Re: [MiNT] XaAES / GEM memory issues

[Sven Karlsson <svenka@it.kth.se>]

Hi Frank

> Maybe you mean the right but write it wrong. The point is that the device
> driver have to operate on the lowest level as it's designed. Every mixup
> with AES/VDI functionality is wrong. These AES/VDI things must be layered
> some levels above (but in the kernel space to get rid of overflous context
> switches). This is simply so because AES/VDI operate at higher levels and
> not on device driver level.

First of all I do not propose to move all AES handling into the driver.
Secondly, I cannot see why moving some selected high level functions
into the device driver is generally wrong in any way. See for instance
drivers for 3Com ethernet cards that implement some IP handling in
hardware and advanced SCSI chipset like those found on Adaptec cards
where the hardware performs large portions of the SCSI handling. One can
also draw parallels to MiNTnet.

Said that I do agree that moving high level functions into device
drivers should be avoided as this could cause respons problems if the
device drivers spend too much time processing data.

But please enlighten me as to if there is any other possiblity of
putting code in kernel space in a clean way without using a device
driver? Surely you agree that a device driver is preferred over a TSR?

Regards
 Sven