jeff (Jeffrey Roberson)
User

Projects

User does not belong to any projects.

User Details

User Since
Aug 6 2017, 12:45 AM (6 w, 4 d)

Recent Activity

Mon, Aug 28

jeff added a comment to D12130: Greatly reduce interrupt latency caused by steal_idle.

This is very interesting. To validate your notion that migrating before iret is the problem you could simply sched_pin() between interrupt and iret or change the SCHED_CAN_MIGRATE test to exclude ithreads. This should be simple to test. I will also reach out to some contacts at AMD.

Mon, Aug 28, 2:59 AM

Sat, Aug 26

jeff added a comment to D12130: Greatly reduce interrupt latency caused by steal_idle.

This is interesting. Do you have any benchmarks or observations about the performance impact? How did you arrive at this needing to be addressed?

Sat, Aug 26, 10:02 PM
jeff added a comment to D11784: Dynamically grow the slab size to control wasted memory.
In D11784#249094, @alc wrote:
In D11784#248990, @jeff wrote:

Three broad comments;

We don't necessarily know which allocations will be handed off to do DMA on. This may be safe as busdma becomes more sophisticated but in the past network drivers definitely assumed mbufs were contiguous, for example. There may still be advantages to allocating aligned kva or possibly just contig physical addresses and using the direct map. You could in principle still use a constant offset from the aligned boundary to find the slab. You would also produce shorter scatter/gather descriptors for those zones that are involved in dma. Eliminating vtoslab may not be that important with the per-cpu caches in play but we do need to consider the implications of contiguous virtual addresses not being physically contiguous.

If we go down this path, I think that it would make sense to have a new flag to uma_zcreate() that guaranteed physical contiguity for the memory underlying each object. This might mean that we sacrifice fragmentation control for these objects.

Sat, Aug 26, 9:53 PM

Aug 13 2017

jeff added a reviewer for D11801: Discover CPU topology on multi-die AMD Zen systems: jeff.

To create the correct topology on my 16 core 2 die system I had to add the following:

Aug 13 2017, 1:23 AM

Aug 12 2017

jeff added a reviewer for D11784: Dynamically grow the slab size to control wasted memory: jeff.

Three broad comments;

Aug 12 2017, 10:31 PM