| Message ID | 20221221090953.341247-2-nrb@linux.ibm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | s390x: test CMM during migration | expand |
On Wed, 21 Dec 2022 10:09:53 +0100 Nico Boehr <nrb@linux.ibm.com> wrote: > Add a test which modifies CMM page states while migration is in > progress. > > This is added to the existing migration-cmm test, which gets a new > command line argument for the sequential and parallel variants. > > Signed-off-by: Nico Boehr <nrb@linux.ibm.com> Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com> > --- > s390x/migration-cmm.c | 258 +++++++++++++++++++++++++++++++++++++----- > s390x/unittests.cfg | 15 ++- > 2 files changed, 240 insertions(+), 33 deletions(-) > > diff --git a/s390x/migration-cmm.c b/s390x/migration-cmm.c > index 43673f18e45a..2d46c6be0ac8 100644 > --- a/s390x/migration-cmm.c > +++ b/s390x/migration-cmm.c > @@ -2,6 +2,12 @@ > /* > * CMM migration tests (ESSA) > * > + * There are two variants of this test: > + * - sequential: sets CMM page states, then migrates the VM and - after > + * migration finished - verifies that page states have been preserved. > + * - parallel: migrate VM and - while migration is in progress - change > + * page states and verify that they are preserved. > + * > * Copyright IBM Corp. 2022 > * > * Authors: > @@ -13,55 +19,249 @@ > #include <asm/interrupt.h> > #include <asm/page.h> > #include <asm/cmm.h> > +#include <asm/barrier.h> > #include <bitops.h> > +#include <smp.h> > + > +struct verify_result { > + bool verify_failed; > + char expected_mask; > + char actual_mask; > + unsigned long page_mismatch_idx; > + unsigned long page_mismatch_addr; > +}; > + > +static enum { > + TEST_INVLALID, > + TEST_SEQUENTIAL, > + TEST_PARALLEL > +} arg_test_to_run; > > #define NUM_PAGES 128 > -static uint8_t pagebuf[NUM_PAGES][PAGE_SIZE] __attribute__((aligned(PAGE_SIZE))); > > -static void test_migration(void) > +/* > + * Allocate 3 pages more than we need so we can start at different offsets. > + * For the parallel test, this ensures page states change on every loop iteration. > + */ > +static uint8_t pagebuf[(NUM_PAGES + 3) * PAGE_SIZE] __attribute__((aligned(PAGE_SIZE))); > + > +static struct verify_result result; > + > +static unsigned int thread_iters; > +static int thread_should_exit; > +static int thread_exited; > + > +/* > + * Maps ESSA actions to states the page is allowed to be in after the > + * respective action was executed. > + */ > +static const unsigned long allowed_essa_state_masks[4] = { > + BIT(ESSA_USAGE_STABLE), /* ESSA_SET_STABLE */ > + BIT(ESSA_USAGE_UNUSED), /* ESSA_SET_UNUSED */ > + BIT(ESSA_USAGE_VOLATILE), /* ESSA_SET_VOLATILE */ > + BIT(ESSA_USAGE_VOLATILE) | BIT(ESSA_USAGE_POT_VOLATILE) /* ESSA_SET_POT_VOLATILE */ > +}; > + > +/* > + * Set CMM page state test pattern on pagebuf. > + * pagebuf must point to page_count consecutive pages. > + * page_count must be a multiple of 4. > + */ > +static void set_test_pattern(uint8_t *pagebuf, size_t page_count) > +{ > + unsigned long addr = (unsigned long)pagebuf; > + size_t i; > + > + assert(page_count % 4 == 0); > + for (i = 0; i < page_count; i += 4) { > + essa(ESSA_SET_STABLE, addr + i * PAGE_SIZE); > + essa(ESSA_SET_UNUSED, addr + (i + 1) * PAGE_SIZE); > + essa(ESSA_SET_VOLATILE, addr + (i + 2) * PAGE_SIZE); > + essa(ESSA_SET_POT_VOLATILE, addr + (i + 