Skip to contentSkip to conversation

RDMA/qedr: Explicitly cast pkt->tx_dest to qed_ll2_tx_dest

PATCH In review Nathan Chancellor 4 participants 8 messages Sep 27, 2018 – Sep 28, 2018

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 RDMA/qedr: Explicitly cast pkt->tx_dest to qed_ll2_tx_dest

Clang warns when one enumerated type is explicitly converted to another.

drivers/infiniband/hw/qedr/qedr_roce_cm.c:198:28: warning: implicit conversion from enumeration type 'enum qed_roce_ll2_tx_dest' to different enumeration type 'enum qed_ll2_tx_dest' [-Wenum-conversion]

        ll2_tx_pkt.tx_dest = pkt->tx_dest;
                           ~ ~~~~~^~~~~~~

1 warning generated.

Avoid this warning by explicitly casting pkt->tx_dest to qed_112_tx_dest, which has the expected values from the type qed_roce_ll2_tx_dest.

---
 drivers/infiniband/hw/qedr/qedr_roce_cm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/infiniband/hw/qedr/qedr_roce_cm.c +1 -1
index 85578887421b..147e0d69003d 100644
@@ -195,7 +195,7 @@ static int qedr_ll2_post_tx(struct qedr_dev *dev,
 
 	ll2_tx_pkt.num_of_bds = 1 /* hdr */  + pkt->n_seg;
 	ll2_tx_pkt.vlan = 0;
-	ll2_tx_pkt.tx_dest = pkt->tx_dest;
+	ll2_tx_pkt.tx_dest = (enum qed_ll2_tx_dest)pkt->tx_dest;
 	ll2_tx_pkt.qed_roce_flavor = roce_flavor;
 	ll2_tx_pkt.first_frag = pkt->header.baddr;
 	ll2_tx_pkt.first_frag_len = pkt->header.len;
Signature · 1 line
2.19.0
#1 lore raw
Quoted 6 lines · Nathan Chancellor

Clang warns when one enumerated type is explicitly converted to another.

drivers/infiniband/hw/qedr/qedr_roce_cm.c:198:28: warning: implicit conversion from enumeration type 'enum qed_roce_ll2_tx_dest' to different enumeration type 'enum qed_ll2_tx_dest' [-Wenum-conversion]

Quoted 7 lines
        ll2_tx_pkt.tx_dest = pkt->tx_dest;
                           ~ ~~~~~^~~~~~~

1 warning generated.

Avoid this warning by explicitly casting pkt->tx_dest to qed_112_tx_dest, which has the expected values from the type qed_roce_ll2_tx_dest.

But these enums are different lengths, which is problematic for this patch. Is this code broken, or that it's ok for ll2_tx_pkt.tx_dest to have a value that's not a valid enumeration value for enum qed_ll2_tx_dest? (QED_LL2_TX_DEST_MAX 's value (3) is outside the enumeration values of enum qed_roce_ll2_tx_dest).

include/linux/qed/qed_rdma_if.h:
42 enum qed_roce_ll2_tx_dest {
43 /* Light L2 TX Destination to the Network */
44 QED_ROCE_LL2_TX_DEST_NW,
45

 46         /* Light L2 TX Destination to the Loopback */
 47         QED_ROCE_LL2_TX_DEST_LB,
 48         QED_ROCE_LL2_TX_DEST_MAX

49 };

include/linux/qed/qed_ll2_if.h:
64 enum qed_ll2_tx_dest {
65 QED_LL2_TX_DEST_NW, /* Light L2 TX Destination to the
Network */
66 QED_LL2_TX_DEST_LB, /* Light L2 TX Destination to the
Loopback */
67 QED_LL2_TX_DEST_DROP, /* Light L2 Drop the TX packet */
68 QED_LL2_TX_DEST_MAX
69 };

Maybe the maintainers can clarify?

