Has anyone noticed that vmalloc()/ioremap() seem to be
overly complicated procedures? Basically, vmalloc_area_pages()/
remap_area_pages() might be getting called on an until now unused
kernel virtual address, for which a kernel page table needs to be
allocated. set_pgdir() takes care of this by:
1. updating the page directory of each running task.
2. updating the cached free page directories on each cpu.
It seems to me that we could do away with the complexity of set_pgdir
if we were to allocate a few kernel page tables at kernel startup
time, have the kernel page directory entries for swapper_pg_dir
mapping the vmalloc area point to these kernel page tables and have
those be copied to every new page directory in the system. Of course,
the whole vmalloc area (VMALLOC_START .. VMALLOC_END) would take too
many kernel page tables probably, so we would need to resize the
vmalloc area to something more easily handled, say 32M on a 64M
system (or any other heuristic), which requires 8 kernel page table
pages.
Has this been discussed before, or are there any problems with this
approach I am not thinking about? This approach could be applied for
other architectures too, since a lot of asm*/pgtable.h's seem to
have the same implementation of set_pgdir. If this is an acceptable
idea, I could code something up ...
Kanoj
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/