Re: raid1 oops in raid1_read_balance on Sparc64
From: Andrew Beresford <hidden>
Date: 2002-04-26 11:00:56
Hi Neil, The problem has gone away compiling with 3.0.4 :) Cheers, Beezly On Thu, Apr 25, 2002 at 08:49:47PM +1000, Neil Brown wrote:
On Thursday April 25, beezly@beezly.org.uk wrote:quoted
I've been trying to get my sparc64 machine to use raid1 but each time I set up the md config with one good disk and one failed disk (so I can get the data off the "failed" disk and into the md device before I do a raidhotadd) I get the following oops (after ksymoops mangling) The kernel is 2.4.19-pre7. md and raid1 are compiled as modules. If any more information is needed please don't hesitate to contact me.That's two reports of this. It looks to me very much like a compiler bug. I'm not an export on Sparc assembler, but I've just been reading through the disassembly listing of read1_read_balance, and at least two things look wrong: 0xc30 <raid1_read_balance+208>: sllx %o2, 3, %g2 0xc34 <raid1_read_balance+212>: add %i1, 0x20, %g3 0xc38 <raid1_read_balance+216>: clr [ %i5 + 0x3f0 ] 0xc3c <raid1_read_balance+220>: add %g2, %o2, %g2 This is the only place that %o2 is used, and it is never initialised. 0xbac <raid1_read_balance+76>: add %i5, 0x18, %o7 0xbb0 <raid1_read_balance+80>: sllx %o7, 3, %g2 %i5 is holding "conf". adding 0x18 gives the address of conf->mirrors[0]->head_position in %o7. But then we shift %o7 left 3 places, multiplying by 8. Multiplying an address by 8 is very unlikely to be correct. Could you both please tell us what compiler version you are using, and see if you can try a different one. NeilBrownquoted
\|/ ____ \|/ "@'/ .. \`@" /_| \__/ |_\ \__U_/Gotta love this!! NeilBrown - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html