linux-toradex.git/drivers/scsi/mac_scsi.c, branch v4.9.31

mac_scsi: Fix pseudo DMA implementation

2016-04-11T20:57:09+00:00

Fix various issues: Comments about bus errors are incorrect. The
PDMA asm must return the size of the memory access that faulted so the
transfer count can be adjusted accordingly. A phase change may cause a
bus error but should not be treated as failure. A bus error does not
always imply a phase change and generally the transfer may continue.
Scatter/gather doesn't seem to work with PDMA due to overruns. This is
a pity because peak throughput seems to double with SG_ALL.
Tested on a Mac LC III and a PowerBook 520.

Signed-off-by: Finn Thain 
Reviewed-by: Hannes Reinecke 
Signed-off-by: Martin K. Petersen

ncr5380: Remove disused atari_NCR5380.c core driver

2016-04-11T20:57:09+00:00

Now that atari_scsi and sun3_scsi have been converted to use the NCR5380.c
core driver, remove atari_NCR5380.c. Also remove the last vestiges of its
Tagged Command Queueing implementation from the wrapper drivers.

The TCQ support in atari_NCR5380.c is abandoned by this patch. It is not
merged into the remaining core driver because,

1) atari_scsi defines SUPPORT_TAGS but leaves FLAG_TAGGED_QUEUING disabled
by default, which indicates that it is mostly undesirable.

2) I'm told that it doesn't work correctly when enabled.

3) The algorithm does not make use of block layer tags which it will have
to do because scmd->tag is deprecated.

4) sun3_scsi doesn't define SUPPORT_TAGS at all, yet the the SUPPORT_TAGS
macro interacts with the CONFIG_SUN3 macro in 'interesting' ways.

5) Compile-time configuration with macros like SUPPORT_TAGS caused the
configuration space to explode, leading to untestable and unmaintainable
code that is too hard to reason about.

The merge_contiguous_buffers() code is also abandoned. This was unused
by sun3_scsi. Only atari_scsi used it and then only on TT, because only TT
supports scatter/gather. I suspect that the TT would work fine with
ENABLE_CLUSTERING instead. If someone can benchmark the difference then
perhaps the merge_contiguous_buffers() code can be be justified. Until
then we are better off without the extra complexity.

Signed-off-by: Finn Thain 
Reviewed-by: Hannes Reinecke 
Tested-by: Michael Schmitz 
Tested-by: Ondrej Zary 
Signed-off-by: Martin K. Petersen

ncr5380: Merge DMA implementation from atari_NCR5380 core driver

2016-04-11T20:57:09+00:00

Adopt the DMA implementation from atari_NCR5380.c. This means that
atari_scsi and sun3_scsi can make use of the NCR5380.c core driver
and the atari_NCR5380.c driver fork can be made redundant.

Signed-off-by: Finn Thain 
Reviewed-by: Hannes Reinecke 
Tested-by: Michael Schmitz 
Tested-by: Ondrej Zary 
Signed-off-by: Martin K. Petersen

ncr5380: Use DMA hooks for PDMA

2016-04-11T20:57:09+00:00

Those wrapper drivers which use DMA define the REAL_DMA macro and
those which use pseudo DMA define PSEUDO_DMA. These macros need to be
removed for a number of reasons, not least of which is to have drivers
share more code.

Redefine the PDMA send and receive hooks as DMA setup hooks, so that the
DMA code can be shared by all 5380 wrapper drivers. This will help to
reunify the forked core driver.

Signed-off-by: Finn Thain 
Reviewed-by: Hannes Reinecke 
Tested-by: Michael Schmitz 
Tested-by: Ondrej Zary 
Signed-off-by: Martin K. Petersen

ncr5380: Remove PSEUDO_DMA macro

2016-04-11T20:57:09+00:00

For those wrapper drivers which only implement Programmed IO, have
NCR5380_dma_xfer_len() evaluate to zero. That allows PDMA to be easily
disabled at run-time and so the PSEUDO_DMA macro is no longer needed.

Also remove the spin counters used for debugging pseudo DMA drivers.

Signed-off-by: Finn Thain 
Reviewed-by: Hannes Reinecke 
Tested-by: Michael Schmitz 
Tested-by: Ondrej Zary 
Signed-off-by: Martin K. Petersen

ncr5380: Remove FLAG_NO_PSEUDO_DMA where possible

2016-04-11T20:57:09+00:00

Drivers that define PSEUDO_DMA also define NCR5380_dma_xfer_len.
The core driver must call NCR5380_dma_xfer_len which means
FLAG_NO_PSEUDO_DMA can be eradicated from the core driver.

dmx3191d doesn't define PSEUDO_DMA and has no use for FLAG_NO_PSEUDO_DMA,
so remove it there also.

Signed-off-by: Finn Thain 
Reviewed-by: Hannes Reinecke 
Tested-by: Michael Schmitz 
Tested-by: Ondrej Zary 
Signed-off-by: Martin K. Petersen

ncr5380: Fix soft lockups

2016-01-07T02:43:09+00:00

Because of the rudimentary design of the chip, it is necessary to poll the
SCSI bus signals during PIO and this tends to hog the CPU. The driver will
accept new commands while others execute, and this causes a soft lockup
because the workqueue item will not terminate until the issue queue is
emptied.

When exercising dmx3191d using sequential IO from dd, the driver is sent
512 KiB WRITE commands and 128 KiB READs. For a PIO transfer, the rate is
is only about 300 KiB/s, so these are long-running commands. And although
PDMA may run at several MiB/s, interrupts are disabled for the duration
of the transfer.

Fix the unresponsiveness and soft lockup issues by calling cond_resched()
after each command is completed and by limiting max_sectors for drivers
that don't implement real DMA.

Signed-off-by: Finn Thain 
Reviewed-by: Hannes Reinecke 
Tested-by: Ondrej Zary 
Tested-by: Michael Schmitz 
Signed-off-by: Martin K. Petersen

ncr5380: Use standard list data structure

2016-01-07T02:43:07+00:00

The NCR5380 drivers have a home-spun linked list implementation for
scsi_cmnd structs that uses cmd->host_scribble as a 'next' pointer. Adopt
the standard list_head data structure and list operations instead. Remove
the eh_abort_handler rather than convert it. Doing the conversion would
only be churn because the existing EH handlers don't work and get replaced
in a subsequent patch.

Signed-off-by: Finn Thain 
Reviewed-by: Hannes Reinecke 
Tested-by: Ondrej Zary 
Tested-by: Michael Schmitz 
Signed-off-by: Martin K. Petersen

ncr5380: Fix and cleanup scsi_host_template initializers

2016-01-07T02:43:03+00:00

Add missing .module initializer. Use distinct .proc_name values for the
g_NCR5380 and g_NCR5380_mmio modules. Remove pointless CAN_QUEUE and
CMD_PER_LUN override macros. Cleanup whitespace and code style.

Signed-off-by: Finn Thain 
Reviewed-by: Hannes Reinecke 
Tested-by: Ondrej Zary 
Tested-by: Michael Schmitz 
Signed-off-by: Martin K. Petersen

ncr5380: Cleanup #include directives

2016-01-07T02:43:02+00:00

Remove unused includes (stat.h, signal.h, proc_fs.h) and move includes
needed by the core drivers into the common header (delay.h etc).

Signed-off-by: Finn Thain 
Reviewed-by: Hannes Reinecke 
Tested-by: Ondrej Zary 
Tested-by: Michael Schmitz 
Signed-off-by: Martin K. Petersen