1=========================================== 2How CPU topology info is exported via sysfs 3=========================================== 4 5CPU topology info is exported via sysfs. Items (attributes) are similar 6to /proc/cpuinfo output of some architectures. They reside in 7/sys/devices/system/cpu/cpuX/topology/. Please refer to the ABI file: 8Documentation/ABI/stable/sysfs-devices-system-cpu. 9 10Architecture-neutral, drivers/base/topology.c, exports these attributes. 11However, the book and drawer related sysfs files will only be created if 12CONFIG_SCHED_BOOK and CONFIG_SCHED_DRAWER are selected, respectively. 13 14CONFIG_SCHED_BOOK and CONFIG_SCHED_DRAWER are currently only used on s390, 15where they reflect the cpu and cache hierarchy. 16 17For an architecture to support this feature, it must define some of 18these macros in include/asm-XXX/topology.h:: 19 20 #define topology_physical_package_id(cpu) 21 #define topology_die_id(cpu) 22 #define topology_core_id(cpu) 23 #define topology_book_id(cpu) 24 #define topology_drawer_id(cpu) 25 #define topology_sibling_cpumask(cpu) 26 #define topology_core_cpumask(cpu) 27 #define topology_die_cpumask(cpu) 28 #define topology_book_cpumask(cpu) 29 #define topology_drawer_cpumask(cpu) 30 31The type of ``**_id macros`` is int. 32The type of ``**_cpumask macros`` is ``(const) struct cpumask *``. The latter 33correspond with appropriate ``**_siblings`` sysfs attributes (except for 34topology_sibling_cpumask() which corresponds with thread_siblings). 35 36To be consistent on all architectures, include/linux/topology.h 37provides default definitions for any of the above macros that are 38not defined by include/asm-XXX/topology.h: 39 401) topology_physical_package_id: -1 412) topology_die_id: -1 423) topology_core_id: 0 434) topology_sibling_cpumask: just the given CPU 445) topology_core_cpumask: just the given CPU 456) topology_die_cpumask: just the given CPU 46 47For architectures that don't support books (CONFIG_SCHED_BOOK) there are no 48default definitions for topology_book_id() and topology_book_cpumask(). 49For architectures that don't support drawers (CONFIG_SCHED_DRAWER) there are 50no default definitions for topology_drawer_id() and topology_drawer_cpumask(). 51 52Additionally, CPU topology information is provided under 53/sys/devices/system/cpu and includes these files. The internal 54source for the output is in brackets ("[]"). 55 56 =========== ========================================================== 57 kernel_max: the maximum CPU index allowed by the kernel configuration. 58 [NR_CPUS-1] 59 60 offline: CPUs that are not online because they have been 61 HOTPLUGGED off or exceed the limit of CPUs allowed by the 62 kernel configuration (kernel_max above). 63 [~cpu_online_mask + cpus >= NR_CPUS] 64 65 online: CPUs that are online and being scheduled [cpu_online_mask] 66 67 possible: CPUs that have been allocated resources and can be 68 brought online if they are present. [cpu_possible_mask] 69 70 present: CPUs that have been identified as being present in the 71 system. [cpu_present_mask] 72 =========== ========================================================== 73 74The format for the above output is compatible with cpulist_parse() 75[see <linux/cpumask.h>]. Some examples follow. 76 77In this example, there are 64 CPUs in the system but cpus 32-63 exceed 78the kernel max which is limited to 0..31 by the NR_CPUS config option 79being 32. Note also that CPUs 2 and 4-31 are not online but could be 80brought online as they are both present and possible:: 81 82 kernel_max: 31 83 offline: 2,4-31,32-63 84 online: 0-1,3 85 possible: 0-31 86 present: 0-31 87 88In this example, the NR_CPUS config option is 128, but the kernel was 89started with possible_cpus=144. There are 4 CPUs in the system and cpu2 90was manually taken offline (and is the only CPU that can be brought 91online.):: 92 93 kernel_max: 127 94 offline: 2,4-127,128-143 95 online: 0-1,3 96 possible: 0-127 97 present: 0-3 98 99See Documentation/core-api/cpu_hotplug.rst for the possible_cpus=NUM 100kernel start parameter as well as more information on the various cpumasks. 101