Help - 2.4.5 SMP doesn't boot (any more)
Hi,
Yesterday I had 2.4.5 running fine. Since today, I can't boot the 2.4.5 SMP kernel any more. I get a kernel panic (invalid page request at 0000000000000010) after SMP is enabled (actually the last message I see is "trying to unmount old root ...", but from kdb messages and the fact that uniprocessor 2.4.5 boots, it is clear that it is a SMP error). The panic happens before syslog startup, so I can't send a backtrace right now (I do not have a serial console ready). kdb backtace leaves me in the start_ap function.
My BIOS has reported "8140 Processor in socket M0 failed FRB level 3 timer" for some time, but that didn't cause problems before and still doesn't with 2.4.4.
Any ideas?
Thanks,
Martin
Signature · 2 lines
Martin Wilck <[email protected]> FSC EP PS DS1, Paderborn Tel. +49 5251 8 15113
Quoted 6 lines · Martin Wilck
Yesterday I had 2.4.5 running fine. Since today, I can't boot the 2.4.5 SMP kernel any more. I get a kernel panic (invalid page request at 0000000000000010) after SMP is enabled (actually the last message I see is "trying to unmount old root ...", but from kdb messages and the fact that uniprocessor 2.4.5 boots, it is clear that it is a SMP error).
That sounds like a known bug in 2.4.5 with undefined data in an inode during root switch. Fixed in 2.4.6-pre2. The symptoms can be weird, other people have blamed it on compilers, it is just luck if the undefined field causes a problem or not.
Index: 5.1/fs/block_dev.c
fs/block_dev.c Sat, 09 Jun 2001 11:25:53 +1000 kaos (linux-2.4/p/b/16_block_dev. 1.2.2.4 644) renamed from fs/block_dev.c Wed, 23 May 2001 11:55:33 +1000 kaos (linux-2.4/p/b/16_block_dev. 1.2.2.3 644) +3 -2
@@ -595,14 +595,15 @@ int check_disk_change(kdev_t dev) int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) { - kdev_t rdev = to_kdev_t(bdev->bd_dev); struct inode inode_fake; int res; mm_segment_t old_fs = get_fs(); if (!bdev->bd_op->ioctl) return -EINVAL; - inode_fake.i_rdev=rdev; + memset(&inode_fake, 0, sizeof(inode_fake)); + inode_fake.i_rdev = to_kdev_t(bdev->bd_dev); + inode_fake.i_bdev = bdev; init_waitqueue_head(&inode_fake.i_wait); set_fs(KERNEL_DS); res = bdev->bd_op->ioctl(&inode_fake, NULL, cmd, arg);
Quoted 2 lines
That sounds like a known bug in 2.4.5 with undefined data in an inode during root switch.
Yes, that was it. Thanks!!!
Martin
Signature · 2 lines
Martin Wilck <[email protected]> FSC EP PS DS1, Paderborn Tel. +49 5251 8 15113