| Message ID | 1510075901-32305-2-git-send-email-ktraynor@redhat.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [ovs-dev,v2,1/2] dpif-netdev: Rename rxq_interval. | expand |
> rxq_cycle_sort summed the latest cycles from each queue for sorting. > While each comparison was correct with the latest cycles, the cycles could > change between calls to rxq_cycle_sort. In order to use consistent values > through each call to rxq_cycle_sort, sum the cycles prior to > rxq_cycle_sort being called. > > Also, change return to 0 when values are equal. > Thanks for this Kevin, I'll validate this patch with a view to add it to the DPDK merge branch, Minor comments below. > Requested-by: Ilya Maximets <i.maximets@samsung.com> @Ilya: do you have any comments? If not I'll look to push this in the next day after testing. > Signed-off-by: Kevin Traynor <ktraynor@redhat.com> > --- > v2: No change, but was previously 3/3. > > lib/dpif-netdev.c | 22 +++++++++++++--------- > 1 file changed, 13 insertions(+), 9 deletions(-) > > diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c index 66c14f9..d9e2912 > 100644 > --- a/lib/dpif-netdev.c > +++ b/lib/dpif-netdev.c > @@ -3458,18 +3458,14 @@ rxq_cycle_sort(const void *a, const void *b) > struct dp_netdev_rxq *qb; > uint64_t total_qa, total_qb; > - unsigned i; > > qa = *(struct dp_netdev_rxq **) a; > qb = *(struct dp_netdev_rxq **) b; > > - total_qa = total_qb = 0; > - for (i = 0; i < PMD_RXQ_INTERVAL_MAX; i++) { > - total_qa += dp_netdev_rxq_get_intrvl_cycles(qa, i); > - total_qb += dp_netdev_rxq_get_intrvl_cycles(qb, i); > - } > - dp_netdev_rxq_set_cycles(qa, RXQ_CYCLES_PROC_HIST, total_qa); > - dp_netdev_rxq_set_cycles(qb, RXQ_CYCLES_PROC_HIST, total_qb); > + total_qa = dp_netdev_rxq_get_cycles(qa, RXQ_CYCLES_PROC_HIST); > + total_qb = dp_netdev_rxq_get_cycles(qb, RXQ_CYCLES_PROC_HIST); > Minor nit but I think the comment for this function should be extended now to give a more detail of what happening and context on possible return values. (3 possible return values now). Maybe the behavior is detailed elsewhere in the code when rxq_cycle_sort is called but I didn't spot it. to I'm thinking for clarity this could be documented in the comment for the function overall? > - if (total_qa >= total_qb) { > + if (total_qa == total_qb) { > + return 0; > + } else if (total_qa > total_qb) { > return -1; > } > @@ -3519,4 +3515,6 @@ rxq_scheduling(struct dp_netdev *dp, bool pinned) > OVS_REQUIRES(dp->port_mutex) > } > } else if (!pinned && q->core_id == OVS_CORE_UNSPEC) { > + uint64_t cycle_hist = 0; > + > if (n_rxqs == 0) { > rxqs = xmalloc(sizeof *rxqs); @@ -3524,4 +3522,10 @@ > rxq_scheduling(struct dp_netdev *dp, bool pinned) OVS_REQUIRES(dp- > >port_mutex) > rxqs = xrealloc(rxqs, sizeof *rxqs * (n_rxqs + 1)); > } > + > + for (unsigned i = 0; i < PMD_RXQ_INTERVAL_MAX; i++) { > + cycle_hist += dp_netdev_rxq_get_intrvl_cycles(q, i); > + } > + dp_netdev_rxq_set_cycles(q, RXQ_CYCLES_PROC_HIST, > + cycle_hist); > + > /* Store the queue. */ > rxqs[n_rxqs++] = q; > -- > 1.8.3.1 > > _______________________________________________ > dev mailing list > dev@openvswitch.org > https://mail.openvswitch.org/mailman/listinfo/ovs-dev
