diff mbox series

[v3] net/i40e: fix inputset field mask

Message ID 20210301070607.24480-1-alvinx.zhang@intel.com (mailing list archive)
State Accepted, archived
Delegated to: Qi Zhang
Headers show
Series [v3] net/i40e: fix inputset field mask | expand

Checks

Context Check Description
ci/intel-Testing success Testing PASS
ci/iol-testing success Testing PASS
ci/iol-intel-Performance success Performance Testing PASS
ci/iol-mellanox-Functional success Functional Testing PASS
ci/iol-mellanox-Performance success Performance Testing PASS
ci/github-robot success github build: passed
ci/travis-robot fail travis build: failed
ci/Intel-compilation success Compilation OK
ci/checkpatch success coding style OK

Commit Message

Alvin Zhang March 1, 2021, 7:06 a.m. UTC
The absolute field offsets of IPv4 or IPv6 header are related to
hardware configuration. The X710 and X722 have different hardware
configurations, and users can even modify the hardware configuration.
Therefore, The default values cannot be used when calculating mask
offset.

The following flows can be created on X722 NIC, but the packet will
not enter the queue 3:
  flow create 0 ingress pattern eth / ipv4 proto is 255  / end
  actions queue index 3 / end
  pkt = Ether()/IP(ttl=63, proto=255)/Raw('X'*40)

  flow create 0 ingress pattern eth / ipv4 tos is 50 / udp / end
  actions queue index 3 / end
  pkt = Ether()/IP(tos=50)/UDP()/Raw('X'*40)

  flow create 0 ingress pattern eth / ipv6 tc is 12 / udp / end
  actions queue index 3 / end
  pkt = Ether()/IPv6(tc=12,hlim=34,fl=0x98765)/UDP()/Raw('X'*40)

  flow create 0 ingress pattern eth / ipv6 hop is 34 / end actions
  queue index 3 / end
  pkt = Ether()/IPv6(tc=12,hlim=34,fl=0x98765)/Raw('X'*40)

This patch read the field offsets from the NIC and return the mask
register value.

Fixes: 98f055707685 ("i40e: configure input fields for RSS or flow director")
Fixes: 92cf7f8ec082 ("i40e: allow filtering on more IP header fields")
Cc: stable@dpdk.org

Signed-off-by: Alvin Zhang <alvinx.zhang@intel.com>
---

v2: Update commit log
v3: Rename 'i40e_fd_get_field_offset' to
    'i40e_get_inset_field_offset'
---
 drivers/net/i40e/i40e_ethdev.c | 158 +++++++++++++++++++++++++++++++----------
 drivers/net/i40e/i40e_ethdev.h |   4 +-
 drivers/net/i40e/i40e_flow.c   |   2 +-
 3 files changed, 125 insertions(+), 39 deletions(-)

Comments

Chen, LingliX March 9, 2021, 2:56 a.m. UTC | #1
> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Alvin Zhang
> Sent: Monday, March 1, 2021 3:06 PM
> To: Guo, Jia <jia.guo@intel.com>; Xing, Beilei <beilei.xing@intel.com>; Zhang,
> Qi Z <qi.z.zhang@intel.com>; Zhou, JunX W <junx.w.zhou@intel.com>
> Cc: dev@dpdk.org; Zhang, AlvinX <alvinx.zhang@intel.com>;
> stable@dpdk.org
> Subject: [dpdk-dev] [PATCH v3] net/i40e: fix inputset field mask
> 
> The absolute field offsets of IPv4 or IPv6 header are related to hardware
> configuration. The X710 and X722 have different hardware configurations,
> and users can even modify the hardware configuration.
> Therefore, The default values cannot be used when calculating mask offset.
> 
> The following flows can be created on X722 NIC, but the packet will not enter
> the queue 3:
>   flow create 0 ingress pattern eth / ipv4 proto is 255  / end
>   actions queue index 3 / end
>   pkt = Ether()/IP(ttl=63, proto=255)/Raw('X'*40)
> 
>   flow create 0 ingress pattern eth / ipv4 tos is 50 / udp / end
>   actions queue index 3 / end
>   pkt = Ether()/IP(tos=50)/UDP()/Raw('X'*40)
> 
>   flow create 0 ingress pattern eth / ipv6 tc is 12 / udp / end
>   actions queue index 3 / end
>   pkt = Ether()/IPv6(tc=12,hlim=34,fl=0x98765)/UDP()/Raw('X'*40)
> 
>   flow create 0 ingress pattern eth / ipv6 hop is 34 / end actions
>   queue index 3 / end
>   pkt = Ether()/IPv6(tc=12,hlim=34,fl=0x98765)/Raw('X'*40)
> 
> This patch read the field offsets from the NIC and return the mask register
> value.
> 
> Fixes: 98f055707685 ("i40e: configure input fields for RSS or flow director")
> Fixes: 92cf7f8ec082 ("i40e: allow filtering on more IP header fields")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Alvin Zhang <alvinx.zhang@intel.com>

