On 03/01/2016 09:31 PM, Izumi, Taku wrote: >> > > I'd like to atten LSF/MM 2016 and I'd like to discuss "Address range mirroring" topic. >> > > The current status of Address range mirroring in Linux is: >> > > - bootmem will be allocated from mirroring range >> > > - kernel memorry will be allocated from mirroring range >> > > by specifying kernelcore=mirror >> > > >> > > I think we can enhance Adderss range mirroring more. >> > > For excample, >> > > - The handling of mirrored memory exhaustion case >> > > - The option any user memory can be allocated from mirrored memory >> > > and so on. It sounds like there was some good discussions of this at LSF/MM: https://lwn.net/Articles/684866/ One thing I wanted to add: There's an Intel platform (Knights Landing aka Xeon Phi) that has some on-package memory. It's higher-bandwidth than normal DRAM, but it shows up as a really slow, remote NUMA node. Instead of needing to come up with new syscalls for allocating from this new memory, applications just use the plain old NUMA APIs. While this doesn't help any of the other issues for mirroring (the fallback and exhaustion problems), is there a reason we shouldn't use the NUMA APIs for this? -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>