Perform a cache flush after ENTER PSCI timestamp capture

Without an explicit cache flush, the next timestamp captured might have
a bogus value.

This can happen if the following operations happen in order,
on a CPU that's being powered down.

1) ENTER PSCI timestamp is captured with caches enabled.

2) The next timestamp (ENTER_HW_LOW_PWR) is captured with caches
   disabled.

3) On a system that uses a write-back cache configuration, the
   cache line that holds the PMF timestamps is evicted.

After step 1), the ENTER_PSCI timestamp is cached and not in main memory.
After step 2), the ENTER_HW_LOW_PWR timestamp is stored in main memory.
Before the CPU power down happens, the hardware evicts the cache line that
contains the PMF timestamps for this service.  As a result, the timestamp
captured in step 2) is overwritten with a bogus value.

Change-Id: Ic1bd816498d1a6d4dc16540208ed3a5efe43f529
Signed-off-by: dp-arm <dimitris.papastamos@arm.com>
This commit is contained in:
dp-arm 2016-10-31 17:17:21 +00:00
parent fc1d1e2df0
commit bfef610667
1 changed files with 6 additions and 1 deletions

View File

@ -81,9 +81,14 @@ uintptr_t std_svc_smc_handler(uint32_t smc_fid,
uint64_t ret;
#if ENABLE_RUNTIME_INSTRUMENTATION
/*
* Flush cache line so that even if CPU power down happens
* the timestamp update is reflected in memory.
*/
PMF_WRITE_TIMESTAMP(rt_instr_svc,
RT_INSTR_ENTER_PSCI,
PMF_NO_CACHE_MAINT,
PMF_CACHE_MAINT,
get_cpu_data(cpu_data_pmf_ts[CPU_DATA_PMF_TS0_IDX]));
#endif