3) * PAGE_SIZE); > + } > +} > + > +/* > + * Verify CMM page states on pagebuf. > + * Page states must have been set by set_test_pattern on pagebuf before. > + * page_count must be a multiple of 4. > + * > + * If page states match the expected result, will return a verify_result > + * with verify_failed false. All other fields are then invalid. > + * If there is a mismatch, the returned struct will have verify_failed true > + * and will be filled with details on the first mismatch encountered. > + */ > +static struct verify_result verify_page_states(uint8_t *pagebuf, size_t page_count) > +{ > + unsigned long expected_mask, actual_mask; > + struct verify_result result = { > + .verify_failed = true > + }; > + unsigned long addr; > + size_t i; > + > + assert(page_count % 4 == 0); > + for (i = 0; i < page_count; i++) { > + addr = (unsigned long)(pagebuf + i * PAGE_SIZE); > + actual_mask = essa(ESSA_GET_STATE, addr); > + /* usage state in bits 60 and 61 */ > + actual_mask = BIT((actual_mask >> 2) & 0x3); > + expected_mask = allowed_essa_state_masks[i % ARRAY_SIZE(allowed_essa_state_masks)]; > + if (!(actual_mask & expected_mask)) { > + result.page_mismatch_idx = i; > + result.page_mismatch_addr = addr; > + result.expected_mask = expected_mask; > + result.actual_mask = actual_mask; > + return result; > + } > + } > + > + result.verify_failed = false; > + return result; > +} > + > +static void report_verify_result(const struct verify_result *result) > +{ > + if (result->verify_failed) > + report_fail("page state mismatch: first page idx = %lu, addr = %lx, " > + "expected_mask = 0x%x, actual_mask = 0x%x", > + result->page_mismatch_idx, result->page_mismatch_addr, > + result->expected_mask, result->actual_mask); > + else > + report_pass("page states match"); > +} > + > +static void test_cmm_migration_sequential(void) > +{ > + report_prefix_push("sequential"); > + > + set_test_pattern(pagebuf, NUM_PAGES); > + > + migrate_once(); > + > + result = verify_page_states(pagebuf, NUM_PAGES); > + report_verify_result(&result); > + > + report_prefix_pop(); > +} > + > +static void set_cmm_thread(void) > { > - int i, state_mask, actual_state; > + uint8_t *pagebuf_start; > /* > - * Maps ESSA actions to states the page is allowed to be in after the > - * respective action was executed. > + * The second CPU must not print to the console, otherwise it will race with > + * the primary CPU on the SCLP buffer. > */ > - int allowed_essa_state_masks[4] = { > - BIT(ESSA_USAGE_STABLE), /* ESSA_SET_STABLE */ > - BIT(ESSA_USAGE_UNUSED), /* ESSA_SET_UNUSED */ > - BIT(ESSA_USAGE_VOLATILE), /* ESSA_SET_VOLATILE */ > - BIT(ESSA_USAGE_VOLATILE) | BIT(ESSA_USAGE_POT_VOLATILE) /* ESSA_SET_POT_VOLATILE */ > - }; > + while (!READ_ONCE(thread_should_exit)) { > + /* > + * Start on a offset different from the last iteration so page states change with > + * every iteration. This is why pagebuf has 3 extra pages. > + */ > + pagebuf_start = pagebuf + (thread_iters % 4) * PAGE_SIZE; > + set_test_pattern(pagebuf_start, NUM_PAGES); > + > + /* > + * Always increment even if the verify fails. This ensures primary CPU knows where > + * we left off and can do an additional verify round after migration finished. > + */ > + thread_iters++; > > - assert(NUM_PAGES % 4 == 0); > - for (i = 0; i < NUM_PAGES; i += 4) { > - essa(ESSA_SET_STABLE, (unsigned