KVM: selftests: randomize memory access of dirty_log_perf_test
v8 cover KVM: selftests: randomize memory access of dirty_log_perf_test
Add the ability to randomize parts of dirty_log_perf_test, specifically the order pages are accessed and whether pages are read or written.
v8:
NOTE: Ricardo and, David please look again. Sean requested I remove your Reviewed-by tags due to changes in the last round. Main change was the interface for the RNG.
David: I made the default seed 1 since 0 isn't a valid input for this pRNG.
Split random number generator from use in dirty_log_perf_test.c and perf_test_utils.c
Colton Lewis (4):
KVM: selftests: implement random number generator for guest code
KVM: selftests: create -r argument to specify random seed
KVM: selftests: randomize which pages are written vs read
KVM: selftests: randomize page access order
.../selftests/kvm/access_tracking_perf_test.c | 2 +- .../selftests/kvm/dirty_log_perf_test.c | 57 ++++++++++++++----- .../selftests/kvm/include/perf_test_util.h | 8 ++- .../testing/selftests/kvm/include/test_util.h | 7 +++ .../selftests/kvm/lib/perf_test_util.c | 32 +++++++++-- tools/testing/selftests/kvm/lib/test_util.c | 17 ++++++ 6 files changed, 99 insertions(+), 24 deletions(-)
--
2.38.1.273.g43a17bfeac-goog
PATCH v8 1/4 KVM: selftests: implement random number generator for guest code
Implement random number generator for guest code to randomize parts of the test, making it less predictable and a more accurate reflection of reality.
The random number generator chosen is the Park-Miller Linear Congruential Generator, a fancy name for a basic and well-understood random number generator entirely sufficient for this purpose. Each vCPU calculates its own seed by adding its index to the seed provided.
--- tools/testing/selftests/kvm/include/test_util.h | 7 +++++++ tools/testing/selftests/kvm/lib/test_util.c | 17 +++++++++++++++++ 2 files changed, 24 insertions(+)
tools/testing/selftests/kvm/include/test_util.h +7 -0
@@ -152,4 +152,11 @@ static inline void *align_ptr_up(void *x, size_t size) return (void *)align_up((unsigned long)x, size); } +struct guest_random_state { +uint32_t seed; +}; + +struct guest_random_state new_guest_random_state (uint32_t seed); +uint32_t guest_random_u32 (struct guest_random_state *state); +#endif /* SELFTEST_KVM_TEST_UTIL_H */
tools/testing/selftests/kvm/lib/test_util.c +17 -0
@@ -17,6 +17,23 @@#include "test_util.h" +/* +* Random number generator that is usable from guest code. This is the +* Park-Miller LCG using standard constants. +*/ + +struct guest_random_state new_guest_random_state (uint32_t seed) +{ +struct guest_random_state s = {.seed = seed}; +return s; +} + +uint32_t guest_random_u32 (struct guest_random_state *state) +{ + state->seed = (uint64_t )state->seed *48271 % ((uint32_t )(1 <<31 ) -1 ); +return state->seed; +} +/* * Parses "[0-9]+[kmgt]?". */
Signature · 1 line
2.38.1.273.g43a17bfeac-goog
PATCH v8 2/4 KVM: selftests: create -r argument to specify random seed
In reply to Colton LewisCreate a -r argument to specify a random seed. If no argument is provided, the seed defaults to 1. The random seed is set with perf_test_set_random_seed() and must be set before guest_code runs to apply.
