PFS3aio and TD64

[Gilloo]

Hello,
Is there a way to force PFS3 to use the TD64 or NSD command set ?
It always use Direct Scsi Commands and I suspect some bugs in my implementation of this, error 170, void sense error.
is it try to use a Write 12 command ?

[Toni Wilen]

If dostype = PFSx: check TD64/NSD first. If failure: try DirectSCSI
If dostype = PDSx: check DirectSCSI first, if failure: try TD64/NSD.

Only SCSI READ(10) and WRITE(10) are used.

But if PFSx also tries to use DirectSCSI: something is wrong with TD64/NSD implementation.

TD64 test: TD_READ64 with zero address and zero length must return success (with io_Actual==0) or IOERR_BADLENGTH or IOERR_BADADDRESS. Partition's last block read must succeed.

ND64 test is same as above and also NSCMD_DEVICEQUERY must return success and at least NSCMD_TD_READ64 must be listed and type must be NSDEVTYPE_TRACKDISK.

DS test: TEST UNIT READY must return success. READ CAPACITY must return sane data. Also fail if partition last block is out of READ CAPACITY returned size. READ(10) last block read must succeed. Last block read of partition must succeed.

Most common bug is probably capacity read returning total size when it should return last block (=total - 1).

[Gilloo]

Thanks you Toni !

before (offset = io_Actual<<32 + io_Offset, before call, and after)
TD_READ64 [24](offset=0x0000000000000000 (0x0000000000000000) size=0)=>0
TD_READ64 [24](offset=0x00000003B07FFE00 (0x00000003B07FFE00) size=512)=>0
DS commands

now
TD_READ64 [24](offset=0x0000000000000000 (0x0000000000000000) size=0)=>0
TD_READ64 [24](offset=0x00000003B07FFE00 (0x00000200B07FFE00) size=512)=>0
TD_xx64 commands

The bug was that the io_Actual was never set to the value of amount of data transfered.
The io_Offset value should be altered to ?

[Gilloo]

PFSaio 19.2, block size 4096

Got some error like
Read 3 Error 42 on block xxx + 255

I suppose 255 x block size = 1044480 ~ 1MB,
The max transfer mask should be ajusted ?

[LIV2]

The error number is coming from the driver itself, whatever is causing it to throw HFERR_Phase is to blame.

Max transfer will depend on the driver and if it even needs it - It's mostly a kludge that shouldn't be needed if the driver is implemented correctly.
What kind of driver are you developing? IDE?

If so, you can transfer up to 256 sectors on a single READ/WRITE ATA command, above that you need to loop and issue another READ/WRITE (or limit max_transfer appropriately)
Here's how lide.device does a read
https://github.com/LIV2/lide.device/...ta.c#L477-L523