On Thu, Dec 28, 2000 at 08:47:05PM -0800, Greywolf wrote:
> On Thu, 28 Dec 2000, Chuck Silvers wrote:
> 
> # the fact that people want this feature right now does not justify
> # making incompatible changes that are clearly not the best solution.
> # back when ISA DMA was still an important issue for many people,
> # people wanted bounce-buffering "right now".  rather than just hack all
> # the relevant drivers to provide that, NetBSD designed and implemented
> # the bus_dma framework, which has been a far superiour solution in the
> # long run.  why would we start making design decisions based on expediency
> # now?  and changing the interpretation of device minor numbers has a far more
> # significant downside than the choice of ISA bounce-buffering schemes,
> # since the changing the interpretation of device minor numbers is visible
> # to applications and it breaks existing usage.
> 
> Chuck, I realise the tone of my most recent message must have sounded
> quite disparaging, and I apologise for that.  It should be POSSIBLE
> at some point to do the volmgr thing, yes, but don't mandate it.  Please.
> It's overhead respective either to space or time, and it would be nice
> to be able to bypass it.  Besides that, it would be something that the
> bootloader would need compiled in?

I sense your frustration.  :-)

the overhead could be minimized for limited feature support with
kernel config options.  the bootloader needs to read whatever disk label
format is in use, but read-only access to data is always much simpler
than read-write.


> With respect to the bounce buffer issue, NetBSD very nearly lost a large
> crowd of people with sub-standard machinery because while we were doing
> The Right Thing, we were doing it at a slow pace.  IIRC, we were working
> on the bounce buffers at the time when PCI was just starting to make serious
> headway, so I think a lot more development effort was going to making the
> PCI stuff work (correct me if I'm wrong; my memory is developing ECC
> errors as I get older).  By the time the bounce buffers were ready to use,
> the hardware on which it would run had long been outstripped.
> 
> I still think it was probably the right thing to do, but how do you weigh
> a success in that manner?

depends on what you consider the goal was.  different people will have
different answers for that, so I'll refrain from debating it.


> On the partitioning issue, that's something that's not going to suddenly
> go away.  16 partitions, for the moment, ought to be enough.

for the moment, for many people, sure.  but eventually it won't be,
and then we'll be having this same discussion again.  I'd like to
having people thinking about the more general solution instead.
and even better, implementing the more general solution.


> [but how's the bootloader going to deal with that one, especially
> on the Suns?]

this is why netbsd will always have to support a variety of disk label
formats, because some platforms require a disk to have a certain label format
in order to boot from it.  the sun platforms will always have to have
a sun label on the boot disk.  plus, we'd like to be able to read disks
created with other operating systems.

-Chuck