We are hedging in serial and in parallel, and more conservative than
the Linux driver from Cavium seems to be, so although I don't know
exactly what the thermal jitter of the device is, this seems like a
reasonable compromise.
As found on the rk3288 and rk3399. This driver only supports the
TRNG, not the rest of the crypto engine, although it uses the AES unit
to do a self-test at attach time to verify that the engine works.
There seem to be two versions of the Rockchip crypto engine, v1 and
v2; this one is for v1. Can't name a driver `rkcryptov1' so we'll
clumsily call it `rkv1crypto' instead to leave room for `rkv2crypto'
later on.
The crypto binding derived from the Rockchip BSP Linux kernel, in the
location it appears on the rk3399, is in rk3399-crypto.dtsi, since
there doesn't seem to be a better place to put it at the moment among
this twisty maze of inclusions, all different.
- Make PGO_LOCKED getpages imply PGO_NOBUSY and remove the latter. Mark
pages busy only when there's actually I/O to do.
- When doing COW on a uvm_object, don't mess with neighbouring pages. In
all likelyhood they're already entered.
- Don't mess with neighbouring VAs that have existing mappings as replacing
those mappings with same can be quite costly.
- Don't enqueue pages for neighbour faults unless not enqueued already, and
don't activate centre pages unless uvmpdpol says its useful.
Also:
- Make PGO_LOCKED getpages on UAOs work more like vnodes: do gang lookup in
the radix tree, and don't allocate new pages.
- Fix many assertion failures around faults/loans with tmpfs.
the allocator's buckets, instead of doing round robin distribution. There
are open questions here but this is better than doing nothing.
- Kernel reserve pages are for the kernel not realtime threads.
Dynamically allocate a page of memory with uvm_km_alloc(kernel_map, ...)
for Hyper-V hypercall. However, this method can no longer be used to
make an executable page.
So we prevent it by using statically allocated memory for text segment.
an ACPI_INTEGER in host byte order but a byte sized buffer with little
endian data.
Extract only the low 8 bits from buffer to get the fan speed again.
The previous version of this test relied on the way how the shell
interprets a lonely backslash at the end of the line. The NetBSD and
FreeBSD shells print the backslash, while Bash doesn't.
While here, make the escaping a bit simpler and align the test
descriptions with the actual test data.
Previously we would read 34 bits of entropy into the pool and stop,
never to try again. This change finishes the conversion to an
on-demand source, although we still use a callout to (a) try again in
another second if the 7811 FIFO was emptied, and (b) wait at least
0.4 sec after we start the RNG to read anything from it as the
alleged documentation allegedly recommends.
Highlights:
- Apply a ten-year-old fix to the driver's misuse of previously
exposed ciphertext as IV for the next ciphertext with CBC.
(Someone^TM should make sure there are no other drivers still doing
this...)
- Various zeroing of buffers.
- Various minor cosmetic changes.
sevan