On 11/12/15 18:38, Richard Weinberger wrote: > Am 11.12.2015 um 12:24 schrieb Anton Ivanov: >> On 11/12/15 08:16, Richard Weinberger wrote: >>> Am 11.12.2015 um 07:58 schrieb Anton Ivanov: >>>>>> 2. I cannot catch what is wrong with the current code in signal.c. When >>>>>> I read it, it should not produce re-entrancy. But it does. >>>>> Sorry for the delay. Until now I did not find the time to dig into that. >>>>> Did you find the offending code in signal.c? >>>> Yes. >>>> >>>> Unblock signals is logically incorrect - it will re-trigger an >>>> interrupts even if there is an interrupt in flight whose processing has >>>> not been finished. >>>> >>>> I tried several approaches both with the original poll() controller and >>>> with my epoll() based version, some show promise. >>>> >>>> I had to put it aside until next Friday as I have some stuff due at work >>>> so I cannot spare time to work on it until then. Once I get that out of >>>> the way I should be able to spare it a day or two which should be enough >>>> to finish it. >>>> >>>> Ditto for the UBD improvements. >>> One thing we have to consider is that's legit to have SIGIO nested. >> Correct. That is considered :) >> >> Both when looking at poll() and epoll() >> >> However, it is not legit to have sigio on a specific fd nested. That is >> mostly safe for the poll() version, but will need to be accounted for in any >> surgery on the irq controller. > So, the current code is fine unless you switch to epoll()?
Correct. It looks OK, though I have not looked at it in depth to make sure there is no race somewhere. > Is it because you use epoll() in edge-triggered mode? For now it is level which means it should be identical to poll(). I want to be able (long term) to use either - have the user set what they want for a particular IRQ. There are a couple of use cases like packet mmap and can-write IRQs where edge allows for better code. I have a working POC which sets per - fd flag to avoid reentrancy on a per-fd basis so it behaves identically to the poll one - similar to the "reactivate fd" semantics. What I see with it however is some weird stalls on ubd and I am not 100% sure that they are epoll specific, it just tends to make them easily reproducible as it is faster. I will pick it Monday-Tuesday the week before Xmas after I get some work stuff out of the way. A. > > Thanks, > //richard > ------------------------------------------------------------------------------ _______________________________________________ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel