| Message ID | 20210825161815.266051-1-alexandru.elisei@arm.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | KVM: arm64: Add Statistical Profiling Extension (SPE) support | expand |
On 25/08/2021 17:17, Alexandru Elisei wrote: > This is v4 of the SPE series posted at [1]. v2 can be found at [2], and the > original series at [3]. > > Statistical Profiling Extension (SPE) is an optional feature added in > ARMv8.2. It allows sampling at regular intervals of the operations executed > by the PE and storing a record of each operation in a memory buffer. A high > level overview of the extension is presented in an article on arm.com [4]. > > This is another complete rewrite of the series, and nothing is set in > stone. If you think of a better way to do things, please suggest it. > > > Features added > ============== > > The rewrite enabled me to add support for several features not > present in the previous iteration: > > - Support for heterogeneous systems, where only some of the CPUs support SPE. > This is accomplished via the KVM_ARM_VCPU_SUPPORTED_CPUS VCPU ioctl. > > - Support for VM migration with the KVM_ARM_VCPU_SPE_CTRL(KVM_ARM_VCPU_SPE_STOP) > VCPU ioctl. > > - The requirement for userspace to mlock() the guest memory has been removed, > and now userspace can make changes to memory contents after the memory is > mapped at stage 2. > > - Better debugging of guest memory pinning by printing a warning when we > get an unexpected read or write fault. This helped me catch several bugs > during development, it has already proven very useful. Many thanks to > James who suggested when reviewing v3. > > > Missing features > ================ > > I've tried to keep the series as small as possible to make it easier to review, > while implementing the core functionality needed for the SPE emulation. As such, > I've chosen to not implement several features: > > - Host profiling a guest which has the SPE feature bit set (see open > questions). > > - No errata workarounds have been implemented yet, and there are quite a few of > them for Neoverse N1 and Neoverse V1. > > - Disabling CONFIG_NUMA_BALANCING is a hack to get KVM SPE to work and I am > investigating other ways to get around automatic numa balancing, like > requiring userspace to disable it via set_mempolicy(). I am also going to > look at how VFIO gets around it. Suggestions welcome. > > - There's plenty of room for optimization. Off the top of my head, using > block mappings at stage 2, batch pinning of pages (similar to what VFIO > does), optimize the way KVM keeps track of pinned pages (using a linked > list triples the memory usage), context-switch the SPE registers on > vcpu_load/vcpu_put on VHE if the host is not profiling, locking > optimizations, etc, etc. > > - ...and others. I'm sure I'm missing at least a few things which are > important for someone. > > > Known issues > ============ > > This is an RFC, so keep in mind that almost definitely there will be scary > bugs. For example, below is a list of known issues which don't affect the > correctness of the emulation, and which I'm planning to fix in a future > iteration: > > - With CONFIG_PROVE_LOCKING=y, lockdep complains about lock contention when > the VCPU executes the dcache clean pending ops. > > - With CONFIG_PROVE_LOCKING=y, KVM will hit a BUG at > kvm_lock_all_vcpus()->mutex_trylock(&vcpu->mutex) with more than 48 > VCPUs. > > This BUG statement can also be triggered with mainline. To reproduce it, > compile kvmtool from this branch [5] and follow the instruction in the > kvmtool commit message. > > One workaround could be to stop trying to lock all VCPUs when locking a > memslot and document the fact that it is required that no VCPUs are run > before the ioctl completes, otherwise bad things might happen to the VM. > > > Open questions > ============== > > 1. Implementing support for host profiling a guest with the SPE feature > means setting the profiling buffer owning regime to EL2. While that is in > effect, PMBIDR_EL1.P will equal 1. This has two consequences: if the guest > probes SPE during this time, the driver will fail; and the guest will be > able to determine when it is profiled. I see two options here: This doesn't mean the EL2 is owning the SPE. It only tells you that a higher level EL is owning the SPE. It could as well be EL3. (e.g, MDCR_EL3.NSPB == 0 or 1). So I think this is architecturally correct, as long as we trap the guest access to other SPE registers and inject and UNDEF. Thanks Suzuki