Quoted 23 lines
---
 drivers/infiniband/hw/qedr/qedr_roce_cm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/infiniband/hw/qedr/qedr_roce_cm.c +1 -2
index 85578887421b..147e0d69003d 100644
@@ -195,7 +195,7 @@ static int qedr_ll2_post_tx(struct qedr_dev *dev,
 
        ll2_tx_pkt.num_of_bds = 1 /* hdr */  + pkt->n_seg;
        ll2_tx_pkt.vlan = 0;
-       ll2_tx_pkt.tx_dest = pkt->tx_dest;
+       ll2_tx_pkt.tx_dest = (enum qed_ll2_tx_dest)pkt->tx_dest;
        ll2_tx_pkt.qed_roce_flavor = roce_flavor;
        ll2_tx_pkt.first_frag = pkt->header.baddr;
        ll2_tx_pkt.first_frag_len = pkt->header.len;
--

2.19.0

Signature · 2 lines
Thanks,
~Nick Desaulniers
#2 lore raw
Quoted 42 lines · Nick Desaulniers
Quoted 6 lines · Nathan Chancellor

Clang warns when one enumerated type is explicitly converted to another.

drivers/infiniband/hw/qedr/qedr_roce_cm.c:198:28: warning: implicit conversion from enumeration type 'enum qed_roce_ll2_tx_dest' to different enumeration type 'enum qed_ll2_tx_dest' [-Wenum-conversion]

Quoted 7 lines
        ll2_tx_pkt.tx_dest = pkt->tx_dest;
                           ~ ~~~~~^~~~~~~

1 warning generated.

Avoid this warning by explicitly casting pkt->tx_dest to qed_112_tx_dest, which has the expected values from the type qed_roce_ll2_tx_dest.

But these enums are different lengths, which is problematic for this patch. Is this code broken, or that it's ok for ll2_tx_pkt.tx_dest to have a value that's not a valid enumeration value for enum qed_ll2_tx_dest? (QED_LL2_TX_DEST_MAX 's value (3) is outside the enumeration values of enum qed_roce_ll2_tx_dest).

include/linux/qed/qed_rdma_if.h:
42 enum qed_roce_ll2_tx_dest {
43 /* Light L2 TX Destination to the Network */
44 QED_ROCE_LL2_TX_DEST_NW,
45

 46         /* Light L2 TX Destination to the Loopback */
 47         QED_ROCE_LL2_TX_DEST_LB,
 48         QED_ROCE_LL2_TX_DEST_MAX

49 };

include/linux/qed/qed_ll2_if.h:
64 enum qed_ll2_tx_dest {
65 QED_LL2_TX_DEST_NW, /* Light L2 TX Destination to the
Network */
66 QED_LL2_TX_DEST_LB, /* Light L2 TX Destination to the
Loopback */
67 QED_LL2_TX_DEST_DROP, /* Light L2 Drop the TX packet */
68 QED_LL2_TX_DEST_MAX
69 };

Quoted 2 lines

Maybe the maintainers can clarify?

My assumption was that if QED_ROCE_LL2_TX_DEST_MAX was used that the packet would be dropped.

Turns out that QED_ROCE_LL2_TX_DEST_MAX isn't actually used anywhere in the tree. I suppose that qed_roce_ll2_tx_dest could just be removed and all other instances of those values could be converted to qed_ll2_tx_dest like the rxe patch. I'll test this now and send a patch along if it works out.

Nathan

Quoted 28 lines
Quoted 23 lines
---
 drivers/infiniband/hw/qedr/qedr_roce_cm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/infiniband/hw/qedr/qedr_roce_cm.c +1 -2
index 85578887421b..147e0d69003d 100644
@@ -195,7 +195,7 @@ static int qedr_ll2_post_tx(struct qedr_dev *dev,
 
        ll2_tx_pkt.num_of_bds = 1 /* hdr */  + pkt->n_seg;
        ll2_tx_pkt.vlan = 0;
-       ll2_tx_pkt.tx_dest = pkt->tx_dest;
+       ll2_tx_pkt.tx_dest = (enum qed_ll2_tx_dest)pkt->tx_dest;
        ll2_tx_pkt.qed_roce_flavor = roce_flavor;
        ll2_tx_pkt.first_frag = pkt->header.baddr;
        ll2_tx_pkt.first_frag_len = pkt->header.len;
--