all regression cases passed.
Tested-by: Chen Lingli <linglix.chen@intel.com>
Qi Zhang March 24, 2021, 11:38 a.m. UTC | #2
> -----Original Message-----
> From: Chen, LingliX <linglix.chen@intel.com>
> Sent: Tuesday, March 9, 2021 10:57 AM
> To: Zhang, AlvinX <alvinx.zhang@intel.com>; Guo, Jia <jia.guo@intel.com>;
> Xing, Beilei <beilei.xing@intel.com>; Zhang, Qi Z <qi.z.zhang@intel.com>; Zhou,
> JunX W <junx.w.zhou@intel.com>
> Cc: dev@dpdk.org; Zhang, AlvinX <alvinx.zhang@intel.com>; stable@dpdk.org
> Subject: RE: [dpdk-dev] [PATCH v3] net/i40e: fix inputset field mask
> 
> > -----Original Message-----
> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Alvin Zhang
> > Sent: Monday, March 1, 2021 3:06 PM
> > To: Guo, Jia <jia.guo@intel.com>; Xing, Beilei
> > <beilei.xing@intel.com>; Zhang, Qi Z <qi.z.zhang@intel.com>; Zhou,
> > JunX W <junx.w.zhou@intel.com>
> > Cc: dev@dpdk.org; Zhang, AlvinX <alvinx.zhang@intel.com>;
> > stable@dpdk.org
> > Subject: [dpdk-dev] [PATCH v3] net/i40e: fix inputset field mask
> >
> > The absolute field offsets of IPv4 or IPv6 header are related to
> > hardware configuration. The X710 and X722 have different hardware
> > configurations, and users can even modify the hardware configuration.
> > Therefore, The default values cannot be used when calculating mask offset.
> >
> > The following flows can be created on X722 NIC, but the packet will
> > not enter the queue 3:
> >   flow create 0 ingress pattern eth / ipv4 proto is 255  / end
> >   actions queue index 3 / end
> >   pkt = Ether()/IP(ttl=63, proto=255)/Raw('X'*40)
> >
> >   flow create 0 ingress pattern eth / ipv4 tos is 50 / udp / end
> >   actions queue index 3 / end
> >   pkt = Ether()/IP(tos=50)/UDP()/Raw('X'*40)
> >
> >   flow create 0 ingress pattern eth / ipv6 tc is 12 / udp / end
> >   actions queue index 3 / end
> >   pkt = Ether()/IPv6(tc=12,hlim=34,fl=0x98765)/UDP()/Raw('X'*40)
> >
> >   flow create 0 ingress pattern eth / ipv6 hop is 34 / end actions
> >   queue index 3 / end
> >   pkt = Ether()/IPv6(tc=12,hlim=34,fl=0x98765)/Raw('X'*40)
> >
> > This patch read the field offsets from the NIC and return the mask
> > register value.
> >
> > Fixes: 98f055707685 ("i40e: configure input fields for RSS or flow
> > director")
> > Fixes: 92cf7f8ec082 ("i40e: allow filtering on more IP header fields")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Alvin Zhang <alvinx.zhang@intel.com>
> 
> all regression cases passed.
> Tested-by: Chen Lingli <linglix.chen@intel.com>

Should be:
 
Lingli Chen <lingli.chen@intel.com>

Applied to dpdk-next-net-intel.

