Now, on the Amiga, we copy a small piece of code to chip memory and enable the
MMU from here.
On the DraCo, we setup %itt0 so that we shadow physical memory without
touching the page tables and very temporarily. The DraCo's memory layout
guarantees that there is no overlap in this case between physical and
virtual addresses.
This fixes a bug where Amiga's with physical memory at 0x200000 wouldn't boot
kernels >2M in size, and possibly other corner cases.
Joint work with Ilpo Ruotsalainen and Ignatios Souvatzis.
and the back end Bluetooth device driver. The device driver now
allocates a suitable buffer on behalf of the bthci driver, and bthci
fills the buffer before despatch.
This saves an ugly temporary allocation (in bthci) and memory copy (in
the USB driver) per write, and also works better with the
sockets-based interface I'm experimenting with.
cpu's who have done so. implement pmap_deactivate() for MULTIPROCESSOR and
call it from cpu_switch() when we are about to switch proces and when we
enter idle().
with this change, i see significantly reduced tlb IPI traffic and fork/exec
bound processes -- such as "configure" -- run significantly faster, upto
15%. i also obvserved a small (0-2%) benefit to CPU bound tasks as well.
external cache enable bit -- this allows software to enable or disable the
(external) L2 cache on the R5k and Rm527x and the (external) L3 cache on
the Rm7k. If the (external) cache is disabled, treat it as if there were
no cache for the purposes of the cache setup code.
Also, update sgimips code to use the new name.
Uploaded scripts work better if they are little endian, as the
card's engine expects, so convert to le first.
Also clean up attach routine a bit (use pa_id and PCI_REVISION
instead of fetching it ourselves).
This makes the driver work on my macppc G4, it can decode and
display video and tuner input. Sound does not seem to work, but
this may be my wonky formac bktr-for-macintosh card.
unless ACPI_FDC_DEBUG
array returned from _FDE contains UINT32 values, not UINT8; also change
the magic number '14' to '5 * sizeof(UINT32)' for clarity
remove XXX for the tape presence comment; it's Just Okay to not use the info
fdc_acpi_getknownfds(): if fdc_acpi_nvtotype() returns NULL, don't
attempt to attach the drive at all
XXX not tested
which is safer than the loop there used to be here.
wi_mwrite_bap: if wi_write_bap fails, don't keep on going: this
way you avoid writing garbage to the radio. First time you see
an odd-length mbuf, copy the remainder of the chain to sc_txbuf
and from there to the MAC. This way, you do not read an mbuf past
the end of its data (occasionally you will cross a page doing
that!) and you avoid expensive, excess seeks in the radio's own
buffer chain.
wi_rx_intr: clamp the frame length told to us by the driver to the
most bytes we can fit in our mbuf cluster.
I am still getting e-mails from my testers telling me how much
better this makes things.
This fixes the problem where no grf* console would be found on startup because
they wouldn't match with any parent.
While here, memset() that structure to zero so that we either pass or crash
right away if the MI code changes again. From Leo Weppelman, as done on atari.
(backout parts of rev.1.40)
otherwise, directory structures can be corrupted because checkpoints can
occur via eg. lfs_vflush before parent directory is written.
major code cleanup esp. types used. Also cleanup up a major BUG that for
some odd reason worked :-/ makes me puzzled. It signifies that there might
be copies around in physical space of the DRAM ??? and thus its function
was motherboard dependent? It must have been old cruft from before the
cleanup of the relocation engine.
/*
* The size of the code/data to be moved is not `end - rmbase' but
* `__bss_start__ - rmbase' for the module is loaded into RISC OS
* based on the filesize where as NetBSD doesn't have to include all
* the bss space into the file itself. In some odd cases the
* relocatable module area can be smaller than the module + bss and
* thus bomb out.
*/
XXX - need to move this (as well as the equivalent sparc stuff added
recently) outa here into sbus_machdep or something. We should not need
to know details of the actual bus_space implementation here.
Shuffle some code around so that this an exist as a method next
to mpbios, both creating the same data structures. If both are
defined, try ACPI first.
* Improve acpi interrupt fixup a bit
* Source is an array, don't compare it to NULL, instead
look for an empty string to denote a link-device-less
entry.
* For root PCI busses, try to use the _BBN method to get
numbering right.
* Add acpi_md_callback() function for MD handling after the init,
but before * at acpi probing.
rather than the level13 software intr xpmsg area. now DDB IPI's don't lock
the xpmsg_lock and we avoid recursion and more.
- don't actually use cpuinfo.msg.lock yet, xpmsg_lock suffices.
- reread the pending register on mbus hypersparc cpus to avoid bugs in the
h/w that cause IPI's to be missed.
