sfc: add CONFIG_INET dependency for TC offload
Subsystems (3)
PATCH v2 1/3 sfc: add CONFIG_INET dependency for TC offload
From: Arnd Bergmann <[email protected]>
The driver now fails to link when CONFIG_INET is disabled, so add an explicit Kconfig dependency:
ld.lld: error: undefined symbol: ip_route_output_flow
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
referenced by
tc_encap_actions.c
drivers/net/ethernet/sfc/tc_encap_actions.o:(efx_tc_flower_create_encap_md) in archive vmlinux.a
ld.lld: error: undefined symbol: ip_send_check
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
referenced by
tc_encap_actions.c
drivers/net/ethernet/sfc/tc_encap_actions.o:(efx_gen_encap_header) in archive vmlinux.a
referenced bytc_encap_actions.c
drivers/net/ethernet/sfc/tc_encap_actions.o:(efx_gen_encap_header) in archive vmlinux.a
ld.lld: error: undefined symbol: arp_tbl
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
referenced by
tc_encap_actions.c
drivers/net/ethernet/sfc/tc_encap_actions.o:(efx_tc_netevent_event) in archive vmlinux.a
referenced bytc_encap_actions.c
drivers/net/ethernet/sfc/tc_encap_actions.o:(efx_tc_netevent_event) in archive vmlinux.a
---
v2: add Fixes and Closes tags
--- drivers/net/ethernet/sfc/Kconfig | 1 + 1 file changed, 1 insertion(+)
drivers/net/ethernet/sfc/Kconfig +1 -0
@@ -50,6 +50,7 @@ config SFC_MCDI_MON config SFC_SRIOV bool "Solarflare SFC9100-family SR-IOV support" depends on SFC && PCI_IOV + depends on INET default y help This enables support for the Single Root I/O Virtualization
Signature · 1 line
2.39.2
PATCH v2 2/3 sfc: fix uninitialized variable use
From: Arnd Bergmann <[email protected]>
The new efx_bind_neigh() function contains a broken code path when IPV6 is disabled:
drivers/net/ethernet/sfc/tc_encap_actions.c:144:7: error: variable 'n' is used uninitialized whenever 'if' condition is true [-Werror,-Wsometimes-uninitialized]
if (encap->type & EFX_ENCAP_FLAG_IPV6) {
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/net/ethernet/sfc/tc_encap_actions.c:184:8: note: uninitialized use occurs here
if (!n) {
^
drivers/net/ethernet/sfc/tc_encap_actions.c:144:3: note: remove the 'if' if its condition is always false
if (encap->type & EFX_ENCAP_FLAG_IPV6) {
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/net/ethernet/sfc/tc_encap_actions.c:141:22: note: initialize the variable 'n' to silence this warning
struct neighbour *n;
^
= NULLChange it to use the existing error handling path here.
---
v2: use 'goto' instead of incorrectly entering another error path
--- drivers/net/ethernet/sfc/tc_encap_actions.c | 1 + 1 file changed, 1 insertion(+)
drivers/net/ethernet/sfc/tc_encap_actions.c +1 -0
@@ -164,6 +164,7 @@ static int efx_bind_neigh(struct efx_nic *efx,*/ rc = -EOPNOTSUPP ;NL_SET_ERR_MSG_MOD (extack,"No IPv6 support (neigh bind)" ); + goto out_free;#endif }else { rt =ip_route_output_key (net, &flow4);
Signature · 1 line
2.39.2
PATCH 3/3 sfc: selftest: fix struct packing
In reply to Arnd BergmannFrom: Arnd Bergmann <[email protected]>
Three of the sfc drivers define a packed loopback_payload structure with an ethernet header followed by an IP header. However, the kernel definition of iphdr specifies that this is 4-byte aligned, causing a W=1 warning:
net/ethernet/sfc/siena/selftest.c:46:15: error: field ip within 'struct efx_loopback_payload' is less aligned than 'struct iphdr' and is usually due to 'struct efx_loopback_payload' being packed, which can lead to unaligned accesses [-Werror,-Wunaligned-access]
struct iphdr ip;
As the iphdr packing is not easily changed without breaking other code, change the three structures to use a local definition instead.
