* Re: [PATCH 19.11 4/6] net/mlx5: fix GRE protocol type translation
2021-12-02 16:03 [PATCH 19.11 4/6] net/mlx5: fix GRE protocol type translation Gregory Etelson
@ 2021-12-06 9:54 ` Christian Ehrhardt
2021-12-06 9:57 ` Christian Ehrhardt
2021-12-06 10:38 ` [PATCH 19.11 v2 1/2] " Gregory Etelson
2021-12-06 10:39 ` [PATCH 19.11 v2 2/2] net/mlx5: fix GRE flow item matching Gregory Etelson
2 siblings, 1 reply; 6+ messages in thread
From: Christian Ehrhardt @ 2021-12-06 9:54 UTC (permalink / raw)
To: Gregory Etelson
Cc: stable, Viacheslav Ovsiienko, Matan Azrad, Shahaf Shuler,
Ori Kam, Yongseok Koh
On Thu, Dec 2, 2021 at 5:04 PM Gregory Etelson <getelson@nvidia.com> wrote:
>
> [ upstream commit be26e81bfc1cb681d3ee0764aa6733000869984a ]
>
> When application creates several flows to match on GRE tunnel
> without explicitly specifying GRE protocol type value in
> flow rules, PMD will translate that to zero mask.
> RDMA-CORE cannot distinguish between different inner flow types and
> produces identical matchers for each zero mask.
>
> The patch extracts inner header type from flow rule and forces it
> in GRE protocol type, if application did not specify
> any without explicitly specifying GRE protocol type value in
> flow rules, protocol type value.
>
> Fixes: fc2c498ccb94 ("net/mlx5: add Direct Verbs translate items")
> Cc: stable@dpdk.org
>
> Signed-off-by: Gregory Etelson <getelson@nvidia.com>
> Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
> ---
> drivers/net/mlx5/mlx5_flow_dv.c | 70 +++++++++++++++++++--------------
> 1 file changed, 40 insertions(+), 30 deletions(-)
>
> diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
> index 8dec8d9ff5..edbdba3e23 100644
> --- a/drivers/net/mlx5/mlx5_flow_dv.c
> +++ b/drivers/net/mlx5/mlx5_flow_dv.c
> @@ -5875,18 +5875,19 @@ flow_dv_translate_item_gre_key(void *matcher, void *key,
> * Flow matcher value.
> * @param[in] item
> * Flow pattern to translate.
> - * @param[in] inner
> - * Item is inner pattern.
> + * @param[in] pattern_flags
> + * Accumulated pattern flags.
> */
> static void
> flow_dv_translate_item_gre(void *matcher, void *key,
> const struct rte_flow_item *item,
> - int inner)
> + uint64_t pattern_flags)
> {
> + static const struct rte_flow_item_gre empty_gre = {0,};
> const struct rte_flow_item_gre *gre_m = item->mask;
> const struct rte_flow_item_gre *gre_v = item->spec;
> - void *headers_m;
> - void *headers_v;
> + void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
> + void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
> void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
> void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
> struct {
> @@ -5903,26 +5904,17 @@ flow_dv_translate_item_gre(void *matcher, void *key,
> uint16_t value;
> };
> } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
> + uint16_t protocol_m, protocol_v;
>
> - if (inner) {
> - headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
> - inner_headers);
> - headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
> - } else {
> - headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
> - outer_headers);
> - headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
> - }
> MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
> MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
> - if (!gre_v)
> - return;
> - if (!gre_m)
> - gre_m = &rte_flow_item_gre_mask;
> - MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
> - rte_be_to_cpu_16(gre_m->protocol));
> - MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
> - rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
> + if (!gre_v) {
> + gre_v = &empty_gre;
> + gre_m = &empty_gre;
> + } else {
> + if (!gre_m)
> + gre_m = &rte_flow_item_gre_mask;
> + }
> gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
> gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
> MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
> @@ -5940,6 +5932,16 @@ flow_dv_translate_item_gre(void *matcher, void *key,
> MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
> gre_crks_rsvd0_ver_v.s_present &
> gre_crks_rsvd0_ver_m.s_present);
> + protocol_m = rte_be_to_cpu_16(gre_m->protocol);
> + protocol_v = rte_be_to_cpu_16(gre_v->protocol);
> + if (!protocol_m) {
> + /* Force next protocol to prevent matchers duplication */
> + protocol_m = 0xFFFF;
> + protocol_v = mlx5_translate_tunnel_etypes(pattern_flags);
> + }
> + MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, protocol_m);
> + MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
> + protocol_m & protocol_v);
> }
>
> /**
> @@ -5951,13 +5953,13 @@ flow_dv_translate_item_gre(void *matcher, void *key,
> * Flow matcher value.
