Re: SYM-2 patches against latest kernels available

[Gérard Roudier]

On Sun, 4 Nov 2001, Jens Axboe wrote:

> On Sun, Nov 04 2001, G?rard Roudier wrote:
> >
> >
> > On Sun, 4 Nov 2001, Jeff Garzik wrote:
> >
> > > Gérard Roudier wrote:
> > > > The patch against linux-2.4.13 has been sent to Alan Cox for inclusion in
> > > > newer stable kernels. Alan wants to test it on his machines which is a
> > > > good thing. Anyway, those patches just add the new driver version to
> > > > kernel tree and leave stock sym53c8xx and ncr53c8xx in place.
> > >
> > > Are the older sym/ncr drivers going away in 2.5?
> > >
> > >
> > > > Any report, especially on large memory machines using 64 bit DMA (2.4
> > > > kernels + PCI DAC capable controllers only), is welcome. I can't test 64
> > > > bit DMA, since my fatest machine has only 512 MB of memory.
> > > >
> > > > To configure the driver, you must select "SYM53C8XX version 2 driver" from
> > > > kernel config. For large memory machines, a new "DMA addressing mode"
> > > > option is to be configured as follows (help texts have been added to
> > > > Configure.help):
> > > >
> > > > Value 0: 32 bit DMA addressing
> > > > Value 1: 40 bit DMA addressing (upper 24 bytes set to zero)
> > > > Value 2: 64 bit DMA addressing limited to 16 segments of 4 GB (64 GB) max.
> > >
> > > Are you using the new pci64 API under 2.4.x?
> >
> > Didn't see any. Only the dma_addr_t thing can be 32 bit or 64 bit
> > depending on some magic. Apart this, the driver is asking for the
> > appropriate dma mask given the configured dma adressing mode.
>
> I've looked over the sym-2 and it is using pci_map_sg so it's 64-bit
> safe for sg transfers at least. For non-sg requests you are using
> pci_map_single, but you can't do any better because the mid layer is
> handing you virtual addresses in request_buffer currently anyways...

If there is some platform-specific thing that allows to handle map_single
more right:), I can go with it. Would be fine to get IA64 and Alpha
actually PCI-64 bit safe even by using some software shoehorn for that. :)

> > PS: There is some pci64* API on some arch., but nobody will want to
> > ever use it, in my opinion.
>
> You are doing it right :-)

You mean as right as possible? :)

  Gérard.