Description of problem: The following hunk from SOURCES/linux-2.6.0-compile.patch causes a huge number (thousands depending on number of disks in the system, repeatedly over time) of messages to be spewn to the console by tools such as Dell OpenManage Storage Administrator due to its (ab)use of /dev/sg to detect devices. Please remove this hunk, it's also not included upstream, or re-architect it to spew far less often (once per open is still way too often for the same app). --- linux-1130/drivers/scsi/sg.c +++ linux-10000/drivers/scsi/sg.c @@ -241,6 +241,12 @@ sg_open(struct inode *inode, struct file return -ENXIO; if (sdp->detached) return -ENODEV; + + /* scsi generic is only for non-disk, non-cd, non-tape devices */ + if ( (sdp->device->type == TYPE_DISK) || (sdp->device->type == TYPE_MOD) || (sdp->device->type == TYPE_ROM) + || (sdp->device->type == TYPE_WORM) || ( sdp->device->type == TYPE_TAPE)) + printk(KERN_WARNING "%s: Using deprecated /dev/sg mechanism instead of SG_IO on the actual device\n",current\ ->comm); + /* This driver's module count bumped by fops_get in <linux/fs.h> */ /* Prevent the device driver from vanishing while we sleep */ d Version-Release number of selected component (if applicable): kernel-2.6.8-1.624.src.rpm How reproducible: always by any app (such as ploink.c given to Doug Ledford to debug a RHEL3 SCSI race condition between various actions (add/remove/open) on SCSI devices. Steps to Reproduce: 1. Run something that opens /dev/sg* repeatedly 2. 3. Actual results: kernel spews to the console Expected results: kernel warns once or occasionally at KERN_DEBUG level, not constantly due to app constantly calling open(/dev/sg). Additional info:
Note, the above patch is provided by Red Hat only, it has not been included upstream. While you can argue that the app should change to not use /dev/sg, forcing all apps to do so to avoid clogging the system with these messages will limit third party app availability on RHEL4. Dell OpenManage apps currently plan on using the RHEL2.1 and RHEL3 compatibility promised by RHEL4 marketing to have their apps ready within a year of RHEL4 launch.
Doug, can we expect this to be fixed for RHEL4 RC please?
Setting state to modified per status from Doug: This one is done (although not by me, as it turns out this message was a Red Hat added message and we just removed it).
Removal confirmed with 2.6.9-1.730_EL.
Verified via code inspection that this has been fixed in kernel 2.6.9- 1.751_EL. The printk KERN_WARNING message has been removed. Thanks. Removing from internal RC Blocker list. Sue: Please do the same from your list.