2.19.0

Signature · 2 lines
Thanks,
~Nick Desaulniers
#3 lore raw
Quoted 47 lines · Nathan Chancellor
Quoted 42 lines · Nick Desaulniers
Quoted 6 lines · Nathan Chancellor

Clang warns when one enumerated type is explicitly converted to another.

drivers/infiniband/hw/qedr/qedr_roce_cm.c:198:28: warning: implicit conversion from enumeration type 'enum qed_roce_ll2_tx_dest' to different enumeration type 'enum qed_ll2_tx_dest' [-Wenum-conversion]

Quoted 7 lines
        ll2_tx_pkt.tx_dest = pkt->tx_dest;
                           ~ ~~~~~^~~~~~~

1 warning generated.

Avoid this warning by explicitly casting pkt->tx_dest to qed_112_tx_dest, which has the expected values from the type qed_roce_ll2_tx_dest.

But these enums are different lengths, which is problematic for this patch. Is this code broken, or that it's ok for ll2_tx_pkt.tx_dest to have a value that's not a valid enumeration value for enum qed_ll2_tx_dest? (QED_LL2_TX_DEST_MAX 's value (3) is outside the enumeration values of enum qed_roce_ll2_tx_dest).

include/linux/qed/qed_rdma_if.h:
42 enum qed_roce_ll2_tx_dest {
43 /* Light L2 TX Destination to the Network */
44 QED_ROCE_LL2_TX_DEST_NW,
45

 46         /* Light L2 TX Destination to the Loopback */
 47         QED_ROCE_LL2_TX_DEST_LB,
 48         QED_ROCE_LL2_TX_DEST_MAX

49 };

include/linux/qed/qed_ll2_if.h:
64 enum qed_ll2_tx_dest {
65 QED_LL2_TX_DEST_NW, /* Light L2 TX Destination to the
Network */
66 QED_LL2_TX_DEST_LB, /* Light L2 TX Destination to the
Loopback */
67 QED_LL2_TX_DEST_DROP, /* Light L2 Drop the TX packet */
68 QED_LL2_TX_DEST_MAX
69 };

Quoted 2 lines

Maybe the maintainers can clarify?

Quoted 8 lines

My assumption was that if QED_ROCE_LL2_TX_DEST_MAX was used that the packet would be dropped.

Turns out that QED_ROCE_LL2_TX_DEST_MAX isn't actually used anywhere in the tree. I suppose that qed_roce_ll2_tx_dest could just be removed and all other instances of those values could be converted to qed_ll2_tx_dest like the rxe patch. I'll test this now and send a patch along if it works out.

Even better. Sometimes removing code is the correct solution. Number of lines of code and number of bugs are positively correlated, so the less code sometimes the better.

Quoted 31 lines

Nathan

Quoted 28 lines
Quoted 21 lines
---
 drivers/infiniband/hw/qedr/qedr_roce_cm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/infiniband/hw/qedr/qedr_roce_cm.c +1 -2
index 85578887421b..147e0d69003d 100644
@@ -195,7 +195,7 @@ static int qedr_ll2_post_tx(struct qedr_dev *dev,
 
        ll2_tx_pkt.num_of_bds = 1 /* hdr */  + pkt->n_seg;
        ll2_tx_pkt.vlan = 0;
-       ll2_tx_pkt.tx_dest = pkt->tx_dest;
+       ll2_tx_pkt.tx_dest = (enum qed_ll2_tx_dest)pkt->tx_dest;
        ll2_tx_pkt.qed_roce_flavor = roce_flavor;
        ll2_tx_pkt.first_frag = pkt->header.baddr;
        ll2_tx_pkt.first_frag_len = pkt->header.len;
--
Quoted 2 lines

2.19.0

--
Thanks,
~Nick Desaulniers

Signature · 2 lines
Thanks,
~Nick Desaulniers
#4 lore raw

PATCH v2 1/1 RDMA/qedr: Remove enumerated type qed_roce_ll2_tx_dest

In reply to Nathan Chancellor

Clang warns when one enumerated type is explicitly converted to another.

drivers/infiniband/hw/qedr/qedr_roce_cm.c:198:28: warning: implicit conversion from enumeration type 'enum qed_roce_ll2_tx_dest' to different enumeration type 'enum qed_ll2_tx_dest' [-Wenum-conversion]

        ll2_tx_pkt.tx_dest = pkt->tx_dest;
                           ~ ~~~~~^~~~~~~

1 warning generated.