--- tools/testing/selftests/kvm/dirty_log_perf_test.c | 12 ++++++++++-- tools/testing/selftests/kvm/include/perf_test_util.h | 2 ++ tools/testing/selftests/kvm/lib/perf_test_util.c | 7 +++++++ 3 files changed, 19 insertions(+), 2 deletions(-)
tools/testing/selftests/kvm/dirty_log_perf_test.c +10 -2
@@ -132,6 +132,7 @@ struct test_params {bool partition_vcpu_memory_access;enum vm_mem_backing_src_type backing_src;int slots; +uint32_t random_seed; };static void toggle_dirty_logging (struct kvm_vm *vm,int slots,bool enable)
@@ -225,6 +226,9 @@ static void run_test(enum vm_guest_mode mode, void *arg) p->slots, p->backing_src, p->partition_vcpu_memory_access); +/* If no argument provided, random seed will be 1. */ +pr_info ("Random seed: %u\n" , p->random_seed); +perf_test_set_random_seed (vm, p->random_seed ? p->random_seed :1 );perf_test_set_wr_fract (vm, p->wr_fract); guest_num_pages = (nr_vcpus * guest_percpu_mem_size) >> vm->page_shift;
@@ -352,7 +356,7 @@ static void help(char *name) {puts ("" );printf ("usage: %s [-h] [-i iterations] [-p offset] [-g] " -"[-m mode] [-n] [-b vcpu bytes] [-v vcpus] [-o] [-s mem type]" +"[-m mode] [-n] [-b vcpu bytes] [-v vcpus] [-o] [-r random seed ] [-s mem type]" "[-x memslots]\n" , name);puts ("" ); printf(" -i: specify iteration counts (default: %"PRIu64")\n",
@@ -380,6 +384,7 @@ static void help(char *name)printf (" -v: specify the number of vCPUs to run.\n" );printf (" -o: Overlap guest memory accesses instead of partitioning\n" " them into a separate region of memory for each vCPU.\n" ); +printf (" -r: specify the starting random seed.\n" );backing_src_help ("-s" );printf (" -x: Split the memory region into this number of memslots.\n" " (default: 1)\n" );
@@ -406,7 +411,7 @@ int main(int argc, char *argv[])guest_modes_append_default (); -while ((opt =getopt (argc, argv,"eghi:p:m:nb:f:v:os:x:" )) !=-1 ) { +while ((opt =getopt (argc, argv,"eghi:p:m:nb:f:v:or:s:x:" )) !=-1 ) {switch (opt) {case 'e' :/* 'e' is for evil. */
@@ -442,6 +447,9 @@ int main(int argc, char *argv[])case 'o' : p.partition_vcpu_memory_access = false;break ; +case 'r' : + p.random_seed =atoi (optarg); +break ;case 's' : p.backing_src =parse_backing_src_type (optarg);break ;
tools/testing/selftests/kvm/include/perf_test_util.h +2 -0
@@ -35,6 +35,7 @@ struct perf_test_args {uint64_t gpa;uint64_t size;uint64_t guest_page_size; +uint32_t random_seed;int wr_fract;/* Run vCPUs in L2 instead of L1, if the architecture supports it. */
@@ -52,6 +53,7 @@ struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int nr_vcpus,void perf_test_destroy_vm (struct kvm_vm *vm);void perf_test_set_wr_fract (struct kvm_vm *vm,int wr_fract); +void perf_test_set_random_seed (struct kvm_vm *vm,uint32_t random_seed);void perf_test_start_vcpu_threads (int vcpus,void (*vcpu_fn)(struct perf_test_vcpu_args *));void perf_test_join_vcpu_threads (int vcpus);
tools/testing/selftests/kvm/lib/perf_test_util.c +7 -0
@@ -49,6 +49,7 @@ void perf_test_guest_code(uint32_t vcpu_idx)uint64_t gva;uint64_t pages;int i; +struct guest_random_state rand_state =new_guest_random_state (pta->random_seed + vcpu_idx); gva = vcpu_args->gva; pages = vcpu_args->pages;
@@ -229,6 +230,12 @@ void perf_test_set_wr_fract(struct kvm_vm *vm, int wr_fract)sync_global_to_guest (vm, perf_test_args); } +void perf_test_set_random_seed (struct kvm_vm *vm,uint32_t random_seed) +{ + perf_test_args.random_seed = random_seed; +sync_global_to_guest (vm, perf_test_args.random_seed); +} +uint64_t __weak perf_test_nested_pages (int nr_vcpus) {return 0 ;
Signature · 1 line
2.38.1.273.g43a17bfeac-goog
Sean Christopherson
reviewerRe: [PATCH v8 2/4] KVM: selftests: create -r argument to specify random seed
tools/testing/selftests/kvm/lib/perf_test_util.c void perf_test_guest_code(uint32_t vcpu_idx) · line 52
Quoted 9 lines · Colton Lewis
tools/testing/selftests/kvm/lib/perf_test_util.c +1
-0@@ -49,6 +49,7 @@ void perf_test_guest_code(uint32_t vcpu_idx)uint64_t gva;uint64_t pages;int i; +struct guest_random_state rand_state =new_guest_random_state (pta->random_seed + vcpu_idx);
This belong in the first patch that consumes rand_state, which I believe is patch 3.