On 11/13/2017 09:42 PM, Stokes, Ian wrote: >> rxq_cycle_sort summed the latest cycles from each queue for sorting. >> While each comparison was correct with the latest cycles, the cycles could >> change between calls to rxq_cycle_sort. In order to use consistent values >> through each call to rxq_cycle_sort, sum the cycles prior to >> rxq_cycle_sort being called. >> >> Also, change return to 0 when values are equal. >> > > Thanks for this Kevin, I'll validate this patch with a view to add it to the DPDK merge branch, Minor comments below. Thanks for reviewing. Actually, hold off on applying this as, based on feedback from Alin, I may need to change the sort a bit to give a consistent result for the unit test on Windows. > >> Requested-by: Ilya Maximets <i.maximets@samsung.com> > > @Ilya: do you have any comments? If not I'll look to push this in the next day after testing. > >> Signed-off-by: Kevin Traynor <ktraynor@redhat.com> >> --- >> v2: No change, but was previously 3/3. >> >> lib/dpif-netdev.c | 22 +++++++++++++--------- >> 1 file changed, 13 insertions(+), 9 deletions(-) >> >> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c index 66c14f9..d9e2912 >> 100644 >> --- a/lib/dpif-netdev.c >> +++ b/lib/dpif-netdev.c >> @@ -3458,18 +3458,14 @@ rxq_cycle_sort(const void *a, const void *b) >> struct dp_netdev_rxq *qb; >> uint64_t total_qa, total_qb; >> - unsigned i; >> >> qa = *(struct dp_netdev_rxq **) a; >> qb = *(struct dp_netdev_rxq **) b; >> >> - total_qa = total_qb = 0; >> - for (i = 0; i < PMD_RXQ_INTERVAL_MAX; i++) { >> - total_qa += dp_netdev_rxq_get_intrvl_cycles(qa, i); >> - total_qb += dp_netdev_rxq_get_intrvl_cycles(qb, i); >> - } >> - dp_netdev_rxq_set_cycles(qa, RXQ_CYCLES_PROC_HIST, total_qa); >> - dp_netdev_rxq_set_cycles(qb, RXQ_CYCLES_PROC_HIST, total_qb); >> + total_qa = dp_netdev_rxq_get_cycles(qa, RXQ_CYCLES_PROC_HIST); >> + total_qb = dp_netdev_rxq_get_cycles(qb, RXQ_CYCLES_PROC_HIST); >> > > Minor nit but I think the comment for this function should be extended now to give a more detail of what happening and context on possible return values. (3 possible return values now). Maybe the behavior is detailed elsewhere in the code when rxq_cycle_sort is called but I didn't spot it. > sure I can add some additional comments > to I'm thinking for clarity this could be documented in the comment for the function overall? >> - if (total_qa >= total_qb) { >> + if (total_qa == total_qb) { >> + return 0; >> + } else if (total_qa > total_qb) { >> return -1; >> } >> @@ -3519,4 +3515,6 @@ rxq_scheduling(struct dp_netdev *dp, bool pinned) >> OVS_REQUIRES(dp->port_mutex) >> } >> } else if (!pinned && q->core_id == OVS_CORE_UNSPEC) { >> + uint64_t cycle_hist = 0; >> + >> if (n_rxqs == 0) { >> rxqs = xmalloc(sizeof *rxqs); @@ -3524,4 +3522,10 @@ >> rxq_scheduling(struct dp_netdev *dp, bool pinned) OVS_REQUIRES(dp- >>> port_mutex) >> rxqs = xrealloc(rxqs, sizeof *rxqs * (n_rxqs + 1)); >> } >> + >> + for (unsigned i = 0; i < PMD_RXQ_INTERVAL_MAX; i++) { >> + cycle_hist += dp_netdev_rxq_get_intrvl_cycles(q, i); >> + } >> + dp_netdev_rxq_set_cycles(q, RXQ_CYCLES_PROC_HIST, >> + cycle_hist); >> + >> /* Store the queue. */ >> rxqs[n_rxqs++] = q; >> -- >> 1.8.3.1 >> >> _______________________________________________ >> dev mailing list >> dev@openvswitch.org >> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