Turns out that QED_ROCE_LL2_TX_DEST_NW and QED_ROCE_LL2_TX_DEST_LB are only used once in the whole tree and QED_ROCE_LL2_TX_DEST_MAX is used nowhere. Remove them and use the equivalent values from qed_ll2_tx_dest in their place.

---

v1 -> v2:

* Rather than using an explicit cast, just convert the uses to the appropriate values and delete the duplicated enum.

 drivers/infiniband/hw/qedr/qedr_roce_cm.c |  4 ++--
 include/linux/qed/qed_rdma_if.h           | 11 +----------
 2 files changed, 3 insertions(+), 12 deletions(-)
drivers/infiniband/hw/qedr/qedr_roce_cm.c +2 -2
index 85578887421b..e1ac2fd60bb1 100644
@@ -519,9 +519,9 @@ static inline int qedr_gsi_build_packet(struct qedr_dev *dev,
 	}
 
 	if (ether_addr_equal(udh.eth.smac_h, udh.eth.dmac_h))
-		packet->tx_dest = QED_ROCE_LL2_TX_DEST_LB;
+		packet->tx_dest = QED_LL2_TX_DEST_LB;
 	else
-		packet->tx_dest = QED_ROCE_LL2_TX_DEST_NW;
+		packet->tx_dest = QED_LL2_TX_DEST_NW;
 
 	packet->roce_mode = roce_mode;
 	memcpy(packet->header.vaddr, ud_header_buffer, header_size);
include/linux/qed/qed_rdma_if.h +1 -10
index df4d13f7e191..d15f8e4815e3 100644
@@ -39,15 +39,6 @@
 #include <linux/qed/qed_ll2_if.h>
 #include <linux/qed/rdma_common.h>
 
-enum qed_roce_ll2_tx_dest {
-	/* Light L2 TX Destination to the Network */
-	QED_ROCE_LL2_TX_DEST_NW,
-
-	/* Light L2 TX Destination to the Loopback */
-	QED_ROCE_LL2_TX_DEST_LB,
-	QED_ROCE_LL2_TX_DEST_MAX
-};
-
 #define QED_RDMA_MAX_CNQ_SIZE               (0xFFFF)
 
 /* rdma interface */
@@ -581,7 +572,7 @@ struct qed_roce_ll2_packet {
 	int n_seg;
 	struct qed_roce_ll2_buffer payload[RDMA_MAX_SGE_PER_SQ_WQE];
 	int roce_mode;
-	enum qed_roce_ll2_tx_dest tx_dest;
+	enum qed_ll2_tx_dest tx_dest;
 };
 
 enum qed_rdma_type {
Signature · 1 line
2.19.0
Reviewed-by on v2 1/1
#5 lore raw

Re: [PATCH v2] RDMA/qedr: Remove enumerated type qed_roce_ll2_tx_dest

Quoted 33 lines · Nathan Chancellor

Clang warns when one enumerated type is explicitly converted to another.

drivers/infiniband/hw/qedr/qedr_roce_cm.c:198:28: warning: implicit conversion from enumeration type 'enum qed_roce_ll2_tx_dest' to different enumeration type 'enum qed_ll2_tx_dest' [-Wenum-conversion]

        ll2_tx_pkt.tx_dest = pkt->tx_dest;
                           ~ ~~~~~^~~~~~~

1 warning generated.

Turns out that QED_ROCE_LL2_TX_DEST_NW and QED_ROCE_LL2_TX_DEST_LB are only used once in the whole tree and QED_ROCE_LL2_TX_DEST_MAX is used nowhere. Remove them and use the equivalent values from qed_ll2_tx_dest in their place.