long)pagebuf[i]); > - essa(ESSA_SET_UNUSED, (unsigned long)pagebuf[i + 1]); > - essa(ESSA_SET_VOLATILE, (unsigned long)pagebuf[i + 2]); > - essa(ESSA_SET_POT_VOLATILE, (unsigned long)pagebuf[i + 3]); > + result = verify_page_states(pagebuf_start, NUM_PAGES); > + if (result.verify_failed) > + break; > } > > + WRITE_ONCE(thread_exited, 1); > +} > + > +static void test_cmm_migration_parallel(void) > +{ > + report_prefix_push("parallel"); > + > + if (smp_query_num_cpus() == 1) { > + report_skip("need at least 2 cpus for this test"); > + goto error; > + } > + > + smp_cpu_setup(1, PSW_WITH_CUR_MASK(set_cmm_thread)); > + > migrate_once(); > > - for (i = 0; i < NUM_PAGES; i++) { > - actual_state = essa(ESSA_GET_STATE, (unsigned long)pagebuf[i]); > - /* extract the usage state in bits 60 and 61 */ > - actual_state = (actual_state >> 2) & 0x3; > - state_mask = allowed_essa_state_masks[i % ARRAY_SIZE(allowed_essa_state_masks)]; > - report(BIT(actual_state) & state_mask, "page %d state: expected_mask=0x%x actual_mask=0x%lx", i, state_mask, BIT(actual_state)); > + WRITE_ONCE(thread_should_exit, 1); > + > + while (!READ_ONCE(thread_exited)) > + ; > + > + /* Ensure thread_iters and result below are read from memory after thread completed */ > + mb(); > + > + report_info("thread completed %u iterations", thread_iters); > + > + report_prefix_push("during migration"); > + report_verify_result(&result); > + report_prefix_pop(); > + > + /* > + * Verification of page states occurs on the thread. We don't know if we > + * were still migrating during the verification. > + * To be sure, make another verification round after the migration > + * finished to catch page states which might not have been migrated > + * correctly. > + */ > + report_prefix_push("after migration"); > + assert(thread_iters > 0); > + result = verify_page_states(pagebuf + ((thread_iters - 1) % 4) * PAGE_SIZE, NUM_PAGES); > + report_verify_result(&result); > + report_prefix_pop(); > + > +error: > + report_prefix_pop(); > +} > + > +static void print_usage(void) > +{ > + report_info("Usage: migration-cmm [--parallel|--sequential]"); > +} > + > +static void parse_args(int argc, char **argv) > +{ > + if (argc < 2) { > + /* default to sequential since it only needs one CPU */ > + arg_test_to_run = TEST_SEQUENTIAL; > + return; > } > + > + if (!strcmp("--parallel", argv[1])) > + arg_test_to_run = TEST_PARALLEL; > + else if (!strcmp("--sequential", argv[1])) > + arg_test_to_run = TEST_SEQUENTIAL; > + else > + arg_test_to_run = TEST_INVLALID; > } > > -int main(void) > +int main(int argc, char **argv) > { > report_prefix_push("migration-cmm"); > - > - if (!check_essa_available()) > + if (!check_essa_available()) { > report_skip("ESSA is not available"); > - else > - test_migration(); > + goto error; > + } > > - migrate_once(); > + parse_args(argc, argv); > + > + switch (arg_test_to_run) { > + case TEST_SEQUENTIAL: > + test_cmm_migration_sequential(); > + break; > + case TEST_PARALLEL: > + test_cmm_migration_parallel(); > + break; > + default: > + print_usage(); > + } > > +error: > + migrate_once(); > report_prefix_pop(); > return report_summary(); > } > diff --git a/s390x/unittests.cfg b/s390x/unittests.cfg > index 3caf81eda396..c7fe1006e25d 100644 > --- a/s390x/unittests.cfg > +++ b/s390x/unittests.cfg > @@ -181,10 +181,6 @@ file = migration.elf > groups = migration > smp = 2 > > -[migration-cmm] > -file = migration-cmm.elf > -groups = migration > - > [migration-skey] > file = migration-skey.elf > groups = migration > @@ -208,3 +204,14 @@ groups = migration > [exittime] > file = exittime.elf > smp = 2 > + > +[migration-cmm-sequential] > +file = migration-cmm.elf > +groups = migration > +extra_params = -append '--sequential' > + > +[migration-cmm-parallel] > +file = migration-cmm.elf > +groups = migration > +smp = 2 > +extra_params = -append '--parallel'
