Migrating the libhugetlbfs/testcases/alloc-instantiate-race.c test
PATCH v3 1/1 Migrating the libhugetlbfs/testcases/alloc-instantiate-race.c test
This test is designed to detect a kernel allocation race introduced with hugepage demand-faulting. The problem is that no lock is held between allocating a hugepage and instantiating it in the pagetables or page cache index. In between the two, the (huge) page is cleared, so there's substantial time. Thus two processes can race instantiating the (same) last available hugepage - one will fail on the allocation, and thus cause an OOM fault even though the page it actually wants is being instantiated by the other racing process.
---
v3: --Addressed the below requested changes
1. Added support to run test cases multiple times using the -iN or -IN option.
2. Removed unnecessary comments.
3. Set "shared" mode as the default for mmap when the user does not specify <shared/private> with the -m option.
4. Updated user-facing help messages to clarify how to use the -m option.
--- runtest/hugetlb | 1 + testcases/kernel/mem/.gitignore | 1 + .../kernel/mem/hugetlb/hugemmap/hugemmap36.c | 300 ++++++++++++++++++ 3 files changed, 302 insertions(+) create mode 100644 testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c
runtest/hugetlb +1 -0
@@ -36,6 +36,7 @@ hugemmap30 hugemmap30 hugemmap31 hugemmap31 hugemmap32 hugemmap32 hugemmap34 hugemmap34 +hugemmap36 hugemmap36 hugemmap05_1 hugemmap05 -m hugemmap05_2 hugemmap05 -s hugemmap05_3 hugemmap05 -s -m
testcases/kernel/mem/.gitignore +1 -0
@@ -36,6 +36,7 @@ /hugetlb/hugemmap/hugemmap31 /hugetlb/hugemmap/hugemmap32 /hugetlb/hugemmap/hugemmap34 +/hugetlb/hugemmap/hugemmap36 /hugetlb/hugeshmat/hugeshmat01 /hugetlb/hugeshmat/hugeshmat02 /hugetlb/hugeshmat/hugeshmat03
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +300 -0
300 lines not on this page · raw message
Signature · 5 lines
2.47.3 -- Mailing list info: https://lists.linux.it/listinfo/ltp
Quoted 55 lines · Samir Mulani
This test is designed to detect a kernel allocation race introduced with hugepage demand-faulting. The problem is that no lock is held between allocating a hugepage and instantiating it in the pagetables or page cache index. In between the two, the (huge) page is cleared, so there's substantial time. Thus two processes can race instantiating the (same) last available hugepage - one will fail on the allocation, and thus cause an OOM fault even though the page it actually wants is being instantiated by the other racing process.
Signed-off-by Samir Mulani <[email protected]>---
v3: --Addressed the below requested changes
1. Added support to run test cases multiple times using the -iN or -IN option.
2. Removed unnecessary comments.
3. Set "shared" mode as the default for mmap when the user does not specify <shared/private> with the -m option.
4. Updated user-facing help messages to clarify how to use the -m option.--- runtest/hugetlb | 1 + testcases/kernel/mem/.gitignore | 1 + .../kernel/mem/hugetlb/hugemmap/hugemmap36.c | 300 ++++++++++++++++++ 3 files changed, 302 insertions(+) create mode 100644 testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.cruntest/hugetlb +1
-0@@ -36,6 +36,7 @@ hugemmap30 hugemmap30 hugemmap31 hugemmap31 hugemmap32 hugemmap32 hugemmap34 hugemmap34 +hugemmap36 hugemmap36 hugemmap05_1 hugemmap05 -m hugemmap05_2 hugemmap05 -s hugemmap05_3 hugemmap05 -s -mtestcases/kernel/mem/.gitignore +1
-0@@ -36,6 +36,7 @@ /hugetlb/hugemmap/hugemmap31 /hugetlb/hugemmap/hugemmap32 /hugetlb/hugemmap/hugemmap34 +/hugetlb/hugemmap/hugemmap36 /hugetlb/hugeshmat/hugeshmat01 /hugetlb/hugeshmat/hugeshmat02 /hugetlb/hugeshmat/hugeshmat03testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +0
-0
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c line 8
Quoted 8 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +8
-0@@ -0,0 +1,300 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* +* Copyright (C) 2005-2006 IBM Corporation +* Author: David Gibson & Adam Litke +*/ + +/*\ +* [Description]
Description tag not needed.
Quoted 10 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +10
-0+* +* This test is designed to detect a kernel allocation race introduced +* with hugepage demand-faulting. The problem is that no lock is held +* between allocating a hugepage and instantiating it in the +* pagetables or page cache index. In between the two, the (huge) +* page is cleared, so there's substantial time. Thus two processes +* can race instantiating the (same) last available hugepage - one +* will fail on the allocation, and thus cause an OOM fault even +* though the page it actually wants is being instantiated by the +* other racing process.
