From: Guo Ren <guoren@xxxxxxxxxxxxxxxxx> Hello everyone, I'm back after one year. This is the tenth version of riscv qspinlock. patch[1 - 8]: Native qspinlock patch[9 -17]: Paravirt qspinlock patch[18-19]: Compact NUMA-awared (CNA) qspinlock This series based on Andrew Jones' pv-time patches and Alex Kogan's CNA qspinlock patches. I merge them into sg2042-master branch, then you could directly try it: https://github.com/guoren83/linux/tree/qspinlock_v10_pvlock_cna_qspinlock_v15 Use sophgo_mango_ubuntu_defconfig for sg2042 64 cores board. Native qspinlock ================ This time we've proved the qspinlock on th1520 [1] & sg2042 [2], which gives stability and performance improvement. All T-HEAD processors have a strong LR/SC forward progress guarantee than the requirements of the ISA, which could satisfy the xchg_tail of native_qspinlock. Now, qspinlock has been run with us for more than 1 year, and we have enough confidence to enable it for all the T-HEAD processors. Of causes, we found a livelock problem with the qspinlock lock torture test from the CPU store merge buffer delay mechanism, which caused the queued spinlock becomes a dead ring and RCU warning to come out. We introduce a custom WRITE_ONCE to solve this. Do we need explicit ISA instruction to signal it? Or let hardware handle this. We've tested the patch on SOPHGO sg2042 & th1520 and passed the stress test on Fedora & Ubuntu & OpenEuler ... Here is the performance comparison between qspinlock and ticket_lock on sg2042 (64 cores): sysbench test=threads threads=32 yields=100 lock=8 (+13.8%): queued_spinlock 0.5109/0.00 ticket_spinlock 0.5814/0.00 perf futex/hash (+6.7%): queued_spinlock 1444393 operations/sec (+- 0.09%) ticket_spinlock 1353215 operations/sec (+- 0.15%) perf futex/wake-parallel (+8.6%): queued_spinlock (waking 1/64 threads) in 0.0253 ms (+-2.90%) ticket_spinlock (waking 1/64 threads) in 0.0275 ms (+-3.12%) perf futex/requeue (+4.2%): queued_spinlock Requeued 64 of 64 threads in 0.0785 ms (+-0.55%) ticket_spinlock Requeued 64 of 64 threads in 0.0818 ms (+-4.12%) System Benchmarks (+6.4%) queued_spinlock: System Benchmarks Index Values BASELINE RESULT INDEX Dhrystone 2 using register variables 116700.0 628613745.4 53865.8 Double-Precision Whetstone 55.0 182422.8 33167.8 Execl Throughput 43.0 13116.6 3050.4 File Copy 1024 bufsize 2000 maxblocks 3960.0 7762306.2 19601.8 File Copy 256 bufsize 500 maxblocks 1655.0 3417556.8 20649.9 File Copy 4096 bufsize 8000 maxblocks 5800.0 7427995.7 12806.9 Pipe Throughput 12440.0 23058600.5 18535.9 Pipe-based Context Switching 4000.0 2835617.7 7089.0 Process Creation 126.0 12537.3 995.0 Shell Scripts (1 concurrent) 42.4 57057.4 13456.9 Shell Scripts (8 concurrent) 6.0 7367.1 12278.5 System Call Overhead 15000.0 33308301.3 22205.5 ======== System Benchmarks Index Score 12426.1 ticket_spinlock: System Benchmarks Index Values BASELINE RESULT INDEX Dhrystone 2 using register variables 116700.0 626541701.9 53688.2 Double-Precision Whetstone 55.0 181921.0 33076.5 Execl Throughput 43.0 12625.1 2936.1 File Copy 1024 bufsize 2000 maxblocks 3960.0 6553792.9 16550.0 File Copy 256 bufsize 500 maxblocks 1655.0 3189231.6 19270.3 File Copy 4096 bufsize 8000 maxblocks 5800.0 7221277.0 12450.5 Pipe Throughput 12440.0 20594018.7 16554.7 