On Thu, 21 Oct 2004 11:17:54 -0400 (EDT), sriram krishnamoorthy <krishnsr@xxxxxxxxxxxxxxxxxx> wrote: > Thanks for the realistic appraisal. I am interested in setting the page > size such that the available TLB entries map the entire physical memory. > The idea is to eliminate TLB misses completely. I don't think you understand how TLBs work. I have to go out now as one of the local sandwhich shops is about to close :-) however, briefly, it would seem that you're confused between the different views of virtual memory seen by the kernel and userspace processes. If you really did understand this then you're not making your intent clear in your post. The kernel has TLB pinnings (or soft mappings which come and go according to time of day, phase of moon, orientation of SCO interbalistic missile silos...) which mean that it can always see all of physical memory - if it's not pinning right now (I forget how 440 is setup by default) then go do that. The userspace processes have their own virtual memory on a per-process basis and corresponding TLB mappings must be setup for each process as it accesses a particular virtual address - just because process A sees something at virtual address X does not mean that it's the same for process B. Jon. -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
