> 在 2018年11月29日，上午3:14，Paul Burton <pburton@xxxxxxxxxxxx> 写道： > > Hi Yunqiang, > > Please send only plain text mail to the linux-mips mailing list, and > don't top-post. > > On Wed, Nov 28, 2018 at 08:42:08AM -0800, Yunqiang Su wrote: >>> 在 2018年11月15日，下午3:53，Huacai Chen <chenhc@xxxxxxxxxx> 写道： >>> >>> On the Loongson-2G/2H/3A/3B there is a hardware flaw that ll/sc and >>> lld/scd is very weak ordering. We should add sync instructions before >>> each ll/lld and after the last sc/scd to workaround. Otherwise, this >>> flaw will cause deadlock occationally (e.g. when doing heavy load test >>> with LTP). >>> >>> This patch is not a minimal change (it is very difficult to make a >>> minimal change), but it is a safest change. >>> >>> Why disable fix-loongson3-llsc in compiler? >>> Because compiler fix will cause problems in kernel's .fixup section. >> >> @Paul Burton Loongson 3A/B 3000 have a bug of ll/sc, which is not in 3A/3B 1000 >> even. > > Do you have any details? Are you saying that the problem Huacai's patch > addresses only applies to those models, or that there's a separate > problem? Yes. 3000 introduces some new bugs, this is one of them. 1000 doesn’t have this problem. Without this patch, 3A 3000 is quite unstable when heavy load, as an personal experience, building kernel, with -j5 works well, while -j9 always makes the machine hangs. 3A 1000 has itself bugs, not the same with 3A 3000, and some of 1000’s bug is fixed in 3000. > > Thanks, > Paul