Thanks
Qi
diff mbox series

Patch

diff --git a/drivers/net/i40e/i40e_ethdev.c b/drivers/net/i40e/i40e_ethdev.c
index d7cd049..18217b8 100644
--- a/drivers/net/i40e/i40e_ethdev.c
+++ b/drivers/net/i40e/i40e_ethdev.c
@@ -202,12 +202,12 @@ 
 #define I40E_TRANSLATE_INSET 0
 #define I40E_TRANSLATE_REG   1
 
-#define I40E_INSET_IPV4_TOS_MASK        0x0009FF00UL
-#define I40E_INSET_IPv4_TTL_MASK        0x000D00FFUL
-#define I40E_INSET_IPV4_PROTO_MASK      0x000DFF00UL
-#define I40E_INSET_IPV6_TC_MASK         0x0009F00FUL
-#define I40E_INSET_IPV6_HOP_LIMIT_MASK  0x000CFF00UL
-#define I40E_INSET_IPV6_NEXT_HDR_MASK   0x000C00FFUL
+#define I40E_INSET_IPV4_TOS_MASK        0x0000FF00UL
+#define I40E_INSET_IPV4_TTL_MASK        0x000000FFUL
+#define I40E_INSET_IPV4_PROTO_MASK      0x0000FF00UL
+#define I40E_INSET_IPV6_TC_MASK         0x0000F00FUL
+#define I40E_INSET_IPV6_HOP_LIMIT_MASK  0x0000FF00UL
+#define I40E_INSET_IPV6_NEXT_HDR_MASK   0x000000FFUL
 
 /* PCI offset for querying capability */
 #define PCI_DEV_CAP_REG            0xA4
@@ -220,6 +220,25 @@ 
 /* Bit mask of Extended Tag enable/disable */
 #define PCI_DEV_CTRL_EXT_TAG_MASK  (1 << PCI_DEV_CTRL_EXT_TAG_SHIFT)
 
+#define I40E_GLQF_PIT_IPV4_START	2
+#define I40E_GLQF_PIT_IPV4_COUNT	2
+#define I40E_GLQF_PIT_IPV6_START	4
+#define I40E_GLQF_PIT_IPV6_COUNT	2
+
+#define I40E_GLQF_PIT_SOURCE_OFF_GET(a)	\
+				(((a) & I40E_GLQF_PIT_SOURCE_OFF_MASK) >> \
+				 I40E_GLQF_PIT_SOURCE_OFF_SHIFT)
+
+#define I40E_GLQF_PIT_DEST_OFF_GET(a) \
+				(((a) & I40E_GLQF_PIT_DEST_OFF_MASK) >> \
+				 I40E_GLQF_PIT_DEST_OFF_SHIFT)
+
+#define I40E_GLQF_PIT_FSIZE_GET(a)	(((a) & I40E_GLQF_PIT_FSIZE_MASK) >> \
+					 I40E_GLQF_PIT_FSIZE_SHIFT)
+
+#define I40E_GLQF_PIT_BUILD(off, mask)	(((off) << 16) | (mask))
+#define I40E_FDIR_FIELD_OFFSET(a)	((a) >> 1)
+
 static int eth_i40e_dev_init(struct rte_eth_dev *eth_dev, void *init_params);
 static int eth_i40e_dev_uninit(struct rte_eth_dev *eth_dev);
 static int i40e_dev_configure(struct rte_eth_dev *dev);
@@ -9417,49 +9436,116 @@  i40e_status_code i40e_replace_gtp_cloud_filter(struct i40e_pf *pf)
 	return val;
 }
 
+static int
+i40e_get_inset_field_offset(struct i40e_hw *hw, uint32_t pit_reg_start,
+			    uint32_t pit_reg_count, uint32_t hdr_off)
+{
+	const uint32_t pit_reg_end = pit_reg_start + pit_reg_count;
+	uint32_t field_off = I40E_FDIR_FIELD_OFFSET(hdr_off);
+	uint32_t i, reg_val, src_off, count;
+
+	for (i = pit_reg_start; i < pit_reg_end; i++) {
+		reg_val = i40e_read_rx_ctl(hw, I40E_GLQF_PIT(i));
+
+		src_off = I40E_GLQF_PIT_SOURCE_OFF_GET(reg_val);
+		count = I40E_GLQF_PIT_FSIZE_GET(reg_val);
+
+		if (src_off <= field_off && (src_off + count) > field_off)
+			break;
+	}
+
+	if (i >= pit_reg_end) {
+		PMD_DRV_LOG(ERR,
+			    "Hardware GLQF_PIT configuration does not support this field mask");
+		return -1;
+	}
+
+	return I40E_GLQF_PIT_DEST_OFF_GET(reg_val) + field_off - src_off;
+}
+
 int
-i40e_generate_inset_mask_reg(uint64_t inset, uint32_t *mask, uint8_t nb_elem)
+i40e_generate_inset_mask_reg(struct i40e_hw *hw, uint64_t inset,
+			     uint32_t *mask, uint8_t nb_elem)
 {
-	uint8_t i, idx = 0;
-	uint64_t inset_need_mask = inset;
+	static const uint64_t mask_inset[] = {
+		I40E_INSET_IPV4_PROTO | I40E_INSET_IPV4_TTL,
+		I40E_INSET_IPV6_NEXT_HDR | I40E_INSET_IPV6_HOP_LIMIT };
 