diff --git a/s390x/migration-cmm.c b/s390x/migration-cmm.c index 43673f18e45a..2d46c6be0ac8 100644 --- a/s390x/migration-cmm.c +++ b/s390x/migration-cmm.c @@ -2,6 +2,12 @@ /* * CMM migration tests (ESSA) * + * There are two variants of this test: + * - sequential: sets CMM page states, then migrates the VM and - after + * migration finished - verifies that page states have been preserved. + * - parallel: migrate VM and - while migration is in progress - change + * page states and verify that they are preserved. + * * Copyright IBM Corp. 2022 * * Authors: @@ -13,55 +19,249 @@ #include <asm/interrupt.h> #include <asm/page.h> #include <asm/cmm.h> +#include <asm/barrier.h> #include <bitops.h> +#include <smp.h> + +struct verify_result { + bool verify_failed; + char expected_mask; + char actual_mask; + unsigned long page_mismatch_idx; + unsigned long page_mismatch_addr; +}; + +static enum { + TEST_INVLALID, + TEST_SEQUENTIAL, + TEST_PARALLEL +} arg_test_to_run; #define NUM_PAGES 128 -static uint8_t pagebuf[NUM_PAGES][PAGE_SIZE] __attribute__((aligned(PAGE_SIZE))); -static void test_migration(void) +/* + * Allocate 3 pages more than we need so we can start at different offsets. + * For the parallel test, this ensures page states change on every loop iteration. + */ +static uint8_t pagebuf[(NUM_PAGES + 3) * PAGE_SIZE] __attribute__((aligned(PAGE_SIZE))); + +static struct verify_result result; + +static unsigned int thread_iters; +static int thread_should_exit; +static int thread_exited; + +/* + * Maps ESSA actions to states the page is allowed to be in after the + * respective action was executed. + */ +static const unsigned long allowed_essa_state_masks[4] = { + BIT(ESSA_USAGE_STABLE), /* ESSA_SET_STABLE */ + BIT(ESSA_USAGE_UNUSED), /* ESSA_SET_UNUSED */ + BIT(ESSA_USAGE_VOLATILE), /* ESSA_SET_VOLATILE */ + BIT(ESSA_USAGE_VOLATILE) | BIT(ESSA_USAGE_POT_VOLATILE) /* ESSA_SET_POT_VOLATILE */ +}; + +/* + * Set CMM page state test pattern on pagebuf. + * pagebuf must point to page_count consecutive pages. + * page_count must be a multiple of 4. + */ +static void set_test_pattern(uint8_t *pagebuf, size_t page_count) +{ + unsigned long addr = (unsigned long)pagebuf; + size_t i; + + assert(page_count % 4 == 0); + for (i = 0; i < page_count; i += 4) { + essa(ESSA_SET_STABLE, addr + i * PAGE_SIZE); + essa(ESSA_SET_UNUSED, addr + (i + 1) * PAGE_SIZE); + essa(ESSA_SET_VOLATILE, addr + (i + 2) * PAGE_SIZE); + essa(ESSA_SET_POT_VOLATILE, addr + (i + 3) * PAGE_SIZE); + } +} + +/* + * Verify CMM page states on pagebuf. + * Page states must have been set by set_test_pattern on pagebuf before. + * page_count must be a multiple of 4. + * + * If page states match the expected result, will return a verify_result + * with verify_failed false. All other fields are then invalid. + * If there is a mismatch, the returned struct will have verify_failed true + * and