> * @param[in] item
> * Flow pattern to translate.
> - * @param[in] inner
> - * Item is inner pattern.
> + * @param[in] pattern_flags
> + * Accumulated pattern flags.
> */
> static void
> flow_dv_translate_item_nvgre(void *matcher, void *key,
> const struct rte_flow_item *item,
> - int inner)
> + unsigned long pattern_flags)
> {
> const struct rte_flow_item_nvgre *nvgre_m = item->mask;
> const struct rte_flow_item_nvgre *nvgre_v = item->spec;
> @@ -5984,7 +5986,7 @@ flow_dv_translate_item_nvgre(void *matcher, void *key,
> .mask = &gre_mask,
> .last = NULL,
> };
> - flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
> + flow_dv_translate_item_gre(matcher, key, &gre_item, pattern_flags);
> if (!nvgre_v)
> return;
> if (!nvgre_m)
> @@ -7788,11 +7790,10 @@ __flow_dv_translate(struct rte_eth_dev *dev,
> MLX5_FLOW_LAYER_OUTER_L4_UDP;
> break;
> case RTE_FLOW_ITEM_TYPE_GRE:
> - flow_dv_translate_item_gre(match_mask, match_value,
> - items, tunnel);
> matcher.priority = flow->rss.level >= 2 ?
> MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
> last_item = MLX5_FLOW_LAYER_GRE;
> + tunnel_item = items;
> break;
> case RTE_FLOW_ITEM_TYPE_GRE_KEY:
> flow_dv_translate_item_gre_key(match_mask,
> @@ -7800,11 +7801,10 @@ __flow_dv_translate(struct rte_eth_dev *dev,
> last_item = MLX5_FLOW_LAYER_GRE_KEY;
> break;
> case RTE_FLOW_ITEM_TYPE_NVGRE:
> - flow_dv_translate_item_nvgre(match_mask, match_value,
> - items, tunnel);
> matcher.priority = flow->rss.level >= 2 ?
> MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
> last_item = MLX5_FLOW_LAYER_GRE;
> + tunnel_item = items;
> break;
> case RTE_FLOW_ITEM_TYPE_VXLAN:
> flow_dv_translate_item_vxlan(match_mask, match_value,
> @@ -7892,6 +7892,16 @@ __flow_dv_translate(struct rte_eth_dev *dev,
> else if (item_flags & MLX5_FLOW_LAYER_GENEVE)
> flow_dv_translate_item_geneve(match_mask, match_value,
> tunnel_item, item_flags);
> + else if (item_flags & MLX5_FLOW_LAYER_GRE) {
> + if (tunnel_item->type == RTE_FLOW_ITEM_TYPE_GRE)
> + flow_dv_translate_item_gre(match_mask, match_value,
> + tunnel_item, item_flags);
> + else if (tunnel_item->type == RTE_FLOW_ITEM_TYPE_NVGRE)
> + flow_dv_translate_item_nvgre(match_mask, match_value,
> + tunnel_item, item_flags);
> + else
> + MLX5_ASSERT(false);
Hi,
build tests have shown that this breaks on 19.11 like:
[1479/2115] Compiling C object
drivers/libtmp_rte_pmd_mlx5.a.p/net_mlx5_mlx5_flow_dv.c.o
../drivers/net/mlx5/mlx5_flow_dv.c: In function ‘__flow_dv_translate’:
../drivers/net/mlx5/mlx5_flow_dv.c:7912:4: warning: implicit
declaration of function ‘MLX5_ASSERT’; did you mean ‘MLX5_SET’?
