Re: [hatari-devel] MODE SENSE patch

[Eero Tamminen <oak@xxxxxxxxxxxxxx>]

Hi,

On 15.3.2022 22.43, Uwe Seimet wrote:
> Obviously you are familiar with some rather exotic DISKUS functionality :-):
>
> I stumbled upon the missing MODE SENSE support for 0x3f when extending my
> SCSI Driver tests, but these tests do not request individual pages. This is
> why I did not look into the details of the existing code for a single page.
> On first sight I agree with what you say. To fix this one can probably just
> call HDC_CmdModeSense0x04 with an offset of 4, similar to how I did it for
> 0x3f.
>
> The current code also does not check the DBD field, which means that more
> changes would be required in order to be compatible at least with SCSI-2.
> On the other hand, Hatari claims to be compatible with SCSI-1 CCS, for which
> I don't have a reliable specification. But from my experience with real
> SCSI-1 drives there is hardly any difference compared to SCSI-2.
>
> But before applying more changes I would like to learn more on how to
> reproduce what Eero reported:
>
> HDC: REQUEST SENSE (SCSI, t=0, lun=0, cdb=03:00:00:00:60:00).
> WARN : HDC: *** Strange REQUEST SENSE ***!
>
> @Eero Can you please provide instructions on how to reproduce this,

Uh, that's a bit involved, as it was from m68k Linux.  Quick instructions...


First build latest Linux version for m68k (on Debian, which has m68k ELF 
cross-compiler):

1. Install build deps:
   $ sudo apt install bc bison flex install gcc-m68k-linux-gnu

2. Get latest upstream kernel release sources (without history):
   $ git clone --depth 1 --branch v5.16 \
     git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
   $ cd linux

3. Use Hatari compatible configuration:
   $ cp /path/to/hatari/tools/linux/kernel.config .config

4. Compile configured kernel:
   $ ARCH=m68k CROSS_COMPILE=m68k-linux-gnu- make -j4 vmlinux

(Build could be in a container, if you do not have Debian installed.  Or 
if you'll just want a Linux binary image, I can send that privately.)


Hatari config for booting that Linux version:
  $ cat > lilo.cnf
[LILO]
Kernel = vmlinux
Symbols = System.map
Args = video=atafb:sthigh console=tty
KernelToFastRam = FALSE
Ramdisk =
^D


Create empty[1] 4 MB EXT2 HD image:
   $ mkdir disk/
   $ /sbin/mkfs.ext2 -t ext2 -L DISK -d disk/ root.img 4M

[1] Because SCSI mount fails, image content does not matter.  More 
instructions are in:
https://hatari.tuxfamily.org/doc/m68k-linux.t


Then run Hatari with these under TT emulation:
  $ hatari --trace scsi_cmd --timer-d on --machine tt --vme none 
--cpu-exact off --compatible off --mmu on -s 14 --ttram 64 --addr24 off 
-c lilo.cfg --natfeats on --scsi 0=root.img --lilo "debug=nfcon 
root=/dev/sda ro init=/init"


NOTEs:

* Hatari's dummy VME emulation needs to be disabled, otherwise Linux 
tries to actually use it, which would obviously fail

* It's better to disable 030 cache & prefetch emulation, they do not 
work well enough with MMU + fast-RAM, for Linux user-space <-> kernel 
transition to work properly (i.e. after init gets booted from root FS)

* If you want to test Linux more, Hatari's tools/linux/ dir has couple 
of useful patches to Linux sources:
  https://git.tuxfamily.org/hatari/hatari.git/tree/tools/linux



> and the Hatari commit ID that is affected?

It was the Hatari git HEAD (commit 418b6bb0a I think) i.e. with your 
patches that Nicolas had pushed to the repo at that time.


	- Eero