will be filled with details on the first mismatch encountered. + */ +static struct verify_result verify_page_states(uint8_t *pagebuf, size_t page_count) +{ + unsigned long expected_mask, actual_mask; + struct verify_result result = { + .verify_failed = true + }; + unsigned long addr; + size_t i; + + assert(page_count % 4 == 0); + for (i = 0; i < page_count; i++) { + addr = (unsigned long)(pagebuf + i * PAGE_SIZE); + actual_mask = essa(ESSA_GET_STATE, addr); + /* usage state in bits 60 and 61 */ + actual_mask = BIT((actual_mask >> 2) & 0x3); + expected_mask = allowed_essa_state_masks[i % ARRAY_SIZE(allowed_essa_state_masks)]; + if (!(actual_mask & expected_mask)) { + result.page_mismatch_idx = i; + result.page_mismatch_addr = addr; + result.expected_mask = expected_mask; + result.actual_mask = actual_mask; + return result; + } + } + + result.verify_failed = false; + return result; +} + +static void report_verify_result(const struct verify_result *result) +{ + if (result->verify_failed) + report_fail("page state mismatch: first page idx = %lu, addr = %lx, " + "expected_mask = 0x%x, actual_mask = 0x%x", + result->page_mismatch_idx, result->page_mismatch_addr, + result->expected_mask, result->actual_mask); + else + report_pass("page states match"); +} + +static void test_cmm_migration_sequential(void) +{ + report_prefix_push("sequential"); + + set_test_pattern(pagebuf, NUM_PAGES); + + migrate_once(); + + result = verify_page_states(pagebuf, NUM_PAGES); + report_verify_result(&result); + + report_prefix_pop(); +} + +static void set_cmm_thread(void) { - int i, state_mask, actual_state; + uint8_t *pagebuf_start; /* - * Maps ESSA actions to states the page is allowed to be in after the - * respective action was executed. + * The second CPU must not print to the console, otherwise it will race with + * the primary CPU on the SCLP buffer. */ - int allowed_essa_state_masks[4] = { - BIT(ESSA_USAGE_STABLE), /* ESSA_SET_STABLE */ - BIT(ESSA_USAGE_UNUSED), /* ESSA_SET_UNUSED */ - BIT(ESSA_USAGE_VOLATILE), /* ESSA_SET_VOLATILE */ - BIT(ESSA_USAGE_VOLATILE) | BIT(ESSA_USAGE_POT_VOLATILE) /* ESSA_SET_POT_VOLATILE */ - }; + while (!READ_ONCE(thread_should_exit)) { + /* + * Start on a offset different from the last iteration so page states change with + * every iteration. This is why pagebuf has 3 extra pages. + */ + pagebuf_start = pagebuf + (thread_iters % 4) * PAGE_SIZE; + set_test_pattern(pagebuf_start, NUM_PAGES); + + /* + * Always increment even if the verify fails. This ensures primary CPU knows where + * we left off and can do an additional verify round after migration finished. + */ + thread_iters++; - assert(NUM_PAGES % 4 == 0); - for (i = 0; i < NUM_PAGES; i += 4) { - essa(ESSA_SET_STABLE, (unsigned long)pagebuf[i]); - essa(ESSA_SET_UNUSED, (unsigned long)pagebuf[i + 1]); - essa(ESSA_SET_VOLATILE, (unsigned long)pagebuf[i + 2]); - essa(ESSA_SET_POT_VOLATILE, (unsigned long)pagebuf[i + 3]); + result = verify_page_states(pagebuf_start, NUM_PAGES); + if (result.verify_failed) + break; } + WRITE_ONCE(thread_exited, 1); +} + +static void test_cmm_migration_parallel(void) +{ + report_prefix_push("parallel"); + + if (smp_query_num_cpus() == 1) { + report_skip("need at least 2 cpus for this test"); + goto error; + } + + smp_cpu_setup(1, PSW_WITH_CUR_MASK(set_cmm_thread)); + migrate_once(); - for (i = 0; i < NUM_PAGES; i++) { - actual_state = essa(ESSA_GET_STATE, (unsigned long)pagebuf[i]); - /* extract the usage state in bits 60 and 61 */ - actual_state = (actual_state >> 2) & 0x3; - state_mask = allowed_essa_state_masks[i % ARRAY_SIZE(allowed_essa_state_masks)]; - report(BIT(actual_state) & state_mask, "page %d state: expected_mask=0x%x actual_mask=0x%lx", i, state_mask, BIT(actual_state)); + WRITE_ONCE(thread_should_exit, 1); + + while (!READ_ONCE(thread_exited)) + ; + + /* Ensure thread_iters and result below are read from memory after thread completed */ + mb(); + + report_info("thread completed %u iterations", thread_iters); + + report_prefix_push("during migration"); + report_verify_result(&result); + report_prefix_pop(); + + /* + * Verification of page states occurs on the thread. We don't know if we + * were still migrating during the verification. + * To be sure, make another verification round after the migration + * finished to catch page states which might not have been migrated + * correctly. + */ + report_prefix_push("after migration"); + assert(thread_iters > 0); + result = verify_page_states(pagebuf + ((thread_iters - 1) % 4) * PAGE_SIZE, NUM_PAGES); + report_verify_result(&result); + report_prefix_pop(); + +error: + report_prefix_pop(); +} + +static void print_usage(void) +{ + report_info("Usage: migration-cmm [--parallel|--sequential]"); +} + +static void parse_args(int argc, char **argv) +{ + if (argc < 2) { + /* default to sequential since it only needs one CPU */ + arg_test_to_run = TEST_SEQUENTIAL; + return; } + + if (!strcmp("--parallel", argv[1])) + arg_test_to_run = TEST_PARALLEL; + else if (!strcmp("--sequential", argv[1])) + arg_test_to_run = TEST_SEQUENTIAL; + else + arg_test_to_run = TEST_INVLALID; } -int main(void) +int main(int argc, char **argv) { report_prefix_push("migration-cmm"); - - if (!check_essa_available()) + if (!check_essa_available()) { report_skip("ESSA is not available"); - else - test_migration(); + goto error; + } - migrate_once(); + parse_args(argc, argv); + + switch (arg_test_to_run) { + case TEST_SEQUENTIAL: + test_cmm_migration_sequential(); + break; + case TEST_PARALLEL: + test_cmm_migration_parallel(); + break; + default: + print_usage(); + } +error: + migrate_once(); report_prefix_pop(); return report_summary(); } diff --git a/s390x/unittests.cfg b/s390x/unittests.cfg index 3caf81eda396..c7fe1006e25d 100644 --- a/s390x/unittests.cfg +++ b/s390x/unittests.cfg @@ -181,10 +181,6 @@ file = migration.elf groups = migration smp = 2 -[migration-cmm] -file = migration-cmm.elf -groups = migration - [migration-skey] file = migration-skey.elf groups = migration @@ -208,3 +204,14 @@ groups = migration [exittime] file = exittime.elf smp = 2 + +[migration-cmm-sequential] +file = migration-cmm.elf +groups = migration +extra_params = -append '--sequential' + +[migration-cmm-parallel] +file = migration-cmm.elf +groups = migration +smp = 2 +extra_params = -append '--parallel'
Add a test which modifies CMM page states while migration is in progress. This is added to the existing migration-cmm test, which gets a new command line argument for the sequential and parallel variants. Signed-off-by: Nico Boehr <nrb@linux.ibm.com> --- s390x/migration-cmm.c | 258 +++++++++++++++++++++++++++++++++++++----- s390x/unittests.cfg | 15 ++- 2 files changed, 240 insertions(+), 33 deletions(-)