Skip to contentSkip to conversation

tap: fix build of tap_bpf_program

PATCH In review Stephen Hemminger 4 participants 9 messages Jul 17, 2023 – Jul 26, 2023 Network drivers

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 tap: fix build of tap_bpf_program

The tap_bpf_program.c is not built as part of normal DPDK EAL environment. It is intended to be built standalone and does not use rte_common.h.

This reverts the related change from
commit ef5baf3486e0 ("replace packed attributes")

Note: this patch will cause expected warnings from checkpatch because the code involved is not used directly in DPDK environment.

---
 drivers/net/tap/tap_bpf_program.c | 4 ++--
 drivers/net/tap/tap_rss.h         | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/net/tap/tap_bpf_program.c +2 -2
index 20c310e5e7ba..532e8838fe27 100644
@@ -75,14 +75,14 @@ struct ipv4_l3_l4_tuple {
 	__u32    dst_addr;
 	__u16    dport;
 	__u16    sport;
-} __rte_packed;
+} __attribute__((packed));
 
 struct ipv6_l3_l4_tuple {
 	__u8        src_addr[16];
 	__u8        dst_addr[16];
 	__u16       dport;
 	__u16       sport;
-} __rte_packed;
+} __attribute__((packed));
 
 static const __u8 def_rss_key[TAP_RSS_HASH_KEY_SIZE] = {
 	0xd1, 0x81, 0xc6, 0x2c,
drivers/net/tap/tap_rss.h +1 -1
index 48c151cf6b68..dff46a012f94 100644
@@ -35,6 +35,6 @@ struct rss_key {
 	__u32 key_size;
 	__u32 queues[TAP_MAX_QUEUES];
 	__u32 nb_queues;
-} __rte_packed;
+} __attribute__((packed));
 
 #endif /* _TAP_RSS_H_ */
Signature · 1 line
2.39.2
Acked-by on v1 1/1
#1 lore raw
Quoted 4 lines · Stephen Hemminger

The tap_bpf_program.c is not built as part of normal DPDK EAL environment. It is intended to be built standalone and does not use rte_common.h.

Quoted 8 lines

This reverts the related change from
commit ef5baf3486e0 ("replace packed attributes")

Note: this patch will cause expected warnings from checkpatch because the code involved is not used directly in DPDK environment.

Agree, this seems done by mistake as part of batch update,

But I can't update the bpf file at all, if I am not missing something I am not sure if we should get just this update or have a patch/patchset that fixes the build.

@Ophir, how the bpf file is compiled? And did you test it recently?

I am using command from the documentation:
clang -O2 -emit-llvm -c tap_bpf_program.c -o - | llc -march=bpf -filetype=obj -o tap_bpf_program.o

@Aaron, should we add a simple build test around it, since this file is not build by default with DPDK build?

Quoted 38 lines
---
 drivers/net/tap/tap_bpf_program.c | 4 ++--
 drivers/net/tap/tap_rss.h         | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/net/tap/tap_bpf_program.c +2 -2
index 20c310e5e7ba..532e8838fe27 100644
@@ -75,14 +75,14 @@ struct ipv4_l3_l4_tuple {
 	__u32    dst_addr;
 	__u16    dport;
 	__u16    sport;
-} __rte_packed;
+} __attribute__((packed));
 
 struct ipv6_l3_l4_tuple {
 	__u8        src_addr[16];
 	__u8        dst_addr[16];
 	__u16       dport;
 	__u16       sport;
-} __rte_packed;
+} __attribute__((packed));
 
 static const __u8 def_rss_key[TAP_RSS_HASH_KEY_SIZE] = {
 	0xd1, 0x81, 0xc6, 0x2c,
drivers/net/tap/tap_rss.h +1 -1
index 48c151cf6b68..dff46a012f94 100644
@@ -35,6 +35,6 @@ struct rss_key {
 	__u32 key_size;
 	__u32 queues[TAP_MAX_QUEUES];
 	__u32 nb_queues;
-} __rte_packed;
+} __attribute__((packed));
 
 #endif /* _TAP_RSS_H_ */
#2 lore raw
Quoted 14 lines · Ferruh Yigit
Quoted 4 lines · Stephen Hemminger

The tap_bpf_program.c is not built as part of normal DPDK EAL environment. It is intended to be built standalone and does not use rte_common.h.

Quoted 8 lines

This reverts the related change from
commit ef5baf3486e0 ("replace packed attributes")

Note: this patch will cause expected warnings from checkpatch because the code involved is not used directly in DPDK environment.

Quoted 6 lines

Agree, this seems done by mistake as part of batch update,

But I can't update the bpf file at all, if I am not missing something I

* I can't *compile* the bpf file ...

