Version 1.7.10 (Mar 18, 2021)¶
IHK major updates¶
N/A
IHK major bug fixes¶
__ihk_device_detect_hungup: detect hungup via device-ioctl
McKernel major updates¶
N/A
McKernel major bug fixes¶
N/A
Version 1.7.9 (Mar 17, 2021)¶
IHK major updates¶
N/A
IHK major bug fixes¶
ihklib: ihk_reserve_mem_conf*: fix default values
smp_ihk_os_shutdown: fix memory leak
smp_ihk_os_shutdown: prevent double free
__ihk_os_shutdown: fix smp_ihk_os_shutdown()-related double free
smp_ihk_os_panic_notifier: exclude memory from Linux dump with default setting
smp_ihk_os_panic_notifier: exclude memory from Linux dump while booting, on timeout
McKernel major updates¶
N/A
McKernel major bug fixes¶
mcctrl_wakeup_desc: refcount and fix timeouts
Version 1.7.8 (Mar 12, 2021)¶
IHK major updates¶
N/A
IHK major bug fixes¶
ihklib: ihk_reserve_cpu: fix job cpu check when using krm
McKernel major updates¶
N/A
McKernel major bug fixes¶
N/A
Version 1.7.7 (Mar 11, 2021)¶
IHK major updates¶
N/A
IHK major bug fixes¶
N/A
McKernel major updates¶
N/A
McKernel major bug fixes¶
mcexec: fput executable just after its contents is transferred
spec: cmake-config cmake parameters
Version 1.7.6 (Mar 11, 2021)¶
IHK major updates¶
N/A
IHK major bug fixes¶
ihklib: ihk_reserve_mem_conf*: apply change only to the next reservation
McKernel major updates¶
N/A
McKernel major bug fixes¶
N/A
Version 1.7.5 (Mar 11, 2021)¶
IHK major updates¶
N/A
IHK major bug fixes¶
ihklib: fix cgroup cpuset.cpus/mems check when using krm
ihklib: ihk_reserve_mem_conf_str: set default values to those not specified
McKernel major updates¶
N/A
McKernel major bug fixes¶
N/A
Version 1.7.4 (Mar 7, 2021)¶
IHK major updates¶
N/A
IHK major bug fixes¶
N/A
McKernel major updates¶
N/A
McKernel major bug fixes¶
N/A
Version 1.7.3 (Mar 5, 2021)¶
IHK major updates¶
N/A
IHK major bug fixes¶
N/A
McKernel major updates¶
N/A
McKernel major bug fixes¶
N/A
Version 1.7.2 (Mar 5, 2021)¶
IHK major updates¶
ihklib: add *_str() functions for reserve, assign, IKC-map, kargs
smp: make smp_call_func() arch independent
IHK major bug fixes¶
ihklib: ihk_reserve_mem: fix capped best-effort
TO RESET: fake missing NUMA node pieces, 90% memory limit
ihklib: ihk_reserve_mem_conf: range-check for IHK_RESERVE_MEM_MAX_SIZE_RATIO_ALL
ihklib: ihk_os_kargs: check if “hidos” is included
SMP: omit slab/slub shrink, use 95% limit by default
check cpu / numa cgroup set by krm
SMP: __ihk_smp_reserve_mem: add __GFP_COMP to __GFP_ATOMIC allocation
ihk_register_device: record minor to IHK device object
McKernel major updates¶
mcexec: memory policy control by environmental variable
mempolicy: Support MPOL_INTERLEAVE
uti: futex call function in mcctrl
uti: integrate libuti and redirect to mck/libuti.so
uti: integrate syscall_intercept
shmobj: support large page
xpmem: support large page
MM: handle zero_at_free in page faults
McKernel major bug fixes¶
TO RESET: stack changes
Tofu: keep track of stags per memory range
Tofu: match page sizes to MBPT and fault PTEs if not present
Tofu: fix phys addr calculation for contiguous pages in MBPT/BCH update
rus_vm_fault: vmf_insert_pfn: treat VM_FAULT_NOPAGE as success
Tofu: mcctrl side MMU notifier and CQ/BCH cleanup
copy_user_ranges: copy straight_start of struct vm_range
mcctrl: abort on invalid addr in mcexec_transfer_image()
mcctrl: fix access to uninitialized usrdata->cpu_topology_list
mcexec: propagate error in __NR_gettid handler
mcexec_transfer_image(): map exact size of remote memory (instead of forcing PAGE_SIZE)
xpmem: fault stack area of remote process if VM range doesn’t yet exist
Tofu: fault stack area if VM range doesn’t exist in STAG registration
__mcctrl_os_read_write_cpu_register: fix timeout
mbind: Use range_policy’s numamask as priority on MPOL_BIND
migrate: Don’t migrate on in-kernel interrupt
Send a signal to mcexec after switching to that process.
