MIME-Version: 1.0 In-Reply-To: X-Mailer: MIME-tools 5.505 (Entity 5.505) Content-Disposition: inline X-RT-Interface: Web References: Content-Type: text/plain; charset="utf-8" Message-ID: Content-Transfer-Encoding: binary X-RT-Original-Encoding: utf-8 RT-Send-CC: X-RT-Encrypt: 0 X-RT-Sign: 0 Content-Length: 780 On Wed Apr 20 11:02:30 2016, muks wrote: > On Wed Apr 20 10:00:21 2016, muks wrote: > > There's a race in the adaptive rwlock implementation in read and > > update of rwl->spins. > > The new code seems to use ISC_PLATFORM_BUSYWAITNOP which seems to be > defined as "rep; nop" for x86 and x86_64. > > rep is an instruction prefix that repeats the next instruction ecx > times, decrementing ecx at each step. So ecx ought to be set to some > deterministic value for it (what is its value here?). Also it clobbers > ecx here.. is that fine? Francis pointed out that the rep prefix does not apply to the nop instruction - the opcode sequence is used as the "pause" instruction. So this part of the bug report is bogus. The part about the race condition in rwl->spins still exists.