On 11/13/2017 09:42 PM, Stokes, Ian wrote: >> rxq_cycle_sort summed the latest cycles from each queue for sorting. >> While each comparison was correct with the latest cycles, the cycles could >> change between calls to rxq_cycle_sort. In order to use consistent values >> through each call to rxq_cycle_sort, sum the cycles prior to >> rxq_cycle_sort being called. >> >> Also, change return to 0 when values are equal. >> > > Thanks for this Kevin, I'll validate this patch with a view to add it to the DPDK merge branch, Minor comments below. > >> Requested-by: Ilya Maximets <i.maximets@samsung.com> > > @Ilya: do you have any comments? If not I'll look to push this in the next day after testing. > >> Signed-off-by: Kevin Traynor <ktraynor@redhat.com> >> --- >> v2: No change, but was previously 3/3. >> >> lib/dpif-netdev.c | 22 +++++++++++++--------- >> 1 file changed, 13 insertions(+), 9 deletions(-) >> >> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c index 66c14f9..d9e2912 >> 100644 >> --- a/lib/dpif-netdev.c >> +++ b/lib/dpif-netdev.c >> @@ -3458,18 +3458,14 @@ rxq_cycle_sort(const void *a, const void *b) >> struct dp_netdev_rxq *qb; >> uint64_t total_qa, total_qb; >> - unsigned i; >> >> qa = *(struct dp_netdev_rxq **) a; >> qb = *(struct dp_netdev_rxq **) b; >> >> - total_qa = total_qb = 0; >> - for (i = 0; i < PMD_RXQ_INTERVAL_MAX; i++) { >> - total_qa += dp_netdev_rxq_get_intrvl_cycles(qa, i); >> - total_qb += dp_netdev_rxq_get_intrvl_cycles(qb, i); >> - } >> - dp_netdev_rxq_set_cycles(qa, RXQ_CYCLES_PROC_HIST, total_qa); >> - dp_netdev_rxq_set_cycles(qb, RXQ_CYCLES_PROC_HIST, total_qb); >> + total_qa = dp_netdev_rxq_get_cycles(qa, RXQ_CYCLES_PROC_HIST); >> + total_qb = dp_netdev_rxq_get_cycles(qb, RXQ_CYCLES_PROC_HIST); >> > > Minor nit but I think the comment for this function should be extended now to give a more detail of what happening and context on possible return values. (3 possible return values now). Maybe the behavior is detailed elsewhere in the code when rxq_cycle_sort is called but I didn't spot it. > Just to follow up - I've taken any changes around the return value out of the patch and sent v3. When Alin and I figure out the best approach for Windows, I will submit a separate patch to update the return value and update the comments then. > to I'm thinking for clarity this could be documented in the comment for the function overall? >> - if (total_qa >= total_qb) { >> + if (total_qa == total_qb) { >> + return 0; >> + } else if (total_qa > total_qb) { >> return -1; >> } >> @@ -3519,4 +3515,6 @@ rxq_scheduling(struct dp_netdev *dp, bool pinned) >> OVS_REQUIRES(dp->port_mutex) >> } >> } else if (!pinned && q->core_id == OVS_CORE_UNSPEC) { >> + uint64_t cycle_hist = 0; >> + >> if (n_rxqs == 0) { >> rxqs = xmalloc(sizeof *rxqs); @@ -3524,4 +3522,10 @@ >> rxq_scheduling(struct