Hi Suzuki, Thank you for having a look! On 9/22/21 11:11, Suzuki K Poulose wrote: > On 25/08/2021 17:17, Alexandru Elisei wrote: >> This is v4 of the SPE series posted at [1]. v2 can be found at [2], and the >> original series at [3]. >> >> Statistical Profiling Extension (SPE) is an optional feature added in >> ARMv8.2. It allows sampling at regular intervals of the operations executed >> by the PE and storing a record of each operation in a memory buffer. A high >> level overview of the extension is presented in an article on arm.com [4]. >> >> This is another complete rewrite of the series, and nothing is set in >> stone. If you think of a better way to do things, please suggest it. >> >> >> Features added >> ============== >> >> The rewrite enabled me to add support for several features not >> present in the previous iteration: >> >> - Support for heterogeneous systems, where only some of the CPUs support SPE. >> This is accomplished via the KVM_ARM_VCPU_SUPPORTED_CPUS VCPU ioctl. >> >> - Support for VM migration with the KVM_ARM_VCPU_SPE_CTRL(KVM_ARM_VCPU_SPE_STOP) >> VCPU ioctl. >> >> - The requirement for userspace to mlock() the guest memory has been removed, >> and now userspace can make changes to memory contents after the memory is >> mapped at stage 2. >> >> - Better debugging of guest memory pinning by printing a warning when we >> get an unexpected read or write fault. This helped me catch several bugs >> during development, it has already proven very useful. Many thanks to >> James who suggested when reviewing v3. >> >> >> Missing features >> ================ >> >> I've tried to keep the series as small as possible to make it easier to review, >> while implementing the core functionality needed for the SPE emulation. As such, >> I've chosen to not implement several features: >> >> - Host profiling a guest which has the SPE feature bit set (see open >> questions). >> >> - No errata workarounds have been implemented yet, and there are quite a few of >> them for Neoverse N1 and Neoverse V1. >> >> - Disabling CONFIG_NUMA_BALANCING is a hack to get KVM SPE to work and I am >> investigating other ways to get around automatic numa balancing, like >> requiring userspace to disable it via set_mempolicy(). I am also going to >> look at how VFIO gets around it. Suggestions welcome. >> >> - There's plenty of room for optimization. Off the top of my head, using >> block mappings at stage 2, batch pinning of pages (similar to what VFIO >> does), optimize the way KVM keeps track of pinned pages (using a linked >> list triples the memory usage), context-switch the SPE registers on >> vcpu_load/vcpu_put on VHE if the host is not profiling, locking >> optimizations, etc, etc. >> >> - ...and others. I'm sure I'm missing at least a few things which are >> important for someone. >> >> >> Known issues >> ============ >> >> This is an RFC, so keep in mind that almost definitely there will be scary >> bugs. For example, below is a list of known issues which don't affect the >> correctness of the emulation, and which I'm planning to fix in a future >> iteration: >> >> - With CONFIG_PROVE_LOCKING=y, lockdep complains about lock contention when >> the VCPU executes the dcache clean pending ops. >> >> - With CONFIG_PROVE_LOCKING=y, KVM will hit a BUG at >> kvm_lock_all_vcpus()->mutex_trylock(&vcpu->mutex) with more than 48 >> VCPUs. >> >> This BUG statement can also be triggered with mainline. To reproduce it, >> compile kvmtool from this branch [5] and follow the instruction in the >> kvmtool commit message. >> >> One workaround could be to stop trying to lock all VCPUs when locking a >> memslot and document the fact that it is required that no VCPUs are run >> before the ioctl completes, otherwise bad things might happen to the VM. >> >> >> Open questions >> ============== >> >> 1. Implementing support for host profiling a guest with the SPE feature >> means setting the profiling buffer owning regime to EL2. While that is in >> effect, PMBIDR_EL1.P will equal 1. This has two consequences: if the guest >> probes SPE during this time, the driver will fail; and the guest will be >> able to determine when it is profiled. I see two options here: > > This doesn't mean the EL2 is owning the SPE. It only tells you that a > higher level EL is owning the SPE. It could as well be EL3. (e.g, MDCR_EL3.NSPB > == 0 or 1). So I