[v2,2/2] net/hns3: add FDIR VLAN match mode runtime config
Checks
Commit Message
From: Huisong Li <lihuisong@huawei.com>
The VLAN number in FDIR meta data is used to enable that hardware
bases on VLAN number to strictly match the input flow. And it is
enabled by default.
For the following two rules:
rule0:
pattern: eth type is 0x0806
actions: queue index 3
rule1:
pattern: eth type is 0x0806 / vlan vid is 20
actions: queue index 4
If enable VLAN number, only the ARP packets with VLAN 20 are directed
to queue 4, and the ARP packets with other VLAN ID cannot be directed
to the specified queue. If app want to all ARP (VLAN or no VLAN)
packets to be directed to the specified queue, app has to set many
rules for VLAN packet. In this case, if driver doesn't enable VLAN
number, app just need to set one rule (rule0).
So this patch adds a "fdir_vlan_match_mode" runtime config which only
can be 'strict' or 'nostrict'. And driver still uses 'strict' mode as
the default mode. Please select 'nostrict' mode if you request all same
ethertype packets with and without VLAN to a specified queue.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
---
doc/guides/nics/hns3.rst | 26 +++++++++++++++++++++++++
drivers/net/hns3/hns3_common.c | 35 ++++++++++++++++++++++++++++++++++
drivers/net/hns3/hns3_common.h | 2 ++
drivers/net/hns3/hns3_fdir.c | 10 +++++++---
drivers/net/hns3/hns3_fdir.h | 8 ++++++++
5 files changed, 78 insertions(+), 3 deletions(-)
Comments
27/06/2023 09:11, Dongdong Liu:
> +- ``fdir_vlan_match_mode`` (default `strict`)
> +
> + Used to select VLAN match mode. This runtime config can be `strict`
> + or `nostrict` and is only valid for PF drives.
drives? Do you mean devices?
> + If driver works on `strict` mode (default mode), hardware does strictly
> + match the input flow base on VLAN number.
> +
> + For the following scenarios with two rules:
> +
> + .. code-block:: console
> +
> + rule0:
> + pattern: eth type is 0x0806
> + actions: queue index 3
> + rule1:
> + pattern: eth type is 0x0806 / vlan vid is 20
> + actions: queue index 4
> +
> + If application select `strict` mode, only the ARP packets with VLAN
> + 20 are directed to queue 4, and the ARP packets with other VLAN ID
> + cannot be directed to the specified queue. If application want to all
> + ARP packets with or without VLAN to be directed to the specified queue,
> + application can select `nostrict` mode and just need to set rule0.
As it is done in the rest of the document, you should use double backquotes
for values of the devargs.
Hi Thomas
Many thanks for your review.
On 2023/6/28 7:12, Thomas Monjalon wrote:
> 27/06/2023 09:11, Dongdong Liu:
>> +- ``fdir_vlan_match_mode`` (default `strict`)
>> +
>> + Used to select VLAN match mode. This runtime config can be `strict`
>> + or `nostrict` and is only valid for PF drives.
>
> drives? Do you mean devices?
for PF devices, will fix.
>
>> + If driver works on `strict` mode (default mode), hardware does strictly
>> + match the input flow base on VLAN number.
>> +
>> + For the following scenarios with two rules:
>> +
>> + .. code-block:: console
>> +
>> + rule0:
>> + pattern: eth type is 0x0806
>> + actions: queue index 3
>> + rule1:
>> + pattern: eth type is 0x0806 / vlan vid is 20
>> + actions: queue index 4
>> +
>> + If application select `strict` mode, only the ARP packets with VLAN
>> + 20 are directed to queue 4, and the ARP packets with other VLAN ID
>> + cannot be directed to the specified queue. If application want to all
>> + ARP packets with or without VLAN to be directed to the specified queue,
>> + application can select `nostrict` mode and just need to set rule0.
>
> As it is done in the rest of the document, you should use double backquotes
> for values of the devargs.
Good point, will fix.
Thanks,
Dongdong.
>
>
>
> .
>
On 6/28/2023 3:59 AM, Dongdong Liu wrote:
> Hi Thomas
>
> Many thanks for your review.
>
> On 2023/6/28 7:12, Thomas Monjalon wrote:
>> 27/06/2023 09:11, Dongdong Liu:
>>> +- ``fdir_vlan_match_mode`` (default `strict`)
>>> +
>>> + Used to select VLAN match mode. This runtime config can be `strict`
>>> + or `nostrict` and is only valid for PF drives.
