|
| 1 | +======================================================= |
| 2 | +Activity Monitors Unit (AMU) extension in AArch64 Linux |
| 3 | +======================================================= |
| 4 | + |
| 5 | +Author: Ionela Voinescu < [email protected]> |
| 6 | + |
| 7 | +Date: 2019-09-10 |
| 8 | + |
| 9 | +This document briefly describes the provision of Activity Monitors Unit |
| 10 | +support in AArch64 Linux. |
| 11 | + |
| 12 | + |
| 13 | +Architecture overview |
| 14 | +--------------------- |
| 15 | + |
| 16 | +The activity monitors extension is an optional extension introduced by the |
| 17 | +ARMv8.4 CPU architecture. |
| 18 | + |
| 19 | +The activity monitors unit, implemented in each CPU, provides performance |
| 20 | +counters intended for system management use. The AMU extension provides a |
| 21 | +system register interface to the counter registers and also supports an |
| 22 | +optional external memory-mapped interface. |
| 23 | + |
| 24 | +Version 1 of the Activity Monitors architecture implements a counter group |
| 25 | +of four fixed and architecturally defined 64-bit event counters. |
| 26 | + - CPU cycle counter: increments at the frequency of the CPU. |
| 27 | + - Constant counter: increments at the fixed frequency of the system |
| 28 | + clock. |
| 29 | + - Instructions retired: increments with every architecturally executed |
| 30 | + instruction. |
| 31 | + - Memory stall cycles: counts instruction dispatch stall cycles caused by |
| 32 | + misses in the last level cache within the clock domain. |
| 33 | + |
| 34 | +When in WFI or WFE these counters do not increment. |
| 35 | + |
| 36 | +The Activity Monitors architecture provides space for up to 16 architected |
| 37 | +event counters. Future versions of the architecture may use this space to |
| 38 | +implement additional architected event counters. |
| 39 | + |
| 40 | +Additionally, version 1 implements a counter group of up to 16 auxiliary |
| 41 | +64-bit event counters. |
| 42 | + |
| 43 | +On cold reset all counters reset to 0. |
| 44 | + |
| 45 | + |
| 46 | +Basic support |
| 47 | +------------- |
| 48 | + |
| 49 | +The kernel can safely run a mix of CPUs with and without support for the |
| 50 | +activity monitors extension. Therefore, when CONFIG_ARM64_AMU_EXTN is |
| 51 | +selected we unconditionally enable the capability to allow any late CPU |
| 52 | +(secondary or hotplugged) to detect and use the feature. |
| 53 | + |
| 54 | +When the feature is detected on a CPU, we flag the availability of the |
| 55 | +feature but this does not guarantee the correct functionality of the |
| 56 | +counters, only the presence of the extension. |
| 57 | + |
| 58 | +Firmware (code running at higher exception levels, e.g. arm-tf) support is |
| 59 | +needed to: |
| 60 | + - Enable access for lower exception levels (EL2 and EL1) to the AMU |
| 61 | + registers. |
| 62 | + - Enable the counters. If not enabled these will read as 0. |
| 63 | + - Save/restore the counters before/after the CPU is being put/brought up |
| 64 | + from the 'off' power state. |
| 65 | + |
| 66 | +When using kernels that have this feature enabled but boot with broken |
| 67 | +firmware the user may experience panics or lockups when accessing the |
| 68 | +counter registers. Even if these symptoms are not observed, the values |
| 69 | +returned by the register reads might not correctly reflect reality. Most |
| 70 | +commonly, the counters will read as 0, indicating that they are not |
| 71 | +enabled. |
| 72 | + |
| 73 | +If proper support is not provided in firmware it's best to disable |
| 74 | +CONFIG_ARM64_AMU_EXTN. To be noted that for security reasons, this does not |
| 75 | +bypass the setting of AMUSERENR_EL0 to trap accesses from EL0 (userspace) to |
| 76 | +EL1 (kernel). Therefore, firmware should still ensure accesses to AMU registers |
| 77 | +are not trapped in EL2/EL3. |
| 78 | + |
| 79 | +The fixed counters of AMUv1 are accessible though the following system |
| 80 | +register definitions: |
| 81 | + - SYS_AMEVCNTR0_CORE_EL0 |
| 82 | + - SYS_AMEVCNTR0_CONST_EL0 |
| 83 | + - SYS_AMEVCNTR0_INST_RET_EL0 |
| 84 | + - SYS_AMEVCNTR0_MEM_STALL_EL0 |
| 85 | + |
| 86 | +Auxiliary platform specific counters can be accessed using |
| 87 | +SYS_AMEVCNTR1_EL0(n), where n is a value between 0 and 15. |
| 88 | + |
| 89 | +Details can be found in: arch/arm64/include/asm/sysreg.h. |
| 90 | + |
| 91 | + |
| 92 | +Userspace access |
| 93 | +---------------- |
| 94 | + |
| 95 | +Currently, access from userspace to the AMU registers is disabled due to: |
| 96 | + - Security reasons: they might expose information about code executed in |
| 97 | + secure mode. |
| 98 | + - Purpose: AMU counters are intended for system management use. |
| 99 | + |
| 100 | +Also, the presence of the feature is not visible to userspace. |
| 101 | + |
| 102 | + |
| 103 | +Virtualization |
| 104 | +-------------- |
| 105 | + |
| 106 | +Currently, access from userspace (EL0) and kernelspace (EL1) on the KVM |
| 107 | +guest side is disabled due to: |
| 108 | + - Security reasons: they might expose information about code executed |
| 109 | + by other guests or the host. |
| 110 | + |
| 111 | +Any attempt to access the AMU registers will result in an UNDEFINED |
| 112 | +exception being injected into the guest. |
0 commit comments