On 02.12.10 at 11:54, Jan Engelhardt <jengelh@medozas.de> wrote:
On Thursday 2010-12-02 09:49, Jan Beulich wrote:
How come? How does code/data mapped in the kernel address space incerease cache/TLB pressure if it is not being accessed?
Unless the used and unused regions of address space are each contiguous (which they certainly aren't), at least the TLBs need to map unused code along with all the used regions. The total size of mapped space is what matters for determining TLB pressure. For the caches, code would likely be separated well enough to mostly avoid bringing in unused bits (apart from the effects of false speculation), but for data the picture may already be different again when considering many small items coming from various components ending up in the same cache line.
Isn't kernel code, like stuff allocated through kmalloc, contiguous in physical memory too?
Sure, but what would this have to do with TLB usage? Jan -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org