lib/perf_test_util.c: In function ‘perf_test_guest_code’:lib/perf_test_util.c:52:35: error: unused variable ‘rand_state’ [-Werror=unused-variable]
52 | struct guest_random_state rand_state = new_guest_random_state(pta->random_seed + vcpu_idx);
| Re: [PATCH v8 2/4] KVM: selftests: create -r argument to specify random seed
Quoted 14 lines · Sean Christopherson
tools/testing/selftests/kvm/lib/perf_test_util.c void perf_test_guest_code(uint32_t vcpu_idx) · line 53
Quoted 10 lines · Colton Lewis
tools/testing/selftests/kvm/lib/perf_test_util.c +1
-0@@ -49,6 +49,7 @@ void perf_test_guest_code(uint32_t vcpu_idx)uint64_t gva;uint64_t pages;int i; +struct guest_random_state rand_state =new_guest_random_state (pta->random_seed + vcpu_idx);This belong in the first patch that consumes
rand_state, which I believe is patch 3.
Quoted 5 lines
lib/perf_test_util.c: In function ‘perf_test_guest_code’:lib/perf_test_util.c:52:35: error: unused variable ‘rand_state’ [-Werror=unused-variable]
52 | structguest_random_staterand_state=new_guest_random_state(pta->random_seed + vcpu_idx); |
Will do. Forgot this part when splitting the first commit. Will also break it into two lines for the line lenght reason your other email.
PATCH v8 3/4 KVM: selftests: randomize which pages are written vs read
In reply to Colton LewisRandomize which pages are written vs read using the random number generator.
Change the variable wr_fract and associated function calls to write_percent that now operates as a percentage from 0 to 100 where X means each page has an X% chance of being written. Change the -f argument to -w to reflect the new variable semantics. Keep the same default of 100% writes.
Population always uses 100% writes to ensure all memory is actually populated and not just mapped to the zero page. The prevents expensive copy-on-write faults from occurring during the dirty memory iterations below, which would pollute the performance results.
--- .../selftests/kvm/access_tracking_perf_test.c | 2 +- .../selftests/kvm/dirty_log_perf_test.c | 38 ++++++++++++------- .../selftests/kvm/include/perf_test_util.h | 4 +- .../selftests/kvm/lib/perf_test_util.c | 10 ++--- 4 files changed, 33 insertions(+), 21 deletions(-)
tools/testing/selftests/kvm/access_tracking_perf_test.c +1 -1
@@ -279,7 +279,7 @@ static void run_iteration(struct kvm_vm *vm, int nr_vcpus, const char *descriptistatic void access_memory (struct kvm_vm *vm,int nr_vcpus,enum access_type access,const char *description) { -perf_test_set_wr_fract (vm, (access ==ACCESS_READ ) ?INT_MAX :1 ); +perf_test_set_write_percent (vm, (access ==ACCESS_READ ) ?0 :100 ); iteration_work =ITERATION_ACCESS_MEMORY ;run_iteration (vm, nr_vcpus, description); }
tools/testing/selftests/kvm/dirty_log_perf_test.c +25 -13
@@ -128,10 +128,10 @@ static void vcpu_worker(struct perf_test_vcpu_args *vcpu_args)struct test_params {unsigned long iterations;uint64_t phys_offset; -int wr_fract;bool partition_vcpu_memory_access;enum vm_mem_backing_src_type backing_src;int slots; +uint32_t write_percent;uint32_t random_seed; };
