Kyle Swenson | 8d8f654 | 2021-03-15 11:02:55 -0600 | [diff] [blame] | 1 | |
| 2 | Export CPU topology info via sysfs. Items (attributes) are similar |
| 3 | to /proc/cpuinfo output of some architectures: |
| 4 | |
| 5 | 1) /sys/devices/system/cpu/cpuX/topology/physical_package_id: |
| 6 | |
| 7 | physical package id of cpuX. Typically corresponds to a physical |
| 8 | socket number, but the actual value is architecture and platform |
| 9 | dependent. |
| 10 | |
| 11 | 2) /sys/devices/system/cpu/cpuX/topology/core_id: |
| 12 | |
| 13 | the CPU core ID of cpuX. Typically it is the hardware platform's |
| 14 | identifier (rather than the kernel's). The actual value is |
| 15 | architecture and platform dependent. |
| 16 | |
| 17 | 3) /sys/devices/system/cpu/cpuX/topology/book_id: |
| 18 | |
| 19 | the book ID of cpuX. Typically it is the hardware platform's |
| 20 | identifier (rather than the kernel's). The actual value is |
| 21 | architecture and platform dependent. |
| 22 | |
| 23 | 4) /sys/devices/system/cpu/cpuX/topology/thread_siblings: |
| 24 | |
| 25 | internal kernel map of cpuX's hardware threads within the same |
| 26 | core as cpuX. |
| 27 | |
| 28 | 5) /sys/devices/system/cpu/cpuX/topology/thread_siblings_list: |
| 29 | |
| 30 | human-readable list of cpuX's hardware threads within the same |
| 31 | core as cpuX. |
| 32 | |
| 33 | 6) /sys/devices/system/cpu/cpuX/topology/core_siblings: |
| 34 | |
| 35 | internal kernel map of cpuX's hardware threads within the same |
| 36 | physical_package_id. |
| 37 | |
| 38 | 7) /sys/devices/system/cpu/cpuX/topology/core_siblings_list: |
| 39 | |
| 40 | human-readable list of cpuX's hardware threads within the same |
| 41 | physical_package_id. |
| 42 | |
| 43 | 8) /sys/devices/system/cpu/cpuX/topology/book_siblings: |
| 44 | |
| 45 | internal kernel map of cpuX's hardware threads within the same |
| 46 | book_id. |
| 47 | |
| 48 | 9) /sys/devices/system/cpu/cpuX/topology/book_siblings_list: |
| 49 | |
| 50 | human-readable list of cpuX's hardware threads within the same |
| 51 | book_id. |
| 52 | |
| 53 | To implement it in an architecture-neutral way, a new source file, |
| 54 | drivers/base/topology.c, is to export the 6 or 9 attributes. The three book |
| 55 | related sysfs files will only be created if CONFIG_SCHED_BOOK is selected. |
| 56 | |
| 57 | For an architecture to support this feature, it must define some of |
| 58 | these macros in include/asm-XXX/topology.h: |
| 59 | #define topology_physical_package_id(cpu) |
| 60 | #define topology_core_id(cpu) |
| 61 | #define topology_book_id(cpu) |
| 62 | #define topology_sibling_cpumask(cpu) |
| 63 | #define topology_core_cpumask(cpu) |
| 64 | #define topology_book_cpumask(cpu) |
| 65 | |
| 66 | The type of **_id macros is int. |
| 67 | The type of **_cpumask macros is (const) struct cpumask *. The latter |
| 68 | correspond with appropriate **_siblings sysfs attributes (except for |
| 69 | topology_sibling_cpumask() which corresponds with thread_siblings). |
| 70 | |
| 71 | To be consistent on all architectures, include/linux/topology.h |
| 72 | provides default definitions for any of the above macros that are |
| 73 | not defined by include/asm-XXX/topology.h: |
| 74 | 1) physical_package_id: -1 |
| 75 | 2) core_id: 0 |
| 76 | 3) sibling_cpumask: just the given CPU |
| 77 | 4) core_cpumask: just the given CPU |
| 78 | |
| 79 | For architectures that don't support books (CONFIG_SCHED_BOOK) there are no |
| 80 | default definitions for topology_book_id() and topology_book_cpumask(). |
| 81 | |
| 82 | Additionally, CPU topology information is provided under |
| 83 | /sys/devices/system/cpu and includes these files. The internal |
| 84 | source for the output is in brackets ("[]"). |
| 85 | |
| 86 | kernel_max: the maximum CPU index allowed by the kernel configuration. |
| 87 | [NR_CPUS-1] |
| 88 | |
| 89 | offline: CPUs that are not online because they have been |
| 90 | HOTPLUGGED off (see cpu-hotplug.txt) or exceed the limit |
| 91 | of CPUs allowed by the kernel configuration (kernel_max |
| 92 | above). [~cpu_online_mask + cpus >= NR_CPUS] |
| 93 | |
| 94 | online: CPUs that are online and being scheduled [cpu_online_mask] |
| 95 | |
| 96 | possible: CPUs that have been allocated resources and can be |
| 97 | brought online if they are present. [cpu_possible_mask] |
| 98 | |
| 99 | present: CPUs that have been identified as being present in the |
| 100 | system. [cpu_present_mask] |
| 101 | |
| 102 | The format for the above output is compatible with cpulist_parse() |
| 103 | [see <linux/cpumask.h>]. Some examples follow. |
| 104 | |
| 105 | In this example, there are 64 CPUs in the system but cpus 32-63 exceed |
| 106 | the kernel max which is limited to 0..31 by the NR_CPUS config option |
| 107 | being 32. Note also that CPUs 2 and 4-31 are not online but could be |
| 108 | brought online as they are both present and possible. |
| 109 | |
| 110 | kernel_max: 31 |
| 111 | offline: 2,4-31,32-63 |
| 112 | online: 0-1,3 |
| 113 | possible: 0-31 |
| 114 | present: 0-31 |
| 115 | |
| 116 | In this example, the NR_CPUS config option is 128, but the kernel was |
| 117 | started with possible_cpus=144. There are 4 CPUs in the system and cpu2 |
| 118 | was manually taken offline (and is the only CPU that can be brought |
| 119 | online.) |
| 120 | |
| 121 | kernel_max: 127 |
| 122 | offline: 2,4-127,128-143 |
| 123 | online: 0-1,3 |
| 124 | possible: 0-127 |
| 125 | present: 0-3 |
| 126 | |
| 127 | See cpu-hotplug.txt for the possible_cpus=NUM kernel start parameter |
| 128 | as well as more information on the various cpumasks. |