kernel/include/linux/perf_event.h, branch linux-5.17.y

perf: Fix perf_event_read_local() time

2022-01-18T11:09:47Z

Time readers that cannot take locks (due to NMI etc..) currently make use of perf_event::shadow_ctx_time, which, for that event gives: time' = now + (time - timestamp) or, alternatively arranged: time' = time + (now - timestamp) IOW, the progression of time since the last time the shadow_ctx_time was updated. There's problems with this: A) the shadow_ctx_time is per-event, even though the ctx_time it reflects is obviously per context. The direct concequence of this is that the context needs to iterate all events all the time to keep the shadow_ctx_time in sync. B) even with the prior point, the context itself might not be active meaning its time should not advance to begin with. C) shadow_ctx_time isn't consistently updated when ctx_time is There are 3 users of this stuff, that suffer differently from this: - calc_timer_values() - perf_output_read() - perf_event_update_userpage() /* A */ - perf_event_read_local() /* A,B */ In particular, perf_output_read() doesn't suffer at all, because it's sample driven and hence only relevant when the event is actually running. This same was supposed to be true for perf_event_update_userpage(), after all self-monitoring implies the context is active *HOWEVER*, as per commit f79256532682 ("perf/core: fix userpage->time_enabled of inactive events") this goes wrong when combined with counter overcommit, in that case those events that do not get scheduled when the context becomes active (task events typically) miss out on the EVENT_TIME update and ENABLED time is inflated (for a little while) with the time the context was inactive. Once the event gets rotated in, this gets corrected, leading to a non-monotonic timeflow. perf_event_read_local() made things even worse, it can request time at any point, suffering all the problems perf_event_update_userpage() does and more. Because while perf_event_update_userpage() is limited by the context being active, perf_event_read_local() users have no such constraint. Therefore, completely overhaul things and do away with perf_event::shadow_ctx_time. Instead have regular context time updates keep track of this offset directly and provide perf_event_time_now() to complement perf_event_time(). perf_event_time_now() will, in adition to being context wide, also take into account if the context is active. For inactive context, it will not advance time. This latter property means the cgroup perf_cgroup_info context needs to grow addition state to track this. Additionally, since all this is strictly per-cpu, we can use barrier() to order context activity vs context time. Fixes: 7d9285e82db5 ("perf/bpf: Extend the perf_event_read_local() interface, a.k.a. "bpf: perf event change needed for subsequent bpf helpers"") Signed-off-by: Peter Zijlstra (Intel) Tested-by: Song Liu Tested-by: Namhyung Kim Link: https://lkml.kernel.org/r/YcB06DasOBtU0b00@hirez.programming.kicks-ass.net

Merge tag 'perf_core_for_v5.17_rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

2022-01-13T00:26:58Z

Pull perf updates from Borislav Petkov: "Cleanup of the perf/kvm interaction." * tag 'perf_core_for_v5.17_rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: perf: Drop guest callback (un)register stubs KVM: arm64: Drop perf.c and fold its tiny bits of code into arm.c KVM: arm64: Hide kvm_arm_pmu_available behind CONFIG_HW_PERF_EVENTS=y KVM: arm64: Convert to the generic perf callbacks KVM: x86: Move Intel Processor Trace interrupt handler to vmx.c KVM: Move x86's perf guest info callbacks to generic KVM KVM: x86: More precisely identify NMI from guest when handling PMI KVM: x86: Drop current_vcpu for kvm_running_vcpu + kvm_arch_vcpu variable perf/core: Use static_call to optimize perf_guest_info_callbacks perf: Force architectures to opt-in to guest callbacks perf: Add wrappers for invoking guest callbacks perf/core: Rework guest callbacks to prepare for static_call support perf: Drop dead and useless guest "support" from arm, csky, nds32 and riscv perf: Stop pretending that perf can handle multiple guest callbacks KVM: x86: Register Processor Trace interrupt hook iff PT enabled in guest KVM: x86: Register perf callbacks after calling vendor's hardware_setup() perf: Protect perf_guest_cbs with RCU

Merge tag '5.17-net-next' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next

