On Wed, 17 Dec 2014 07:57:02 -0600
Jay Rolette <rolette at infiniteio.com> wrote:
> Signed-off-by: Jay Rolette <rolette at infiniteio.com>
> ---
> lib/librte_kni/rte_kni.c | 21 ++++++++++++++++++++-
> 1 file changed, 20 insertions(+), 1 deletion(-)
>
> diff --git a/lib/librte_kni/rte_kni.c b/lib/librte_kni/rte_kni.c
> index fdb7509..f89319c 100644
> --- a/lib/librte_kni/rte_kni.c
> +++ b/lib/librte_kni/rte_kni.c
> @@ -40,6 +40,7 @@
> #include <unistd.h>
> #include <sys/ioctl.h>
>
> +#include <rte_cycles.h>
> #include <rte_spinlock.h>
> #include <rte_string_fns.h>
> #include <rte_ethdev.h>
> @@ -61,6 +62,9 @@
>
> #define KNI_MEM_CHECK(cond) do { if (cond) goto kni_fail; } while (0)
>
> +// Configure how often we log "out of memory" messages (in seconds)
> +#define KNI_SPAM_SUPPRESSION_PERIOD 60*10
> +
> /**
> * KNI context
> */
> @@ -592,6 +596,10 @@ kni_free_mbufs(struct rte_kni *kni)
> static void
> kni_allocate_mbufs(struct rte_kni *kni)
> {
> + static uint64_t no_mbufs = 0;
> + static uint64_t spam_filter = 0;
> + static uint64_t delayPeriod = 0;
> +
> int i, ret;
> struct rte_mbuf *pkts[MAX_MBUF_BURST_NUM];
>
> @@ -620,7 +628,18 @@ kni_allocate_mbufs(struct rte_kni *kni)
> pkts[i] = rte_pktmbuf_alloc(kni->pktmbuf_pool);
> if (unlikely(pkts[i] == NULL)) {
> /* Out of memory */
> - RTE_LOG(ERR, KNI, "Out of memory\n");
> + no_mbufs++;
> +
> + // Memory leak or need to tune? Regardless, if we get
> here once,
> + // we will get here a *lot*. Don't spam the logs!
> + now = rte_get_tsc_cycles();
> + if (!delayPeriod)
> + delayPeriod = rte_get_tsc_hz() *
> KNI_SPAM_SUPPRESSION_PERIOD;
> +
> + if (!spam_filter || (now - spam_filter) > delayPeriod) {
> + RTE_LOG(ERR, KNI, "No mbufs available
> (%llu)\n", (unsigned long long)no_mbufs);
> + spam_filter = now;
> + }
> break;
> }
> }
I agree whole completely with the intent of this.
But just remove the log message completely. It doesn't
help at all, use a statistic instead.
If you want to do ratelimiting, then it is better to create
a common function (see net_ratelimit() in Linux kernel) to
have all code use the same method, rather than reinventing private
code to do it.
Minor style complaints:
* don't use camelCase, it goes against the style of the rest of the code.
* don't use C++ style comments.
* always use rte_cycles() not TSC for things like this.
Please resubmit removing the log message and adding a statistic.
