SMT being hyperthreading, right? Do the default conf files in ./config/ have SMT enabled? If not, then yes, I have seen this in non-SMT configurations. Though I typically see this problem only at high core counts. I rarely see it happens at 8 cores or less.
On Fri, Dec 6, 2013 at 9:40 AM, <[email protected]> wrote: > I'm experiencing this problem with newer, untested versions of qemu as > well. I've only seen it happen with SMT configurations, however -- have > either of you seen it without SMT? > > Tyler > > > We sometimes had this happen too, but I think no one has found a reliable > > way to figure out where the problem happens. > > > > When you say 'never', how long have you waited for it? In my experience > > sometimes when this happens, it can stay in kernel mode a long long time > > but then eventually come back to user space. > > > > > > On Wed, Dec 4, 2013 at 10:27 PM, Hongsun An <[email protected]> wrote: > > > >> Hello, > >> > >> I'm using marss86+dramsim2 > >> I made a checkpoint of sphinx3 after fastfowarding 10 billions > >> instructions > >> and simulated the checkpoint. > >> > >> here is the middle of the log > >> > >> Completed 1060261000 cycles, 415478808 commits: 940320 Hz, > >> 332565 insns/sec: rip 000000000041e5cf > >> Completed 1060450000 cycles, 415546041 commits: 940386 Hz, > >> 334523 insns/sec: rip 000000000041e5cf > >> Completed 1060633000 cycles, 415610543 commits: 912773 Hz, > >> 321725 insns/sec: rip 0000000000402a25 > >> Completed 1060811000 cycles, 415672043 commits: 889292 Hz, > >> 307255 insns/sec: rip 0000000000415c42 > >> Completed 1061031000 cycles, 415712907 commits: 1099278 Hz, > >> 204185 insns/sec: rip ffffffff8105f411 > >> Completed 1061175000 cycles, 415825905 commits: 715988 Hz, > >> 561841 insns/sec: rip ffffffff810c043a > >> Completed 1061237000 cycles, 416011905 commits: 307139 Hz, > >> 921417 insns/sec: rip ffffffff810c043a > >> Completed 1061299000 cycles, 416197905 commits: 306858 Hz, > >> 920576 insns/sec: rip ffffffff810c043a > >> Completed 1061361000 cycles, 416383905 commits: 306683 Hz, > >> 920050 insns/sec: rip ffffffff810c043a > >> ....... > >> ....... > >> user.base_machine.ooo_0_0.thread0.commit.ipc = 0.391764 > >> kernel.base_machine.ooo_0_0.thread0.commit.ipc = 2.99783 > >> total.base_machine.ooo_0_0.thread0.commit.ipc = 2.29421 > >> > >> At some point, rip is changed to ffffffff810c043a and never turns back > >> to > >> 0x000000000xxxxx. > >> Every benchmark has simliar result. > >> > >> I simulated soplex, libquantum with test input and they worked well. > >> What I was missing.. > >> > >> > >> > >> _______________________________________________ > >> http://www.marss86.org > >> Marss86-Devel mailing list > >> [email protected] > >> https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel > >> > >> > > _______________________________________________ > > http://www.marss86.org > > Marss86-Devel mailing list > > [email protected] > > https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel > > > > >
_______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
