On Fri, 21 Feb 2025 17:21:46 +0100 Sebastian Andrzej Siewior <[email protected]> wrote:
> On 2025-02-20 11:54:00 [-0500], Steven Rostedt wrote: > > > kworker/0:5-88 [000] D.Zf2 211.983391: igbvf_reset_L14: > > > (igbvf_reset+0x62/0x120 [igbvf]) > > > kworker/0:5-88 [000] .N... 211.983396: igbvf_reset_L16: > > > (igbvf_reset+0x7b/0x120 [igbvf]) > … > > > (Except in line 6 where it show 'f', which seems to be an unrelated > > > issue) > > > > But it appears to be consistent. > > > > Let me go and add some probes and see what they show. The output you gave > > me that I requested didn't show anything unusual. > > Since I just noticed that you posted the patches, did you find anything? > From looking at it, if we preemption at 0, migrate disable at 3 then a > substraction of one would underflow the preemption counter to f and > migrate disable to 2. But based on code inspection mg should be at 1. > And this does not explain where D & Z is coming from. I haven't found the bug for this yet, but I need to send out other fixes. The preempt count issue you posted is agnostic to this issue, so I just took it and added it to my current queue for this rc release. -- Steve