---

v1 -> v2:

* Rather than using an explicit cast, just convert the uses to the appropriate values and delete the duplicated enum.

 drivers/infiniband/hw/qedr/qedr_roce_cm.c |  4 ++--
 include/linux/qed/qed_rdma_if.h           | 11 +----------
 2 files changed, 3 insertions(+), 12 deletions(-)
drivers/infiniband/hw/qedr/qedr_roce_cm.c +0 -0
index 85578887421b..e1ac2fd60bb1 100644

drivers/infiniband/hw/qedr/qedr_roce_cm.c static inline int qedr_gsi_build_packet(struct qedr_dev *dev, · line 524

Quoted 8 lines
drivers/infiniband/hw/qedr/qedr_roce_cm.c +2 -2
@@ -519,9 +519,9 @@ static inline int qedr_gsi_build_packet(struct qedr_dev *dev,
        }
 
        if (ether_addr_equal(udh.eth.smac_h, udh.eth.dmac_h))
-               packet->tx_dest = QED_ROCE_LL2_TX_DEST_LB;
+               packet->tx_dest = QED_LL2_TX_DEST_LB;
        else
-               packet->tx_dest = QED_ROCE_LL2_TX_DEST_NW;
+               packet->tx_dest = QED_LL2_TX_DEST_NW;

Yep, looks like these were the only two sites using enum qed_roce_ll2_tx_dest in the whole kernel. The new enum values match the previous enum values, so this is no functional change while simplifying the code (one less enum to get wrong).

Quoted 35 lines
drivers/infiniband/hw/qedr/qedr_roce_cm.c +0 -0
        packet->roce_mode = roce_mode;
        memcpy(packet->header.vaddr, ud_header_buffer, header_size);
include/linux/qed/qed_rdma_if.h +1 -11
index df4d13f7e191..d15f8e4815e3 100644
@@ -39,15 +39,6 @@
 #include <linux/qed/qed_ll2_if.h>
 #include <linux/qed/rdma_common.h>
 
-enum qed_roce_ll2_tx_dest {
-       /* Light L2 TX Destination to the Network */
-       QED_ROCE_LL2_TX_DEST_NW,
-
-       /* Light L2 TX Destination to the Loopback */
-       QED_ROCE_LL2_TX_DEST_LB,
-       QED_ROCE_LL2_TX_DEST_MAX
-};
-
 #define QED_RDMA_MAX_CNQ_SIZE               (0xFFFF)
 
 /* rdma interface */
@@ -581,7 +572,7 @@ struct qed_roce_ll2_packet {
        int n_seg;
        struct qed_roce_ll2_buffer payload[RDMA_MAX_SGE_PER_SQ_WQE];
        int roce_mode;
-       enum qed_roce_ll2_tx_dest tx_dest;
+       enum qed_ll2_tx_dest tx_dest;
 };
 
 enum qed_rdma_type {
--

2.19.0

Signature · 2 lines
Thanks,
~Nick Desaulniers
#6 lore raw

RE: [PATCH v2] RDMA/qedr: Remove enumerated type qed_roce_ll2_tx_dest

Quoted 91 lines

From: Nick Desaulniers <[email protected]>
Sent: Friday, September 28, 2018 1:30 AM

External Email

Quoted 11 lines · Nathan Chancellor

Clang warns when one enumerated type is explicitly converted to another.

drivers/infiniband/hw/qedr/qedr_roce_cm.c:198:28: warning: implicit conversion from enumeration type 'enum qed_roce_ll2_tx_dest' to different enumeration type 'enum qed_ll2_tx_dest' [-Wenum-conversion]

        ll2_tx_pkt.tx_dest = pkt->tx_dest;
                           ~ ~~~~~^~~~~~~

1 warning generated.

Turns out that QED_ROCE_LL2_TX_DEST_NW and

QED_ROCE_LL2_TX_DEST_LB are

Quoted 1 line

only used once in the whole tree and QED_ROCE_LL2_TX_DEST_MAX is

used

drivers/infiniband/hw/qedr/qedr_roce_cm.c static inline int qedr_gsi_build_packet(struct

Quoted 22 lines

nowhere. Remove them and use the equivalent values from qed_ll2_tx_dest in their place.

