Change Details

Wire page containing the sigfastblock word, and access it through direct map or sf instead of performing fueword(9) or casueword(9). I do not like two things in this approach, this is why I did not implemented it that way. First, if the page gets unmapped or remapped, kernel does not catch it automatically. Right now I only handle fork(2) for CoW, where the page is recalculated. I might add some code to vm_map_delete() which would ultimately send a signal, before the patch is considered finished. Second, the fact that the page is wired adds one wired page per thread. Anyway, this should eliminate fueword(2) and CLAC/STAC around accesses, which are indeed serializing on some modern CPUs (undocumented).

Only read it on the attempt to deliver signal in ast(). If the word is set, signal is not delivered and word is kept, preventing interruption of interruptible sleeps by signals until userspace calls sigfastblock(UNBLOCK). This way, the only spurious EINTR that userspace can see while in critical section is on first interruptible sleep, if a signal is pending. It is believed that it is not important for rtld and lbithr critical sections. It might be visible for the application code e.g. for the callback of dl_iterate_phdr(3), but again the belief is that the non-compliance is acceptable. For now I added the knob to enable the word read on syscall entry to be able to diagnose possible issues due to spurious EINTR. [This is not tested]

Wire page containing theOnly read it on the attempt to deliver signal in ast(). If the word is set, sigfastblocknal is not delivered and word is kept, and access it through direct map or sf instead of performing fueword(9) or casueword(9preventing interruption of interruptible sleeps by signals until userspace calls sigfastblock(UNBLOCK). I do not like two things in tThis approachway, this is why I did not implemented it that way. Firstthe only spurious EINTR that userspace can see while in critical section is on first interruptible sleep, if the page gets unmapped or remapped, kernel doeif a signal is pending. It is believed that it is not catch it automaimportant for rtld and lbithr critically sections. Right now I only handle fork(2) for CoW, where the page is recalculated. I might add some code to vm_map_delete() which would ultimately send a signalIt might be visible for the application code e.g. for the callback of dl_iterate_phdr(3), beforebut again the patchbelief is considered finishedthat the non-compliance is acceptable. Second,For now I added the fact thatknob to enable the page is wired adds one wired page per threadword read on syscall entry to be able to diagnose possible issues due to spurious EINTR. Anyway, this should eliminate fueword(2) and CLAC/STAC around accesses, which are indeed serializing on some modern CPUs (undocumented).[This is not tested]