[-Wimplicit-function-declaration]
7912 | MLX5_ASSERT(false);
| ^~~~~~~~~~~
| MLX5_SET
../drivers/net/mlx5/mlx5_flow_dv.c:7912:4: warning: nested extern
declaration of ‘MLX5_ASSERT’ [-Wnested-externs]
[1538/2115] Linking target drivers/librte_pmd_mlx5.so.20.0
FAILED: drivers/librte_pmd_mlx5.so.20.0
For now - I'll remove the patch from the WIP 19.11 branch, please
consider submitting a backport adapted to work on
https://github.com/cpaelzer/dpdk-stable-queue/tree/19.11
> + }
> assert(!flow_dv_check_valid_spec(matcher.mask.buf,
> dev_flow->dv.value.buf));
> /*
> --
> 2.34.0
>
--
Christian Ehrhardt
Staff Engineer, Ubuntu Server
Canonical Ltd
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [PATCH 19.11 4/6] net/mlx5: fix GRE protocol type translation
2021-12-06 9:54 ` Christian Ehrhardt
@ 2021-12-06 9:57 ` Christian Ehrhardt
2021-12-06 10:10 ` Gregory Etelson
0 siblings, 1 reply; 6+ messages in thread
From: Christian Ehrhardt @ 2021-12-06 9:57 UTC (permalink / raw)
To: Gregory Etelson
Cc: stable, Viacheslav Ovsiienko, Matan Azrad, Shahaf Shuler,
Ori Kam, Yongseok Koh
On Mon, Dec 6, 2021 at 10:54 AM Christian Ehrhardt
<christian.ehrhardt@canonical.com> wrote:
>
> On Thu, Dec 2, 2021 at 5:04 PM Gregory Etelson <getelson@nvidia.com> wrote:
> >
> > [ upstream commit be26e81bfc1cb681d3ee0764aa6733000869984a ]
> >
> > When application creates several flows to match on GRE tunnel
> > without explicitly specifying GRE protocol type value in
> > flow rules, PMD will translate that to zero mask.
> > RDMA-CORE cannot distinguish between different inner flow types and
> > produces identical matchers for each zero mask.
> >
> > The patch extracts inner header type from flow rule and forces it
> > in GRE protocol type, if application did not specify
> > any without explicitly specifying GRE protocol type value in
> > flow rules, protocol type value.
> >
> > Fixes: fc2c498ccb94 ("net/mlx5: add Direct Verbs translate items")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Gregory Etelson <getelson@nvidia.com>
> > Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
> > ---
> > drivers/net/mlx5/mlx5_flow_dv.c | 70 +++++++++++++++++++--------------
> > 1 file changed, 40 insertions(+), 30 deletions(-)
> >
> > diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
> > index 8dec8d9ff5..edbdba3e23 100644
> > --- a/drivers/net/mlx5/mlx5_flow_dv.c
> > +++ b/drivers/net/mlx5/mlx5_flow_dv.c
> > @@ -5875,18 +5875,19 @@ flow_dv_translate_item_gre_key(void *matcher, void *key,
> > * Flow matcher value.
> > * @param[in] item
> > * Flow pattern to translate.
> > - * @param[in] inner
> > - * Item is inner pattern.
> > + * @param[in] pattern_flags
> > + * Accumulated pattern flags.
> > */
> > static void
> > flow_dv_translate_item_gre(void *matcher, void *key,
> > const struct rte_flow_item *item,
> > - int inner)
> > + uint64_t pattern_flags)
> > {
> > + static const struct rte_flow_item_gre empty_gre = {0,};
> > const struct rte_flow_item_gre *gre_m = item->mask;
> > const struct rte_flow_item_gre *gre_v = item->spec;
> > - void *headers_m;
> > - void *headers_v;
> > + void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
> > + void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
> > void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
> > void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
> > struct {
> > @@ -5903,26 +5904,17 @@ flow_dv_translate_item_gre(void *matcher, void *key,
> > uint16_t value;
> > };
> > } gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
> > + uint16_t protocol_m, protocol_v;
> >
> > - if (inner) {
> > - headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
> > - inner_headers);
> > - headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
> > - } else {
> > - headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
> > - outer_headers);
> > - headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
> > - }
> > MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
> > MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
> > - if (!gre_v)
> > - return;
> > - if (!gre_m)
> > - gre_m = &rte_flow_item_gre_mask;
> > - MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
> > - rte_be_to_cpu_16(gre_m->protocol));
> > - MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
> > - rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
> > + if (!gre_v) {
> > + gre_v = &empty_gre;
> > + gre_m = &empty_gre;
> > + } else {
> > + if (!gre_m)
> > + gre_m = &rte_flow_item_gre_mask;
> > + }
> > gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
> > gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
> > MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
> > @@ -5940,6 +5932,16 @@ flow_dv_translate_item_gre(void *matcher, void *key,
> > MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
> > gre_crks_rsvd0_ver_v.s_present &
> > gre_crks_rsvd0_ver_m.s_present);
> > + protocol_m = rte_be_to_cpu_16(gre_m->protocol);
> > + protocol_v = rte_be_to_cpu_16(gre_v->protocol);
> > + if (!protocol_m) {
> > + /* Force next protocol to prevent matchers duplication */
> > + protocol_m = 0xFFFF;
> > + protocol_v = mlx5_translate_tunnel_etypes(pattern_flags);
> > + }
> > + MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, protocol_m);
> > + MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
> > + protocol_m & protocol_v);
> > }
> >
> > /**
> > @@ -5951,13 +5953,13 @@ flow_dv_translate_item_gre(void *matcher, void *key,
> > * Flow matcher value.