--- drivers/net/ethernet/sfc/falcon/selftest.c | 21 ++++++++++++++++++++- drivers/net/ethernet/sfc/selftest.c | 21 ++++++++++++++++++++- drivers/net/ethernet/sfc/siena/selftest.c | 21 ++++++++++++++++++++- 3 files changed, 60 insertions(+), 3 deletions(-)
drivers/net/ethernet/sfc/falcon/selftest.c +20 -1
@@ -40,7 +40,26 @@*/ struct ef4_loopback_payload {struct ethhdr header; -struct iphdr ip; +struct { +#if defined(__LITTLE_ENDIAN_BITFIELD) +__u8 ihl:4 , + version:4 ; +#elif defined (__BIG_ENDIAN_BITFIELD) +__u8 version:4 , + ihl:4 ; +#else +#error "Please fix <asm/byteorder.h>" +#endif +__u8 tos; +__be16 tot_len; +__be16 id; +__be16 frag_off; +__u8 ttl; +__u8 protocol; +__sum16 check; +__be32 saddr; +__be32 daddr; + } __packed ip; /* unaligned struct iphdr */struct udphdr udp;__be16 iteration;char msg[64 ];
drivers/net/ethernet/sfc/selftest.c +20 -1
@@ -43,7 +43,26 @@*/ struct efx_loopback_payload {struct ethhdr header; -struct iphdr ip; +struct { +#if defined(__LITTLE_ENDIAN_BITFIELD) +__u8 ihl:4 , + version:4 ; +#elif defined (__BIG_ENDIAN_BITFIELD) +__u8 version:4 , + ihl:4 ; +#else +#error "Please fix <asm/byteorder.h>" +#endif +__u8 tos; +__be16 tot_len; +__be16 id; +__be16 frag_off; +__u8 ttl; +__u8 protocol; +__sum16 check; +__be32 saddr; +__be32 daddr; + } __packed ip; /* unaligned struct iphdr */struct udphdr udp;__be16 iteration;char msg[64 ];
drivers/net/ethernet/sfc/siena/selftest.c +20 -1
@@ -43,7 +43,26 @@*/ struct efx_loopback_payload {struct ethhdr header; -struct iphdr ip; +struct { +#if defined(__LITTLE_ENDIAN_BITFIELD) +__u8 ihl:4 , + version:4 ; +#elif defined (__BIG_ENDIAN_BITFIELD) +__u8 version:4 , + ihl:4 ; +#else +#error "Please fix <asm/byteorder.h>" +#endif +__u8 tos; +__be16 tot_len; +__be16 id; +__be16 frag_off; +__u8 ttl; +__u8 protocol; +__sum16 check; +__be32 saddr; +__be32 daddr; + } __packed ip; /* unaligned struct iphdr */struct udphdr udp;__be16 iteration;char msg[64 ];
Signature · 1 line
2.39.2
Edward Cree
maintainerQuoted 21 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
The driver now fails to link when
CONFIG_INETis disabled, so add an explicitKconfigdependency:ld.lld: error: undefined symbol:
ip_route_output_flowQuoted 2 lines
Quoted 2 lines
Quoted 2 lines
referenced by
tc_encap_actions.c
drivers/net/ethernet/sfc/tc_encap_actions.o:(efx_tc_flower_create_encap_md) in archive vmlinux.ald.lld: error: undefined symbol:
ip_send_checkQuoted 4 lines
Quoted 4 lines
Quoted 4 lines
referenced by
tc_encap_actions.c
drivers/net/ethernet/sfc/tc_encap_actions.o:(efx_gen_encap_header) in archive vmlinux.a
referenced bytc_encap_actions.c
drivers/net/ethernet/sfc/tc_encap_actions.o:(efx_gen_encap_header) in archive vmlinux.ald.lld: error: undefined symbol:
arp_tblQuoted 4 lines
Quoted 4 lines
Quoted 4 lines
referenced by
tc_encap_actions.c
drivers/net/ethernet/sfc/tc_encap_actions.o:(efx_tc_netevent_event) in archive vmlinux.a
referenced bytc_encap_actions.c
drivers/net/ethernet/sfc/tc_encap_actions.o:(efx_tc_netevent_event) in archive vmlinux.a
Quoted 5 lines
Fixes a1e82162af0b8 ("sfc: generate encap headers for TC offload")Reviewed-by Edward Cree <[email protected]>Reviewed-by Simon Horman <[email protected]>Signed-off-by Arnd Bergmann <[email protected]>
Additionally
The subject line doesn't specify a target tree; this should be for net-next.