think this is architecturally correct, > as long as we trap the guest access to other SPE registers and inject > and UNDEF. You are right, I was wrong about the part about the guest being able to determine when it is profiled, I forgot that PMBIDR_EL1.P can also be 1 if EL3 is owning the buffer. But I don't understand why you are suggesting that KVM injects UNDEF in this case. I was thinking that when the host is profiling the guest, KVM can store guest writes to the SPE registers to the shadow copy of the registers. On a world switch to guest, KVM will not restore the registers onto the hardware as to not interfere with the profiling operation performed by the host. When profiling ends, KVM can then let the guest use SPE again by restoring the guest register values onto the hardware at every world switch and setting the buffer owning regime to EL1 again. I put this as an open question because if the guest sees PMBIDR_EL1.P set when the guest SPE driver probes, the driver will likely decide that SPE is not available. But the VM has been created with SPE enabled because the host userspace wants the guest to have access to SPE, and the host userspace might not realize that the act of profiling a guest can make SPE unusable by that guest. I am inclined now to let the host's userspace do whatever it wishes and allow it to profile a guest if it so desires, and mention this possible side effect in the KVM documentation, as it might be surprising for someone who isn't familiar with the inner workings of KVM's SPE emulation. Thanks, Alex > > > Thanks > Suzuki
This is v4 of the SPE series posted at [1]. v2 can be found at [2], and the original series at [3]. Statistical Profiling Extension (SPE) is an optional feature added in ARMv8.2. It allows sampling at regular intervals of the operations executed by the PE and storing a record of each operation in a memory buffer. A high level overview of the extension is presented in an article on arm.com [4]. This is another complete rewrite of the series, and nothing is set in stone. If you think of a better way to do things, please suggest it. Features added ============== The rewrite enabled me to add support for several features not present in the previous iteration: - Support for heterogeneous systems, where only some of the CPUs support SPE. This is accomplished via the KVM_ARM_VCPU_SUPPORTED_CPUS VCPU ioctl. - Support for VM migration with the KVM_ARM_VCPU_SPE_CTRL(KVM_ARM_VCPU_SPE_STOP) VCPU ioctl. - The requirement for userspace to mlock() the guest memory has been removed, and now userspace can make changes to memory contents after the memory is mapped at stage 2. - Better debugging of guest memory pinning by printing a warning when we get an unexpected read or write fault. This helped me catch several bugs during development, it has already proven very useful. Many thanks to James who suggested when reviewing v3. Missing features ================ I've tried to keep the series as small as possible to make it easier to review, while implementing the core functionality needed for the SPE emulation. As such, I've chosen to not implement several features: - Host profiling a guest which has the SPE feature bit set (see open questions). - No errata workarounds have been implemented yet, and there are quite a few of them for Neoverse N1 and Neoverse V1. - Disabling CONFIG_NUMA_BALANCING is a hack to get KVM SPE to work and I am investigating other ways to get around automatic numa balancing, like requiring userspace to disable it via set_mempolicy(). I am also going to look at how VFIO gets around it. Suggestions welcome. - There's plenty of room for optimization. Off the top of my head, using block mappings at stage 2, batch pinning of pages (similar to what VFIO does), optimize the way KVM keeps track of pinned pages (using a linked list triples the memory usage), context-switch the SPE registers on vcpu_load/vcpu_put on VHE if the host is not profiling, locking optimizations, etc, etc. - ...and others. I'm sure I'm missing at least a few things which are important for someone. Known issues ============ This is an RFC, so keep in mind that almost definitely there will be scary bugs. For example, below is a list of known issues which don't affect the correctness of the emulation, and which I'm planning to fix in a future iteration: - With CONFIG_PROVE_LOCKING=y, lockdep complains about lock contention when the VCPU executes the dcache