1.\" Copyright (c) 2008, Sun Microsystems, Inc. All Rights Reserved. 2.\" The contents of this file are subject to the terms of the Common Development and Distribution License (the "License"). You may not use this file except in compliance with the License. 3.\" You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE or http://www.opensolaris.org/os/licensing. See the License for the specific language governing permissions and limitations under the License. 4.\" When distributing Covered Code, include this CDDL HEADER in each file and include the License file at usr/src/OPENSOLARIS.LICENSE. If applicable, add the following below this CDDL HEADER, with the fields enclosed by brackets "[]" replaced with your own identifying information: Portions Copyright [yyyy] [name of copyright owner] 5.\" Copyright (c) 2019, Joyent, Inc. 6.\" Copyright 2021 Oxide Computer Company 7.Dd January 22, 2021 8.Dt CPC 3CPC 9.Os 10.Sh NAME 11.Nm cpc 12.Nd hardware performance counters 13.Sh DESCRIPTION 14Modern microprocessors contain 15.Em hardware performance counters 16that allow 17the measurement of many different hardware events related to CPU behavior, 18including instruction and data cache misses as well as various internal states 19of the processor. 20The counters can be configured to count user events, system 21events, or both. 22Data from the performance counters can be used to analyze and 23tune the behavior of software on a particular type of processor. 24.Pp 25Most processors are able to generate an interrupt on counter overflow, allowing 26the counters to be used for various forms of profiling. 27.Pp 28This manual page describes a set of APIs that allow illumos applications to use 29these counters. 30Applications can measure their own behavior, the behavior of 31other applications, or the behavior of the whole system. 32.Ss Shared Counters or Private Counters 33There are two principal models for using these performance counters. 34Some users of these statistics want to observe system-wide behavior. 35Other users want to view the performance counters as part of the register set 36exported by each 37.Sy LWP . 38On a machine performing more than one activity, these two models are 39in conflict because the counters represent a critical hardware resource that 40cannot simultaneously be both shared and private. 41.Ss Configuration Interfaces 42The following configuration interfaces are provided: 43.Bl -tag -width Xr 44.It Xr cpc_open 3CPC 45Check the version the application was compiled with against the version of the 46.It Xr cpc_cciname 3CPC 47Return a printable string to describe the performance counters of the 48processor. 49.It Xr cpc_npic 3CPC 50Return the number of performance counters on the processor. 51.It cpc_cpuref 3CPC 52Return a reference to documentation that should be consulted to understand how 53to use and interpret data from the performance counters. 54.El 55.Ss Performance Counter Access 56Performance counters can be present in hardware but not acccessible because 57either some of the necessary system software components are not available or 58not installed, or the counters might be in use by other processes. 59The 60.Xr cpc_open 3CPC 61function determines the accessibility of the counters and 62must be invoked before any attempt to program the counters. 63.Ss "Finding Events" 64Each different type of processor has its own set of events available for 65measurement. 66The 67.Xr cpc_walk_events_all 3CPC 68and 69.Xr cpc_walk_events_pic 3CPC 70functions allow an application to determine the 71names of events supported by the underlying processor. 72A collection of generic, platform independent event names are defined by 73.Xr generic_events 3CPC . 74Each generic event maps to an underlying hardware event specific to the 75underlying processor and any optional attributes. 76The 77.Xr cpc_walk_generic_events_all 3CPC 78and 79.Xr cpc_walk_generic_events_pic 3CPC 80functions allow an application to determine the generic events supported 81on the underlying platform. 82.Ss Intel Processor Specific Events 83The following manual pages provide more detailed information on the 84events available for the specific Intel processor models. 85The covered processor models are listed in hexadecimal. 