> > * @param[in] item
> > * Flow pattern to translate.
> > - * @param[in] inner
> > - * Item is inner pattern.
> > + * @param[in] pattern_flags
> > + * Accumulated pattern flags.
> > */
> > static void
> > flow_dv_translate_item_nvgre(void *matcher, void *key,
> > const struct rte_flow_item *item,
> > - int inner)
> > + unsigned long pattern_flags)
> > {
> > const struct rte_flow_item_nvgre *nvgre_m = item->mask;
> > const struct rte_flow_item_nvgre *nvgre_v = item->spec;
> > @@ -5984,7 +5986,7 @@ flow_dv_translate_item_nvgre(void *matcher, void *key,
> > .mask = &gre_mask,
> > .last = NULL,
> > };
> > - flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
> > + flow_dv_translate_item_gre(matcher, key, &gre_item, pattern_flags);
> > if (!nvgre_v)
> > return;
> > if (!nvgre_m)
> > @@ -7788,11 +7790,10 @@ __flow_dv_translate(struct rte_eth_dev *dev,
> > MLX5_FLOW_LAYER_OUTER_L4_UDP;
> > break;
> > case RTE_FLOW_ITEM_TYPE_GRE:
> > - flow_dv_translate_item_gre(match_mask, match_value,
> > - items, tunnel);
> > matcher.priority = flow->rss.level >= 2 ?
> > MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
> > last_item = MLX5_FLOW_LAYER_GRE;
> > + tunnel_item = items;
> > break;
> > case RTE_FLOW_ITEM_TYPE_GRE_KEY:
> > flow_dv_translate_item_gre_key(match_mask,
> > @@ -7800,11 +7801,10 @@ __flow_dv_translate(struct rte_eth_dev *dev,
> > last_item = MLX5_FLOW_LAYER_GRE_KEY;
> > break;
> > case RTE_FLOW_ITEM_TYPE_NVGRE:
> > - flow_dv_translate_item_nvgre(match_mask, match_value,
> > - items, tunnel);
> > matcher.priority = flow->rss.level >= 2 ?
> > MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
> > last_item = MLX5_FLOW_LAYER_GRE;
> > + tunnel_item = items;
> > break;
> > case RTE_FLOW_ITEM_TYPE_VXLAN:
> > flow_dv_translate_item_vxlan(match_mask, match_value,
> > @@ -7892,6 +7892,16 @@ __flow_dv_translate(struct rte_eth_dev *dev,
> > else if (item_flags & MLX5_FLOW_LAYER_GENEVE)
> > flow_dv_translate_item_geneve(match_mask, match_value,
> > tunnel_item, item_flags);
> > + else if (item_flags & MLX5_FLOW_LAYER_GRE) {
> > + if (tunnel_item->type == RTE_FLOW_ITEM_TYPE_GRE)
> > + flow_dv_translate_item_gre(match_mask, match_value,
> > + tunnel_item, item_flags);
> > + else if (tunnel_item->type == RTE_FLOW_ITEM_TYPE_NVGRE)
> > + flow_dv_translate_item_nvgre(match_mask, match_value,
> > + tunnel_item, item_flags);
> > + else
> > + MLX5_ASSERT(false);
>
> Hi,
> build tests have shown that this breaks on 19.11 like:
>
> [1479/2115] Compiling C object
> drivers/libtmp_rte_pmd_mlx5.a.p/net_mlx5_mlx5_flow_dv.c.o
> ../drivers/net/mlx5/mlx5_flow_dv.c: In function ‘__flow_dv_translate’:
> ../drivers/net/mlx5/mlx5_flow_dv.c:7912:4: warning: implicit
> declaration of function ‘MLX5_ASSERT’; did you mean ‘MLX5_SET’?