@@ -229,7 +229,7 @@ static void run_test(enum vm_guest_mode mode, void *arg)/* If no argument provided, random seed will be 1. */ pr_info ("Random seed: %u\n" , p->random_seed);perf_test_set_random_seed (vm, p->random_seed ? p->random_seed :1 ); -perf_test_set_wr_fract (vm, p->wr_fract); +perf_test_set_write_percent (vm, p->write_percent); guest_num_pages = (nr_vcpus * guest_percpu_mem_size) >> vm->page_shift; guest_num_pages =vm_adjust_num_guest_pages (mode, guest_num_pages);
@@ -252,6 +252,14 @@ static void run_test(enum vm_guest_mode mode, void *arg)for (i =0 ; i < nr_vcpus; i++) vcpu_last_completed_iteration[i] =-1 ; +/* +* Use 100% writes during the population phase to ensure all +* memory is actually populated and not just mapped to the zero +* page. The prevents expensive copy-on-write faults from +* occurring during the dirty memory iterations below, which +* would pollute the performance results. +*/ +perf_test_set_write_percent (vm,100 );perf_test_start_vcpu_threads (nr_vcpus, vcpu_worker);/* Allow the vCPUs to populate memory */
@@ -273,6 +281,8 @@ static void run_test(enum vm_guest_mode mode, void *arg)pr_info ("Enabling dirty logging time: %ld.%.9lds\n\n" , ts_diff.tv_sec, ts_diff.tv_nsec); +perf_test_set_write_percent (vm, p->write_percent); +while (iteration < p->iterations) {/* * Incrementing the iteration number will start the vCPUs
@@ -357,7 +367,7 @@ static void help(char *name)puts ("" );printf ("usage: %s [-h] [-i iterations] [-p offset] [-g] " "[-m mode] [-n] [-b vcpu bytes] [-v vcpus] [-o] [-r random seed ] [-s mem type]" -"[-x memslots]\n" , name); +"[-x memslots] [-w percentage]\n" , name);puts ("" );printf (" -i: specify iteration counts (default: %" PRIu64")\n" ,TEST_HOST_LOOP_N );
@@ -377,10 +387,6 @@ static void help(char *name)printf (" -b: specify the size of the memory region which should be\n" " dirtied by each vCPU. e.g. 10M or 3G.\n" " (default: 1G)\n" ); -printf (" -f: specify the fraction of pages which should be written to\n" -" as opposed to simply read, in the form\n" -" 1/<fraction of pages to write>.\n" -" (default: 1 i.e. all pages are written to.)\n" );printf (" -v: specify the number of vCPUs to run.\n" );printf (" -o: Overlap guest memory accesses instead of partitioning\n" " them into a separate region of memory for each vCPU.\n" );
@@ -388,6 +394,11 @@ static void help(char *name)backing_src_help ("-s" );printf (" -x: Split the memory region into this number of memslots.\n" " (default: 1)\n" ); +printf (" -w: specify the percentage of pages which should be written to\n" +" as an integer from 0-100 inclusive. This is probabalistic,\n" +" so -w X means each page has an X%% chance of writing\n" +" and a (100-X)%% chance of reading.\n" +" (default: 100 i.e. all pages are written to.)\n" );puts ("" );exit (0 ); }
@@ -397,10 +408,10 @@ int main(int argc, char *argv[])int max_vcpus =kvm_check_cap (KVM_CAP_MAX_VCPUS );struct test_params p = { .iterations =TEST_HOST_LOOP_N , - .wr_fract =1 , .partition_vcpu_memory_access = true, .backing_src =DEFAULT_VM_MEM_SRC , .slots =1 , + .write_percent =100 , };int opt;