Quoted 19 lines
---
v2: add Fixes and Closes tags--- drivers/net/ethernet/sfc/Kconfig | 1 + 1 file changed, 1 insertion(+)drivers/net/ethernet/sfc/Kconfig +1
-0@@ -50,6 +50,7 @@ config SFC_MCDI_MON config SFC_SRIOV bool "Solarflare SFC9100-family SR-IOV support" depends on SFC && PCI_IOV + depends on INET default y help This enables support for the Single Root I/O Virtualization
Re: [PATCH 2/3] [v2] sfc: fix uninitialized variable use
In reply to Arnd BergmannQuoted 19 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
The new
efx_bind_neigh()function contains a broken code path when IPV6 is disabled:
drivers/net/ethernet/sfc/tc_encap_actions.c:144:7: error: variable 'n' is used uninitialized whenever 'if' condition is true [-Werror,-Wsometimes-uninitialized]if (encap->type & EFX_ENCAP_FLAG_IPV6) { ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ drivers/net/ethernet/sfc/tc_encap_actions.c:184:8: note: uninitialized use occurs here if (!n) { ^ drivers/net/ethernet/sfc/tc_encap_actions.c:144:3: note: remove the 'if' if its condition is always false if (encap->type & EFX_ENCAP_FLAG_IPV6) { ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ drivers/net/ethernet/sfc/tc_encap_actions.c:141:22: note: initialize the variable 'n' to silence this warning struct neighbour *n; ^ = NULL
Quoted 5 lines
Change it to use the existing error handling path here.
Fixes 7e5e7d800011a ("sfc: neighbour lookup for TC encap action offload")Suggested-by Edward Cree <[email protected]>Signed-off-by Arnd Bergmann <[email protected]>
Quoted 19 lines
---
v2: use 'goto' instead of incorrectly entering another error path--- drivers/net/ethernet/sfc/tc_encap_actions.c | 1 + 1 file changed, 1 insertion(+)drivers/net/ethernet/sfc/tc_encap_actions.c +1
-0@@ -164,6 +164,7 @@ static int efx_bind_neigh(struct efx_nic *efx,*/ rc = -EOPNOTSUPP ;NL_SET_ERR_MSG_MOD (extack,"No IPv6 support (neigh bind)" ); + goto out_free;#endif }else { rt =ip_route_output_key (net, &flow4);
Edward Cree
maintainerRe: [PATCH 3/3] sfc: selftest: fix struct packing
In reply to Arnd BergmannQuoted 6 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
Three of the sfc drivers define a packed
loopback_payloadstructure with an ethernet header followed by an IP header. However, the kernel definition of iphdr specifies that this is 4-byte aligned, causing a W=1 warning:
Quoted 7 lines
net/ethernet/sfc/siena/selftest.c:46:15: error: field ip within 'structefx_loopback_payload' is less aligned than 'struct iphdr' and is usually due to 'structefx_loopback_payload' being packed, which can lead to unaligned accesses [-Werror,-Wunaligned-access]struct iphdr ip;As the iphdr packing is not easily changed without breaking other code, change the three structures to use a local definition instead.
Signed-off-by Arnd Bergmann <[email protected]>
Duplicating the definition isn't the prettiest thing in the world; it'd
do for a quick fix if needed but I assume W=1 warnings aren't blocking anyone, so maybe defer this one for now and I'll follow up soon with a rewrite that fixes this more cleanly? My idea is to drop the __packed from the containing struct, make efx_begin_loopback() copy the layers separately, and efx_loopback_rx_packet() similarly do something less direct than casting the packet data to the struct.
But I don't insist on it; if you want this fix in immediately then I'm
okay with that too.
