пет, 22. мај 2020. у 05:12 Richard Henderson <[email protected]> је написао/ла: > > On 5/21/20 4:32 AM, Ahmed Karaman wrote: > > Does this mean that there is a bug > > in this function for the other targets? > > Yes, I think so. > > > That we have to do the page crossings check in both modes to avoid the > > user-mode failures that you have mentioned above? > > Well, that or we need to fix linux-user/mmap.c to do all the invalidations > required. >
Hi, Mr. Richard, :) Many thanks for diagnosis, and, more than this, presenting two alternatives for resolution! That mmap()... It has been giving us the hardest time since forever, and it looks it continues to torture us. It looks we are now between a rock and a hard place. I slightly prefer the later (fixing mmap.c) alternative, just from the gut feeling that it is better to fix the problem at its source, rather than to apply (easy, but possibly performance-wise costly) band-aid to some later consequence. But it seem to me that this is as if I say I choose Charybdis, between Scylla and Charybdis. We also risk jumping out of the frying pan into the fire. Adding Laurent, since this is, in essence, a linux-user issue. If nobody objects, I will instruct Ahmed to file a bug in QEMU Bugzilla. Wishing to improve the performance, we found the hard bug... eh... Thanks again to Richard, Aleksandar > > r~ >
