Hi Finn,
On Mon, 11 Aug 2014, Sam Creasey wrote:
On Fri, Aug 08, 2014 at 08:46:08PM +1200, Michael Schmitz wrote:
IIRC, the DMA controller for the sun3's NCR5380 implementation is
extremely fussy about what happens in which phase, and it's quick to
anger if you don't handle everything exactly how it expects.
Does the DMA controller also sit in between the bus and the NCR chip,
as on Falcon? Otherwise, I can't see how it would matter if we just
bypass it and use PIO instead.
The PIO code from the Atari driver worked just fine for the Mac 5380
(in fact, the first Mac driver used PIO for everything, including data
in/out).
PIO definitely works on the Sun3 implementation under the right
circumstances, I ran it in PIO mode only for a while (much like the Mac
driver).
I can't remember the bus configuration offhand.
It might not be a big problem: sun3_scsi does a lot of PIO a lot as it is.
For NCR5380_reselect(), atari_NCR5380.c apparently uses only PIO whereas
sun3_NCR5380.c will use PIO up to a 128 byte size limit (beyond which,
only DMA is used).
Transfer sizes for message in during reselection are only a few bytes.
According to atari_dma_xfer_len(), DMA reads must be multiple of 512
bytes (writes can be rounded up, reads cannot). PIO is the only option
for Atari, I'm afraid.
For NCR5380_information_transfer() and PHASE_DATAIN, atari will use PIO
for transfersize <= 31 bytes whereas sun3 will use PIO for <= 128 bytes.
IIRC the ST-DMA can't be programmed for shorter transfers than 512 bytes.
The condition
(transfersize = NCR5380_dma_xfer_len(instance,cmd,phase)) > 31)
enforces that - transfersize will be 0 on reads that are not multiples
of 512 bytes (and even some where the advertised transfer size does meet
that condition, but the command is not guranteed to be a block mode
command (tape reads).
However, atari never uses DMA here if cmd->device->borken.
Different semantics of borken, I presume. On Atari, this means a DMA
transfer has failed earlier.
When cmd->device->borken, I assume sun3_NCR5380 inhibits PIO because PIO
was itself expected to be problematic?
OTOH, if PIO works up to 128 bytes in all cases, why not larger transfers?
Performance, perhaps. PIO would work, but hog the CPU. In fact, PIO for
everything _does_ work, on Falcon as well as Mac (I tried the PIO only
mode on the Falcon before trying on Mac). Did I eve mention how reallly
incredibly painfully slow that is?
Does the sun3_scsi driver still work after removing #define REAL_DMA?
For NCR5380_information_transfer() and PHASE_CMDOUT, the sun3 version
applies the same size limit but only does DMA setup if cmd type is
REQ_TYPE_FS (i.e. filesystem request). This command is then sent by PIO,
so the DMA setup here seems to assume that the next phase will always be
PHASE_DATAIN...
Might be a similar limitation to certain transfer sizes here. Block mode
vs. char mode.
This would seem to imply that PIO always works when not REQ_TYPE_FS, such
that no size limit is applicable...
BTW, testing sun3_dma_setup_done against cmd pointers looks very
unreliable to me: unique cmds only have unique pointers until freed by the
scsi mid-layer. After that, the same pointer is likely be re-used.
I definitely remember that the DMA setup logic from the atari version
did not work at all on sun3.
No surprise there - DMA is implementation specific and needs to be
handled on a per-case basis.
Perhaps atari_NCR5380 could use some of the sun3_NCR5380 code. E.g. atari
avoids DMA entirely for reselect, and (only in some phases) when
cmd->device->borken.
I don't think we can use DMA for the reselect message in transfers. Data
phase transfer is handled by the coroutine after reselection anyway -
that one will use DMA.
Cheers,
Michael
--
To unsubscribe from this list: send the line "unsubscribe linux-m68k" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