Quoted 54 lines

am not sure if we should get just this update or have a patch/patchset that fixes the build.

@Ophir, how the bpf file is compiled? And did you test it recently?

I am using command from the documentation:
clang -O2 -emit-llvm -c tap_bpf_program.c -o - | llc -march=bpf -filetype=obj -o tap_bpf_program.o

@Aaron, should we add a simple build test around it, since this file is not build by default with DPDK build?

Quoted 38 lines
---
 drivers/net/tap/tap_bpf_program.c | 4 ++--
 drivers/net/tap/tap_rss.h         | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/net/tap/tap_bpf_program.c +2 -2
index 20c310e5e7ba..532e8838fe27 100644
@@ -75,14 +75,14 @@ struct ipv4_l3_l4_tuple {
 	__u32    dst_addr;
 	__u16    dport;
 	__u16    sport;
-} __rte_packed;
+} __attribute__((packed));
 
 struct ipv6_l3_l4_tuple {
 	__u8        src_addr[16];
 	__u8        dst_addr[16];
 	__u16       dport;
 	__u16       sport;
-} __rte_packed;
+} __attribute__((packed));
 
 static const __u8 def_rss_key[TAP_RSS_HASH_KEY_SIZE] = {
 	0xd1, 0x81, 0xc6, 0x2c,
drivers/net/tap/tap_rss.h +1 -1
index 48c151cf6b68..dff46a012f94 100644
@@ -35,6 +35,6 @@ struct rss_key {
 	__u32 key_size;
 	__u32 queues[TAP_MAX_QUEUES];
 	__u32 nb_queues;
-} __rte_packed;
+} __attribute__((packed));
 
 #endif /* _TAP_RSS_H_ */
#3 lore raw
Quoted 15 lines · Ferruh Yigit
Quoted 14 lines · Ferruh Yigit
Quoted 12 lines · Stephen Hemminger

The tap_bpf_program.c is not built as part of normal DPDK EAL environment. It is intended to be built standalone and does not use rte_common.h.

This reverts the related change from
commit ef5baf3486e0 ("replace packed attributes")

Note: this patch will cause expected warnings from checkpatch because the code involved is not used directly in DPDK environment.

Quoted 23 lines
Quoted 6 lines

Agree, this seems done by mistake as part of batch update,

But I can't update the bpf file at all, if I am not missing something I

* I can't *compile* the bpf file ...

Quoted 14 lines

am not sure if we should get just this update or have a patch/patchset that fixes the build.

@Ophir, how the bpf file is compiled? And did you test it recently?

I am using command from the documentation:
clang -O2 -emit-llvm -c tap_bpf_program.c -o - | llc -march=bpf -filetype=obj -o tap_bpf_program.o

@Aaron, should we add a simple build test around it, since this file is not build by default with DPDK build?

I saw Stephen already touched to the topic:
https://inbox.dpdk.org/dev/[email protected]/

Quoted 41 lines
Quoted 40 lines
Quoted 39 lines
---
 drivers/net/tap/tap_bpf_program.c | 4 ++--
 drivers/net/tap/tap_rss.h         | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/net/tap/tap_bpf_program.c +2 -2
index 20c310e5e7ba..532e8838fe27 100644
@@ -75,14 +75,14 @@ struct ipv4_l3_l4_tuple {
 	__u32    dst_addr;
 	__u16    dport;
 	__u16    sport;
-} __rte_packed;
+} __attribute__((packed));
 
 struct ipv6_l3_l4_tuple {
 	__u8        src_addr[16];
 	__u8        dst_addr[16];
 	__u16       dport;
 	__u16       sport;
-} __rte_packed;
+} __attribute__((packed));
 
 static const __u8 def_rss_key[TAP_RSS_HASH_KEY_SIZE] = {
 	0xd1, 0x81, 0xc6, 0x2c,
drivers/net/tap/tap_rss.h +1 -1
index 48c151cf6b68..dff46a012f94 100644
@@ -35,6 +35,6 @@ struct rss_key {
 	__u32 key_size;
 	__u32 queues[TAP_MAX_QUEUES];
 	__u32 nb_queues;
-} __rte_packed;
+} __attribute__((packed));
 
 #endif /* _TAP_RSS_H_ */
#4 lore raw
In reply to Ferruh Yigit
Quoted 15 lines · On Wed, 19 Jul 2023 11:03:36 +0100 Ferruh Yigit <[email protected]> wrote:
Quoted 14 lines · Ferruh Yigit
Quoted 12 lines · Stephen Hemminger

The tap_bpf_program.c is not built as part of normal DPDK EAL environment. It is intended to be built standalone and does not use rte_common.h.

This reverts the related change from
commit ef5baf3486e0 ("replace packed attributes")

Note: this patch will cause expected warnings from checkpatch because the code involved is not used directly in DPDK environment.