>>
>> drives? Do you mean devices?
> for PF devices, will fix.
>>
>>> + If driver works on `strict` mode (default mode), hardware does
>>> strictly
>>> + match the input flow base on VLAN number.
>>> +
>>> + For the following scenarios with two rules:
>>> +
>>> + .. code-block:: console
>>> +
>>> + rule0:
>>> + pattern: eth type is 0x0806
>>> + actions: queue index 3
>>> + rule1:
>>> + pattern: eth type is 0x0806 / vlan vid is 20
>>> + actions: queue index 4
>>> +
>>> + If application select `strict` mode, only the ARP packets with VLAN
>>> + 20 are directed to queue 4, and the ARP packets with other VLAN ID
>>> + cannot be directed to the specified queue. If application want to all
>>> + ARP packets with or without VLAN to be directed to the specified
>>> queue,
>>> + application can select `nostrict` mode and just need to set rule0.
>>
>> As it is done in the rest of the document, you should use double
>> backquotes
>> for values of the devargs.
> Good point, will fix.
>
I dropped set from next-net, please send a new version with fix.
Updating patchwork status as "Change Requested".
On 2023/6/29 19:08, Ferruh Yigit wrote:
> On 6/28/2023 3:59 AM, Dongdong Liu wrote:
>> Hi Thomas
>>
>> Many thanks for your review.
>>
>> On 2023/6/28 7:12, Thomas Monjalon wrote:
>>> 27/06/2023 09:11, Dongdong Liu:
>>>> +- ``fdir_vlan_match_mode`` (default `strict`)
>>>> +
>>>> + Used to select VLAN match mode. This runtime config can be `strict`
>>>> + or `nostrict` and is only valid for PF drives.
>>>
>>> drives? Do you mean devices?
>> for PF devices, will fix.
>>>
>>>> + If driver works on `strict` mode (default mode), hardware does
>>>> strictly
>>>> + match the input flow base on VLAN number.
>>>> +
>>>> + For the following scenarios with two rules:
>>>> +
>>>> + .. code-block:: console
>>>> +
>>>> + rule0:
>>>> + pattern: eth type is 0x0806
>>>> + actions: queue index 3
>>>> + rule1:
>>>> + pattern: eth type is 0x0806 / vlan vid is 20
>>>> + actions: queue index 4
>>>> +
>>>> + If application select `strict` mode, only the ARP packets with VLAN
>>>> + 20 are directed to queue 4, and the ARP packets with other VLAN ID
>>>> + cannot be directed to the specified queue. If application want to all
>>>> + ARP packets with or without VLAN to be directed to the specified
>>>> queue,
>>>> + application can select `nostrict` mode and just need to set rule0.
>>>
>>> As it is done in the rest of the document, you should use double
>>> backquotes
>>> for values of the devargs.
>> Good point, will fix.
>>
>
> I dropped set from next-net, please send a new version with fix.
> Updating patchwork status as "Change Requested".
PATCH V3 has been sent out, thanks for doing this.
Thanks,
Dongdong.
>
> .
>
@@ -140,6 +140,32 @@ Runtime Configuration
For example::
-a 0000:7d:00.0,mbx_time_limit_ms=600
+- ``fdir_vlan_match_mode`` (default `strict`)
+
+ Used to select VLAN match mode. This runtime config can be `strict`
+ or `nostrict` and is only valid for PF drives.
+ If driver works on `strict` mode (default mode), hardware does strictly
+ match the input flow base on VLAN number.
+
+ For the following scenarios with two rules:
+
+ .. code-block:: console
+
+ rule0:
+ pattern: eth type is 0x0806
+ actions: queue index 3
+ rule1:
+ pattern: eth type is 0x0806 / vlan vid is 20
+ actions: queue index 4
+
+ If application select `strict` mode, only the ARP packets with VLAN
+ 20 are directed to queue 4, and the ARP packets with other VLAN ID
+ cannot be directed to the specified queue. If application want to all
+ ARP packets with or without VLAN to be directed to the specified queue,
+ application can select `nostrict` mode and just need to set rule0.
