On 02/20/2014 03:00 PM, Matthew Garrett wrote: > On Thu, 2014-02-20 at 14:59 -0600, Russ Anderson wrote: >> On Thu, Feb 20, 2014 at 08:46:04PM +0000, Matthew Garrett wrote: >>> On Thu, 2014-02-20 at 14:40 -0600, Russ Anderson wrote: >>> >>>> This is also a problem for systems with functional BMCs. Our >>>> large cluster systems do all IPMI traffic (monitoring) through >>>> a system controller back door. We do not want the kernel >>>> doing IPMI commands on those systems. >>> Why not? >> Because some customers want to use cpu cycles for their >> application and let the ipmi monitoring go on through >> the system controller network. > Why is it generating any significant amount of CPU load? We're not > talking about a high-bandwidth interface here. >
I believe that problem is fixed now, at least the one with kipmid using lots of CPU. However, the basic problem is that hardware vendors produce hardware that sucks and then expect software to fix all the problems. Most IPMI interfaces don't have interrupts, so they have to be polled. Then they add important interfaces on top of it like firmware upgrade and ACPI and expect it to perform well. If vendors would just have an interrupt for IPMI, 99% of these problems would go away. If there are still issues with lots of CPU being used, then the problem is most likely non-compliant or just broken hardware. I've seen enough of that. One thing we can do is remove the default interface probing for IPMI. Even though the spec has it, all modern hardware should have it specified in ACPI or device tree. That should fix all the slow boot problems, at least. If a user wants to add a default interface, they can use the interface to dynamically add it after boot time. -corey -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