uti: fix syscall response is mis-consumed by __do_in_kernel_irq_syscall
uti: fix handling UTI_CPU_SET env
do_execveat: kill instead of panic when init_process_stack fails
remote_page_fault is handled by the offloaded thread.
coredump: fix behavior when gencore fail
xpmem: truncates the size of xpmem_attach at the page boundary (workaround for fjmpi)
__mcctrl_os_read_write_cpu_register: spin timeout in mcctrl_ikc_send_wait()
Version 1.7.1 (Dec 23, 2020)¶
IHK major updates¶
d5d5c23 Tofu: support for barrier gate
Tofu: proper cleanup of premapped DMA regions
Tofu: initial version
SMP: try with GFP_ATOMIC as well in mem reserve
IHK major bug fixes¶
ihklib: ihk(_os)_query_{cpu,mem}: allow to pass empty array
SMP: non compound page free and GFP_ATOMIC
ihk_get_num_os_instances: don’t open /dev/mcdN
ihklib: ihk_create_os_str: fix variable prefix
McKernel major updates¶
stragiht map: creates a straight map covering the whole physical memory, and gives virtual address ranges out of it to maps to which physical pages are allocated on map
free-time, lazy, potentially Linux-side page-zeroing
Tofu built-in driver: supports memory registration and barrier gate setup
kmalloc cache
McKernel major bug fixes¶
mmap: return -EINVAL for non-anonymous, MAP_HUGETLB map
kernel: increase stack size
Tofu: proper cleanup of device files when mcexec gets killed
Version 1.7.0 (Nov 25, 2020)¶
IHK major updates¶
ihklib: add ihk_create_os_str
ihklib: ihk_reserve_mem: add capped best effort to balanced
IHK major bug fixes¶
make /dev/mcdN sharable
acpi: compat: RHEL-8.2
gic_chip_data: compat: RHEL-8.3
McKernel major updates¶
arm64: Contiguous PTE support
arm64: Scalable Vector Extension (SVE) support
arm64: PMU overflow interrupt support
arm64 port: Direct access to Mckernel memory from Linux
arm64 port: utility thread offloading, which spawns thread onto Linux CPU
eclair: support for live debug
Crash utility extension
Replace mcoverlayfs with a soft userspace overlay
Build system is switched to cmake
Core dump includes thread information
mcinspect and mcps: DWARF based LWK inspection
McKernel major bug fixes¶
shmobj: Fix rusage counting for large page
mcctrl control: task start_time changed to u64 nsec
mcctrl: add handling for one more level of page tables
Add kernel argument to turn on/off time sharing
flatten_string / process env: realign env and clear trailing bits
madvise: Add MADV_HUGEPAGE support
mcctrl: remove in-kernel calls to syscalls
arch_cpu_read_write_register: error return fix.
set_cputime(): interrupt enable/disable fix.
set_mempolicy(): Add mode check.
mbind(): Fix memory_range_lock deadlock.
ihk_ikc_recv: Record channel to packet for release
Add set_cputime() kernel to kernel case and mode enum.