Pipe-based Context Switching 4000.0 2571117.7 6427.8 Process Creation 126.0 10798.4 857.0 Shell Scripts (1 concurrent) 42.4 57227.5 13497.1 Shell Scripts (8 concurrent) 6.0 7329.2 12215.3 System Call Overhead 15000.0 30766778.4 20511.2 ======== System Benchmarks Index Score 11670.7 The qspinlock has a significant improvement on SOPHGO SG2042 64 cores platform than the ticket_lock. Paravirt qspinlock ================== Based on Andrew Jones' "Add skeleton for pv-time support" patches to unify the paravirt framework. We implemented kvm_kick_cpu/kvm_wait_cpu and add tracepoints to observe the behaviors. Also, introduce a new SBI extension SBI_EXT_PVLOCK (0xAB0401). If the name and number are approved, I will send a formal proposal to the SBI spec. Compact NUMA-awared (CNA) qspinlock =================================== Based on Alex Kogan's "Add NUMA-awareness to qspinlock" patches. The multi sg2042 chips could compose 2/4 NUMA nodes, and these pathes are the preparation for the next test. I hope "CNA v.s. native" could have a better improvement than "qspinlock v.s. ticket_lock." We tested it on sg2042 hardware with "numa_spinlock=on" to ensure the software is bug-free before the next multi-nodes hardware comes out. Changlog: V10: - Using an alternative framework instead of static_key_branch in the asm/spinlock.h. - Fixup store merge buffer problem, which causes qspinlock lock torture test livelock. - Add paravirt qspinlock support, include KVM backend - Add Compact NUMA-awared qspinlock support V9: https://lore.kernel.org/linux-riscv/20220808071318.3335746-1-guoren@xxxxxxxxxx/ - Cleanup generic ticket-lock code, (Using smp_mb__after_spinlock as RCsc) - Add qspinlock and combo-lock for riscv - Add qspinlock to openrisc - Use generic header in csky - Optimize cmpxchg & atomic code V8: https://lore.kernel.org/linux-riscv/20220724122517.1019187-1-guoren@xxxxxxxxxx/ - Coding convention ticket fixup - Move combo spinlock into riscv and simply asm-generic/spinlock.h - Fixup xchg16 with wrong return value - Add csky qspinlock - Add combo & qspinlock & ticket-lock comparison - Clean up unnecessary riscv acquire and release definitions - Enable ARCH_INLINE_READ*/WRITE*/SPIN* for riscv & csky V7: https://lore.kernel.org/linux-riscv/20220628081946.1999419-1-guoren@xxxxxxxxxx/ - Add combo spinlock (ticket & queued) support - Rename ticket_spinlock.h - Remove unnecessary atomic_read in ticket_spin_value_unlocked V6: https://lore.kernel.org/linux-riscv/20220621144920.2945595-1-guoren@xxxxxxxxxx/ - Fixup Clang compile problem Reported-by: kernel test robot - Cleanup asm-generic/spinlock.h - Remove changelog in patch main comment part, suggested by Conor.Dooley - Remove "default y if NUMA" in Kconfig V5: https://lore.kernel.org/linux-riscv/20220620155404.1968739-1-guoren@xxxxxxxxxx/ - Update comment with RISC-V forward guarantee feature. - Back to V3 direction and optimize asm code. V4: https://lore.kernel.org/linux-riscv/1616868399-82848-4-git-send-email-guoren@xxxxxxxxxx/ - Remove custom sub-word xchg implementation - Add ARCH_USE_QUEUED_SPINLOCKS_XCHG32 in locking/qspinlock V3: https://lore.kernel.org/linux-riscv/1616658937-82063-1-git-send-email-guoren@xxxxxxxxxx/ - Coding convention by Peter Zijlstra's advices V2: https://lore.kernel.org/linux-riscv/1606225437-22948-2-git-send-email-guoren@xxxxxxxxxx/ - Coding convention in cmpxchg.h - Re-implement short xchg - Remove char & cmpxchg implementations V1: https://lore.kernel.org/linux-riscv/20190211043829.30096-1-michaeljclark@xxxxxxx/ - Using cmpxchg loop to implement sub-word atomic Guo Ren (19): asm-generic: ticket-lock: Reuse arch_spinlock_t of qspinlock asm-generic: ticket-lock: Move into ticket_spinlock.h riscv: qspinlock: errata: Add ERRATA_THEAD_WRITE_ONCE fixup riscv: qspinlock: Add basic queued_spinlock support riscv: qspinlock: Introduce combo spinlock riscv: qspinlock: Allow force qspinlock from the command line riscv: qspinlock: errata: Introduce ERRATA_THEAD_QSPINLOCK riscv: qspinlock: Use new static key for controlling call of virt_spin_lock() RISC-V: paravirt: pvqspinlock: Add paravirt qspinlock skeleton RISC-V: paravirt: pvqspinlock: KVM: Add paravirt qspinlock skeleton RISC-V: paravirt: pvqspinlock: KVM: Implement kvm_sbi_ext_pvlock_kick_cpu() RISC-V: paravirt: pvqspinlock: Add nopvspin kernel parameter RISC-V: paravirt: pvqspinlock: Remove unnecessary definitions of cmpxchg & xchg RISC-V: paravirt: pvqspinlock: Add xchg8 & cmpxchg_small support RISC-V: paravirt: pvqspinlock: Add SBI implementation RISC-V: paravirt: pvqspinlock: Add kconfig entry RISC-V: paravirt: pvqspinlock: Add trace point for pv_kick/wait locking/qspinlock: Move pv_ops into x86 directory locking/qspinlock: riscv: Add Compact NUMA-aware lock support .../admin-guide/kernel-parameters.txt | 5 +- arch/riscv/Kconfig | 54 ++++ arch/riscv/Kconfig.errata | 32 ++ arch/riscv/errata/thead/errata.c | 44 +++ arch/riscv/include/asm/Kbuild | 2 +- arch/riscv/include/asm/cmpxchg.h | 286 ++++++++++++------ arch/riscv/include/asm/cpufeature.h | 2 + arch/riscv/include/asm/errata_list.h | 33 +- arch/riscv/include/asm/hwcap.h | 1 + arch/riscv/include/asm/kvm_vcpu_sbi.h | 1 + arch/riscv/include/asm/paravirt.h | 20 ++ arch/riscv/include/asm/qspinlock.h | 34 +++ arch/riscv/include/asm/qspinlock_paravirt.h | 7 + arch/riscv/include/asm/rwonce.h | 24 ++ arch/riscv/include/asm/sbi.h | 14 + arch/riscv/include/asm/spinlock.h | 122 ++++++++ arch/riscv/include/asm/vendorid_list.h | 15 + arch/riscv/include/uapi/asm/kvm.h | 1 + arch/riscv/kernel/cpufeature.c | 26 ++ arch/riscv/kernel/paravirt.c | 86 ++++++ arch/riscv/kernel/sbi.c | 2 +- arch/riscv/kernel/setup.c | 22 ++ .../kernel/trace_events_filter_paravirt.h | 60 ++++ arch/riscv/kvm/Makefile | 1 + arch/riscv/kvm/vcpu_sbi.c | 4 + arch/riscv/kvm/vcpu_sbi_pvlock.c | 57 ++++ arch/x86/include/asm/qspinlock.h | 3 +- arch/x86/kernel/alternative.c | 6 +- include/asm-generic/rwonce.h | 2 + include/asm-generic/spinlock.h | 87 +----- include/asm-generic/spinlock_types.h | 12 +- include/asm-generic/ticket_spinlock.h | 103 +++++++ kernel/locking/qspinlock_cna.h | 14 +- 33 files changed, 971 insertions(+), 211 deletions(-) create mode 100644 arch/riscv/include/asm/qspinlock.h create mode 100644 arch/riscv/include/asm/qspinlock_paravirt.h create mode 100644 arch/riscv/include/asm/rwonce.h create mode 100644 arch/riscv/include/asm/spinlock.h create mode 100644 arch/riscv/kernel/trace_events_filter_paravirt.h create mode 100644 arch/riscv/kvm/vcpu_sbi_pvlock.c create mode 100644 include/asm-generic/ticket_spinlock.h -- 2.36.1