86.Bl -tag -width Xr 87.It Xr bdw_de_events 3CPC 88Intel Broadwell-DE events; covers model 56h. 89.It Xr bdw_events 3CPC 90Intel Broadwell client events; covers models 3dh and 47h. 91.It Xr bdx_events 3CPC 92Intel Broadwell server events; covers model 4fh. 93.It Xr bnl_events 3CPC 94Intel Atom Bonnell events; covers models 35h, 36h, 27h, 26h, and 1ch. 95.It Xr clx_events 3CPC 96Intel Cascade Lake server events; covers model 55h, steppings 5-fh. 97.It Xr glm_events 3CPC 98Intel Goldmont SoC events; covers models 5fh and 5ch. 99.It Xr glp_events 3CPC 100Intel Goldmont Plus SoC events; covers model 7ah. 101.It Xr hsw_events 3CPC 102Intel Haswell client events; covers models 46h, 45h, and 3ch. 103.It Xr hsx_events 3CPC 104Intel Haswell server events; covers model 3fh. 105.It xr icl_events 3CPC 106Intel Ice Lake client events; covers model 7eh. 107.It Xr ivb_events 3CPC 108Intel Ivy Bridge client events; covers model 3ah. 109.It Xr ivt_events 3CPC 110Intel Ivy Bridge server events; covers model 3eh. 111.It Xr jkt_events 3CPC 112Intel Sandy Bridge server events; covers model 2dh. 113.It Xr nhm_ep_events 3CPC 114Intel Nehalem-EP events; covers models, 1ah, 1fh, and 1eh. 115.It Xr nhm_ex_events 3CPC 116Intel Sandy Bridge server events; covers model 23h. 117.It Xr skl_events 3CPC 118Intel Skylake client events; covers model a6h, a5h, 9eh, 8eh, 5e, and 4eh. 119.It Xr skx_events 3CPC 120Intel Skylake server events; covers model 55h, steppings 0-4h. 121.It Xr slm_events 3CPC 122Intel Atom Silvermont events; covers models 4ch, 4dh, and 37h. 123.It Xr snr_Events 3CPC 124Intel Atom Snow Ridge events; covers model 86h. 125.It Xr snb_events 3CPC 126Intel Sandy Bridge client events; covers model 2ah. 127.It Xr tgl_events 3CPC 128Intel Tiger Lake client events; covers models 8ch and 8dh. 129.It Xr wsm_ep_dp_events 3CPC 130Intel Westmere-EP-DP events; covers model 2ch. 131.It Xr wsm_ep_sp_events 3CPC 132Intel Westmere-EP-SP events; covers model 25h. 133.It Xr wsm_ex_events 3CPC 134Intel Westmere-EX events; covers model 2fh. 135.El 136.Ss AMD Processor Specific Events 137The following manual pages provide more detailed information on the 138events available for the specific AMD processor models. 139The covered processor families are listed in hexadecimal. 140.Bl -tag -width Xr 141.It Xr amd_f17h_zen1_events 3CPC 142AMD Family 17h Zen 1 processors, including models 00-2fh. 143Include Ryzen, ThreadRipper, and EPYC branded processors. 144.It Xr amd_f17h_zen2_events 3CPC 145AMD Family 17h Zen 2 processors, including models 30-7fh. 146Include Ryzen, ThreadRipper, and EPYC branded processors. 147.El 148.Ss Using Attributes 149Some processors have advanced performance counter capabilities that are 150configured with attributes. 151The 152.Xr cpc_walk_attrs 3CPC 153function can be used to determine the names of attributes supported by 154the underlying processor. 155The documentation referenced by 156.Xr cpc_cpuref 3CPC 157should be consulted to understand the meaning of a processor's performance 158counter attributes. 159.Ss Performance Counter Context 160Each processor on the system possesses its own set of performance counter 161registers. 162For a single process, it is often desirable to maintain the illusion 163that the counters are an intrinsic part of that process (whichever processors 164it runs on), since this allows the events to be directly attributed to the 165process without having to make passive all other activity on the system. 166.Pp 167To achieve this behavior, the library associates 168.Em performance counter context 169with each 170.Sy LWP 171in the process. 172The context consists of a small amount of kernel memory to hold the counter 173values when the 174.Sy BLWP 175is not running, and some simple kernel functions to save and restore those counter 176values from and to the hardware registers when the 177.Sy LWP 178performs a normal context switch. 179A process can only observe and manipulate its own copy of the 180performance counter control and data registers. 181.Ss Performance Counters \&In Other Processes 182Though applications can be modified to instrument themselves as demonstrated 183above, it is frequently useful to be able to examine the behavior of an 184existing application without changing the source code. 185A separate library, 186.Sy libpctx , 187provides a simple set of interfaces that use the facilities of 188.Xr proc 4 189to control a target process, and together with functions in 190.Sy libcpc , 191allow 192.Sy truss No -like 193tools to be constructed to measure the performance counters in other 194applications. 195An example of one such application is 196.Xr cputrack 1 . 197.Pp 198The functions in 199.Sy libpctx 200are independent of those in 201.Sy libcpc . 202These functions manage a process using an event-loop paradigm \(em that is, the 203execution of certain system calls by the controlled process cause the library 204to stop the controlled process and execute callback functions in the context of 205the controlling process. 206These handlers can perform various operations on the target process using APIs 207in 208.Sy libpctx 209and 210.Sy libcpc 211that consume 212.Vt pctx_t 213handles. 214.Sh SEE ALSO 215.Xr cputrack 1 , 216.Xr cpustat 1M , 217.Xr cpc_bind_curlwp 3CPC , 218.Xr cpc_buf_create 3CPC , 219.Xr cpc_enable 3CPC , 220.Xr cpc_npic 3CPC , 221.Xr cpc_open 3CPC , 222.Xr cpc_set_create 3CPC , 223.Xr cpc_seterrhndlr 3CPC , 224.Xr generic_events 3CPC , 225.Xr pctx_capture 3CPC , 226.Xr pctx_set_events 3CPC , 227.Xr libcpc 3LIB , 228.Xr proc 4 229