execve: Call preempt_enable() before error-exit
memory/x86_64: fix linux safe_kernel_map
do_kill(): fix pids table when nr of threads is larger than num_processors
shmget: Use transparent huge pages when page size isn’t specified
prctl: Add support for PR_SET_THP_DISABLE and PR_GET_THP_DISABLE
monitor_init: fix undetected hang on highest numbered core
init_process_stack: change premapped stack size based on arch
x86 syscalls: add a bunch of XXat() delegated syscalls
do_pageout: fix direct kernel-user access
stack: add hwcap auxval
perf counters: add arch-specific perf counters
Added check of nohost to terminate_host().
kmalloc: Fix address order in free list
sysfs: use nr_cpu_ids for cpumasks (fixes libnuma parsing error on ARM)
monitor_init: Use ihk_mc_cpu_info()
Fix ThunderX2 write-combined PTE flag insanity
ARM: eliminate zero page mapping (i.e, init_low_area())
eliminate futex_cmpxchg_enabled check (not used and dereffed a NULL pointer)
page_table: Fix return value of lookup_pte when ptl4 is blank
sysfs: add missing symlinks for cpu/node
Make Linux handler run when mmap to procfs.
Separate mmap area from program loading (relocation) area
move rusage into kernel ELF image (avoid dynamic alloc before NUMA init)
arm: turn off cpu on panic
page fault handler: protect thread accesses
Register PPD and release_handler at the same time.
fix to missing exclusive processing between terminate() and finalize_process().
perfctr_stop: add flags to no ‘disable_intens’
fileobj, shmobj: free pages in object destructor (as opposed to page_unmap())
clear_range_l1, clear_range_middle: Fix handling contiguous PTE
do_mmap: don’t pre-populate the whole file when asked for smaller segment
invalidate_one_page: Support shmobj and contiguous PTE
ubsan: fix undefined shifts
x86: disable zero mapping and add a boot pt for ap trampoline
rusage: Don’t count PF_PATCH change
Fixed time processing.
copy_user_pte: vmap area not owned by McKernel
gencore: Zero-clear ELF header and memory range table
rpm: ignore CMakeCache.txt in dist and relax BuildRequires on cross build
gencore: Allocate ELF header to heap instead of stack
nanosleep: add cpu_pause() in spinwait loop
init_process: add missing initializations to proc struct
rus_vm_fault: always use a packet on the stack
process stack: use PAGE_SIZE in aux vector
copy_user_pte: base memobj copy on range & VR_PRIVATE
arm64: ptrace: Fix overwriting 1st argument with return value
page fault: use cow for private device mappings
reproductible builds: remove most install paths in c code
page fault: clear writable bit for non-dirtying access to shared ranges
mcreboot/mcstop+release: support for regular user execution
irqbalance_mck: replace extra service with service drop-in
do_mmap: give addr argument a chance even if not MAP_FIXED
x86: fix xchg() and cmpxchg() macros
IHK: support for using Linux work IRQ as IKC interrupt (optional)
MCS: fix ARM64 issue by using smp_XXX() functions (i.e., barrier()s)
procfs: add number of threads to stat and status
memory_range_lock: Fix deadlock in procfs/sysfs handler
flush instruction cache at context switch time if necessary
arm64: Fix PMU related functions
page_fault_process_memory_range: Disable COW for VM region with zeroobj
extend_process_region: Fall back to demand paging when not contiguous
munmap: fix deadlock with remote pagefault on vm range lock
procfs: if memory_range_lock fails, process later
migrate-cpu: Prevent migration target from calling schedule() twice
sched_request_migrate(): fix race condition between migration req and IRQs
get_one_cpu_topology: Renumber core_id (physical core id)
bb7e140 procfs cpuinfo: use sequence number as processor
set_host_vma(): do NOT read protect Linux VMA
hugefileobj: rewrite page allocation/handling
VM: use RW spinlock for vm_range_lock
/dev/shm: use Linux PFNs and populate mappings
Make struct ihk_os_rusage compatible with mckernel_rusage (workaround for Fugaku)
Record pthread routine address in clone(), keep helper threads on caller CPU core (workaround for Fugaku)
struct process: fix type of group_exit_status
tgkill: Fix argument validatation
set_robust_list: Add error check
mcexec: Don’t forward SIGTSTP SIGTTIN SIGTTOUT to mckernel
syscall: add prlimit64
stack: grow on page fault
mcexec: use FLIB_NUM_PROCESS_ON_NODE when -n not specified (Fugaku specific)
Version 1.6.0 (Nov 11, 2018)¶
McKernel major updates¶
McKernel and Linux share one unified kernel virtual address space. That is, McKernel sections resides in Linux sections spared for modules. In this way, Linux can access the McKernel kernel memory area.
hugetlbfs support
IHK is now included as a git submodule
Debug messages are turned on/off in per souce file basis at run-time.
