[dpdk-dev] lib/table: fix wrong type of nht field

  From: Michal Kobylinski <michalx.kobylinski@intel.com>

Change type of nht field from uint32_t to uint8_t and increase max of
next hops.

Fixes: dc81ebbacaeb ("lpm: extend IPv4 next hop field")

Signed-off-by: Michal Kobylinski <michalx.kobylinski@intel.com>
Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
---
 examples/ip_pipeline/pipeline/pipeline_routing_be.h | 2 +-
 lib/librte_table/rte_table_lpm.c                    | 8 ++++----
 2 files changed, 5 insertions(+), 5 deletions(-)
  

2016-06-20 12:10, Michal Jastrzebski:
> From: Michal Kobylinski <michalx.kobylinski@intel.com>
> 
> Change type of nht field from uint32_t to uint8_t and increase max of
> next hops.
> 
> Fixes: dc81ebbacaeb ("lpm: extend IPv4 next hop field")

Why the type is wrong?
  

> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Thomas Monjalon
> Sent: Monday, June 20, 2016 11:14 AM
> To: Jastrzebski, MichalX K <michalx.k.jastrzebski@intel.com>; Kobylinski,
> MichalX <michalx.kobylinski@intel.com>
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH] lib/table: fix wrong type of nht field
> 
> 2016-06-20 12:10, Michal Jastrzebski:
> > From: Michal Kobylinski <michalx.kobylinski@intel.com>
> >
> > Change type of nht field from uint32_t to uint8_t and increase max of
> > next hops.
> >
> > Fixes: dc81ebbacaeb ("lpm: extend IPv4 next hop field")
> 
> Why the type is wrong?

The lpm->nht is simply some raw memory allocated at the end of the table context using the usual pattern:
struct rte_table_lpm {
	... 
	uint8_t nht[0] __rte_cache_aligned;
}

Therefore, when we do: 
	nht_entry = &lpm->nht[i * lpm->entry_size];
in several places, it makes big difference whether nht_entry is declared as uin8_t * (correct) or uint32_t * (incorrect), as the position computed by the latter is 4 times the position computed by the former ;(

Michal K and Michal J,
	I just realized we still need to do a small change to this patch, I will reply to the original mail now. So Thomas, sorry, there is one small change, we'll send new version soon.

Thanks,
Cristian
  

> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Michal Jastrzebski
> Sent: Monday, June 20, 2016 11:10 AM
> To: dev@dpdk.org
> Cc: Kobylinski, MichalX <michalx.kobylinski@intel.com>
> Subject: [dpdk-dev] [PATCH] lib/table: fix wrong type of nht field
> 
> From: Michal Kobylinski <michalx.kobylinski@intel.com>
> 
> Change type of nht field from uint32_t to uint8_t and increase max of
> next hops.
> 
> Fixes: dc81ebbacaeb ("lpm: extend IPv4 next hop field")
> 
> Signed-off-by: Michal Kobylinski <michalx.kobylinski@intel.com>
> Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
> ---
>  examples/ip_pipeline/pipeline/pipeline_routing_be.h | 2 +-
>  lib/librte_table/rte_table_lpm.c                    | 8 ++++----
>  2 files changed, 5 insertions(+), 5 deletions(-)
> 
> diff --git a/examples/ip_pipeline/pipeline/pipeline_routing_be.h
> b/examples/ip_pipeline/pipeline/pipeline_routing_be.h
> index 1276342..ea50896 100644
> --- a/examples/ip_pipeline/pipeline/pipeline_routing_be.h
> +++ b/examples/ip_pipeline/pipeline/pipeline_routing_be.h
> @@ -42,7 +42,7 @@
>   * Pipeline argument parsing
>   */
>  #ifndef PIPELINE_ROUTING_N_ROUTES_DEFAULT
> -#define PIPELINE_ROUTING_N_ROUTES_DEFAULT                  4096
> +#define PIPELINE_ROUTING_N_ROUTES_DEFAULT                  65536
>  #endif
> 

Changing the PIPELINE_ROUTING_N_ROUTES_DEFAULT  is actually not required, this is simply the default value which can be changed through the configuration file. Please remove this.

>  enum pipeline_routing_encap {
> diff --git a/lib/librte_table/rte_table_lpm.c
> b/lib/librte_table/rte_table_lpm.c
> index cdeb0f5..f2eaed5 100644
> --- a/lib/librte_table/rte_table_lpm.c
> +++ b/lib/librte_table/rte_table_lpm.c
> @@ -44,7 +44,7 @@
> 
>  #include "rte_table_lpm.h"
> 
> -#define RTE_TABLE_LPM_MAX_NEXT_HOPS                        256
> +#define RTE_TABLE_LPM_MAX_NEXT_HOPS                        65536
> 

With the next hop size of 24 bits, we can now make this configurable, so please use:

#ifndef RTE_TABLE_LPM_MAX_NEXT_HOPS
#define RTE_TABLE_LPM_MAX_NEXT_HOPS                        65536
#endif

>  #ifdef RTE_TABLE_STATS_COLLECT
> 
> @@ -74,7 +74,7 @@ struct rte_table_lpm {
> 
>  	/* Next Hop Table (NHT) */
>  	uint32_t nht_users[RTE_TABLE_LPM_MAX_NEXT_HOPS];
> -	uint32_t nht[0] __rte_cache_aligned;
> +	uint8_t nht[0] __rte_cache_aligned;
>  };
> 
>  static void *
> @@ -188,7 +188,7 @@ nht_find_existing(struct rte_table_lpm *lpm, void
> *entry, uint32_t *pos)
>  	uint32_t i;
> 
>  	for (i = 0; i < RTE_TABLE_LPM_MAX_NEXT_HOPS; i++) {
> -		uint32_t *nht_entry = &lpm->nht[i * lpm->entry_size];
> +		uint8_t *nht_entry = &lpm->nht[i * lpm->entry_size];
> 
>  		if ((lpm->nht_users[i] > 0) && (memcmp(nht_entry, entry,
>  			lpm->entry_unique_size) == 0)) {
> @@ -242,7 +242,7 @@ rte_table_lpm_entry_add(
> 
>  	/* Find existing or free NHT entry */
>  	if (nht_find_existing(lpm, entry, &nht_pos) == 0) {
> -		uint32_t *nht_entry;
> +		uint8_t *nht_entry;
> 
>  		if (nht_find_free(lpm, &nht_pos) == 0) {
>  			RTE_LOG(ERR, TABLE, "%s: NHT full\n", __func__);
> --
> 1.9.1