> [-Wimplicit-function-declaration]
> 7912 | MLX5_ASSERT(false);
> | ^~~~~~~~~~~
> | MLX5_SET
> ../drivers/net/mlx5/mlx5_flow_dv.c:7912:4: warning: nested extern
> declaration of ‘MLX5_ASSERT’ [-Wnested-externs]
> [1538/2115] Linking target drivers/librte_pmd_mlx5.so.20.0
> FAILED: drivers/librte_pmd_mlx5.so.20.0
>
>
> For now - I'll remove the patch from the WIP 19.11 branch, please
> consider submitting a backport adapted to work on
> https://github.com/cpaelzer/dpdk-stable-queue/tree/19.11
And dropping that makes the subsequent "net/mlx5: fix GRE flow item
matching" fail to apply well, so I had to drop that as well for now.
Let me know if you consider re-submitting updated backports of those two.
> > + }
> > assert(!flow_dv_check_valid_spec(matcher.mask.buf,
> > dev_flow->dv.value.buf));
> > /*
> > --
> > 2.34.0
> >
>
>
> --
> Christian Ehrhardt
> Staff Engineer, Ubuntu Server
> Canonical Ltd
--
Christian Ehrhardt
Staff Engineer, Ubuntu Server
Canonical Ltd
^ permalink raw reply [flat|nested] 6+ messages in thread* RE: [PATCH 19.11 4/6] net/mlx5: fix GRE protocol type translation
2021-12-06 9:57 ` Christian Ehrhardt
@ 2021-12-06 10:10 ` Gregory Etelson
0 siblings, 0 replies; 6+ messages in thread
From: Gregory Etelson @ 2021-12-06 10:10 UTC (permalink / raw)
To: Christian Ehrhardt
Cc: stable, Slava Ovsiienko, Matan Azrad, Shahaf Shuler, Ori Kam,
Yongseok Koh
Hello Christian,
.. snip ..
> > Hi,
> > build tests have shown that this breaks on
> 19.11 like:
> >
> > [1479/2115] Compiling C object
> >
> drivers/libtmp_rte_pmd_mlx5.a.p/net_mlx5_ml
> x5_flow_dv.c.o
> > ../drivers/net/mlx5/mlx5_flow_dv.c: In
> function '__flow_dv_translate':
> > ../drivers/net/mlx5/mlx5_flow_dv.c:7912:4:
> warning: implicit
> > declaration of function 'MLX5_ASSERT'; did
> you mean 'MLX5_SET'?
> > [-Wimplicit-function-declaration]
> > 7912 | MLX5_ASSERT(false);
> > | ^~~~~~~~~~~
> > | MLX5_SET
> > ../drivers/net/mlx5/mlx5_flow_dv.c:7912:4:
> warning: nested extern
> > declaration of 'MLX5_ASSERT' [-Wnested-
> externs]
> > [1538/2115] Linking target
> drivers/librte_pmd_mlx5.so.20.0
> > FAILED: drivers/librte_pmd_mlx5.so.20.0
> >
> >
> > For now - I'll remove the patch from the WIP
> 19.11 branch, please
> > consider submitting a backport adapted to
> work on
> > https://github.com/cpaelzer/dpdk-stable-
> queue/tree/19.11
>
>
> And dropping that makes the subsequent
> "net/mlx5: fix GRE flow item
> matching" fail to apply well, so I had to drop that
> as well for now.
> Let me know if you consider re-submitting
> updated backports of those two.
>
>
I reproduced the fault and prepare a fix.
I'll post v2 later today.
Regards,
Gregory
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH 19.11 v2 1/2] net/mlx5: fix GRE protocol type translation
2021-12-02 16:03 [PATCH 19.11 4/6] net/mlx5: fix GRE protocol type translation Gregory Etelson
2021-12-06 9:54 ` Christian Ehrhardt
@ 2021-12-06 10:38 ` Gregory Etelson
2021-12-06 10:39 ` [PATCH 19.11 v2 2/2] net/mlx5: fix GRE flow item matching Gregory Etelson
2 siblings, 0 replies; 6+ messages in thread
From: Gregory Etelson @ 2021-12-06 10:38 UTC (permalink / raw)
To: getelson; +Cc: matan, orika, shahafs, stable, viacheslavo, Christian Ehrhardt
[ upstream commit be26e81bfc1cb681d3ee0764aa6733000869984a ]
When application creates several flows to match on GRE tunnel
without explicitly specifying GRE protocol type value in
flow rules, PMD will translate that to zero mask.