It’s prohibited for McKernel to access physical memory ranges which Linux didn’t give to McKernel.
UTI (capability to spawn a thread on Linux CPU) improvement:
System calls issued from the thread are hooked by modifying binary in memory.
McKernel major bug fixes¶
#<digits> below denotes the redmine issue number (https://postpeta.pccluster.org/redmine/).
#926: shmget: Hide object with IPC_RMID from shmget
#1028: init_process: Inherit parent cpu_set
#995: Fix shebang recorded in argv[0]
#1024: Fix VMAP virtual address leak
#1109: init_process_stack: Support “ulimit -s unlimited”
x86 mem init: do not map identity mapping
mcexec_wait_syscall: requeue potential request on interrupted wait
mcctrl_ikc_send_wait: fix interrupt with do_frees == NULL
pager_req_read: handle short read
kprintf: only call eventfd() if it is safe to interrupt
process_procfs_request: Add Pid to /proc/<PID>/status
terminate: fix oversubscribe hang when waiting for other threads on same CPU to die
mcexec: Do not close fd returned to mckernel side
#976: execve: Clear sigaltstack and fp_regs
#1002: perf_event: Specify counter by bit_mask on start/stop
#1027: schedule: Don’t reschedule immediately when wake up on migrate
#mcctrl: lookup unexported symbols at runtime
__sched_wakeup_thread: Notify interrupt_exit() of re-schedule
futex_wait_queue_me: Spin-sleep when timeout and idle_halt is specified
#1167: ihk_os_getperfevent,setperfevent: Timeout IKC sent by mcctrl
devobj: fix object size (POSTK_DEBUG_TEMP_FIX_36)
mcctrl: remove rus page cache
#1021: procfs: Support multiple reads of e.g.
/proc/*/maps
#1006: wait: Delay wake-up parent within switch context
#1164: mem: Check if phys-mem is within the range of McKernel memory
#1039: page_fault_process_memory_range: Remove ihk_mc_map_virtual for CoW of device map
partitioned execution: pass process rank to LWK
process/vm: implement access_ok()
spinlock: rewrite spinlock to use Linux ticket head/tail format
#986: Fix deadlock involving mmap_sem and memory_range_lock
Prevent one CPU from getting chosen by concurrent forks
#1009: check_signal: system call restart is done only once
#1176: syscall: the signal received during system call processing is not processed.
#1036 syscall_time: Handle by McKernel
#1165 do_syscall: Delegate system calls to the mcexec with the same pid
#1194 execve: Fix calling ptrace_report_signal after preemption is disabled
#1005 coredump: Exclude special areas
#1018 procfs: Fix pread/pwrite to procfs fail when specified size is bigger than 4MB
#1180 sched_setaffinity: Check migration after decrementing in_interrupt
#771, #1179, #1143 ptrace supports threads
#1189 procfs/do_fork: wait until procfs entries are registered
#1114 procfs: add ‘/proc/pid/stat’ to mckernel side and fix its comm
#1116 mcctrl procfs: check entry was returned before using it
#1167 ihk_os_getperfevent,setperfevent: Return -ETIME when IKC timeouts
mcexec/execve: fix shebangs handling
procfs: handle ‘comm’ on mckernel side
ihk_os_setperfevent: Return number of registered events
mcexec: fix terminating zero after readlink()
Version 1.5.1 (July 9, 2018)¶
McKernel major updates¶
Watchdog timer to detect hang of McKernel¶
mcexec prints out the following line to its stderr when a hang of McKernel is detected.
mcexec detected hang of McKernel
The watchdog timer is enabled by passing -i <timeout_in_sec> option to mcreboot.sh. <timeout_in_sec> specifies the interval of checking if McKernel is alive.