+
+ For example::
+ -a 0000:7d:00.0,fdir_vlan_match_mode=nostrict
Driver compilation and testing
------------------------------
@@ -238,6 +238,34 @@ hns3_parse_mbx_time_limit(const char *key, const char *value, void *extra_args)
return 0;
}
+static int
+hns3_parse_vlan_match_mode(const char *key, const char *value, void *args)
+{
+ uint8_t mode;
+
+ RTE_SET_USED(key);
+
+ if (value == NULL) {
+ PMD_INIT_LOG(WARNING, "no value for key:\"%s\"", key);
+ return -1;
+ }
+
+ if (strcmp(value, "strict") == 0) {
+ mode = HNS3_FDIR_VLAN_STRICT_MATCH;
+ } else if (strcmp(value, "nostrict") == 0) {
+ mode = HNS3_FDIR_VLAN_NOSTRICT_MATCH;
+ } else {
+ PMD_INIT_LOG(WARNING, "invalid value:\"%s\" for key:\"%s\", "
+ "value must be 'strict' or 'nostrict'",
+ value, key);
+ return -1;
+ }
+
+ *(uint8_t *)args = mode;
+
+ return 0;
+}
+
void
hns3_parse_devargs(struct rte_eth_dev *dev)
{
@@ -254,6 +282,8 @@ hns3_parse_devargs(struct rte_eth_dev *dev)
hns->tx_func_hint = HNS3_IO_FUNC_HINT_NONE;
hns->dev_caps_mask = 0;
hns->mbx_time_limit_ms = HNS3_MBX_DEF_TIME_LIMIT_MS;
+ if (!hns->is_vf)
+ hns->pf.fdir.vlan_match_mode = HNS3_FDIR_VLAN_STRICT_MATCH;
if (dev->device->devargs == NULL)
return;
@@ -270,6 +300,11 @@ hns3_parse_devargs(struct rte_eth_dev *dev)
&hns3_parse_dev_caps_mask, &dev_caps_mask);
(void)rte_kvargs_process(kvlist, HNS3_DEVARG_MBX_TIME_LIMIT_MS,
&hns3_parse_mbx_time_limit, &mbx_time_limit_ms);
+ if (!hns->is_vf)
+ (void)rte_kvargs_process(kvlist,
+ HNS3_DEVARG_FDIR_VALN_MATCH_MODE,
+ &hns3_parse_vlan_match_mode,
+ &hns->pf.fdir.vlan_match_mode);
rte_kvargs_free(kvlist);
@@ -27,6 +27,8 @@ enum {
#define HNS3_DEVARG_MBX_TIME_LIMIT_MS "mbx_time_limit_ms"
+#define HNS3_DEVARG_FDIR_VALN_MATCH_MODE "fdir_vlan_match_mode"
+
#define MSEC_PER_SEC 1000L
#define USEC_PER_MSEC 1000L
@@ -355,9 +355,13 @@ int hns3_init_fd_config(struct hns3_adapter *hns)
/* roce_type is used to filter roce frames
* dst_vport is used to specify the rule
*/
- key_cfg->meta_data_active = BIT(DST_VPORT) | BIT(TUNNEL_PACKET) |
- BIT(VLAN_NUMBER);
- hns3_dbg(hw, "fdir meta data: dst_vport tunnel_packet vlan_number");
+ key_cfg->meta_data_active = BIT(DST_VPORT) | BIT(TUNNEL_PACKET);
+ if (pf->fdir.vlan_match_mode)
+ key_cfg->meta_data_active |= BIT(VLAN_NUMBER);
+
+ hns3_dbg(hw, "fdir meta data: dst_vport tunnel_packet %s",
+ (pf->fdir.vlan_match_mode == HNS3_FDIR_VLAN_STRICT_MATCH) ?
+ "vlan_number" : "");
ret = hns3_get_fd_allocation(hw,
&pf->fdir.fd_cfg.rule_num[HNS3_FD_STAGE_1],
@@ -170,6 +170,13 @@ struct hns3_fdir_rule_ele {
TAILQ_HEAD(hns3_fdir_rule_list, hns3_fdir_rule_ele);
+/*
+ * On 'strict' mode, hardware bases on VLAN number to exactly match the
+ * input flow.
+ */
+#define HNS3_FDIR_VLAN_STRICT_MATCH 1
+#define HNS3_FDIR_VLAN_NOSTRICT_MATCH 0
+
/*
* A structure used to define fields of a FDIR related info.
*/
@@ -178,6 +185,7 @@ struct hns3_fdir_info {
struct hns3_fdir_rule_ele **hash_map;
struct rte_hash *hash_handle;
struct hns3_fd_cfg fd_cfg;
+ uint8_t vlan_match_mode;
};
struct hns3_adapter;