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c line 28
Quoted 10 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +10
-0+*/ + +#define _GNU_SOURCE +#include <stdio.h> +#include <pthread.h> +#include "tst_safe_pthread.h" +#include "hugetlb.h" +#define SYSFS_CPU_ONLINE_FMT "/sys/devices/system/cpu/cpu%d/online" +#define MNTPOINT "hugetlbfs/" +unsigned long totpages;
missing static definition. also it's used only in run() so there's no need to have it here. Please verify the same for the other variables as well.
Quoted 31 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +31
-0+static long hpage_size; +static char *str_op; +static int child1, child2, race_type, fd_sync, test_flag; +static pthread_t thread1, thread2; +static void *p_sync, *q_sync; + +struct racer_info { +void *p;/* instantiation address */ +int cpu; +volatile int *mytrigger; +volatile int *othertrigger; +int status; +}; + +static int one_racer (void *p,int cpu, +volatile int *mytrigger,volatile int *othertrigger) +{ +volatile int *pi = p; +cpu_set_t *cpuset; +size_t mask_size; +int err; + + cpuset =CPU_ALLOC (cpu +1 ); +if (!cpuset) +tst_brk (TBROK |TERRNO ,"CPU_ALLOC() failed" ); + + mask_size =CPU_ALLOC_SIZE (cpu +1 ); + +/* Split onto different CPUs to encourage the race */ +CPU_ZERO_S (mask_size, cpuset); +CPU_SET_S (cpu, mask_size, cpuset);
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c line 69
Quoted 10 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +10
-0+ + err =sched_setaffinity (getpid (), mask_size, cpuset); +if (err ==-1 ) +tst_brk (TBROK |TERRNO ,"sched_setaffinity() failed" ); + +/* Ready */ + *mytrigger =1 ; +/* Wait for the other trigger to be set */ +while (!*othertrigger) + ;
This is a busy loop that is gonna create many issues. You probably need to take a look at the checkpoints if you need to synchronize threads.
Quoted 8 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +8
-0+ +/* Set the shared value */ + *pi =1 ; + +CPU_FREE (cpuset); +return 0 ; +} +
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c line 87
Quoted 10 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +10
-0+static void proc_racer (void *p,int cpu, +volatile int *mytrigger,volatile int *othertrigger) +{ +exit (one_racer (p, cpu, mytrigger, othertrigger)); +} + +static void *thread_racer (void *info) +{ +struct racer_info *ri = info; +ri->status =one_racer (ri->p, ri->cpu, ri->mytrigger, ri->othertrigger);
indentation.
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c line 91
Quoted 4 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +4
-0+return ri; +} + +void check_online_cpus (int online_cpus[],int nr_cpus_needed)
should be a static function
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c line 94
Quoted 3 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +3
-0+{ +cpu_set_t cpuset; +int total_cpus, cpu_idx, i;
i can be declared inside the loop.
Quoted 5 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +4
-0+// Initialize the CPU set +CPU_ZERO (&cpuset); + +for (int i =0 ; i <CPU_SETSIZE ; i++)
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c line 108
Quoted 10 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +10
-0+CPU_SET (i, &cpuset); +if (sched_setaffinity (0 ,sizeof (cpuset), &cpuset) ==-1 ) +tst_brk (TBROK |TERRNO ,"sched_setaffinity() reset failed" ); +// Get the total number of configured CPUs + total_cpus =get_nprocs_conf (); +// Get the CPU affinity mask of the calling process +if (sched_getaffinity (0 ,sizeof (cpu_set_t), &cpuset) ==-1 ) +tst_brk (TBROK |TERRNO ,"sched_getaffinity() failed" ); + +int available =CPU_COUNT (&cpuset);
Just use CPU_COUNT() it in the tst_res(..), there's no need to have a variable here.
Quoted 1 line
+tst_res(TINFO, "Online CPUs needed: %d, available: %d",
nr_cpus_needed, available);
Wrong indentation.
