Hi folks,
in the next 4 emails from me (which of course, you may receive
out-of-order) there is the next version of the Large Block Device
patch against 2.5.40 that allows linux to transition to 64-bit
sector_t, and so escape the 2TB block device size limit.
This patch fixes the block layer, the SCSI and IDE midlayers, the
loop device, and MD. It has been tested in previous incarnations
with the megaraid controller and up to 3.6TB of attached storage,
and reviewed by various people including Jens Axboe, Christopher
Hellwig, Neil Brown and Martin Dalecki. I have not tested this
version (although I'm reasonably confident it'll work) because
software RAID is currently broken wrt recent block layer changes
(MD can't cope with I/O requests that cross one of its zone
boundaries).
The only part I find horribly kludgy (a better solution is
requested) is that to support RAID levels 0 through 5 with large
block devices you really really need 64-bit division. As nothing
else needs it, I've linked gcc's division and remainder routines
into md.o, and exported them as symbols (if you're using modules).
Nothing else uses 64-bit division directly; for the few cases where
it's needed I've included a sector_div() macro based on the div64()
macro needed for printk.
There are a couple of workarounds for gcc 2.96: gcc3.2 works a lot better.
Still to do:
-- 16-byte SCSI commands (READ_CAPACITY_16, READ16, WRITE16 and
the corresponding error paths). I have an untested patch for
that, but as far as I know, none of the drivers for cards I
have access to at present understand the 16-byte command set.
-- Better 64-bit support for the megaraid controller (that
should be relatively easy)
-- A major rewrite of MD (someone else do this please? NeilB?)
to allow software RAID members bigger than 2TB.
-- the original (old) HD driver.
You can get the patch directly from the bk tree at
bk://gelato.unsw.edu.au:2023, or
pick it up from http://www.gelato.unsw.edu.au/patches
Note to Neil and Jens: I'm not going to send the patches directly to you
--
Dr Peter Chubb [email protected]
You are lost in a maze of BitKeeper repositories, all almost the same.
On Thu, 3 Oct 2002 [email protected] wrote:
> Hi folks,
Hi Peter,
>...
> There are a couple of workarounds for gcc 2.96: gcc3.2 works a lot better.
>...
does it work correctly with 2.95? 2.95 is still the recommended compiler
as of 2.5.40 .
cu
Adrian
--
You only think this is a free country. Like the US the UK spends a lot of
time explaining its a free country because its a police state.
Alan Cox