Quoted 17 lines
Quoted 6 lines

Agree, this seems done by mistake as part of batch update,

But I can't update the bpf file at all, if I am not missing something I

* I can't *compile* the bpf file ...

Quoted 8 lines

am not sure if we should get just this update or have a patch/patchset that fixes the build.

@Ophir, how the bpf file is compiled? And did you test it recently?

I am using command from the documentation:
clang -O2 -emit-llvm -c tap_bpf_program.c -o - | llc -march=bpf -filetype=obj -o tap_bpf_program.o

It looks like this won't work because it was expecting to be able to find header files from older version of iproute2. These are not distributed, and the change to support libbpf in iproute2 makes the current versions not work.

As a stopgap, will look back in history and see what version of header files will at least get a working build.

From there, need to replace how the conversion of .o to array works. Would prefer to use dlopen() to read the ELF file rather than expecting developers to hack together their own tools.

Not sure how much effort is really needed here. This is only being used for the case of rte_flow with multiq RSS. Probably, no one ever used it.

#5 lore raw
Quoted 44 lines · Stephen Hemminger
Quoted 15 lines · On Wed, 19 Jul 2023 11:03:36 +0100 Ferruh Yigit <[email protected]> wrote:
Quoted 14 lines · Ferruh Yigit
Quoted 12 lines · Stephen Hemminger

The tap_bpf_program.c is not built as part of normal DPDK EAL environment. It is intended to be built standalone and does not use rte_common.h.

This reverts the related change from
commit ef5baf3486e0 ("replace packed attributes")

Note: this patch will cause expected warnings from checkpatch because the code involved is not used directly in DPDK environment.

Quoted 17 lines
Quoted 6 lines

Agree, this seems done by mistake as part of batch update,

But I can't update the bpf file at all, if I am not missing something I

* I can't *compile* the bpf file ...

Quoted 8 lines

am not sure if we should get just this update or have a patch/patchset that fixes the build.

@Ophir, how the bpf file is compiled? And did you test it recently?

I am using command from the documentation:
clang -O2 -emit-llvm -c tap_bpf_program.c -o - | llc -march=bpf -filetype=obj -o tap_bpf_program.o

It looks like this won't work because it was expecting to be able to find header files from older version of iproute2. These are not distributed, and the change to support libbpf in iproute2 makes the current versions not work.

As a stopgap, will look back in history and see what version of header files will at least get a working build.

Quoted 8 lines

From there, need to replace how the conversion of .o to array works. Would prefer to use dlopen() to read the ELF file rather than expecting developers to hack together their own tools.

Not sure how much effort is really needed here. This is only being used for the case of rte_flow with multiq RSS. Probably, no one ever used it.

Should we remove the file, instead of fixing '__rte_packed'?

#6 lore raw
Quoted 49 lines · Ferruh Yigit
Quoted 48 lines · Stephen Hemminger
Quoted 15 lines · On Wed, 19 Jul 2023 11:03:36 +0100 Ferruh Yigit <[email protected]> wrote:
Quoted 14 lines · Ferruh Yigit
Quoted 12 lines · Stephen Hemminger

The tap_bpf_program.c is not built as part of normal DPDK EAL environment. It is intended to be built standalone and does not use rte_common.h.

This reverts the related change from
commit ef5baf3486e0 ("replace packed attributes")

Note: this patch will cause expected warnings from checkpatch because the code involved is not used directly in DPDK environment.

Quoted 17 lines
Quoted 6 lines

Agree, this seems done by mistake as part of batch update,

But I can't update the bpf file at all, if I am not missing something I

* I can't *compile* the bpf file ...

Quoted 8 lines

am not sure if we should get just this update or have a patch/patchset that fixes the build.

@Ophir, how the bpf file is compiled? And did you test it recently?

I am using command from the documentation:
clang -O2 -emit-llvm -c tap_bpf_program.c -o - | llc -march=bpf -filetype=obj -o tap_bpf_program.o

It looks like this won't work because it was expecting to be able to find header files from older version of iproute2. These are not distributed, and the change to support libbpf in iproute2 makes the current versions not work.

As a stopgap, will look back in history and see what version of header files will at least get a working build.

From there, need to replace how the conversion of .o to array works. Would prefer to use dlopen() to read the ELF file rather than expecting developers to hack together their own tools.

Quoted 7 lines
Quoted 4 lines

Not sure how much effort is really needed here. This is only being used for the case of rte_flow with multiq RSS. Probably, no one ever used it.

Should we remove the file, instead of fixing '__rte_packed'?

+Long, and af_xdp maintainers,

@Long, do you know if this bfp code is still in use somewhere, if so is the user interested in fixing/maintaining the code?