Quoted 122 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +122
-0+ +// Check if there are enough online CPUs +if (CPU_COUNT (&cpuset) < nr_cpus_needed) +tst_brk (TBROK |TERRNO ,"At least %d online CPUs are required " , nr_cpus_needed); + + cpu_idx =0 ; +// Find the first `nr_cpus_needed` CPUs in the affinity mask +for (i =0 ; i < total_cpus && cpu_idx < nr_cpus_needed; i++) { +if (CPU_ISSET (i, &cpuset)) + online_cpus[cpu_idx++] = i; + } +if (cpu_idx < nr_cpus_needed) +tst_brk (TBROK |TERRNO ,"Unable to find enough online CPUs" ); +} + +static void run_race (void *syncarea,int race_type) +{ +volatile int *trigger1, *trigger2; +int fd; +void *p, *tret1, *tret2; +int status1, status2; +int online_cpus[2 ]; + +check_online_cpus (online_cpus,2 ); +memset (syncarea,0 ,sizeof (*trigger1) +sizeof (*trigger2)); + trigger1 = syncarea; + trigger2 = trigger1 +1 ; + +/* Get a new file for the final page */ + fd =tst_creat_unlinked (MNTPOINT ,0 ,0600 ); +tst_res (TINFO ,"Mapping final page.. " ); + + + p =SAFE_MMAP (NULL , hpage_size,PROT_READ |PROT_WRITE , race_type, fd,0 ); +if (race_type ==MAP_SHARED ) { + child1 =SAFE_FORK (); +if (child1 ==0 ) +proc_racer (p, online_cpus[0 ], trigger1, trigger2); + + child2 =SAFE_FORK (); +if (child2 ==0 ) +proc_racer (p, online_cpus[1 ], trigger2, trigger1); + +/* wait() calls */ +SAFE_WAITPID (child1, &status1,0 ); +tst_res (TINFO ,"Child 1 status: %x" , status1); + + +SAFE_WAITPID (child2, &status2,0 ); +tst_res (TINFO ,"Child 2 status: %x" , status2); + +if (WIFSIGNALED (status1)) +tst_res (TFAIL ,"Child 1 killed by signal %s" , +strsignal (WTERMSIG (status1))); +if (WIFSIGNALED (status2)) +tst_res (TFAIL ,"Child 2 killed by signal %s" , +strsignal (WTERMSIG (status2))); + }else { +struct racer_info ri1 = { + .p = p, + .cpu = online_cpus[0 ], + .mytrigger = trigger1, + .othertrigger = trigger2, + .status =-1 , + }; +struct racer_info ri2 = { + .p = p, + .cpu = online_cpus[1 ], + .mytrigger = trigger2, + .othertrigger = trigger1, + .status =-1 , + }; +SAFE_PTHREAD_CREATE (&thread1,NULL , thread_racer, &ri1); +SAFE_PTHREAD_CREATE (&thread2,NULL , thread_racer, &ri2); +SAFE_PTHREAD_JOIN (thread1, &tret1); +if (tret1 != &ri1) { + test_flag =-1 ; +tst_res (TFAIL ,"Thread 1 returned %p not %p, killed?\n" , + tret1, &ri1); + } +SAFE_PTHREAD_JOIN (thread2, &tret2); + +if (tret2 != &ri2) { + test_flag =-1 ; +tst_res (TFAIL ,"Thread 2 returned %p not %p, killed?\n" , + tret2, &ri2); + } + status1 = ri1.status; + status2 = ri2.status; + } + +if (status1 !=0 ) { + test_flag =-1 ; +tst_res (TFAIL ,"Racer 1 terminated with code %d" , status1); + } + +if (status2 !=0 ) { + test_flag =-1 ; +tst_res (TFAIL ,"Racer 2 terminated with code %d" , status2); + } +if (test_flag !=-1 ) + test_flag =0 ; + +if (fd >=0 ) +SAFE_CLOSE (fd); + +if (p !=MAP_FAILED ) +SAFE_MUNMAP (p, hpage_size); + +if (q_sync !=MAP_FAILED ) { +SAFE_MUNMAP (q_sync,getpagesize ()); + q_sync =NULL ; + } +} + +static void run_test (void ) +{ + totpages =SAFE_READ_MEMINFO (MEMINFO_HPAGE_FREE ); + hpage_size =tst_get_hugepage_size (); + +/* Allocate all save one of the pages up front */ +tst_res (TINFO ,"instantiating.. " );
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c line 241
Quoted 10 lines
testcases/kernel/mem/hugetlb/hugemmap/hugemmap36.c new file +10
-0+ + fd_sync =tst_creat_unlinked (MNTPOINT ,0 ,0600 ); +/* Get a shared normal page for synchronization */ + q_sync =SAFE_MMAP (NULL ,getpagesize (),PROT_READ |PROT_WRITE , +MAP_SHARED |MAP_ANONYMOUS ,-1 ,0 ); +tst_res (TINFO ,"Mapping %ld/%ld pages.. " , totpages-1 , totpages); + p_sync =SAFE_MMAP (NULL , (totpages-1 )*hpage_size,PROT_READ |PROT_WRITE , +MAP_SHARED , fd_sync,0 ); + +run_race (q_sync, race_type);
This q_sync is passed around in the wrong way. Please take a look at its usage because q_sync is used in run_race() but initialized in run_test(). If run_test() fails before run_race(), q_sync could be uninitialized.
Kind regards,
Signature · 6 lines
Andrea Cervesato SUSE QE Automation Engineer Linux [email protected] -- Mailing list info: https://lists.linux.it/listinfo/ltp