On 5/29/07, Mulyadi Santosa <mulyadi.santosa@xxxxxxxxx> wrote:
Hello Robert Hopefully, I still can help you... > right, but what is the formula that determines how *much* of RAM is > allocated for that vmalloc area? IMHO, no formula. As long as they fit into VMALLOC adress range and there are free pages, you can allocate as many as you need. Notice that vmalloc is just a way to allocate non contigous pages and make them virtually contigous. > so far, so good. but if i move to another system that has only 512M > of RAM, i read from the same places: > > ... > MemTotal: 448928 kB > ... > VmallocTotal: 573432 kB > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ are you sure about that? hm , all are vmalloc()-ed? possible, but weird for me. > so with only 512M of RAM, clearly i can't use all of that for the > "NORMAL" zone since some of it has to be reserved for kernel tables > that normally reside in "high" memory, right? > yeah, they do need space. > i'm just trying to get straight the algorithm the kernel uses to > allocate what RAM i have to the various zones, and what that space > will eventually be used for. > > I think that really depends on how kernel developer decides the weight factor of pressures that happen toward each zones. IMSO (In My Stupid Opinion), certainly DMA zone should be less occupied, because DMA operation needs them badly. Then, instead of pressing NORMAL zone, if the pages aren't so frequently accessed or could be accessed in non atomic style, HIGHMEM zone is perfect choice. So the order is, stomp HIGHMEM first, then NORMAL, finally DMA zone.
Don't you think it should be NORMAL zone first instead of highmem, unless normal zone is under pressure though? Because accessing HIGHMEM is indeed a small hit because it involves setting temporary kernel mappings and resetting them after use. Should you use the NORMAL zone absolutely, unless you know NORMAL zone is already under heavy pressure? Thoughts?? thanks --psr I hope this is what you ask.
> sadly, the phrase "logical address" is already in common use to refer > to those offset-based addresses, as opposed to the addresses that > actually use the kernel page tables. > > OK... well then, I hope I help you... regards, Mulyadi -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
-- play the game -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