dp_netdev *dp, bool pinned) OVS_REQUIRES(dp- >>> port_mutex) >> rxqs = xrealloc(rxqs, sizeof *rxqs * (n_rxqs + 1)); >> } >> + >> + for (unsigned i = 0; i < PMD_RXQ_INTERVAL_MAX; i++) { >> + cycle_hist += dp_netdev_rxq_get_intrvl_cycles(q, i); >> + } >> + dp_netdev_rxq_set_cycles(q, RXQ_CYCLES_PROC_HIST, >> + cycle_hist); >> + >> /* Store the queue. */ >> rxqs[n_rxqs++] = q; >> -- >> 1.8.3.1 >> >> _______________________________________________ >> dev mailing list >> dev@openvswitch.org >> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
> On 11/13/2017 09:42 PM, Stokes, Ian wrote: > >> rxq_cycle_sort summed the latest cycles from each queue for sorting. > >> While each comparison was correct with the latest cycles, the cycles > >> could change between calls to rxq_cycle_sort. In order to use > >> consistent values through each call to rxq_cycle_sort, sum the cycles > >> prior to rxq_cycle_sort being called. > >> > >> Also, change return to 0 when values are equal. > >> > > > > Thanks for this Kevin, I'll validate this patch with a view to add it to > the DPDK merge branch, Minor comments below. > > > >> Requested-by: Ilya Maximets <i.maximets@samsung.com> > > > > @Ilya: do you have any comments? If not I'll look to push this in the > next day after testing. > > > >> Signed-off-by: Kevin Traynor <ktraynor@redhat.com> > >> --- > >> v2: No change, but was previously 3/3. > >> > >> lib/dpif-netdev.c | 22 +++++++++++++--------- > >> 1 file changed, 13 insertions(+), 9 deletions(-) > >> > >> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c index > >> 66c14f9..d9e2912 > >> 100644 > >> --- a/lib/dpif-netdev.c > >> +++ b/lib/dpif-netdev.c > >> @@ -3458,18 +3458,14 @@ rxq_cycle_sort(const void *a, const void *b) > >> struct dp_netdev_rxq *qb; > >> uint64_t total_qa, total_qb; > >> - unsigned i; > >> > >> qa = *(struct dp_netdev_rxq **) a; > >> qb = *(struct dp_netdev_rxq **) b; > >> > >> - total_qa = total_qb = 0; > >> - for (i = 0; i < PMD_RXQ_INTERVAL_MAX; i++) { > >> - total_qa += dp_netdev_rxq_get_intrvl_cycles(qa, i); > >> - total_qb += dp_netdev_rxq_get_intrvl_cycles(qb, i); > >> - } > >> - dp_netdev_rxq_set_cycles(qa, RXQ_CYCLES_PROC_HIST, total_qa); > >> - dp_netdev_rxq_set_cycles(qb, RXQ_CYCLES_PROC_HIST, total_qb); > >> + total_qa = dp_netdev_rxq_get_cycles(qa, RXQ_CYCLES_PROC_HIST); > >> + total_qb = dp_netdev_rxq_get_cycles(qb, RXQ_CYCLES_PROC_HIST); > >> > > > > Minor nit but I think the comment for this function should be extended > now to give a more detail of what happening and context on possible return > values. (3 possible return values now). Maybe the behavior is detailed > elsewhere in the code when rxq_cycle_sort is called but I didn't spot it. > > > > Just to follow up - I've taken any changes around the return value out of > the patch and sent v3. When Alin and I figure out the best approach for > Windows, I will submit a separate patch to update the return value and > update the comments then. > Sounds Good, I'm starting to look at the v3 rxq patch now. Thanks Ian > > to I'm thinking for clarity this could be documented in the comment for > the function overall? > >> - if (total_qa >= total_qb) { > >> + if (total_qa == total_qb) { > >> + return 0; > >> + } else if (total_qa > total_qb) { > >> return -1; > >> } > >> @@ -3519,4 +3515,6 @@ rxq_scheduling(struct dp_netdev *dp, bool > >> pinned) > >> OVS_REQUIRES(dp->port_mutex) > >> } > >> } else if (!pinned && q->core_id == OVS_CORE_UNSPEC) { > >> + uint64_t cycle_hist = 0; > >> + > >> if (n_rxqs == 0) { > >> rxqs = xmalloc(sizeof *rxqs); @@ -3524,4 > >> +3522,10 @@ rxq_scheduling(struct dp_netdev *dp, bool pinned) > >> OVS_REQUIRES(dp- > >>> port_mutex) > >> rxqs = xrealloc(rxqs, sizeof *rxqs * (n_rxqs + > 1)); > >> } > >> + > >> + for (unsigned i = 0; i < PMD_RXQ_INTERVAL_MAX; i++) { > >> + cycle_hist += dp_netdev_rxq_get_intrvl_cycles(q, > i); > >> + } > >> + dp_netdev_rxq_set_cycles(q, RXQ_CYCLES_PROC_HIST, > >> + cycle_hist); > >> + > >> /* Store the queue. */ > >> rxqs[n_rxqs++] = q; > >> -- > >> 1.8.3.1 > >> > >> _______________________________________________ > >> dev mailing list > >> dev@openvswitch.org > >> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c index 66c14f9..d9e2912 100644 --- a/lib/dpif-netdev.c +++ b/lib/dpif-netdev.c @@ -3458,18 +3458,14 @@ rxq_cycle_sort(const void *a, const void *b) struct dp_netdev_rxq *qb; uint64_t total_qa, total_qb; - unsigned i; qa = *(struct dp_netdev_rxq **) a; qb = *(struct dp_netdev_rxq **) b; - total_qa = total_qb = 0; - for (i = 0; i < PMD_RXQ_INTERVAL_MAX; i++) { - total_qa += dp_netdev_rxq_get_intrvl_cycles(qa, i); - total_qb += dp_netdev_rxq_get_intrvl_cycles(qb, i); - } - dp_netdev_rxq_set_cycles(qa, RXQ_CYCLES_PROC_HIST, total_qa); - dp_netdev_rxq_set_cycles(qb, RXQ_CYCLES_PROC_HIST, total_qb); + total_qa = dp_netdev_rxq_get_cycles(qa, RXQ_CYCLES_PROC_HIST); + total_qb = dp_netdev_rxq_get_cycles(qb, RXQ_CYCLES_PROC_HIST); - if (total_qa >= total_qb) { + if (total_qa == total_qb) { + return 0; + } else if (total_qa > total_qb) { return -1; } @@ -3519,4 +3515,6 @@ rxq_scheduling(struct dp_netdev *dp, bool pinned) OVS_REQUIRES(dp->port_mutex) } } else if (!pinned && q->core_id == OVS_CORE_UNSPEC) { + uint64_t cycle_hist = 0; + if (n_rxqs == 0) { rxqs = xmalloc(sizeof *rxqs); @@ -3524,4 +3522,10 @@ rxq_scheduling(struct dp_netdev *dp, bool pinned) OVS_REQUIRES(dp->port_mutex) rxqs = xrealloc(rxqs, sizeof *rxqs * (n_rxqs + 1)); } + + for (unsigned i = 0; i < PMD_RXQ_INTERVAL_MAX; i++) { + cycle_hist += dp_netdev_rxq_get_intrvl_cycles(q, i); + } + dp_netdev_rxq_set_cycles(q, RXQ_CYCLES_PROC_HIST, cycle_hist); + /* Store the queue. */ rxqs[n_rxqs++] = q;
rxq_cycle_sort summed the latest cycles from each queue for sorting. While each comparison was correct with the latest cycles, the cycles could change between calls to rxq_cycle_sort. In order to use consistent values through each call to rxq_cycle_sort, sum the cycles prior to rxq_cycle_sort being called. Also, change return to 0 when values are equal. Requested-by: Ilya Maximets <i.maximets@samsung.com> Signed-off-by: Kevin Traynor <ktraynor@redhat.com> --- v2: No change, but was previously 3/3. lib/dpif-netdev.c | 22 +++++++++++++--------- 1 file changed, 13 insertions(+), 9 deletions(-)