A level-triggered interrupt handler for a particular vector (aka, a given device & driver) should only be invoked on one CPU. Drivers should be able to safely assume that one one instance of a given interrupt handler is running on the system at a time.
However, due to a VMware ESXi software defect, a level-triggered interrupt handler may be invoked on two different CPUs in parallel at the same time, resulting in multiple different failures. When a level-triggered interrupt is being serviced by a CPU, subsequent interrupts on the same vector for that CPU are queued on the Local APIC. However, the ESXi Interrupt Tracker can asynchronously change the vector's destination to a new CPU, which may immediately handle it. If this occurs, the interrupt handler on the first and second CPUs may interact in unforeseen and detrimental ways.
The specific symptoms observed depend on the driver which is affected, and are outside scope of this article. See Impact/Risks section for known issues.
Unfortunately, this is an issue that was suspected to be a hardware fault for some time, leading to cases being opened with hardware partners. If misbehaving hardware does cause double-firing of interrupts in the future, this fix should address it as well.
( Try at your own risk )
VMware KB: Configuring advanced driver module parameters in ESX/ESXi
http://www.myitblog.in/
