If a segment starts exactly at 16M don't try to split it because that causes
the code to try to call uvm_page_physload with a zero length segment.
(kern/17199)
- Set up a Panic Stack,
- Don't pre-map the sysfpga at this time.
- Calculate the CPU speed only if SH5_CPU_SPEED isn't defined.
- On a related note, make the CPU speed probing code more accurate.
- Print the CPU speed at startup.
- Force RB_SINGLE for now, at least until I get a bootloader written.
- in _EXCEPTION_EXIT, copy the current ASID to the pre-exception
context before we switch.
- fix the pteg hash generation code and EPN masking in the tlb
miss handler.
Sprinkle some DIAGNOSTIC checks.
to ensure the callee-saved set will be restored when we switch to it.
(It doesn't actually matter to the new process; it just inherits some
crud in those registers from the kernel if we don't set the bit).
Also ensure the strings pointer in r7 is sign-extended.
Seems that we assume that the dram blocks are sorted, and that the first/lowest address is also where the kernel is.
If the above is not true, then we're on a kinetic (probably should make a better way to indicate this) So search for all dram blocks < with starting addr lower than the first block and remove them.
Currently there's minimal performance gain (which is odd as the SDRAM is meant to be faster, I'm wondering if we need to prod some hidden registers to set timing information.
Note that I still get 16MB/s compared with 7MB/s on RiscStation and 93MB/s on my cats. I'm thinking that something else is seriously nasty on acorn32.
any time we remove all access to a given virtual page,
we must invalidate the (write-through) L1 dcache.
pmap_remove() still had it, but pmap_kremove(), pmap_clear_reference()
and pmap_page_protect(VM_PROT_NONE) didn't and needed it.
fixes PR 18040.
section of another exception. This is likely to happen if the kernel
stack is misaligned, has dropped off the bottom of the PCB, or has
otherwise gone into orbit.
In this case, switch to a safe stack, save as much of the machine
state as possible and dump it to the console.
accessing the kernel stack, since a TLB miss on the kernel stack
will result in r24 being trashed.
Also clear the ES_CRITICAL flag just before returning to the
previous context.
/dev/ttyS0 crashed the kernel. This is because sacom_filltx uses some
uninitialized static variables. Pulling the salues from softc instead
fixes the problem (this is what was done before the drver was moved
from /sys/arch/hpcarm to /sys/arch/arm, anyway).
as intrframe and trapframe are concerned.
According to the ABI, only the low 32-bits of these registers are
guaranteed to be preserved by the callee. Therefore, we need to
preserve all 64-bits of them in the interrupt trampoline.
ToDo:
- Symbol support (can't test as yet, due to lack of symbols),
- Take notice of adjacent "movi/shori" instructions in order to display
the resulting 32/64-bit value, with symbol lookup if possible.
At the very least, this will dump the machine state. At best,
we get into ddb().
This provides a useful way to regain control using an NMI button
if the cpu decides to spin at a high ipl.