2022-01-11T03:06:09Z

Pull networking updates from Jakub Kicinski: "Core ---- - Defer freeing TCP skbs to the BH handler, whenever possible, or at least perform the freeing outside of the socket lock section to decrease cross-CPU allocator work and improve latency. - Add netdevice refcount tracking to locate sources of netdevice and net namespace refcount leaks. - Make Tx watchdog less intrusive - avoid pausing Tx and restarting all queues from a single CPU removing latency spikes. - Various small optimizations throughout the stack from Eric Dumazet. - Make netdev->dev_addr[] constant, force modifications to go via appropriate helpers to allow us to keep addresses in ordered data structures. - Replace unix_table_lock with per-hash locks, improving performance of bind() calls. - Extend skb drop tracepoint with a drop reason. - Allow SO_MARK and SO_PRIORITY setsockopt under CAP_NET_RAW. BPF --- - New helpers: - bpf_find_vma(), find and inspect VMAs for profiling use cases - bpf_loop(), runtime-bounded loop helper trading some execution time for much faster (if at all converging) verification - bpf_strncmp(), improve performance, avoid compiler flakiness - bpf_get_func_arg(), bpf_get_func_ret(), bpf_get_func_arg_cnt() for tracing programs, all inlined by the verifier - Support BPF relocations (CO-RE) in the kernel loader. - Further the support for BTF_TYPE_TAG annotations. - Allow access to local storage in sleepable helpers. - Convert verifier argument types to a composable form with different attributes which can be shared across types (ro, maybe-null). - Prepare libbpf for upcoming v1.0 release by cleaning up APIs, creating new, extensible ones where missing and deprecating those to be removed. Protocols --------- - WiFi (mac80211/cfg80211): - notify user space about long "come back in N" AP responses, allow it to react to such temporary rejections - allow non-standard VHT MCS 10/11 rates - use coarse time in airtime fairness code to save CPU cycles - Bluetooth: - rework of HCI command execution serialization to use a common queue and work struct, and improve handling errors reported in the middle of a batch of commands - rework HCI event handling to use skb_pull_data, avoiding packet parsing pitfalls - support AOSP Bluetooth Quality Report - SMC: - support net namespaces, following the RDMA model - improve connection establishment latency by pre-clearing buffers - introduce TCP ULP for automatic redirection to SMC - Multi-Path TCP: - support ioctls: SIOCINQ, OUTQ, and OUTQNSD - support socket options: IP_TOS, IP_FREEBIND, IP_TRANSPARENT, IPV6_FREEBIND, and IPV6_TRANSPARENT, TCP_CORK and TCP_NODELAY - support cmsgs: TCP_INQ - improvements in the data scheduler (assigning data to subflows) - support fastclose option (quick shutdown of the full MPTCP connection, similar to TCP RST in regular TCP) - MCTP (Management Component Transport) over serial, as defined by DMTF spec DSP0253 - "MCTP Serial Transport Binding". Driver API ---------- - Support timestamping on bond interfaces in active/passive mode. - Introduce generic phylink link mode validation for drivers which don't have any quirks and where MAC capability bits fully express what's supported. Allow PCS layer to participate in the validation. Convert a number of drivers. - Add support to set/get size of buffers on the Rx rings and size of the tx copybreak buffer via ethtool. - Support offloading TC actions as first-class citizens rather than only as attributes of filters, improve sharing and device resource utilization. - WiFi (mac80211/cfg80211): - support forwarding offload (ndo_fill_forward_path) - support for background radar detection hardware - SA Query Procedures offload on the AP side New hardware / drivers ---------------------- - tsnep - FPGA based TSN endpoint Ethernet MAC used in PLCs with real-time requirements for isochronous communication with protocols like OPC UA Pub/Sub. - Qualcomm BAM-DMUX WWAN - driver for data channels of modems integrated into many older Qualcomm SoCs, e.g. MSM8916 or MSM8974 (qcom_bam_dmux). - Microchip LAN966x multi-port Gigabit AVB/TSN