 	static const struct {
 		uint64_t inset;
 		uint32_t mask;
-	} inset_mask_map[] = {
-		{I40E_INSET_IPV4_TOS, I40E_INSET_IPV4_TOS_MASK},
-		{I40E_INSET_IPV4_PROTO | I40E_INSET_IPV4_TTL, 0},
-		{I40E_INSET_IPV4_PROTO, I40E_INSET_IPV4_PROTO_MASK},
-		{I40E_INSET_IPV4_TTL, I40E_INSET_IPv4_TTL_MASK},
-		{I40E_INSET_IPV6_TC, I40E_INSET_IPV6_TC_MASK},
-		{I40E_INSET_IPV6_NEXT_HDR | I40E_INSET_IPV6_HOP_LIMIT, 0},
-		{I40E_INSET_IPV6_NEXT_HDR, I40E_INSET_IPV6_NEXT_HDR_MASK},
-		{I40E_INSET_IPV6_HOP_LIMIT, I40E_INSET_IPV6_HOP_LIMIT_MASK},
+		uint32_t offset;
+	} inset_mask_offset_map[] = {
+		{ I40E_INSET_IPV4_TOS, I40E_INSET_IPV4_TOS_MASK,
+		  offsetof(struct rte_ipv4_hdr, type_of_service) },
+
+		{ I40E_INSET_IPV4_PROTO, I40E_INSET_IPV4_PROTO_MASK,
+		  offsetof(struct rte_ipv4_hdr, next_proto_id) },
+
+		{ I40E_INSET_IPV4_TTL, I40E_INSET_IPV4_TTL_MASK,
+		  offsetof(struct rte_ipv4_hdr, time_to_live) },
+
+		{ I40E_INSET_IPV6_TC, I40E_INSET_IPV6_TC_MASK,
+		  offsetof(struct rte_ipv6_hdr, vtc_flow) },
+
+		{ I40E_INSET_IPV6_NEXT_HDR, I40E_INSET_IPV6_NEXT_HDR_MASK,
+		  offsetof(struct rte_ipv6_hdr, proto) },
+
+		{ I40E_INSET_IPV6_HOP_LIMIT, I40E_INSET_IPV6_HOP_LIMIT_MASK,
+		  offsetof(struct rte_ipv6_hdr, hop_limits) },
 	};
 
-	if (!inset || !mask || !nb_elem)
+	uint32_t i;
+	int idx = 0;
+
+	assert(mask);
+	if (!inset)
 		return 0;
 
