> On Nov. 18, 2014, 10:43 p.m., Gabe Black wrote: > > util/m5/m5ops.h, line 57 > > <http://reviews.gem5.org/r/2313/diff/4/?file=42082#file42082line57> > > > > Why do we need psuedo ops for syscalls when there are actual syscall > > instructions? The same goes for page faults. I'm not saying I know that we > > don't, I'm just surprised that that would be necessary. > > > > It seems unlikely that a program is going to realize it's about to > > cause a pagefault and obligingly call a special instruction first. > > Andreas Sandberg wrote: > The problem is that KVM doesn't normally intercept syscalls. This means > that they need a small syscall handler and fault handler stub to support SE > mode in KVM (see RB #2322). There are other ways of communicating with the > host (e.g., IO ports, hypervisor calls). The beauty of this solution is that > it is supported seamlessly by the simulated CPUs, which makes it possible to > do CPU switching reliably (we need to handle situation where the switch takes > place while the stub code is executing). It should be fairly straight forward > to extend this to simulated CPUs, making mixed FS/SE mode simulation possible.
Initially I have implemented syscall and pagefault intercepts using IO ports. This implementation imposes all CPU models to be aware of the magic ports KvmCPU is using to trap syscalls and page faults. Andreas suggested the MMIO interface as it is already commun to all CPU models. - Alexandru ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: http://reviews.gem5.org/r/2313/#review5480 ----------------------------------------------------------- On Sept. 30, 2014, 4:21 p.m., Alexandru Dutu wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > http://reviews.gem5.org/r/2313/ > ----------------------------------------------------------- > > (Updated Sept. 30, 2014, 4:21 p.m.) > > > Review request for Default. > > > Repository: gem5 > > > Description > ------- > > Changeset 10421:d4162e1b1c56 > --------------------------- > kvm, x86: Adding support for SE mode execution > This patch adds methods in KvmCPU model to handle KVM exits caused by syscall > instructions and page faults. These types of exits will be encountered if > KvmCPU is run in SE mode. > > > Diffs > ----- > > src/SConscript 28b31101d9e6e5e75d04448451986d6318383f3c > src/arch/SConscript 28b31101d9e6e5e75d04448451986d6318383f3c > src/arch/alpha/SConscript 28b31101d9e6e5e75d04448451986d6318383f3c > src/arch/arm/SConscript 28b31101d9e6e5e75d04448451986d6318383f3c > src/arch/arm/pseudo_inst.hh PRE-CREATION > src/arch/arm/pseudo_inst.cc PRE-CREATION > src/arch/mips/SConscript 28b31101d9e6e5e75d04448451986d6318383f3c > src/arch/power/SConscript 28b31101d9e6e5e75d04448451986d6318383f3c > src/arch/sparc/SConscript 28b31101d9e6e5e75d04448451986d6318383f3c > src/arch/x86/SConscript 28b31101d9e6e5e75d04448451986d6318383f3c > src/arch/x86/pseudo_inst.hh PRE-CREATION > src/arch/x86/pseudo_inst.cc PRE-CREATION > src/arch/x86/tlb.cc 28b31101d9e6e5e75d04448451986d6318383f3c > src/arch/x86/utility.cc 28b31101d9e6e5e75d04448451986d6318383f3c > src/cpu/kvm/base.cc 28b31101d9e6e5e75d04448451986d6318383f3c > src/sim/pseudo_inst.cc 28b31101d9e6e5e75d04448451986d6318383f3c > src/sim/system.cc 28b31101d9e6e5e75d04448451986d6318383f3c > util/m5/m5ops.h 28b31101d9e6e5e75d04448451986d6318383f3c > > Diff: http://reviews.gem5.org/r/2313/diff/ > > > Testing > ------- > > Quick regressions passed. > > > Thanks, > > Alexandru Dutu > > _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
