> From: Stephen Hemminger [mailto:step...@networkplumber.org] > Sent: Tuesday, 4 July 2023 01.24 > > RTE_ALIGN_MUL_NEAR is a macro so the cycle argument could > get evaluated twice causing some potential skew. Fix by > computing value once. > > Suggested by patch to fix side effects. > > Fixes: 5cbd14b3e5f9 ("eal: roundup TSC frequency when estimating") > Cc: pbhagavat...@marvell.com > Signed-off-by: Stephen Hemminger <step...@networkplumber.org> > --- > v2 - fix spelling error in commit message > > lib/eal/common/eal_common_timer.c | 8 ++++++-- > 1 file changed, 6 insertions(+), 2 deletions(-) > > diff --git a/lib/eal/common/eal_common_timer.c > b/lib/eal/common/eal_common_timer.c > index 5686a5102b66..05614b0503cf 100644 > --- a/lib/eal/common/eal_common_timer.c > +++ b/lib/eal/common/eal_common_timer.c > @@ -42,10 +42,14 @@ estimate_tsc_freq(void) > RTE_LOG(WARNING, EAL, "WARNING: TSC frequency estimated roughly" > " - clock timings may be less accurate.\n"); > /* assume that the rte_delay_us_sleep() will sleep for 1 second */ > - uint64_t start = rte_rdtsc(); > + uint64_t start, elapsed; > + > + start = rte_rdtsc(); > rte_delay_us_sleep(US_PER_S); > + elapsed = rte_rdtsc() - start; > + > /* Round up to 10Mhz. 1E7 ~ 10Mhz */ > - return RTE_ALIGN_MUL_NEAR(rte_rdtsc() - start, CYC_PER_10MHZ); > + return RTE_ALIGN_MUL_NEAR(elapsed, CYC_PER_10MHZ); > } > > void > -- > 2.39.2
Please fix the RTE_ALIGN_MUL_NEAR() macro instead. It already uses temporary variables with typeof() anyway. Other macros might have similar behavior of using their parameters more than once, and could be improved too.
