IDE emu: corrupted inquiry data after CMD_READ more than 16 sectors

[thomas]

If I do a CMD_READ from IDE scsi.device with more than 16 sectors (8 KB) at once, a subsequent HD_SCSICMD with INQUIRY or READ CAPACITY returns completely corrupted data. And a following CMD_READ crashes the Amiga.

This started in 3.2.0, 3.1.0 is ok. And of course it does not happen on a real Amiga.

To reproduce use the attached test program:

- select a4000 quickstart
- add test.lha as boot drive
- add just any HDF or VHD to IDE:0
- set serial port to tcp::1234/wait
- start
- run telnet localhost 1234 in a DOS window.

- at the AmigaDOS prompt enter

test_ser scsi.device 0 cache 16

to see a successful test and enter

test_ser scsi.device 0 cache 17

or any other number higher than 16 to see the problem.

(test_con writes to the CLI window, test_ser uses kprintf to write to the serial port, otherwise they are identical)

If you are lucky and the emulated Amiga does not crash, you can "repair" it by running the test again with 16 sectors.

Example output:

begin of test
try 1 of 2
do_scsi_cmd: Actual      = 36
do_scsi_cmd: CmdActual   = 6
do_scsi_cmd: SenseActual = 0
scsi_inquiry: Vendor   = <UAE-IDE >
scsi_inquiry: Product  = <wb31.vhd        >
scsi_inquiry: Revision = <0.7 >
do_scsi_cmd: Actual      = 8
do_scsi_cmd: CmdActual   = 10
do_scsi_cmd: SenseActual = 0
scsi_capacity: BlockAddress = 2047751
scsi_capacity: BlockLength  = 512
cache size = 17 * 512 = 8704
cmd_read: offset = 0; length = 8704; data = 78791AC
0: <    >
1: <    >
2: <RDSK>
3: <PART>
4: <PART>
5: <PART>
6: <FSHD>
7: <LSEG>
try 2 of 2
do_scsi_cmd: Actual      = 36
do_scsi_cmd: CmdActual   = 6
do_scsi_cmd: SenseActual = 0
scsi_inquiry: Vendor   = gl,(   >
scsi_inquiry: Product  = <>■R"            >
scsi_inquiry: Revision = <g@">
do_scsi_cmd: Actual      = 8
do_scsi_cmd: CmdActual   = 10
do_scsi_cmd: SenseActual = 0
scsi_capacity: BlockAddress = 4294967295
scsi_capacity: BlockLength  = 512
cache size = 17 * 512 = 8704
cmd_read: offset = 0; length = 8704; data = 78791AC

[Toni Wilen]

Fixed (and thanks for perfect instructions. This is how it should be always done!), multiple transfer offset counter was not reset causing identify to return some previously loaded data block if previous transfer size was larger than multiple transfer count (which KS3.x at least sets to 16 by default).