benchmark: change ts buffer allocation flow
1. Change timestamp buffer allocation/mapping flow
2. Add misc cosmetic fixes
In case if timestamp buffer is allocated in userspace and new register
user memory API is used for its registering in OP-TEE (introduced in
optee_client commit 27888d73d156 ("tee_client_api: register user memory")),
there is no possibility to keep this mapping permanent among different
TEEC_InvokeCommand invocations. All all SHM are automatically unmapped from
OP-TEE VA space after TEEC_InvokeCommand is handled by OP-TEE.
Fixes: https://github.com/OP-TEE/optee_os/issues/1979
Acked-by: Joakim Bech <joakim.bech@linaro.org>
Signed-off-by: Igor Opaniuk <igor.opaniuk@linaro.org>
diff --git a/main.c b/main.c
index f6c0dde..ffbb4a5 100644
--- a/main.c
+++ b/main.c
@@ -49,13 +49,10 @@
static struct tee_ts_global *bench_ts_global;
static const TEEC_UUID pta_benchmark_uuid = PTA_BENCHMARK_UUID;
-static TEEC_SharedMemory ts_buf_shm = {
- .flags = TEEC_MEM_INPUT | TEEC_MEM_OUTPUT
-};
static TEEC_Context ctx;
static TEEC_Session sess;
-static volatile sig_atomic_t is_running;
+static sig_atomic_t is_running;
static yaml_emitter_t emitter;
@@ -82,52 +79,42 @@
static void close_bench_pta(void)
{
/* release benchmark timestamp shm */
- TEEC_ReleaseSharedMemory(&ts_buf_shm);
TEEC_CloseSession(&sess);
TEEC_FinalizeContext(&ctx);
}
-static void init_ts_global(void *ts_global, uint32_t cores)
-{
- unsigned int i;
- struct tee_ts_cpu_buf *cpu_buf;
-
- /* init global timestamp buffer */
- bench_ts_global = (struct tee_ts_global *)ts_global;
- bench_ts_global->cores = cores;
-
- /* init per-cpu timestamp buffers */
- for (i = 0; i < cores; i++) {
- cpu_buf = &bench_ts_global->cpu_buf[i];
- memset(cpu_buf, 0, sizeof(struct tee_ts_cpu_buf));
- }
-}
-
-static void register_bench_buf(uint32_t cores)
+static void alloc_bench_buf(uint32_t cores)
{
TEEC_Result res;
TEEC_Operation op = { 0 };
uint32_t ret_orig;
+ intptr_t paddr_ts_buf = 0;
+ size_t size;
- ts_buf_shm.size = sizeof(struct tee_ts_global) +
- sizeof(struct tee_ts_cpu_buf) * cores;
+ op.paramTypes = TEEC_PARAM_TYPES(TEEC_VALUE_INOUT,
+ TEEC_VALUE_INPUT, TEEC_NONE, TEEC_NONE);
- /* allocate global timestamp buffer */
- res = TEEC_AllocateSharedMemory(&ctx, &ts_buf_shm);
- tee_check_res(res, "TEEC_AllocateSharedMemory");
+ op.params[1].value.a = cores;
- init_ts_global(ts_buf_shm.buffer, cores);
-
- op.paramTypes = TEEC_PARAM_TYPES(TEEC_MEMREF_PARTIAL_INOUT,
- TEEC_NONE, TEEC_NONE, TEEC_NONE);
- op.params[0].memref.parent = &ts_buf_shm;
-
- TEEC_InvokeCommand(&sess, BENCHMARK_CMD_REGISTER_MEMREF,
+ res = TEEC_InvokeCommand(&sess, BENCHMARK_CMD_REGISTER_MEMREF,
&op, &ret_orig);
tee_check_res(res, "TEEC_InvokeCommand");
+
+ paddr_ts_buf = op.params[0].value.a;
+ size = op.params[0].value.b;
+
+ INFO("ts buffer paddr = %x, size = %d\n", paddr_ts_buf, size);
+ if (paddr_ts_buf) {
+
+ bench_ts_global = mmap_paddr(paddr_ts_buf, size);
+ if (!bench_ts_global)
+ ERROR_EXIT("Failed to allocate timestamp buffer");
+ } else {
+ ERROR_EXIT("Failed to allocate timestamp buffer");
+ }
}
-static void unregister_bench(void)
+static void free_bench_buf(void)
{
TEEC_Result res;
TEEC_Operation op = { 0 };
@@ -154,7 +141,7 @@
}
static int timestamp_pop(struct tee_ts_cpu_buf *cpu_buf,
- struct tee_time_st *ts)
+ struct tee_time_st *ts)
{
uint64_t ts_tail;
@@ -358,13 +345,15 @@
&ts_data);
if (!ret) {
ts_received = true;
- DBG("Timestamp: core = %u; tick = %lld; pc = 0x%"
- PRIx64 ";system = %s",
- i, ts_data.cnt, ts_data.addr,
- bench_str_src(ts_data.src));
+ DBG("Timestamp: core = %u; tick = %lld; "
+ "pc = 0x%" PRIx64 "; system = %s",
+ i, ts_data.cnt, ts_data.addr,
+ bench_str_src(ts_data.src));
if (!fill_timestamp(i, ts_data.cnt,
- ts_data.addr, bench_str_src(ts_data.src)))
- ERROR_GOTO(deinit_yaml, "Adding timestamp failed");
+ ts_data.addr,
+ bench_str_src(ts_data.src)))
+ ERROR_GOTO(deinit_yaml,
+ "Adding timestamp failed");
}
}
@@ -375,7 +364,8 @@
sched_yield();
} else {
ERROR_GOTO(deinit_yaml,
- "No new data in the per-cpu ringbuffers, closing ts file");
+ "No new data in the per-cpu ringbuffers"
+ );
}
}
}
@@ -424,7 +414,7 @@
INFO("2. Allocating per-core buffers, cores detected = %d",
cores);
- register_bench_buf(cores);
+ alloc_bench_buf(cores);
res = realpath(argv[1], testapp_path);
if (!res)
@@ -479,7 +469,7 @@
INFO("4. Done benchmark");
dealloc_argv(argc-1, testapp_argv);
- unregister_bench();
+ free_bench_buf();
close_bench_pta();
return 0;
}