Re: 2.3.43 alpha broken with >2g of ram

[Andrea Arcangeli]

On Sat, 12 Feb 2000, Richard Henderson wrote:

>I'm not sure how highmem would work with the pci dma mapping 

I just designed in mind all the details of how they have to work together.

I finished now to fix up all the blockdevice layer breakage I did in the
last two days. All returned to compile now on top of 2.3.43 but I don't
expect it to work at the first try (and I also have to still implement a
few debugging functions to check the values I now calculate in O(1)
remains coherent to what I would have obtained with the previous O(n)
implementation, this in order to catch places that I may have missed). If
all goes fine after the weekend I could be able to work on the alpha
highmem.

>functions.  Cause I wouldn't want bounce buffers to be used with
>any driver that had been converted.

For this specific problem related to the I/O bounces the solution I just
planned is a per blk_dev[major].need_bounces bitlfag and run
create_bounces only if it's set to 1. Set 1 in the ll_rw_block init
function and allow the aware updated drivers to clear it back to zero. The
not updated drivers run slower but running slower is way faster than not
running at all! ;)

Andrea