@Ciara, @Qi, do you see any benefit to keep/extend this kind of bfp file usage? Do you think is this something to invest more?

#7 lore raw
Quoted 58 lines

Subject: Re: [PATCH] tap: fix build of tap_bpf_program

Quoted 55 lines · Ferruh Yigit
Quoted 53 lines · Stephen Hemminger
Quoted 17 lines · On Wed, 19 Jul 2023 11:03:36 +0100 Ferruh Yigit <[email protected]> wrote:
Quoted 16 lines · Ferruh Yigit
Quoted 4 lines · Stephen Hemminger

The tap_bpf_program.c is not built as part of normal DPDK EAL environment. It is intended to be built standalone and does not use rte_common.h.

Quoted 8 lines

This reverts the related change from commit ef5baf3486e0 ("replace packed attributes")

Note: this patch will cause expected warnings from checkpatch because the code involved is not used directly in DPDK environment.

Agree, this seems done by mistake as part of batch update,

Quoted 16 lines
Quoted 5 lines

But I can't update the bpf file at all, if I am not missing something I

* I can't *compile* the bpf file ...

Quoted 8 lines

am not sure if we should get just this update or have a patch/patchset that fixes the build.

@Ophir, how the bpf file is compiled? And did you test it recently?

I am using command from the documentation:
clang -O2 -emit-llvm -c tap_bpf_program.c -o - | llc -march=bpf -filetype=obj -o tap_bpf_program.o

It looks like this won't work because it was expecting to be able to find header files from older version of iproute2. These are not distributed, and the change to support libbpf in iproute2 makes the current versions not work.

As a stopgap, will look back in history and see what version of header files will at least get a working build.

From there, need to replace how the conversion of .o to array works. Would prefer to use dlopen() to read the ELF file rather than expecting developers to hack together their own tools.

Not sure how much effort is really needed here. This is only being used for the case of rte_flow with multiq RSS. Probably, no one ever used it.

Quoted 7 lines
Quoted 2 lines

Should we remove the file, instead of fixing '__rte_packed'?

+Long, and af_xdp maintainers,

@Long, do you know if this bfp code is still in use somewhere, if so is the user interested in fixing/maintaining the code?

I haven't seen usage in the support cases we have handled. Given support cases represent very small percent of actual usage, I suspect it's still being used.

Quoted 4 lines

@Ciara, @Qi, do you see any benefit to keep/extend this kind of bfp file usage? Do you think is this something to invest more?

#8 lore raw
In reply to Ferruh Yigit
Quoted 58 lines · Ferruh Yigit
Quoted 49 lines · Stephen Hemminger
Quoted 48 lines
Quoted 15 lines · On Wed, 19 Jul 2023 11:03:36 +0100 Ferruh Yigit <[email protected]> wrote:
Quoted 14 lines · Ferruh Yigit
Quoted 12 lines · Stephen Hemminger

The tap_bpf_program.c is not built as part of normal DPDK EAL environment. It is intended to be built standalone and does not use rte_common.h.

This reverts the related change from
commit ef5baf3486e0 ("replace packed attributes")

Note: this patch will cause expected warnings from checkpatch because the code involved is not used directly in DPDK environment.

Quoted 17 lines
Quoted 6 lines

Agree, this seems done by mistake as part of batch update,

But I can't update the bpf file at all, if I am not missing something I

* I can't *compile* the bpf file ...

Quoted 8 lines

am not sure if we should get just this update or have a patch/patchset that fixes the build.

@Ophir, how the bpf file is compiled? And did you test it recently?

I am using command from the documentation:
clang -O2 -emit-llvm -c tap_bpf_program.c -o - | llc -march=bpf -filetype=obj -o tap_bpf_program.o

It looks like this won't work because it was expecting to be able to find header files from older version of iproute2. These are not distributed, and the change to support libbpf in iproute2 makes the current versions not work.

As a stopgap, will look back in history and see what version of header files will at least get a working build.

From there, need to replace how the conversion of .o to array works. Would prefer to use dlopen() to read the ELF file rather than expecting developers to hack together their own tools.

Quoted 7 lines
Quoted 4 lines

Not sure how much effort is really needed here. This is only being used for the case of rte_flow with multiq RSS. Probably, no one ever used it.

Should we remove the file, instead of fixing '__rte_packed'?

Quoted 8 lines

+Long, and af_xdp maintainers,

@Long, do you know if this bfp code is still in use somewhere, if so is the user interested in fixing/maintaining the code?

@Ciara, @Qi, do you see any benefit to keep/extend this kind of bfp file usage? Do you think is this something to invest more?

If the code is still being used I would agree with Stephen that using dlopen or libbpf to load the eBPF code would be preferable. The current steps are difficult to follow.