@@ -411,7 +422,7 @@ int main(int argc, char *argv[])guest_modes_append_default (); -while ((opt =getopt (argc, argv,"eghi:p:m:nb:f:v:or:s:x:" )) !=-1 ) { +while ((opt =getopt (argc, argv,"eghi:p:m:nb:v:or:s:x:w:" )) !=-1 ) {switch (opt) {case 'e' :/* 'e' is for evil. */
@@ -434,10 +445,11 @@ int main(int argc, char *argv[])case 'b' : guest_percpu_mem_size =parse_size (optarg);break ; -case 'f' : - p.wr_fract =atoi (optarg); -TEST_ASSERT (p.wr_fract >=1 , -"Write fraction cannot be less than one" ); +case 'w' : + p.write_percent =atoi (optarg); +TEST_ASSERT (p.write_percent >=0 + && p.write_percent <=100 , +"Write percentage must be between 0 and 100" );break ;case 'v' : nr_vcpus =atoi (optarg);
tools/testing/selftests/kvm/include/perf_test_util.h +2 -2
@@ -36,7 +36,7 @@ struct perf_test_args {uint64_t size;uint64_t guest_page_size;uint32_t random_seed; -int wr_fract; +uint32_t write_percent;/* Run vCPUs in L2 instead of L1, if the architecture supports it. */ bool nested;
@@ -52,7 +52,7 @@ struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int nr_vcpus, bool partition_vcpu_memory_access);void perf_test_destroy_vm (struct kvm_vm *vm); -void perf_test_set_wr_fract (struct kvm_vm *vm,int wr_fract); +void perf_test_set_write_percent (struct kvm_vm *vm,uint32_t write_percent);void perf_test_set_random_seed (struct kvm_vm *vm,uint32_t random_seed);void perf_test_start_vcpu_threads (int vcpus,void (*vcpu_fn)(struct perf_test_vcpu_args *));
tools/testing/selftests/kvm/lib/perf_test_util.c +5 -5
@@ -61,7 +61,7 @@ void perf_test_guest_code(uint32_t vcpu_idx)for (i =0 ; i < pages; i++) {uint64_t addr = gva + (i * pta->guest_page_size); -if (i % pta->wr_fract ==0 ) +if (guest_random_u32 (&rand_state) %100 < pta->write_percent) *(uint64_t *)addr =0x0123456789ABCDEF ;else READ_ONCE (*(uint64_t *)addr);
@@ -122,7 +122,7 @@ struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int nr_vcpus,pr_info ("Testing guest mode: %s\n" ,vm_guest_mode_string (mode));/* By default vCPUs will write to memory. */ - pta->wr_fract =1 ; + pta->write_percent =100 ;/* * Snapshot the non-huge page size. This is used by the guest code to
@@ -224,10 +224,10 @@ void perf_test_destroy_vm(struct kvm_vm *vm)kvm_vm_free (vm); } -void perf_test_set_wr_fract (struct kvm_vm *vm,int wr_fract) +void perf_test_set_write_percent (struct kvm_vm *vm,uint32_t write_percent) { - perf_test_args.wr_fract = wr_fract; -sync_global_to_guest (vm, perf_test_args); + perf_test_args.write_percent = write_percent; +sync_global_to_guest (vm, perf_test_args.write_percent); }void perf_test_set_random_seed (struct kvm_vm *vm,uint32_t random_seed)
Signature · 1 line
2.38.1.273.g43a17bfeac-goog
PATCH v8 4/4 KVM: selftests: randomize page access order
In reply to Colton LewisCreate the ability to randomize page access order with the -a argument. This includes the possibility that the same pages may be hit multiple times during an iteration or not at all.
Population has random access as false to ensure all pages will be touched by population and avoid page faults in late dirty memory that would pollute the test results.
