Messages in this thread | | | Date | Fri, 01 May 2015 12:22:43 -0400 | From | Rik van Riel <> | Subject | Re: [PATCH 3/3] context_tracking,x86: remove extraneous irq disable & enable from context tracking on syscall entry |
| |
On 05/01/2015 12:03 PM, Andy Lutomirski wrote:
> The last time I asked, the impression I got was that we needed two things: > > 1. We can't pluck things from the RCU list without knowing whether the > CPU is in an RCU read-side critical section, and we can't know that > unless we have regular grade periods or we know that the CPU is idle. > To make the CPU detectably idle, we need to set a bit somewhere.
More than that. We also need a way for another CPU to identify the callbacks they could run for us, without confusing them with new callbacks queued after we transitioned back from USER to KERNEL context.
> 2. To suppress the timing tick, we need to get some timing for, um, > the scheduler? I wasn't really sure about this one. > > Could we reduce the overhead by making the IN_USER vs IN_KERNEL > indication be a single bit and, worst case, an rdtsc and maybe a > subtraction? We could probably get away with banning full nohz on > non-invariant tsc systems.
I suspect we can.
There is no need to update the vtime every single time we enter vtime_user_enter and functions like it.
We can keep a buffer, which: 1) keeps values in TSC cycles (or whatever unit local_clock does) 2) is only ever accessed by the current task, so it requires no locking 3) values can be actually folded into vtime periodically, when they exceed a certain threshold (1 second ?)
That means the vtime_seqlock is something that we would only take once a second or so, and the calculations in account_user_time would only be done on a fairly infrequent basis. That has the potential to reduce overhead by a lot.
If nobody has any objections to that kind of change, I would be happy to implement it.
-- All rights reversed
| |