Re: Interface for SCSI transactions ?

[Thomas Schmitt]

Hi,

me:
> >   routine device_transact_native(
Samuel Thibault:
> Why "native"?

Because the transactions are of a genuine kind that is specified only
for particular device types.

I dug deeper in the existing code. It seems that my desired transaction
applies only to these storage device types:
- SCSI hard disk
- SCSI CD-ROM
- IDE/ATAPI CD-ROM
(IDE disks do not use SCSI commands. SCSI tapes seem not have the properties
 of a classical block device as they are covered by SSC rather than SBC.
 With ATAPI tapes i expect the same.)

Are there proposals for a better name which describes the general purpose ?
(Reminder: It is so general because the method will occupy a slot in
 struct device_emulation_ops of all device classes.)


There is another publicly visible name to decide:
The function_code for my desired transaction type.
Currently it is

  #define DEVICE_TRANSACT_NATIVE_SCSI_PKT         0x00010001

Other kinds of transaction to the same class or to other classes of devices
would have public names like
  #define DEVICE_TRANSACT_NATIVE_...

I did not look yet, how public Hurd macros names usually look like.
(Those in the driver code are often short english words.)


> > So do i have to feed this by two C variables ?
> Mmm, I think the simplest way is to just rebuild the kernel with the new
> RPC, and look at the generated prototype.

I think the example of device_get_types() already confirms it.
"^array[]" translates to a C pair of pointer and mach_msg_type_number_t.


I am now collecting a sketch of code snippets which shall connect the RPC
receiver function with the IDE and SCSI drivers.
Desperation and optimism alternate in the usual way.


Have a nice day :)

Thomas