PCI64 + block zero-bounce highmem v11

Hi,

Uploaded a new version. Changes since block-highmem-all-10:

- Merge pci64-1 from davem (me)
- Merge pci64-2 from davem (me)
- printk type warning on x86 when HIGHMEM wasn't enabled (me)
- Better comment asm-i386/scatterlist.h
- Add single_segment/highmem comment in scsi_merge (me)
- Include Mark Hemment's no __save_flags/__cli in read bounce
- Convert cciss to pci64 interface (me)

HIGHMEM I/O capable hardware now includes:

- IDE (DMA, PIO will revert to bounce) Basically all low level drivers
- SCSI
- aic7xxx (new)
- qlogicfc (64-bit too)
- sym53c8xx (64-bit too)
- megaraid
- IPS
- cpqarray
- cciss (64-bit too)

SCSI drivers are typically easy to convert to support < 4GB I/O, usually just the can_dma_32 flag needs to be added. Some need a bit more work, like IPS. Look for references to virt_to_bus on the scatterlist ->address and/or bh->b_data to find obvious highmem breakage. For 64-bit PCI DMA, take a look at David's modifications to sym53c8xx as a good reference.

I'll include I2O highmem support in the next release -- haven't started it yet, but it should be a breeze.

DAC960 would be nice to have as well. Leonard, we talked about converting it to the pci dma interface some time ago -- did you make progress on that?

*.kernel.org/pub/linux/kernel/people/axboe/patches/2.4.9/block-highmem-all-11

Jens Axboe

*.kernel.org/pub/linux/kernel/people/axboe/patches/2.4.8-ac9/block-highmem-all-11-ac9

contains a 2.4.8-ac9 variant, changes are identical.

Jens Axboe

It should be, however the amount of i2o firmware that gets 64bit right is unknown and I fear quite minimal.

Alan

DAC960 would be nice to have as well. Leonard, we talked about converting it to the pci dma interface some time ago -- did you make progress on that?

I think late next week I should have some time to start looking at this.

Leonard

Probably, however just getting 4GB right would be a really good start.

Jens Axboe