set the ccb state when it comes off the free list. this isnt used anywhere
yet, but i like to be ready when the time comes.

fix a panic string to mention the corrent place its freaking out in.

leave the reply_dva address alone so we can post it back to the ioc rather
than posting back the offset of the reply frame to the start of the reply
space.

nobody likes panics, even if theyre free.

fix the reply handling on crazy machines that give me bits in the high part
of the address.

64 bit dva addresses so we can >> 32 later; ok dlg

byteswap the ioc_status field so we can respond to scsi things properly on
bigendian archs. this lets ses attach now.

remove the fetching of the manufacturing page. it was just there to see if
i got the page fetching right, its not really useful for anything in the
real world.

mpi hardware uses an 8 bit field to describe the number of devices it has
on a port. since 256 wont fit into 8 bits they say 0 means 256. this diff
does the appropriate interpretation. it also avoids

mpi hardware uses an 8 bit field to describe the number of devices it has
on a port. since 256 wont fit into 8 bits they say 0 means 256. this diff
does the appropriate interpretation. it also avoids a divide by zero when
we figure the openings out by dividing the number of commands the
controller can support by the number of devices it supports.

panic found while testing the fc controller at home.

show more ...

dont display the first manufacturing page

c++ style comments shouldnt be in the tree

int32 i >> 32 is undefined, so cast to unsigned long long first, ok dlg

implement chained scatter gather lists.

raise the timeout on the init commands (especially port enable) to give
the sas controller currently being used in jasons blade 2000 a chance to
respond.

hush; ok dlg

increase the timeout on the enabling commands during attach so that sas
controllers have a chance to respond.

limit the number of scatter gather entries sent with the scsi_io commands
so it fits in the maximum request frame size. this will do until i can
write sgl chaining in a nice way.

put the 64bit flag in the right place on the scatter gather list.

split mpi_poll out into mpi_complete, and rewrite poll to use start and
complete to do its job. enforce the use of a timeout on polled commands
and make the callers check if the timeout happened.

remove some if 0 code

bump mpi up to using 64bit for all dva

junk to read and write configuration pages.

pass the reply from the hardware via the ccb rather than on the stack to
the function handling the completion. this means that the completion
function can do nothing with the reply, instead leaving i

pass the reply from the hardware via the ccb rather than on the stack to
the function handling the completion. this means that the completion
function can do nothing with the reply, instead leaving it up to the
requester to deal with.

show more ...

rename mpi_complete to mpi_poll and make it call mpi_start itself.

make debug output during attach quiet unless you want it by raising
mpidebug

remove dead code