sili: Add MSI support; not enabled by default

MSI does not seem to work at all on 3132, sigh

sili(4): Maximum I/O size is 256KB

kernel - Run AHCI and SILI disk drivers MPSAFE

* These drivers now pass a port-disk-port lock in the cam sim registration,
which should result in CAM callbacks being MPSAFE.

* Add a separate sign

kernel - Run AHCI and SILI disk drivers MPSAFE

* These drivers now pass a port-disk-port lock in the cam sim registration,
which should result in CAM callbacks being MPSAFE.

* Add a separate signalling interlock for the port threads.

* The devices were otherwise already MPSAFE, with per-port locking.

show more ...

SILI - Fix assertion panic during error handling.

* The error handling code could not distinguish between the use of the
error ccb as part of a softreset sequence and the use of the error
ccb as

SILI - Fix assertion panic during error handling.

* The error handling code could not distinguish between the use of the
error ccb as part of a softreset sequence and the use of the error
ccb as part of a error recovery sequence and asserted in the former
case.

* Ensure that no commands are active on the port prior to initiating
a hard reset through the port multiplier.

show more ...

SILI - Cleanup, remove broken KKASSERT and remove DIAGNOSTICs

Reported-by: Sascha Wildner <swildner@crater.dragonflybsd.org>

SILI - Minor fixes and adjustments.

* Aligned scratch space for READ LOG

* Better exclusivity code. Minor fix to only copy-back the rfis
if the CCB was on-chip, instead of unconditionally.

* Pr

SILI - Minor fixes and adjustments.

* Aligned scratch space for READ LOG

* Better exclusivity code. Minor fix to only copy-back the rfis
if the CCB was on-chip, instead of unconditionally.

* Probe all ports reported by the PM, we'll have to adjust for
'fake' disks at the end of the PM port chain more specifically.

show more ...

SILI - Reorganize the probe code, fix CAM startup issues, remove debugging

* Do a major reorganization of the probe code. It is MUCH faster now and
should reliably detect PMs and direct-attached

SILI - Reorganize the probe code, fix CAM startup issues, remove debugging

* Do a major reorganization of the probe code. It is MUCH faster now and
should reliably detect PMs and direct-attached disks.

* Fix an issue in the CAM code.

* Remove a lot of the debugging kprintfs.

show more ...

SILI - Get read log page 10 working, improve error handling.

* Finish up implementing the NCQ log page handling code.

* The recursion check for sili_port_reinit() was insufficient, fix it.

SILI - Work around hardware LRAM bugs part 2/2.

* Add ATA_F_AUTOSENSE and ATA_F_EXCLUSIVE to deal with accessing the
D2H RFIS.

ATA_F_AUTOSENSE tells sili.c that the caller wants to access the r

SILI - Work around hardware LRAM bugs part 2/2.

* Add ATA_F_AUTOSENSE and ATA_F_EXCLUSIVE to deal with accessing the
D2H RFIS.

ATA_F_AUTOSENSE tells sili.c that the caller wants to access the rfis.
The sili core is then responsible for copying the rfis out of the LRAM
and into host-memory.

ATA_F_EXCLUSIVE tells sili.c that the caller wants to be the only
command queued to the chip for that particular command.

* Currently ATA_F_AUTOSENSE implies ATA_F_EXCLUSIVE, but ultimately we will
want to do the auto-sense as a separate step so things like ATAPI farms
can run commands in parallel.

show more ...

SILI - Work around hardware LRAM bugs part 1/2.

After tearing my hair out a bit and wondering why the Sili chip couldn't
have more then one command pending on a port at a time when talking to a PM,

SILI - Work around hardware LRAM bugs part 1/2.

After tearing my hair out a bit and wondering why the Sili chip couldn't
have more then one command pending on a port at a time when talking to a PM,
I found an issue reported in the Linux sili device driver related to its
LRAM. Basically, reading from the Sili chip's LRAM while commands are active
will corrupt it.

The solution is to use a completely external PRB rather then the PRB in
the LRAM, and use the command activation register to start the command
instead of the command FIFO (which only works for the fixed PRBs in the LRAM).

This commit fixes basic command dispatch but does not yet handle situations
where the RFIS must be read from the LRAM... for example, when reading data
from a PM SCA register or reading sense data from an ATAPI command. With
this commit I can dd from 5 drives behind a PM in parallel.

show more ...

SILI - Silicon Image 3132 driver, initial commit

This driver was derived from the AHCI driver and retooled extensively
for the Silicon Image chipset. It retains all the cool features from the
AHCI

SILI - Silicon Image 3132 driver, initial commit

This driver was derived from the AHCI driver and retooled extensively
for the Silicon Image chipset. It retains all the cool features from the
AHCI driver including port multiplier (PM) support, NCQ, and hot-plug for
both direct-attached and PM-attached targets.

This initial commit is very alpha and the driver has significant PM-related
bugs. It is not yet ready for prime time.

show more ...