On Tue, Jan 08, 2002 at 11:07:53AM -0800, Kevin Paul Herbert wrote:
> I'm working with a somewhat dated kernel (2.4.2+patches) and have
> discovered that there are problems with physical memory that does not
> map into KSEG0/KSEG1. I looked over the list archives (manually, I
> couldn't find a search interface) and it appears that this is still
> the case for current kernels (at least as of a note from last summer,
> the last time the issue seems to have come up.)
> Obviously, phys_to_virt() is going to be a problem but besides this
> I'm wondering what anybody may have done to support physical memory
> that is not always mapped into the virtual address space, so that I
> don't have to reinvent the wheel.
> When I tell the kernel about the memory above 0x20000000, userland
> fails to start; the kernel gets as far as execve()'ing init, but
> nothing happens (interrupts are enabled; I get echo on the console,
> but nothing from userland).
Correct. There are two ways to solve this problem. For the 32-bit kernel
I've got a highmem patch and the 64-bit kernel memory limits are the
hardware's memory limits. I'll post the highmem patch soon. I was
planning to have it ready by now but a flu sent me to bed instead ...