jeff (Jeffrey Roberson)
User

Projects

User does not belong to any projects.

User Details

User Since
Aug 6 2017, 12:45 AM (15 w, 23 h)

Recent Activity

Yesterday

jeff committed rS325992: Use a vmem per-domain kernel vmem to ensure that reservations and domain.
Use a vmem per-domain kernel vmem to ensure that reservations and domain
Sun, Nov 19, 3:18 AM

Mon, Nov 13

jeff committed rS325784: Use a soft limit for kmem implemented within uma. Part of r325754.
Use a soft limit for kmem implemented within uma. Part of r325754
Mon, Nov 13, 11:33 PM
jeff committed rS325754: Eliminate kmem_arena to simplify the kmem_ api for forthcoming NUMA support.
Eliminate kmem_arena to simplify the kmem_ api for forthcoming NUMA support
Mon, Nov 13, 3:42 AM
jeff committed rS325752: Make a staging branch for numa patches.
Make a staging branch for numa patches
Mon, Nov 13, 3:38 AM
jeff committed rS325753: Move NUMA policy iterators into the page allocator layer.
Move NUMA policy iterators into the page allocator layer
Mon, Nov 13, 3:35 AM
jeff committed rS325751: Make a directory for my projects.
Make a directory for my projects
Mon, Nov 13, 3:25 AM

Sat, Nov 11

jeff added inline comments to D12873: Move cpuset's parselist function into libutil.
Sat, Nov 11, 8:55 PM
jeff added inline comments to D13014: Move NUMA policy iterators into the page allocator layer.
Sat, Nov 11, 8:50 PM

Thu, Nov 9

jeff added inline comments to D13014: Move NUMA policy iterators into the page allocator layer.
Thu, Nov 9, 5:26 AM
jeff added a comment to D13014: Move NUMA policy iterators into the page allocator layer.

I forgot to mention; This partially came from markj's updated numa2 branch which came from my original numa branch. I then refactored slightly and committed this version. I'm next going to figure out the kmem_ layer and then pick up UMA from the numa2 branch. It probably will be difficult to merge those two and I may abandon the numa branch if I can keep doing patches this small on head.

Thu, Nov 9, 5:22 AM
jeff created D13014: Move NUMA policy iterators into the page allocator layer.
Thu, Nov 9, 5:19 AM

Wed, Nov 8

jeff committed rS325530: Replace manyinstances of VM_WAIT with blocking page allocation flags.
Replace manyinstances of VM_WAIT with blocking page allocation flags
Wed, Nov 8, 2:55 AM

Tue, Oct 31

jeff added inline comments to D12838: Replace many instances of VM_WAIT with blocking page allocation functions.
Tue, Oct 31, 11:37 PM

Mon, Oct 30

jeff created D12838: Replace many instances of VM_WAIT with blocking page allocation functions.
Mon, Oct 30, 11:17 PM

Aug 28 2017

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.

Aug 28 2017, 2:59 AM

Aug 26 2017

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?

Aug 26 2017, 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.

Aug 26 2017, 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