From: Eric Botcazou <ebotcazou@xxxxxxxxxxxxxx> Date: Fri, 23 Jun 2017 19:50:17 +0200 >> Thanks for the explanation. We're now just disabling cilk on Debian >> for gcc-7 and gcc-snapshot for the time being. > > I just fixed the problem on the mainline and 7 branch: > https://gcc.gnu.org/ml/gcc-patches/2017-06/msg01818.html > > The testsuite now flawlessly passes at -j32 for me. > >> Ok, I hope that will help the kernel guys to figure out what the >> problem is. > > libcilkrts does low-level fiddling with the stack pointer and threads, and > overlooked the stack bias in 64-bit mode. This resulted in reproducible hangs > for some tests of the testsuite and I guess that a high level of parallelism > led to a form of congestion somewhere; I'm afraid I can't say more. Unaligned stack accesses, which is what I can gather was going on looking at your fix, could point to a myriad of areas where bugs could live. If somehow the stack pointer itself got misaligned, we could be looking at a window spill/fill bug or similar. I'll try to see if I can figure out what might have gone wrong. It could simply be a bug in the unaligned memory access trap handler. -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
