MTD CVS update: 'mtd/grub doc_stage1.S'
dwmw2 at infradead.org
Wed May 31 20:49:18 EDT 2000
Update of /home/cvs/mtd/grub
In directory infradead.org:/tmp/cvs-serv17114
Fix the setup of the INT 19h handler so we are no longer using the
previously initialised %es register as the segment address of the vector.
Also fix the segment registers in the actual copy, so that we put the
right stuff in place, rather than starting our copy at 0000:0000.
I can hardly believe that this has ever worked. It seems that by an
amazing coincidence, the IPL ROM sets %es to 0x2000 before jumping to
the code - so even though we were using an uninitialised segment
address pointer, and our copy of the IRQ handler to the _intended_
location went AWOL, we ended up with the INT 19h handler pointing at
the location in memory at which the IPL ROM loads the SPL, before the
SPL relocates itself into the reserved area.
So as long as nothing touched that memory later in the startup sequence,
it would still be there when the BIOS got round to calling INT 19h to
boot. This is why it worked on simple machines like my 386 testbed, but
I hereby declare myself to be Today's Official Mr F*** All Good.
To unsubscribe, send "unsubscribe mtd" to majordomo at infradead.org
More information about the linux-mtd