Age | Commit message (Collapse) | Author |
|
tested by RD Thrush on Lynx Point
|
|
the scsi_cmd and intr paths. scsi_done can restart io though, which means
the mutex will be tried recursively which panics.
defer scsi_done calls after mtx_leave by putting completed ccbs on a list
on the stack that we can walk.
i hate locks.
|
|
|
|
sizes are much much smaller. you need to mask the req ring producer
with its size to get its actual entry. also stops scribbling over
memory you dont really own.
lets esxi guests boot without panicking now. maybe itll fix jsgs
workstation panic too. fusion and esxi can do some io now, but not
get I/O errors writing to write an mbr with fdisk -i.
|
|
i got it working on the eurostar while bored, and its at a point where it
will work given a bit more attention rather than just being a way to occupy
my time.
there's some cool use after frees and a lack of hotplug support to work on
ok krw@ jsg@
|
|
ok ratchov@
|
|
ok kettenis@
|
|
Pointed out by Pedro
ok krw@
|
|
branche maintained by the Ubuntu folks. Original linux commit message:
'Fix incoherence with fence updates on Sandybridge+'.
ok jsg@
|
|
|
|
ok sthen@ deraadt@
|
|
ok mpi@, jsg@
|
|
while there, fix a few %d into %u
|
|
|
|
|
|
noteably this uses the gt lock and force wake in some cases
and includes a gen5/ironlake errata.
ok kettenis@
|
|
Some free functions (radeon) grab sleeping locks. Instead when we ahve
taken the object off the tree we unlock and unref, then regrab before
looking again.
from oga in bitrig, ok kettenis@
|
|
expresscard serial card
|
|
|
|
expresscard serial card
|
|
found out the hard way by chris@ and mpi@
|
|
Potentially fixes a race/panic reported by claudio@
ok jsg@
|
|
Potentially fixes a race/panic reported by claudio@
ok jsg@
|
|
|
|
|
|
|
|
|
|
(markus.uhlin; bredband dot net), thanks!
|
|
|
|
xeon e3-1200v2 pcie ids while here as suggested by jsg@
ok jsg@
|
|
with the fastpath and cpu relocs disabled for now.
eb_* functions based on code in FreeBSD.
ok kettenis@
|
|
|
|
|
|
a bunch of AMD Family 15h Models 10h-1Fh Processor entries.
|
|
ok jsg@
|
|
|
|
|
|
|
|
Crystalwell/eDRAM Haswell parts. Original linux commit mesage:
'This magic brings stability to HSW CRW machines.'
|
|
fixes some omissions and flag errors for radeon
and removes the duplicate table for i915
|
|
register (driver shared address high) exceeded the width of the 32bit
bus address; casting the address to a 64bit type will correctly result
in a zero value on i386 and the high bits on amd64.
|
|
ok jsg@
|
|
of the linux kref functions.
Switch to using the FreeBSD refcount api for the fb refcount while here.
|
|
|
|
reduces the diff to linux
|
|
We need to allocate line buffer to each display when
setting up the watermarks. Failure to do so can lead
to a blank screen. This fixes blank screen problems
on dce6 asics.
Fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=64850
from a proposed Linux patch by Alex Deucher of AMD.
Fixes problems with multiple displays on pitcairn
reported by Alexis de BRUYN.
|
|
We need to allocate line buffer to each display when
setting up the watermarks. Failure to do so can lead
to a blank screen. This fixes blank screen problems
on dce4.1/5 asics.
from a proposed Linux patch by Alex Deucher of AMD.
|
|
There is a family of Supermicro boards where the apic pin is
incorrectly mapped on acpi, it tells us the pin for "Intel boot
interrupts". Since this is a fairly new chip, lets use MSI as no one
else is probably using it via apic, this fixes the routing issues.
Machines/Motherboards seen so far with incorrect routing:
Supermicro X9DR3-F
Supermicro X9DRH + Symbios Logic MegaRAID SAS2208
Fujitsu primergy RX300 S7 + Symbios Logic MegaRAID SAS2208
ok dlg@
|
|
but it reduces the diff against the FreeBSD version.
Tested on:
My Wifi Link 130.
jca@ Wifi Link 5300.
Jan Stary Wifi Link 4965 (thanks).
Adapted from FreeBSD r220894.
ok mpi@.
|
|
format for Rx and Tx. Replace the bit fields in the descriptor
structs with 32bit words to access them with traditional bit
operations using shifts and masks. We try to avoid bit fields in
OpenBSD. For consistence with other drivers, this change also uses
letoh32/htole32 endianess conversions even if it is very unlikely that
vmx will ever run on a big-endian VM/host.
discussed with uebayasi@ and deraadt@
|