For example, specify -i 600
to detect the hang with 10 minutes interval:
mcreboot.sh -i 600
- The detailed step of the hang detection is as follows.
mcexec acquires eventfd for notification from IHK and perform epoll() on it.
A daemon called ihkmond monitors the state of McKernel periodically with the interval specified by the -i option. It judges that McKernel is hanging and notifies mcexec by the eventfd if its state hasn’t changed since the last check.
McKernel major bug fixes¶
#1146: pager_req_map(): do not take mmap_sem if not needed
#1135: prepare_process_ranges_args_envs(): fix saving cmdline
#1144: fileobj/devobj: record path name
#1145: fileobj: use MCS locks for per-file page hash
#1076: mcctrl: refactor prepare_image into new generic ikc send&wait
#1072: execve: fix execve with oversubscribing
#1132: execve: use thread variable instead of cpu_local_var(current)
#1117: mprotect: do not set page table writable for cow pages
#1143: syscall wait4: add _WALL (POSTK_DEBUG_ARCH_DEP_44)
#1064: rusage: Fix initialization of rusage->num_processors
#1133: pager_req_unmap: Put per-process data at exit
#731: do_fork: Propagate error code returned by mcexec
#1149: execve: Reinitialize vm_regions’s map area on execve
#1065: procfs: Show file names in /proc/<PID>/maps
#1112: mremap: Fix type of size arguments (from ssize_t to size_t)
#1121: sched_getaffinity: Check arguments in the same order as in Linux
#1137: mmap, mremap: Check arguments in the same order as in Linux
#1122: fix return value of sched_getaffinity
#732: fix: /proc/<PID>/maps outputs a unnecessary NULL character
Version 1.5.0 (Apr 5, 2018)¶
McKernel major updates¶
Aid for Linux version migration: Detect /proc, /sys format change between two kernel verions
Swap out * Only swap-out anonymous pages for now
Improve support of /proc/maps
mcstat: Linux tool to show resource usage
McKernel major bug fixes¶
#727: execve: Fix memory leak when receiving SIGKILL
#829: perf_event_open: Support PERF_TYPE_HARDWARE and PERF_TYPE_HW_CACHE
#906: mcexec: Check return code of fork()
#1038: mcexec: Timeout when incorrect value is given to -n option
#943 #945 #946 #960 #961: mcexec: Support strace
#1029: struct thread is not released with stress-test involving signal and futex
#863 #870 : Respond immediately to terminating signal when offloading system call
#1119: translate_rva_to_rpa(): use 2MB blocks in 1GB pages on x86
#898: Shutdown OS only after no in-flight IKC exist
#882: release_handler: Destroy objects as the process which opened it
#882: mcexec: Make child process exit if the parent is killed during fork()
#925: XPMEM: Don’t destroy per-process object of the parent
#885: ptrace: Support the case where a process attaches its child
#1031: sigaction: Support SA_RESETHAND
#923: rus_vm_fault: Return error when a thread not performing system call offloading causes remote page fault
#1032 #1033 #1034: getrusage: Fix ru_maxrss, RUSAGE_CHILDREN, ru_stime related bugs
#1120: getrusage: Fix deadlock on thread->times_update
#1123: Fix deadlock related to wait_queue_head_list_node
#1124: Fix deadlock of calling terminate() from terminate()
#1125: Fix deadlock related to thread status
Related functions are: hold_thread(), do_kill() and terminate()
#1126: uti: Fix uti thread on the McKernel side blocks others in do_syscall()
#1066: procfs: Show Linux /proc/self/cgroup
#1127: prepare_process_ranges_args_envs(): fix generating saved_cmdline to avoid PF in strlen()
#1128: ihk_mc_map/unmap_virtual(): do proper TLB invalidation
#1043: terminate(): fix update_lock and threads_lock order to avoid deadlock
#1129: mcreboot.sh: Save
/proc/irq/*/smp_affinity
to/tmp/mcreboot
#1130: mcexec: drop READ_IMPLIES_EXEC from personality
McKernel workarounds¶
Forbid CPU oversubscription
It can be turned on by mcreboot.sh -O option
Version 1.4.0 (Oct 30, 2017)¶
Abstracted event type support in perf_event_open()¶
PERF_TYPE_HARDWARE and PERF_TYPE_CACHE types are supported.
