArch64 platforms just because there will be 48-bit VA in the future.
It's 8KB for 42-bit VA, and 512KB for 48-bit. And that's .bss space that is never written to, so no extra physical memory use at all. I'm not sure we can make bytemap allocation dynamic without sacrificing performance. > For now, the reference platform is still Juno, which is 39-bit. It > also works in the APMs, which have just joined with Gigabyte for > server grade ATX boards. If we want to make it work in the AMD (which > is the only 42-bit so far, and it's not even public), we'll need a way > to pick the mmap range at run time. I remember Kostya saying there was > a way, we should at least do that. For 48-bit we must absolutely do > that, and only when someone has real hardware to test on. > > But that's all after we're sure that the patch is safe for LLVM. > > cheers, > --renato > > [1] > http://pkgs.fedoraproject.org/cgit/gcc.git/tree/gcc5-libsanitize-aarch64-va42.patch -- You received this message because you are subscribed to the Google Groups "address-sanitizer" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