-	for (i = 0, idx = 0; i < RTE_DIM(inset_mask_map); i++) {
+	for (i = 0; i < RTE_DIM(mask_inset); i++) {
 		/* Clear the inset bit, if no MASK is required,
 		 * for example proto + ttl
 		 */
-		if ((inset & inset_mask_map[i].inset) ==
-		     inset_mask_map[i].inset && inset_mask_map[i].mask == 0)
-			inset_need_mask &= ~inset_mask_map[i].inset;
-		if (!inset_need_mask)
-			return 0;
+		if ((mask_inset[i] & inset) == mask_inset[i]) {
+			inset &= ~mask_inset[i];
+			if (!inset)
+				return 0;
+		}
 	}
-	for (i = 0, idx = 0; i < RTE_DIM(inset_mask_map); i++) {
-		if ((inset_need_mask & inset_mask_map[i].inset) ==
-		    inset_mask_map[i].inset) {
-			if (idx >= nb_elem) {
-				PMD_DRV_LOG(ERR, "exceed maximal number of bitmasks");
-				return -EINVAL;
-			}
-			mask[idx] = inset_mask_map[i].mask;
-			idx++;
+
+	for (i = 0; i < RTE_DIM(inset_mask_offset_map); i++) {
+		uint32_t pit_start, pit_count;
+		int offset;
+
+		if (!(inset_mask_offset_map[i].inset & inset))
+			continue;
+
+		if (inset_mask_offset_map[i].inset &
+		    (I40E_INSET_IPV4_TOS | I40E_INSET_IPV4_PROTO |
+		     I40E_INSET_IPV4_TTL)) {
+			pit_start = I40E_GLQF_PIT_IPV4_START;
+			pit_count = I40E_GLQF_PIT_IPV4_COUNT;
+		} else {
+			pit_start = I40E_GLQF_PIT_IPV6_START;
+			pit_count = I40E_GLQF_PIT_IPV6_COUNT;
+		}
+
+		offset = i40e_get_inset_field_offset(hw, pit_start, pit_count,
+				inset_mask_offset_map[i].offset);
+
+		if (offset < 0)
+			return -EINVAL;
+
+		if (idx >= nb_elem) {
+			PMD_DRV_LOG(ERR,
+				    "Configuration of inset mask out of range %u",
+				    nb_elem);
+			return -ERANGE;
 		}
+
+		mask[idx] = I40E_GLQF_PIT_BUILD((uint32_t)offset,
+						inset_mask_offset_map[i].mask);
+		idx++;
 	}
 
 	return idx;
@@ -9513,7 +9599,7 @@  i40e_status_code i40e_replace_gtp_cloud_filter(struct i40e_pf *pf)
 
 		input_set = i40e_get_default_input_set(pctype);
 
-		num = i40e_generate_inset_mask_reg(input_set, mask_reg,
+		num = i40e_generate_inset_mask_reg(hw, input_set, mask_reg,
 						   I40E_INSET_MASK_NUM_REG);
 		if (num < 0)
 			return;
@@ -9593,7 +9679,7 @@  i40e_status_code i40e_replace_gtp_cloud_filter(struct i40e_pf *pf)
 		inset_reg |= i40e_read_rx_ctl(hw, I40E_GLQF_HASH_INSET(0, pctype));
 		input_set |= pf->hash_input_set[pctype];
 	}
-	num = i40e_generate_inset_mask_reg(input_set, mask_reg,
+	num = i40e_generate_inset_mask_reg(hw, input_set, mask_reg,
 					   I40E_INSET_MASK_NUM_REG);
 	if (num < 0)
 		return -EINVAL;
diff --git a/drivers/net/i40e/i40e_ethdev.h b/drivers/net/i40e/i40e_ethdev.h
index 1e8f5d3..faf6896 100644
--- a/drivers/net/i40e/i40e_ethdev.h
+++ b/drivers/net/i40e/i40e_ethdev.h
@@ -1458,8 +1458,8 @@  void i40e_set_symmetric_hash_enable_per_port(struct i40e_hw *hw,
 					     uint8_t enable);
 int i40e_validate_input_set(enum i40e_filter_pctype pctype,
 			    enum rte_filter_type filter, uint64_t inset);
-int i40e_generate_inset_mask_reg(uint64_t inset, uint32_t *mask,
-				 uint8_t nb_elem);
+int i40e_generate_inset_mask_reg(struct i40e_hw *hw, uint64_t inset,
+				 uint32_t *mask, uint8_t nb_elem);
 uint64_t i40e_translate_input_set_reg(enum i40e_mac_type type, uint64_t input);
 void i40e_check_write_reg(struct i40e_hw *hw, uint32_t addr, uint32_t val);
 void i40e_check_write_global_reg(struct i40e_hw *hw,
diff --git a/drivers/net/i40e/i40e_flow.c b/drivers/net/i40e/i40e_flow.c
index 4d44282..05842bb 100644
--- a/drivers/net/i40e/i40e_flow.c
+++ b/drivers/net/i40e/i40e_flow.c
@@ -2269,7 +2269,7 @@  static int i40e_flow_parse_l4_cloud_filter(struct rte_eth_dev *dev,
 	    !memcmp(&pf->fdir.input_set[pctype], &input_set, sizeof(uint64_t)))
 		return 0;
 
-	num = i40e_generate_inset_mask_reg(input_set, mask_reg,
+	num = i40e_generate_inset_mask_reg(hw, input_set, mask_reg,
 					   I40E_INSET_MASK_NUM_REG);
 	if (num < 0)
 		return -EINVAL;