Ethernet Switch driver with support for bridging, VLANs and multicast forwarding (lan966x). - iwlmei driver for co-operating between Intel's WiFi driver and Intel's Active Management Technology (AMT) devices. - mse102x - Vertexcom MSE102x Homeplug GreenPHY chips - Bluetooth: - MediaTek MT7921 SDIO devices - Foxconn MT7922A - Realtek RTL8852AE Drivers ------- - Significantly improve performance in the datapaths of: lan78xx, ax88179_178a, lantiq_xrx200, bnxt. - Intel Ethernet NICs: - igb: support PTP/time PEROUT and EXTTS SDP functions on 82580/i354/i350 adapters - ixgbevf: new PF -> VF mailbox API which avoids the risk of mailbox corruption with ESXi - iavf: support configuration of VLAN features of finer granularity, stacked tags and filtering - ice: PTP support for new E822 devices with sub-ns precision - ice: support firmware activation without reboot - Mellanox Ethernet NICs (mlx5): - expose control over IRQ coalescing mode (CQE vs EQE) via ethtool - support TC forwarding when tunnel encap and decap happen between two ports of the same NIC - dynamically size and allow disabling various features to save resources for running in embedded / SmartNIC scenarios - Broadcom Ethernet NICs (bnxt): - use page frag allocator to improve Rx performance - expose control over IRQ coalescing mode (CQE vs EQE) via ethtool - Other Ethernet NICs: - amd-xgbe: add Ryzen 6000 (Yellow Carp) Ethernet support - Microsoft cloud/virtual NIC (mana): - add XDP support (PASS, DROP, TX) - Mellanox Ethernet switches (mlxsw): - initial support for Spectrum-4 ASICs - VxLAN with IPv6 underlay - Marvell Ethernet switches (prestera): - support flower flow templates - add basic IP forwarding support - NXP embedded Ethernet switches (ocelot & felix): - support Per-Stream Filtering and Policing (PSFP) - enable cut-through forwarding between ports by default - support FDMA to improve packet Rx/Tx to CPU - Other embedded switches: - hellcreek: improve trapping management (STP and PTP) packets - qca8k: support link aggregation and port mirroring - Qualcomm 802.11ax WiFi (ath11k): - qca6390, wcn6855: enable 802.11 power save mode in station mode - BSS color change support - WCN6855 hw2.1 support - 11d scan offload support - scan MAC address randomization support - full monitor mode, only supported on QCN9074 - qca6390/wcn6855: report signal and tx bitrate - qca6390: rfkill support - qca6390/wcn6855: regdb.bin support - Intel WiFi (iwlwifi): - support SAR GEO Offset Mapping (SGOM) and Time-Aware-SAR (TAS) in cooperation with the BIOS - support for Optimized Connectivity Experience (OCE) scan - support firmware API version 68 - lots of preparatory work for the upcoming Bz device family - MediaTek WiFi (mt76): - Specific Absorption Rate (SAR) support - mt7921: 160 MHz channel support - RealTek WiFi (rtw88): - Specific Absorption Rate (SAR) support - scan offload - Other WiFi NICs - ath10k: support fetching (pre-)calibration data from nvmem - brcmfmac: configure keep-alive packet on suspend - wcn36xx: beacon filter support" * tag '5.17-net-next' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next: (2048 commits) tcp: tcp_send_challenge_ack delete useless param `skb` net/qla3xxx: Remove useless DMA-32 fallback configuration rocker: Remove useless DMA-32 fallback configuration hinic: Remove useless DMA-32 fallback configuration lan743x: Remove useless DMA-32 fallback configuration net: enetc: Remove useless DMA-32 fallback configuration cxgb4vf: Remove useless DMA-32 fallback configuration cxgb4: Remove useless DMA-32 fallback configuration cxgb3: Remove useless DMA-32 fallback configuration bnx2x: Remove useless DMA-32 fallback configuration et131x: Remove useless DMA-32 fallback configuration be2net: Remove useless DMA-32 fallback configuration vmxnet3: Remove useless DMA-32 fallback configuration bna: Simplify DMA setting net: alteon: Simplify DMA setting myri10ge: Simplify DMA setting qlcnic: Simplify DMA setting net: allwinner: Fix print format page_pool: remove spinlock in page_pool_refill_alloc_cache() amt: fix wrong return type of amt_send_membership_update() ...