Direct user-space access¶
Code lines using direct user-space access (e.g. passing user-space pointer to memcpy()) becomes more portable across processor architectures. The modification follows the following rules.
Move the code section as it is to the architecture dependent directory if it is a part of the critical-path.
Otherwise, rewrite the code section by using the portable methods. The methods include copy_from_user(), copy_to_user(), pte_get_phys() and phys_to_virt().
MPI and OpenMP micro-bench tests¶
The performance figures of MPI and OpenMP primitives are compared with those of Linux by using Intel MPI Benchmarks and EPCC OpenMP Micro Benchmark.
Version 1.3.0 (Sep 30, 2017)¶
Kernel dump¶
A dump level of “only kernel memory” is added.
The following two levels are available now:
0 |
Dump all |
24 |
Dump only kernel memory |
The dump level can be set by -d option in ihkosctl or the argument for ihk_os_makedumpfile(), as shown in the following examples:
Command: ihkosctl 0 dump -d 24
Function call: ihk_os_makedumpfile(0, NULL, 24, 0);
Dump file is created when Linux panics.
The dump level can be set by dump_level kernel argument, as shown in the following example:
ihkosctl 0 kargs "hidos dump_level=24"
The IHK dump function is registered to panic_notifier_list when creating /dev/mcdX and called when Linux panics.
Quick Process Launch¶
MPI process launch time and some of the initialization time can be reduced in application consisting of multiple MPI programs which are launched in turn in the job script.
The following two steps should be performed to use this feature: #. Replace mpiexec with ql_mpiexec_start and add some lines for ql_mpiexec_finalize in the job script #. Modify the app so that it can repeat calculations and wait for the instructions from ql_mpiexec_{start,finalize} at the end of the loop
The first step is explained using an example. Assume the original job script looks like this:
/* Execute ensamble simulation and then data assimilation, and repeat this ten times */
for i in {1..10}; do
/* Each ensamble simulation execution uses 100 nodes, launch ten of them in parallel */
for j in {1..10}; do
mpiexec -n 100 -machinefile ./list1_$j p1.out a1 & pids[$i]=$!;
done
/* Wait until the ten ensamble simulation programs finish */
for j in {1..10}; do wait ${pids[$j]}; done
/* Launch one data assimilation program using 1000 nodes */
mpiexec -n 1000 -machinefile ./list2 p2.out a2
done
The job script should be modified like this:
for i in {1..10}; do
for j in {1..10}; do
/* Replace mpiexec with ql_mpiexec_start */
ql_mpiexec_start -n 100 -machinefile ./list1_$j p1.out a1 & pids[$j]=$!;
done
for j in {1..10}; do wait ${pids[$j]}; done
ql_mpiexec_start -n 1000 -machinefile ./list2 p2.out a2
done
/* p1.out and p2.out don't exit but are waiting for the next calculation. So tell them to exit */
for j in {1..10}; do
ql_mpiexec_finalize -machinefile ./list1_$i p1.out a1;
done
ql_mpiexec_finalize -machinefile ./list2 p2.out a2;
The second step is explained using a pseudo-code.
MPI_Init();
Prepare data exchange with preceding / following MPI programs
loop:
foreach Fortran module
Initialize data using command-line argments, parameter files, environment variables
Input data from preceding MPI programs / Read snap-shot
Perform main calculation
Output data to following MPI programs / Write snap-shot
/* ql_client() waits for command of ql_mpiexec_{start,finish} */
if (ql_client() == QL_CONTINUE) { goto loop; }
MPI_Finalize();
qlmpilib.h should be included in the code and libql{mpi,fort}.so should be linked to the executable file.