2.6.31 kernel on VMWare

ang123 · 01-05-2010, 12:15 PM

Hi - is anyone running a 2.6.31 kernel as a VMWare ESX 3.5.0 guest-OS?

I've just tried upgrading the guest-OS kernel from 2.6.30.10 to 2.6.31.7 and now the virtual-CD cannot be detected. The distro is LFS-based. Boot-time kernel logs illustrating the problem are included below. Basically after boot there is no /dev/hda and no /dev/cdrom. Tried rebuilding the kernel with various config adjustments, to no avail.

Thanks in advance for any tips or advice.

2.6.30.10:
Jan 4 13:54:07 virtapp kernel: Probing IDE interface ide0...
Jan 4 13:54:07 virtapp kernel: hda: VMware Virtual IDE CDROM Drive, ATAPI CD/DVD-ROM drive
Jan 4 13:54:07 virtapp kernel: hda: host max PIO4 wanted PIO255(auto-tune) selected PIO4
Jan 4 13:54:07 virtapp kernel: hda: UDMA/33 mode selected
Jan 4 13:54:07 virtapp kernel: ide0 at 0x1f0-0x1f7,0x3f6 on irq 14

2.6.31.7:
Jan 4 13:41:09 virtapp kernel: Probing IDE interface ide0...
Jan 4 13:41:09 virtapp kernel: ide0: no devices on the port
Jan 4 13:41:09 virtapp kernel: ide0 at 0x1f0-0x1f7,0x3f6 on irq 14

ang123 · 01-08-2010, 04:17 AM

Hi again ... I have added some debug logging to the kernel and discovered that it is detecting /dev/hda (the cdrom) as a device - but is subsequently ignoring it...

Code:

Jan  7 18:18:46 virtapp kernel: Probing IDE interface ide0...
Jan  7 18:18:46 virtapp kernel: ANG: ide_wait_not_busy...
Jan  7 18:18:46 virtapp kernel: ANG: ide_port_for_each_dev (loop)...
Jan  7 18:18:46 virtapp kernel: ANG: device=hda rc=0
Jan  7 18:18:46 virtapp kernel: ANG: ide_port_for_each_dev (loop)...
Jan  7 18:18:46 virtapp kernel: ANG: device=hdb rc=-19
Jan  7 18:18:46 virtapp kernel: ANG: returning rc=-19
Jan  7 18:18:46 virtapp kernel: ide0: no devices on the port
Jan  7 18:18:46 virtapp kernel: ANG: ide_probe_port() returning -19

The root cause of this seems to be an explicit check for ENODEV that was added to ide_probe_port() in drivers/ide/ide-probe.c in the 2.6.31.1 kernel which unfortunately jumps over the extra check needed to detect the CDROM drive (at least, in our VM setup). I've tested this by rebuilding the kernel without the check - and the virtual CDROM is now fully functional.

Code:

        rc = ide_port_wait_ready(hwif);

>        if (rc == -ENODEV) {
>                printk(KERN_INFO "%s: no devices on the port\n", hwif->name);
>                goto out;
        } else if (rc == -EBUSY)
                printk(KERN_ERR "%s: not ready before the probe\n", hwif->name);
        else
                rc = -ENODEV;

        /*
         * Second drive should only exist if first drive was found,
         * but a lot of cdrom drives are configured as single slaves.
         */
        ide_port_for_each_dev(i, drive, hwif) {
                (void) probe_for_drive(drive);
                if (drive->dev_flags & IDE_DFLAG_PRESENT)
                        rc = 0;
        }
out: ...

So now my question is ... can I safely patch this out of my kernel?
Does anyone know why the explicit ENODEV check was added in the first place?

Thanks in advance.

ang123 · 01-11-2010, 11:57 AM

Incase anyone else has the same problem.
The patch we've now tried and tested for this is:

Code:

--- linux-2.6.31.7.orig/drivers/ide/ide-probe.c 2009-12-08 19:13:50.000000000 +0000
+++ linux-2.6.31.7/drivers/ide/ide-probe.c      2010-01-08 12:59:51.000000000 +0000
@@ -685,10 +685,7 @@
                disable_irq(hwif->irq);

        rc = ide_port_wait_ready(hwif);
-       if (rc == -ENODEV) {
-               printk(KERN_INFO "%s: no devices on the port\n", hwif->name);
-               goto out;
-       } else if (rc == -EBUSY)
+       if (rc == -EBUSY)
                printk(KERN_ERR "%s: not ready before the probe\n", hwif->name);
        else
                rc = -ENODEV;
@@ -702,7 +699,11 @@
                if (drive->dev_flags & IDE_DFLAG_PRESENT)
                        rc = 0;
        }
-out:
+
+       /* if still ENODEV - then definitely no devices found   */
+       if (rc == -ENODEV)
+               printk(KERN_INFO "%s: no devices on the port\n", hwif->name);
+
        /*
         * Use cached IRQ number. It might be (and is...) changed by probe
         * code above

mhwood@Ameritech.Net · 01-11-2010, 03:48 PM

I have been *trying* to install 2.6.31 (Gentoo install-amd64-minimal-20091203) on ESXi 3.5.0 123629, but although the CDROM image boots just fine, the kernel thinks there is nothing on the IDE controller and so the initrd can't mount the CDROM it just booted from. :-{ Time to find an earlier CD image, I guess. Thanks for your notes -- at least now I have some reason to believe it's nothing *I* did wrong, and something specific to try.