--- tools/testing/selftests/kvm/dirty_log_perf_test.c | 11 +++++++++-- .../selftests/kvm/include/perf_test_util.h | 2 ++ tools/testing/selftests/kvm/lib/perf_test_util.c | 15 ++++++++++++++- 3 files changed, 25 insertions(+), 3 deletions(-)
tools/testing/selftests/kvm/dirty_log_perf_test.c +9 -2
@@ -133,6 +133,7 @@ struct test_params {int slots;uint32_t write_percent;uint32_t random_seed; +bool random_access; };static void toggle_dirty_logging (struct kvm_vm *vm,int slots,bool enable)
@@ -260,6 +261,7 @@ static void run_test(enum vm_guest_mode mode, void *arg)* would pollute the performance results. */ perf_test_set_write_percent (vm,100 ); +perf_test_set_random_access (vm, false);perf_test_start_vcpu_threads (nr_vcpus, vcpu_worker);/* Allow the vCPUs to populate memory */
@@ -282,6 +284,7 @@ static void run_test(enum vm_guest_mode mode, void *arg) ts_diff.tv_sec, ts_diff.tv_nsec);perf_test_set_write_percent (vm, p->write_percent); +perf_test_set_random_access (vm, p->random_access);while (iteration < p->iterations) {/*
@@ -365,10 +368,11 @@ static void run_test(enum vm_guest_mode mode, void *arg)static void help (char *name) {puts ("" ); -printf ("usage: %s [-h] [-i iterations] [-p offset] [-g] " +printf ("usage: %s [-h] [-a] [-i iterations] [-p offset] [-g] " "[-m mode] [-n] [-b vcpu bytes] [-v vcpus] [-o] [-r random seed ] [-s mem type]" "[-x memslots] [-w percentage]\n" , name);puts ("" ); +printf (" -a: access memory randomly rather than in order.\n" );printf (" -i: specify iteration counts (default: %" PRIu64")\n" ,TEST_HOST_LOOP_N ); printf(" -g: Do not enable KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2. This\n"
@@ -422,8 +426,11 @@ int main(int argc, char *argv[])guest_modes_append_default (); -while ((opt =getopt (argc, argv,"eghi:p:m:nb:v:or:s:x:w:" )) !=-1 ) { +while ((opt =getopt (argc, argv,"aeghi:p:m:nb:v:or:s:x:w:" )) !=-1 ) {switch (opt) { +case 'a' : + p.random_access = true; +break ;case 'e' :/* 'e' is for evil. */ run_vcpus_while_disabling_dirty_logging = true;
tools/testing/selftests/kvm/include/perf_test_util.h +2 -0
@@ -40,6 +40,7 @@ struct perf_test_args {/* Run vCPUs in L2 instead of L1, if the architecture supports it. */ bool nested; +bool random_access;struct perf_test_vcpu_args vcpu_args[KVM_MAX_VCPUS ]; };
@@ -54,6 +55,7 @@ void perf_test_destroy_vm(struct kvm_vm *vm);void perf_test_set_write_percent (struct kvm_vm *vm,uint32_t write_percent);void perf_test_set_random_seed (struct kvm_vm *vm,uint32_t random_seed); +void perf_test_set_random_access (struct kvm_vm *vm,bool random_access);void perf_test_start_vcpu_threads (int vcpus,void (*vcpu_fn)(struct perf_test_vcpu_args *));void perf_test_join_vcpu_threads (int vcpus);
tools/testing/selftests/kvm/lib/perf_test_util.c +14 -1
@@ -48,6 +48,8 @@ void perf_test_guest_code(uint32_t vcpu_idx)struct perf_test_vcpu_args *vcpu_args = &pta->vcpu_args[vcpu_idx];uint64_t gva;uint64_t pages; +uint64_t addr; +uint64_t page;int i;struct guest_random_state rand_state =new_guest_random_state (pta->random_seed + vcpu_idx);
@@ -59,7 +61,12 @@ void perf_test_guest_code(uint32_t vcpu_idx)while (true) {for (i =0 ; i < pages; i++) { -uint64_t addr = gva + (i * pta->guest_page_size); +if (pta->random_access) + page =guest_random_u32 (&rand_state) % pages; +else + page = i; + + addr = gva + (page * pta->guest_page_size);if (guest_random_u32 (&rand_state) %100 < pta->write_percent) *(uint64_t *)addr =0x0123456789ABCDEF ;
@@ -236,6 +243,12 @@ void perf_test_set_random_seed(struct kvm_vm *vm, uint32_t random_seed)sync_global_to_guest (vm, perf_test_args.random_seed); } +void perf_test_set_random_access (struct kvm_vm *vm,bool random_access) +{ + perf_test_args.random_access = random_access; +sync_global_to_guest (vm, perf_test_args.random_access); +} +uint64_t __weak perf_test_nested_pages (int nr_vcpus) {return 0 ;
Signature · 1 line
2.38.1.273.g43a17bfeac-goog