RDMA-CORE cannot distinguish between different inner flow types and
produces identical matchers for each zero mask.
The patch extracts inner header type from flow rule and forces it
in GRE protocol type, if application did not specify
any without explicitly specifying GRE protocol type value in
flow rules, protocol type value.
Fixes: fc2c498ccb94 ("net/mlx5: add Direct Verbs translate items")
Cc: stable@dpdk.org
Signed-off-by: Gregory Etelson <getelson@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
---
drivers/net/mlx5/mlx5_flow_dv.c | 70 +++++++++++++++++++--------------
1 file changed, 40 insertions(+), 30 deletions(-)
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 8dec8d9ff5..a65a670045 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -5875,18 +5875,19 @@ flow_dv_translate_item_gre_key(void *matcher, void *key,
* Flow matcher value.
* @param[in] item
* Flow pattern to translate.
- * @param[in] inner
- * Item is inner pattern.
+ * @param[in] pattern_flags
+ * Accumulated pattern flags.
*/
static void
flow_dv_translate_item_gre(void *matcher, void *key,
const struct rte_flow_item *item,
- int inner)
+ uint64_t pattern_flags)
{
+ static const struct rte_flow_item_gre empty_gre = {0,};
const struct rte_flow_item_gre *gre_m = item->mask;
const struct rte_flow_item_gre *gre_v = item->spec;
- void *headers_m;
- void *headers_v;
+ void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
+ void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
struct {
@@ -5903,26 +5904,17 @@ flow_dv_translate_item_gre(void *matcher, void *key,
uint16_t value;
};
} gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
+ uint16_t protocol_m, protocol_v;
- if (inner) {
- headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
- inner_headers);
- headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
- } else {
- headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
- outer_headers);
- headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
- }
MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
- if (!gre_v)
- return;
- if (!gre_m)
- gre_m = &rte_flow_item_gre_mask;
- MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
- rte_be_to_cpu_16(gre_m->protocol));
- MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
- rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
+ if (!gre_v) {
+ gre_v = &empty_gre;
+ gre_m = &empty_gre;
+ } else {
+ if (!gre_m)
+ gre_m = &rte_flow_item_gre_mask;
+ }
gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
@@ -5940,6 +5932,16 @@ flow_dv_translate_item_gre(void *matcher, void *key,
MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
gre_crks_rsvd0_ver_v.s_present &
gre_crks_rsvd0_ver_m.s_present);
+ protocol_m = rte_be_to_cpu_16(gre_m->protocol);
+ protocol_v = rte_be_to_cpu_16(gre_v->protocol);
+ if (!protocol_m) {
+ /* Force next protocol to prevent matchers duplication */
+ protocol_m = 0xFFFF;
+ protocol_v = mlx5_translate_tunnel_etypes(pattern_flags);
+ }
+ MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, protocol_m);
+ MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
+ protocol_m & protocol_v);
}
/**
@@ -5951,13 +5953,13 @@ flow_dv_translate_item_gre(void *matcher, void *key,
* Flow matcher value.
* @param[in] item
* Flow pattern to translate.
- * @param[in] inner
- * Item is inner pattern.
+ * @param[in] pattern_flags
+ * Accumulated pattern flags.
*/
static void
flow_dv_translate_item_nvgre(void *matcher, void *key,
const struct rte_flow_item *item,
- int inner)
+ unsigned long pattern_flags)
{
const struct rte_flow_item_nvgre *nvgre_m = item->mask;
const struct rte_flow_item_nvgre *nvgre_v = item->spec;
@@ -5984,7 +5986,7 @@ flow_dv_translate_item_nvgre(void *matcher, void *key,
.mask = &gre_mask,
.last = NULL,
};
- flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
+ flow_dv_translate_item_gre(matcher, key, &gre_item, pattern_flags);
if (!nvgre_v)
return;
if (!nvgre_m)
@@ -7788,11 +7790,10 @@ __flow_dv_translate(struct rte_eth_dev *dev,
MLX5_FLOW_LAYER_OUTER_L4_UDP;
break;
case RTE_FLOW_ITEM_TYPE_GRE:
- flow_dv_translate_item_gre(match_mask, match_value,
- items, tunnel);
matcher.priority = flow->rss.level >= 2 ?
MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
last_item = MLX5_FLOW_LAYER_GRE;
+ tunnel_item = items;
break;
case RTE_FLOW_ITEM_TYPE_GRE_KEY:
flow_dv_translate_item_gre_key(match_mask,
@@ -7800,11 +7801,10 @@ __flow_dv_translate(struct rte_eth_dev *dev,
last_item = MLX5_FLOW_LAYER_GRE_KEY;
break;
case RTE_FLOW_ITEM_TYPE_NVGRE:
- flow_dv_translate_item_nvgre(match_mask, match_value,
- items, tunnel);
matcher.priority = flow->rss.level >= 2 ?
MLX5_PRIORITY_MAP_L2 : MLX5_PRIORITY_MAP_L4;
last_item = MLX5_FLOW_LAYER_GRE;
+ tunnel_item = items;
break;
case RTE_FLOW_ITEM_TYPE_VXLAN:
flow_dv_translate_item_vxlan(match_mask, match_value,
@@ -7892,6 +7892,16 @@ __flow_dv_translate(struct rte_eth_dev *dev,
else if (item_flags & MLX5_FLOW_LAYER_GENEVE)
flow_dv_translate_item_geneve(match_mask, match_value,
tunnel_item, item_flags);
+ else if (item_flags & MLX5_FLOW_LAYER_GRE) {
+ if (tunnel_item->type == RTE_FLOW_ITEM_TYPE_GRE)
+ flow_dv_translate_item_gre(match_mask, match_value,
+ tunnel_item, item_flags);
+ else if (tunnel_item->type == RTE_FLOW_ITEM_TYPE_NVGRE)
+ flow_dv_translate_item_nvgre(match_mask, match_value,
+ tunnel_item, item_flags);
+ else
+ RTE_VERIFY(false);
+ }
assert(!flow_dv_check_valid_spec(matcher.mask.buf,
dev_flow->dv.value.buf));
/*
--
2.34.0
^ permalink raw reply [flat|nested] 6+ messages in thread* [PATCH 19.11 v2 2/2] net/mlx5: fix GRE flow item matching
2021-12-02 16:03 [PATCH 19.11 4/6] net/mlx5: fix GRE protocol type translation Gregory Etelson
2021-12-06 9:54 ` Christian Ehrhardt
2021-12-06 10:38 ` [PATCH 19.11 v2 1/2] " Gregory Etelson
@ 2021-12-06 10:39 ` Gregory Etelson
2 siblings, 0 replies; 6+ messages in thread
From: Gregory Etelson @ 2021-12-06 10:39 UTC (permalink / raw)
To: getelson
Cc: matan, orika, shahafs, stable, viacheslavo, Christian Ehrhardt,
Alexander Kozyrev
From: Alexander Kozyrev <akozyrev@nvidia.com>
[ upstream commit 9e61533df27fc4d76080da20a7e8f9c9e7e1b90f ]
GRE protocol type is implicitly set in the matching translation in case
an application doesn't specify any type explicitly in a flow rule.
It is extracted from the inner header type, but this type may be absent.
In this case, GRE item matching is broken. Check if we have the inner
header type before setting it to allow matching on all GRE packets.
Fixes: be26e81bfc1c ("net/mlx5: fix GRE protocol type translation")
Cc: stable@dpdk.org
Signed-off-by: Alexander Kozyrev <akozyrev@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
---
drivers/net/mlx5/mlx5_flow_dv.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index a65a670045..2c7d5c7c02 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -5936,8 +5936,9 @@ flow_dv_translate_item_gre(void *matcher, void *key,
protocol_v = rte_be_to_cpu_16(gre_v->protocol);
if (!protocol_m) {
/* Force next protocol to prevent matchers duplication */
- protocol_m = 0xFFFF;
protocol_v = mlx5_translate_tunnel_etypes(pattern_flags);
+ if (protocol_v)
+ protocol_m = 0xFFFF;
}
MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, protocol_m);
MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
--
2.34.0
^ permalink raw reply [flat|nested] 6+ messages in thread