clean pending ops. - With CONFIG_PROVE_LOCKING=y, KVM will hit a BUG at kvm_lock_all_vcpus()->mutex_trylock(&vcpu->mutex) with more than 48 VCPUs. This BUG statement can also be triggered with mainline. To reproduce it, compile kvmtool from this branch [5] and follow the instruction in the kvmtool commit message. One workaround could be to stop trying to lock all VCPUs when locking a memslot and document the fact that it is required that no VCPUs are run before the ioctl completes, otherwise bad things might happen to the VM. Open questions ============== 1. Implementing support for host profiling a guest with the SPE feature means setting the profiling buffer owning regime to EL2. While that is in effect, PMBIDR_EL1.P will equal 1. This has two consequences: if the guest probes SPE during this time, the driver will fail; and the guest will be able to determine when it is profiled. I see two options here: - Do not allow the host's userspace to profile a guest where at least one VCPU has SPE. - Document the effects somewhere and let userspace do whatever it likes. No preference for either. 2. Userspace is not allowed to profile a CPU event (not bound to a task) is !perf_allow_cpu(). It is my understanding that this is because of security reasons, as we don't want a task to profile another task. Because a VM will only be able to profile itself, I don't think it's necessary to restrict the VM in any way based on perf_allow_cpu(), like we do with perfmon_capable() and physical timer timestamps. 3. How to handle guest triggered SPE SErrors. Right now the Linux SPE drivers doesn't do anything special SErrors reported by SPE, and I've done the same when a guest manages to trigger one. Should I do something more? Disabling SPE emulation for the entire VM is one option. Summary of the patches ====================== Below is a short summary of the patches. For a more detailed explanation of how SPE works, please see version 3 of the series [1]. Patches 1-11 implement the memslot locking functionality. Patch 12 implements the KVM_ARM_VCPU_SUPPORTED_CPUS ioctl. Patches 13-14 are preparatory for KVM SPE. Patches 15-19 makes it possible for KVM to deny running a SPE enabled VCPU on CPUs which don't have the SPE hardware. Patches 20-22 add the userspace interface to configure SPE. Patches 23-32 implement context switching of the SPE registers. Patch 33 allows a guest to use physical timestamps only if the VMM is perfmon_capable(). Patch 34 add the emulation for the SPE buffer management interrupts. Patches 35-37 add the userspace API to temprorarily stop profiling so memory can be unlocked and the VM migrated. Patch 38 is a hack to get KVM SPE emulation going on NUMA systems (like the Altra server which I used for testing). Patch 39 finally enables SPE. Testing ======= Testing was done on Altra server with two sockets, both populated. The Linux patches are based on v5.14-rc5 and can also be found on gitlab [6]. For testing, I've used an SPE enabled version of kvmtool, which can be found at [7]; the kvmtool patches will also be sent upstream. To test the SPE_STOP API, I used a special version of kvmtool which starts the guest in one of the stopped states; that can be found at [8] (compile from a different commit if a different state and/or transition is desired). Finally, in the VM I used defconfig Linux guest compiled from v5.15-rc5 and some kvm-unit-tests patches which I wrote to test SPE [9]. All tests were run three times: once with VHE enabled, once in NVHE mode (kvm-arm.mode=nvhe) and once in protected mode (kvm-arm.mode=protected). The first test that I ran was the kvm-unit-tests test. This is also the test that I used to check that KVM_ARM_VCPU_SPE_STOP_{TRAP,EXIT,RESUME} works correctly with kvmtool. Then I profiled iperf3 in the guest (16 VCPUs to limit the size of perf.data, 32GiB memory), while concurrently profiling in the host. This is the command that I used: # perf record -ae arm_spe/ts_enable=1,pa_enable=1,pct_enable=1/ -- iperf3 -c 127.0.0.1 -t 30 Everything looked right to me and I didn't see any kernel warnings or bugs. [1] https://lore.kernel.org/linux-arm-kernel/20201027172705.15181-1-alexandru.elisei@arm.com/ [2] https://www.spinics.net/lists/arm-kernel/msg776228.html [3] https://lists.cs.columbia.edu/pipermail/kvmarm/2019-February/034887.html [4] https://community.arm.com/developer/ip-products/processors/b/processors-ip-blog/posts/statistical-profiling-extension-for-armv8-a [5] https://gitlab.arm.com/linux-arm/kvmtool-ae/-/tree/vgic-lock-all-vcpus-lockdep-bug-v1 [6] https://gitlab.arm.com/linux-arm/linux-ae/-/tree/kvm-spe-v4 [7] https://gitlab.arm.com/linux-arm/kvmtool-ae/-/tree/kvm-spe-v4 [8] https://gitlab.arm.com/linux-arm/kvmtool-ae/-/tree/kvm-spe-v4-spe-stop-tests [9] https://gitlab.arm.com/linux-arm/kvm-unit-tests-ae/-/tree/kvm-spe-v4 Alexandru Elisei (35): KVM: arm64: Make lock_all_vcpus() available to the rest of KVM KVM: arm64: Add lock/unlock memslot user API KVM: arm64: Implement the memslot lock/unlock functionality KVM: arm64: Defer CMOs for locked memslots until a VCPU is run KVM: arm64: Perform CMOs on locked memslots when userspace resets VCPUs KVM: arm64: Delay tag scrubbing for locked memslots until a VCPU runs KVM: arm64: Unlock memslots after stage 2 tables are freed KVM: arm64: Deny changes to locked memslots KVM: Add kvm_warn{,_ratelimited} macros KVM: arm64: Print a warning for unexpected faults on locked memslots KVM: arm64: Allow userspace to lock and unlock memslots KVM: arm64: Add the KVM_ARM_VCPU_SUPPORTED_CPUS VCPU ioctl KVM: arm64: Add CONFIG_KVM_ARM_SPE Kconfig option KVM: arm64: Add SPE capability and VCPU feature drivers/perf: Expose the cpumask of CPUs that support SPE KVM: arm64: Make SPE available when at least one CPU supports it KVM: arm64: Set the VCPU SPE feature bit when SPE is available KVM: arm64: Expose SPE version to guests KVM: arm64: Do not emulate SPE on CPUs which don't have SPE KVM: arm64: debug: Configure MDCR_EL2 when a VCPU has SPE KVM: arm64: Move the write to MDCR_EL2 out of __activate_traps_common() KVM: arm64: VHE: Change MDCR_EL2 at world switch if VCPU has SPE KVM: arm64: Add SPE system registers to VCPU context KVM: arm64: nVHE: Save PMSCR_EL1 to the host context KVM: arm64: Rename DEBUG_STATE_SAVE_SPE -> DEBUG_SAVE_SPE_BUFFER flags KVM: arm64: nVHE: Context switch SPE state if VCPU has SPE KVM: arm64: VHE: Context switch SPE state if VCPU has SPE KVM: arm64: Save/restore PMSNEVFR_EL1 on VCPU put/load KVM: arm64: Allow guest to use physical timestamps if perfmon_capable() KVM: arm64: Emulate SPE buffer management interrupt KVM: arm64: Add an userspace API to stop a VCPU profiling KVM: arm64: Implement userspace API to stop a VCPU profiling KVM: arm64: Add PMSIDR_EL1 to the SPE register context KVM: arm64: Make CONFIG_KVM_ARM_SPE depend on !CONFIG_NUMA_BALANCING KVM: arm64: Allow userspace to enable SPE for guests Sudeep Holla (4): KVM: arm64: Add a new VCPU device control group for SPE KVM: arm64: Add SPE VCPU device attribute to set the interrupt number KVM: arm64: Add SPE VCPU device attribute to initialize SPE KVM: arm64: VHE: Clear MDCR_EL2.E2PB in vcpu_put() Documentation/virt/kvm/api.rst | 87 ++++- Documentation/virt/kvm/devices/vcpu.rst | 76 ++++ arch/arm64/include/asm/kvm_arm.h | 1 + arch/arm64/include/asm/kvm_host.h | 75 +++- arch/arm64/include/asm/kvm_hyp.h | 51 ++- arch/arm64/include/asm/kvm_mmu.h | 8 + arch/arm64/include/asm/kvm_spe.h | 96 ++++++ arch/arm64/include/asm/sysreg.h | 3 + arch/arm64/include/uapi/asm/kvm.h | 11 + arch/arm64/kvm/Kconfig | 8 + arch/arm64/kvm/Makefile | 1 + arch/arm64/kvm/arm.c | 140 +++++++- arch/arm64/kvm/debug.c | 55 ++- arch/arm64/kvm/guest.c | 10 + arch/arm64/kvm/hyp/include/hyp/spe-sr.h | 32 ++ arch/arm64/kvm/hyp/include/hyp/switch.h | 1 - arch/arm64/kvm/hyp/nvhe/Makefile | 1 + arch/arm64/kvm/hyp/nvhe/debug-sr.c | 24 +- arch/arm64/kvm/hyp/nvhe/spe-sr.c | 133 +++++++ arch/arm64/kvm/hyp/nvhe/switch.c | 35 +- arch/arm64/kvm/hyp/vhe/Makefile | 1 + arch/arm64/kvm/hyp/vhe/spe-sr.c | 193 +++++++++++ arch/arm64/kvm/hyp/vhe/switch.c | 43 ++- arch/arm64/kvm/hyp/vhe/sysreg-sr.c | 2 +- arch/arm64/kvm/mmu.c | 441 +++++++++++++++++++++++- arch/arm64/kvm/reset.c | 23 ++ arch/arm64/kvm/spe.c | 381 ++++++++++++++++++++ arch/arm64/kvm/sys_regs.c | 77 ++++- arch/arm64/kvm/vgic/vgic-init.c | 4 +- arch/arm64/kvm/vgic/vgic-its.c | 8 +- arch/arm64/kvm/vgic/vgic-kvm-device.c | 50 +-- arch/arm64/kvm/vgic/vgic.h | 3 - drivers/perf/arm_spe_pmu.c | 30 +- include/linux/kvm_host.h | 4 + include/linux/perf/arm_pmu.h | 7 + include/uapi/linux/kvm.h | 13 + 36 files changed, 1983 insertions(+), 145 deletions(-) create mode 100644 arch/arm64/include/asm/kvm_spe.h create mode 100644 arch/arm64/kvm/hyp/include/hyp/spe-sr.h create mode 100644 arch/arm64/kvm/hyp/nvhe/spe-sr.c create mode 100644 arch/arm64/kvm/hyp/vhe/spe-sr.c create mode 100644 arch/arm64/kvm/spe.c