[dpdk-dev] [PATCH] rte_delay_us can be replaced with user function

Wiles, Keith keith.wiles at intel.com
Tue Jul 19 15:17:28 CEST 2016


Hi Jozef,

I have two quick comments inline.
> On Jul 19, 2016, at 7:42 AM, jozmarti at cisco.com wrote:
> 
> From: Jozef Martiniak <jozmarti at cisco.com>
> 
> when running single-core, some drivers tend to call rte_delay_us for a
> long time, and that is causing packet drops.
> Attached patch introduces 2 new functions:
> 
> void rte_delay_us_callback_register(void(*userfunc)(unsigned));
> void rte_delay_us_callback_unregister(void);
> 
> First one replaces rte_delay_us with userfunc and second one restores
> original rte_delay_us.
> Test user_delay_us is included.
> 
> Signed-off-by: Jozef Martiniak <jozmarti at cisco.com>
> ---
> app/test/test_cycles.c                             | 39 ++++++++++++++++++++++
> lib/librte_eal/common/eal_common_timer.c           | 19 +++++++++++
> lib/librte_eal/common/include/generic/rte_cycles.h | 13 ++++++++
> 3 files changed, 71 insertions(+)
> 
> diff --git a/app/test/test_cycles.c b/app/test/test_cycles.c
> index f6c043a..2b44a53 100644
> --- a/app/test/test_cycles.c
> +++ b/app/test/test_cycles.c
> @@ -90,3 +90,42 @@ test_cycles(void)
> }
> 
> REGISTER_TEST_COMMAND(cycles_autotest, test_cycles);
> +
> +/*
> + * rte_delay_us_callback test
> + *
> + * - check if callback is correctly registered/unregistered
> + *
> + */
> +
> +static int pattern;
> +static void my_rte_delay_us(unsigned us)
> +{
> +    pattern += us;
> +}
> +
> +static int
> +test_user_delay_us(void)
> +{
> +    pattern = 0;
> +
> +    rte_delay_us_callback_register(my_rte_delay_us);
> +
> +    rte_delay_us(2);
> +    if (pattern != 2)
> +        return -1;
> +
> +    rte_delay_us(3);
> +    if (pattern != 5)
> +        return -1;
> +
> +    rte_delay_us_callback_unregister();
> +
> +    rte_delay_us(3);
> +    if (pattern != 5)
> +        return -1;
> +
> +    return 0;
> +}
> +
> +REGISTER_TEST_COMMAND(user_delay_us, test_user_delay_us);
> diff --git a/lib/librte_eal/common/eal_common_timer.c b/lib/librte_eal/common/eal_common_timer.c
> index c4227cd..a982562 100644
> --- a/lib/librte_eal/common/eal_common_timer.c
> +++ b/lib/librte_eal/common/eal_common_timer.c
> @@ -47,9 +47,18 @@
> /* The frequency of the RDTSC timer resolution */
> static uint64_t eal_tsc_resolution_hz;
> 
> +/* User function which replaces rte_delay_us function */
> +static void (*rte_delay_us_override)(unsigned) = NULL;
> +
> void
> rte_delay_us(unsigned us)
> {
> +	if (unlikely(rte_delay_us_override != NULL))
> +	{
> +	    rte_delay_us_override(us);
> +	    return;
> +	}
> +
> 	const uint64_t start = rte_get_timer_cycles();
> 	const uint64_t ticks = (uint64_t)us * rte_get_timer_hz() / 1E6;
> 	while ((rte_get_timer_cycles() - start) < ticks)
> @@ -84,3 +93,13 @@ set_tsc_freq(void)
> 	RTE_LOG(DEBUG, EAL, "TSC frequency is ~%" PRIu64 " KHz\n", freq / 1000);
> 	eal_tsc_resolution_hz = freq;
> }
> +
> +void rte_delay_us_callback_register(void (*userfunc)(unsigned))
> +{
> +    rte_delay_us_override = userfunc;
> +}
> +
> +void rte_delay_us_callback_unregister(void)
> +{
> +    rte_delay_us_override = NULL;
> +}

I guess I would have used the rte_delay_us_callback_register(NULL) to unregister, but this is fine.

> diff --git a/lib/librte_eal/common/include/generic/rte_cycles.h b/lib/librte_eal/common/include/generic/rte_cycles.h
> index 8cc21f2..274f798 100644
> --- a/lib/librte_eal/common/include/generic/rte_cycles.h
> +++ b/lib/librte_eal/common/include/generic/rte_cycles.h
> @@ -202,4 +202,17 @@ rte_delay_ms(unsigned ms)
> 	rte_delay_us(ms * 1000);
> }
> 
> +/**
> + * Replace rte_delay_us with user defined function.
> + *
> + * @param userfunc
> + *   User function which replaces rte_delay_us.
> + */
> +void rte_delay_us_callback_register(void(*userfunc)(unsigned));
> +
> +/**
> + * Unregister user callback function. Restores original rte_delay_us.
> + */
> +void rte_delay_us_callback_unregister(void);

Just a note we need to add these two new APIs to the map file for ABI checking.

Other then these two comments I would give this one a +1 unless someone else has some comments.

> +
> #endif /* _RTE_CYCLES_H_ */
> -- 
> 2.1.4
> 



More information about the dev mailing list