[v2] lib/librte_meter: fix divide by zero for RFC4115 meter

Message ID 154903721465.46418.12675922836824541692.stgit@dbuild
State Changes Requested
Delegated to: Cristian Dumitrescu
Headers show
Series
  • [v2] lib/librte_meter: fix divide by zero for RFC4115 meter
Related show

Checks

Context Check Description
ci/intel-Performance-Testing success Performance Testing PASS
ci/mellanox-Performance-Testing success Performance Testing PASS
ci/Intel-compilation success Compilation OK

Commit Message

Eelco Chaudron Feb. 1, 2019, 4:07 p.m.
RFC 4115 allows a meter with either cir and/or eir configured.
When only one is configured a divide by zero would occur.

Fixes: 655796d2b5fb ("meter: support RFC4115 trTCM")
Cc: echaudro@redhat.com

Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
---

 v2 - Removed configuration change that got included by accident

 lib/librte_meter/rte_meter.h |    8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

Comments

Jens Freimann Feb. 5, 2019, 8:40 a.m. | #1
On Fri, Feb 01, 2019 at 04:07:16PM +0000, Eelco Chaudron wrote:
>RFC 4115 allows a meter with either cir and/or eir configured.
>When only one is configured a divide by zero would occur.
>
>Fixes: 655796d2b5fb ("meter: support RFC4115 trTCM")
>Cc: echaudro@redhat.com
>
>Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
>---
>
> v2 - Removed configuration change that got included by accident
>
> lib/librte_meter/rte_meter.h |    8 ++++----
> 1 file changed, 4 insertions(+), 4 deletions(-)
>

Reviewed-by: Jens Freimann <jfreimann@redhat.com>
Dumitrescu, Cristian Feb. 28, 2019, 6:50 p.m. | #2
Hi Eelco,

Sorry for my delayed reply,

> -----Original Message-----
> From: Eelco Chaudron [mailto:echaudro@redhat.com]
> Sent: Friday, February 1, 2019 4:07 PM
> To: Dumitrescu, Cristian <cristian.dumitrescu@intel.com>
> Cc: dev@dpdk.org
> Subject: [PATCH v2] lib/librte_meter: fix divide by zero for RFC4115 meter
> 
> RFC 4115 allows a meter with either cir and/or eir configured.
> When only one is configured a divide by zero would occur.
> 
> Fixes: 655796d2b5fb ("meter: support RFC4115 trTCM")
> Cc: echaudro@redhat.com
> 
> Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
> ---
> 
>  v2 - Removed configuration change that got included by accident
> 
>  lib/librte_meter/rte_meter.h |    8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/lib/librte_meter/rte_meter.h b/lib/librte_meter/rte_meter.h
> index 005e4eeee..56d85ecf0 100644
> --- a/lib/librte_meter/rte_meter.h
> +++ b/lib/librte_meter/rte_meter.h
> @@ -597,8 +597,8 @@ rte_meter_trtcm_rfc4115_color_blind_check(
>  	/* Bucket update */
>  	time_diff_tc = time - m->time_tc;
>  	time_diff_te = time - m->time_te;
> -	n_periods_tc = time_diff_tc / p->cir_period;
> -	n_periods_te = time_diff_te / p->eir_period;
> +	n_periods_tc = p->cir_period != 0 ? time_diff_tc / p->cir_period : 0;
> +	n_periods_te = p->eir_period != 0 ? time_diff_te / p->eir_period : 0;
>  	m->time_tc += n_periods_tc * p->cir_period;
>  	m->time_te += n_periods_te * p->eir_period;
> 
> @@ -641,8 +641,8 @@ rte_meter_trtcm_rfc4115_color_aware_check(
>  	/* Bucket update */
>  	time_diff_tc = time - m->time_tc;
>  	time_diff_te = time - m->time_te;
> -	n_periods_tc = time_diff_tc / p->cir_period;
> -	n_periods_te = time_diff_te / p->eir_period;
> +	n_periods_tc = p->cir_period != 0 ? time_diff_tc / p->cir_period : 0;
> +	n_periods_te = p->eir_period != 0 ? time_diff_te / p->eir_period : 0;
>  	m->time_tc += n_periods_tc * p->cir_period;
>  	m->time_te += n_periods_te * p->eir_period;
> 

Yes, this is indeed an issue, good catch!

For performance reasons, we'd like to avoid a test on the fast path (rte_meter_xyz_check) and replace it with more work done on the configuration stage (rte_meter_profile_xyz_config), if possible.

We can intercept the null CIR or EIR cases very early in rte_meter_trtcm_rfc4115_profile_config() and deal with them separately by setting the CIR/EIR period and bytes_per_period to some neutral values and skipping the call to rte_meter_get_tb_params(): period = RTE_METER_TB_PERIOD_MIN, bytes_per_period = 0.

Makes sense?