add includes masked by cgroup -> bpf dependency

2021-12-16T22:57:09Z

cgroup pulls in BPF which pulls in a lot of includes. We're about to break that chain so fix those who were depending on it. Signed-off-by: Jakub Kicinski Signed-off-by: Alexei Starovoitov Link: https://lore.kernel.org/bpf/20211216025538.1649516-2-kuba@kernel.org

perf: Add a counter for number of user access events in context

2021-12-14T11:30:54Z

On arm64, user space counter access will be controlled differently compared to x86. On x86, access in the strictest mode is enabled for all tasks in an MM when any event is mmap'ed. For arm64, access is explicitly requested for an event and only enabled when the event's context is active. This avoids hooks into the arch context switch code and gives better control of when access is enabled. In order to configure user space access when the PMU is enabled, it is necessary to know if any event (currently active or not) in the current context has user space accessed enabled. Add a counter similar to other counters in the context to avoid walking the event list every time. Reviewed-by: Mark Rutland Reviewed-by: Thomas Gleixner Signed-off-by: Rob Herring Link: https://lore.kernel.org/r/20211208201124.310740-3-robh@kernel.org Signed-off-by: Will Deacon

x86: perf: Move RDPMC event flag to a common definition

2021-12-14T11:30:54Z

In preparation to enable user counter access on arm64 and to move some of the user access handling to perf core, create a common event flag for user counter access and convert x86 to use it. Since the architecture specific flags start at the LSB, starting at the MSB for common flags. Cc: Peter Zijlstra Cc: Ingo Molnar Cc: Arnaldo Carvalho de Melo Cc: Alexander Shishkin Cc: Jiri Olsa Cc: Namhyung Kim Cc: Kan Liang Cc: Thomas Gleixner Cc: Borislav Petkov Cc: x86@kernel.org Cc: "H. Peter Anvin" Cc: linux-perf-users@vger.kernel.org Reviewed-by: Mark Rutland Reviewed-by: Thomas Gleixner Signed-off-by: Rob Herring Link: https://lore.kernel.org/r/20211208201124.310740-2-robh@kernel.org Signed-off-by: Will Deacon

perf: Drop guest callback (un)register stubs

2021-11-17T13:49:12Z

Drop perf's stubs for (un)registering guest callbacks now that KVM registration of callbacks is hidden behind GUEST_PERF_EVENTS=y. The only other user is x86 XEN_PV, and x86 unconditionally selects PERF_EVENTS. No functional change intended. Signed-off-by: Sean Christopherson Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Paolo Bonzini Link: https://lore.kernel.org/r/20211111020738.2512932-18-seanjc@google.com

perf/core: Use static_call to optimize perf_guest_info_callbacks

2021-11-17T13:49:09Z

Use static_call to optimize perf's guest callbacks on arm64 and x86, which are now the only architectures that define the callbacks. Use DEFINE_STATIC_CALL_RET0 as the default/NULL for all guest callbacks, as the callback semantics are that a return value '0' means "not in guest". static_call obviously avoids the overhead of CONFIG_RETPOLINE=y, but is also advantageous versus other solutions, e.g. per-cpu callbacks, in that a per-cpu memory load is not needed to detect the !guest case. Based on code from Peter and Like. Suggested-by: Peter Zijlstra (Intel) Signed-off-by: Sean Christopherson Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Paolo Bonzini Link: https://lore.kernel.org/r/20211111020738.2512932-10-seanjc@google.com

perf: Force architectures to opt-in to guest callbacks

2021-11-17T13:49:08Z

Introduce GUEST_PERF_EVENTS and require architectures to select it to allow registering and using guest callbacks in perf. This will hopefully make it more difficult for new architectures to add useless "support" for guest callbacks, e.g. via copy+paste. Stubbing out the helpers has the happy bonus of avoiding a load of perf_guest_cbs when GUEST_PERF_EVENTS=n on arm64/x86. Signed-off-by: Sean Christopherson Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Paolo Bonzini Link: https://lore.kernel.org/r/20211111020738.2512932-9-seanjc@google.com

perf: Add wrappers for invoking guest callbacks

2021-11-17T13:49:08Z

Add helpers for the guest callbacks to prepare for burying the callbacks behind a Kconfig (it's a lot easier to provide a few stubs than to #ifdef piles of code), and also to prepare for converting the callbacks to static_call(). perf_instruction_pointer() in particular will have subtle semantics with static_call(), as the "no callbacks" case will return 0 if the callbacks are unregistered between querying guest state and getting the IP. Implement the change now to avoid a functional change when adding static_call() support, and because the new helper needs to return _something_ in this case. Signed-off-by: Sean Christopherson Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Paolo Bonzini Link: https://lore.kernel.org/r/20211111020738.2512932-8-seanjc@google.com