Thomas Gleixner <tglx@xxxxxxxxxxxxx> writes: > On Tue, 18 Sep 2018, Eric W. Biederman wrote: >> I have been slowly going thought and reworking the arch specific >> functions that generate siginfo. The problems I have been addressing >> is that using siginfo directly is error prone. Using siginfo directly >> makes it easy to leave fields initialized, and get confused about >> which fields need to be filled in. >> >> To address this I have added a series of helper functions to >> kernel/signal.c, that are specific to exactly one use of struct siginfo >> and take the parameters they need. >> >> To use these functions the x86 signal handling needs some cleanups but >> the net result appears to be less code that is easier to follow. >> >> If while looking over these patches you see anything please let me know. > > Only nitpicks. > >> I don't think I missed something but to err is human. > > I went through the changes a couple of times, but failed to spot > something. Was pleasure to read that set! > >> Likewise if you would like to merge these patches via the tip tree >> let me know. Otherwise after the review is complete I plan on merging >> these into my siginfo tree. At this point I believe all of the >> prerequisite patches are merged so it should not make a difference. > > Works either way. Ingo? If I manage to get through all of the architecture specific code I can shrink the in-kernel version of struct siginfo. So there is a slight advantage to having it all in my tree. But worst case I just have to wait another cycle which doesn't look like a particularly long wait at this point. Eric