Regards,
Cristian
Eelco Chaudron March 19, 2019, 11 a.m. | #3
On 28 Feb 2019, at 19:50, Dumitrescu, Cristian wrote:

> Hi Eelco,
>
> Sorry for my delayed reply

No problem I was OOO also so hence my late reply ;)

>> -----Original Message-----
>> From: Eelco Chaudron [mailto:echaudro@redhat.com]
>> Sent: Friday, February 1, 2019 4:07 PM
>> To: Dumitrescu, Cristian <cristian.dumitrescu@intel.com>
>> Cc: dev@dpdk.org
>> Subject: [PATCH v2] lib/librte_meter: fix divide by zero for RFC4115 
>> meter
>>
>> RFC 4115 allows a meter with either cir and/or eir configured.
>> When only one is configured a divide by zero would occur.
>>
>> Fixes: 655796d2b5fb ("meter: support RFC4115 trTCM")
>> Cc: echaudro@redhat.com
>>
>> Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
>> ---
>>
>>  v2 - Removed configuration change that got included by accident
>>
>>  lib/librte_meter/rte_meter.h |    8 ++++----
>>  1 file changed, 4 insertions(+), 4 deletions(-)
>>
>> diff --git a/lib/librte_meter/rte_meter.h 
>> b/lib/librte_meter/rte_meter.h
>> index 005e4eeee..56d85ecf0 100644
>> --- a/lib/librte_meter/rte_meter.h
>> +++ b/lib/librte_meter/rte_meter.h
>> @@ -597,8 +597,8 @@ rte_meter_trtcm_rfc4115_color_blind_check(
>>  	/* Bucket update */
>>  	time_diff_tc = time - m->time_tc;
>>  	time_diff_te = time - m->time_te;
>> -	n_periods_tc = time_diff_tc / p->cir_period;
>> -	n_periods_te = time_diff_te / p->eir_period;
>> +	n_periods_tc = p->cir_period != 0 ? time_diff_tc / p->cir_period : 
>> 0;
>> +	n_periods_te = p->eir_period != 0 ? time_diff_te / p->eir_period : 
>> 0;
>>  	m->time_tc += n_periods_tc * p->cir_period;
>>  	m->time_te += n_periods_te * p->eir_period;
>>
>> @@ -641,8 +641,8 @@ rte_meter_trtcm_rfc4115_color_aware_check(
>>  	/* Bucket update */
>>  	time_diff_tc = time - m->time_tc;
>>  	time_diff_te = time - m->time_te;
>> -	n_periods_tc = time_diff_tc / p->cir_period;
>> -	n_periods_te = time_diff_te / p->eir_period;
>> +	n_periods_tc = p->cir_period != 0 ? time_diff_tc / p->cir_period : 
>> 0;
>> +	n_periods_te = p->eir_period != 0 ? time_diff_te / p->eir_period : 
>> 0;
>>  	m->time_tc += n_periods_tc * p->cir_period;
>>  	m->time_te += n_periods_te * p->eir_period;
>>
>
> Yes, this is indeed an issue, good catch!
>
> For performance reasons, we'd like to avoid a test on the fast path 
> (rte_meter_xyz_check) and replace it with more work done on the 
> configuration stage (rte_meter_profile_xyz_config), if possible.
>
> We can intercept the null CIR or EIR cases very early in 
> rte_meter_trtcm_rfc4115_profile_config() and deal with them separately 
> by setting the CIR/EIR period and bytes_per_period to some neutral 
> values and skipping the call to rte_meter_get_tb_params(): period = 
> RTE_METER_TB_PERIOD_MIN, bytes_per_period = 0.

Excellent idea, will sent out a v3 soon, after some testing…

//Eelco

Patch

diff --git a/lib/librte_meter/rte_meter.h b/lib/librte_meter/rte_meter.h
index 005e4eeee..56d85ecf0 100644
--- a/lib/librte_meter/rte_meter.h
+++ b/lib/librte_meter/rte_meter.h
@@ -597,8 +597,8 @@  rte_meter_trtcm_rfc4115_color_blind_check(
 	/* Bucket update */
 	time_diff_tc = time - m->time_tc;
 	time_diff_te = time - m->time_te;
-	n_periods_tc = time_diff_tc / p->cir_period;
-	n_periods_te = time_diff_te / p->eir_period;
+	n_periods_tc = p->cir_period != 0 ? time_diff_tc / p->cir_period : 0;
+	n_periods_te = p->eir_period != 0 ? time_diff_te / p->eir_period : 0;
 	m->time_tc += n_periods_tc * p->cir_period;
 	m->time_te += n_periods_te * p->eir_period;
 
@@ -641,8 +641,8 @@  rte_meter_trtcm_rfc4115_color_aware_check(
 	/* Bucket update */
 	time_diff_tc = time - m->time_tc;
 	time_diff_te = time - m->time_te;
-	n_periods_tc = time_diff_tc / p->cir_period;
-	n_periods_te = time_diff_te / p->eir_period;
+	n_periods_tc = p->cir_period != 0 ? time_diff_tc / p->cir_period : 0;
+	n_periods_te = p->eir_period != 0 ? time_diff_te / p->eir_period : 0;
 	m->time_tc += n_periods_tc * p->cir_period;
 	m->time_te += n_periods_te * p->eir_period;