---

v1 -> v2:

* Rather than using an explicit cast, just convert the uses to the appropriate values and delete the duplicated enum.

 drivers/infiniband/hw/qedr/qedr_roce_cm.c |  4 ++--
 include/linux/qed/qed_rdma_if.h           | 11 +----------
 2 files changed, 3 insertions(+), 12 deletions(-)
drivers/infiniband/hw/qedr/qedr_roce_cm.c +0 -0
index 85578887421b..e1ac2fd60bb1 100644
@@ -519,9 +519,9 @@ static inline int qedr_gsi_build_packet(struct

qedr_dev *dev,

drivers/infiniband/hw/qedr/qedr_roce_cm.c static inline int qedr_gsi_build_packet(struct qedr_dev *dev, · line 524

Quoted 8 lines
drivers/infiniband/hw/qedr/qedr_roce_cm.c +2 -2
        }
 
        if (ether_addr_equal(udh.eth.smac_h, udh.eth.dmac_h))
-               packet->tx_dest = QED_ROCE_LL2_TX_DEST_LB;
+               packet->tx_dest = QED_LL2_TX_DEST_LB;
        else
-               packet->tx_dest = QED_ROCE_LL2_TX_DEST_NW;
+               packet->tx_dest = QED_LL2_TX_DEST_NW;

Yep, looks like these were the only two sites using enum qed_roce_ll2_tx_dest in the whole kernel. The new enum values match the previous enum values, so this is no functional change while simplifying the code (one less enum to get wrong).

Quoted 32 lines
drivers/infiniband/hw/qedr/qedr_roce_cm.c +0 -0
        packet->roce_mode = roce_mode;
        memcpy(packet->header.vaddr, ud_header_buffer, header_size);
(quoted diff) +0 -0

b/include/linux/qed/qed_rdma_if.h index df4d13f7e191..d15f8e4815e3 100644

include/linux/qed/qed_rdma_if.h +1 -10
@@ -39,15 +39,6 @@
 #include <linux/qed/qed_ll2_if.h>
 #include <linux/qed/rdma_common.h>
 
-enum qed_roce_ll2_tx_dest {
-       /* Light L2 TX Destination to the Network */
-       QED_ROCE_LL2_TX_DEST_NW,
-
-       /* Light L2 TX Destination to the Loopback */
-       QED_ROCE_LL2_TX_DEST_LB,
-       QED_ROCE_LL2_TX_DEST_MAX
-};
-
 #define QED_RDMA_MAX_CNQ_SIZE               (0xFFFF)
 
 /* rdma interface */
@@ -581,7 +572,7 @@ struct qed_roce_ll2_packet {
        int n_seg;
        struct qed_roce_ll2_buffer payload[RDMA_MAX_SGE_PER_SQ_WQE];
        int roce_mode;
-       enum qed_roce_ll2_tx_dest tx_dest;
+       enum qed_ll2_tx_dest tx_dest;
 };
 
Quoted 9 lines
Quoted 4 lines
include/linux/qed/qed_rdma_if.h +0 -1
 enum qed_rdma_type {
--

2.19.0

--
Thanks,
~Nick Desaulniers

Thanks, definitely looks cleaner.

#7 lore raw

Re: [PATCH v2] RDMA/qedr: Remove enumerated type qed_roce_ll2_tx_dest

In reply to Nathan Chancellor
Quoted 18 lines · Nathan Chancellor

Clang warns when one enumerated type is explicitly converted to another.

drivers/infiniband/hw/qedr/qedr_roce_cm.c:198:28: warning: implicit conversion from enumeration type 'enum qed_roce_ll2_tx_dest' to different enumeration type 'enum qed_ll2_tx_dest' [-Wenum-conversion]

        ll2_tx_pkt.tx_dest = pkt->tx_dest;
                           ~ ~~~~~^~~~~~~

1 warning generated.

Turns out that QED_ROCE_LL2_TX_DEST_NW and QED_ROCE_LL2_TX_DEST_LB are only used once in the whole tree and QED_ROCE_LL2_TX_DEST_MAX is used nowhere. Remove them and use the equivalent values from qed_ll2_tx_dest in their place.

Quoted 6 lines

---

v1 -> v2:

* Rather than using an explicit cast, just convert the uses to the appropriate values and delete the duplicated enum.

Applied to for-next, thanks

Jason