Mark Johnson writes: > > > Andrew Gallatin wrote: > > Mark Johnson writes: > > > > > > > > > Andrew Gallatin wrote: > > > > Masa Murayama writes: > > > > > Does anyone know whether large pages are available for solaris > > x86? > > > > > > > > > > masa > > > > > > > > On amd64, pagesize -a shows 4KB and 2MB pages are available. > > > > Hey, 2MB pages might solve some problems for me! :) > > > > > > > > How do I tell the kernel to use 2MB pages for things like kmem_alloc > > > > and ddi_umem_alloc? Or will I run into the same apparent > > > > fragmentation problem that I have with allocating DMAable memory? > > > > > > There isn't a way to allocate large pages from the > > > kernel today, only from user applications. > > > > Large pages are physically contiguous, though, right? > > Yes, they have to be :-) > > > > I guess I'll > > need to see if I have more luck allocating them than allocating contig > > memory via ddi_dma_mem_alloc() on x86. If so, I suppose I could > > allocate the memory in userspace, and then pin it via an ioctl. > > a 512MB system is going to be a lot harder since memory > will fragment pretty fast.
This is sort of a small scale test case for what we expect the customer to actually use (many gigs of ram, many gigs - epsilon used by these buffers). How can I help the kernel defrag it? I've tried icky stuff like kmem_alloc()'ing as much memory as I can get my hands on, sorting it by physical address and freeing it. No dice. Drew _______________________________________________ driver-discuss mailing list driver-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/driver-discuss