Quoted 103 lines
--- drivers/net/ethernet/sfc/falcon/selftest.c | 21 ++++++++++++++++++++- drivers/net/ethernet/sfc/selftest.c | 21 ++++++++++++++++++++- drivers/net/ethernet/sfc/siena/selftest.c | 21 ++++++++++++++++++++- 3 files changed, 60 insertions(+), 3 deletions(-)drivers/net/ethernet/sfc/falcon/selftest.c +20
-1@@ -40,7 +40,26 @@*/ struct ef4_loopback_payload {struct ethhdr header; -struct iphdr ip; +struct { +#if defined(__LITTLE_ENDIAN_BITFIELD) +__u8 ihl:4 , + version:4 ; +#elif defined (__BIG_ENDIAN_BITFIELD) +__u8 version:4 , + ihl:4 ; +#else +#error "Please fix <asm/byteorder.h>" +#endif +__u8 tos; +__be16 tot_len; +__be16 id; +__be16 frag_off; +__u8 ttl; +__u8 protocol; +__sum16 check; +__be32 saddr; +__be32 daddr; + } __packed ip; /* unaligned struct iphdr */struct udphdr udp;__be16 iteration;char msg[64 ];drivers/net/ethernet/sfc/selftest.c +20
-1@@ -43,7 +43,26 @@*/ struct efx_loopback_payload {struct ethhdr header; -struct iphdr ip; +struct { +#if defined(__LITTLE_ENDIAN_BITFIELD) +__u8 ihl:4 , + version:4 ; +#elif defined (__BIG_ENDIAN_BITFIELD) +__u8 version:4 , + ihl:4 ; +#else +#error "Please fix <asm/byteorder.h>" +#endif +__u8 tos; +__be16 tot_len; +__be16 id; +__be16 frag_off; +__u8 ttl; +__u8 protocol; +__sum16 check; +__be32 saddr; +__be32 daddr; + } __packed ip; /* unaligned struct iphdr */struct udphdr udp;__be16 iteration;char msg[64 ];drivers/net/ethernet/sfc/siena/selftest.c +20
-1@@ -43,7 +43,26 @@*/ struct efx_loopback_payload {struct ethhdr header; -struct iphdr ip; +struct { +#if defined(__LITTLE_ENDIAN_BITFIELD) +__u8 ihl:4 , + version:4 ; +#elif defined (__BIG_ENDIAN_BITFIELD) +__u8 version:4 , + ihl:4 ; +#else +#error "Please fix <asm/byteorder.h>" +#endif +__u8 tos; +__be16 tot_len; +__be16 id; +__be16 frag_off; +__u8 ttl; +__u8 protocol; +__sum16 check; +__be32 saddr; +__be32 daddr; + } __packed ip; /* unaligned struct iphdr */struct udphdr udp;__be16 iteration;char msg[64 ];
Re: [PATCH 3/3] sfc: selftest: fix struct packing
Quoted 129 lines · Edward Cree
Quoted 6 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
Three of the sfc drivers define a packed
loopback_payloadstructure with an ethernet header followed by an IP header. However, the kernel definition of iphdr specifies that this is 4-byte aligned, causing a W=1 warning:Quoted 7 lines
net/ethernet/sfc/siena/selftest.c:46:15: error: field ip within 'structefx_loopback_payload' is less aligned than 'struct iphdr' and is usually due to 'structefx_loopback_payload' being packed, which can lead to unaligned accesses [-Werror,-Wunaligned-access]struct iphdr ip;As the iphdr packing is not easily changed without breaking other code, change the three structures to use a local definition instead.
Signed-off-by Arnd Bergmann <[email protected]>Duplicating the definition isn't the prettiest thing in the world; it'd
do for a quick fix if needed but I assume W=1 warnings aren't blocking anyone, so maybe defer this one for now and I'll follow up soon with a rewrite that fixes this more cleanly? My idea is to drop the__packedfrom the containing struct, makeefx_begin_loopback()copy the layers separately, andefx_loopback_rx_packet()similarly do something less direct than casting the packet data to the struct.But I don't insist on it; if you want this fix in immediately then I'm
okay with that too.Quoted 103 lines
--- drivers/net/ethernet/sfc/falcon/selftest.c | 21 ++++++++++++++++++++- drivers/net/ethernet/sfc/selftest.c | 21 ++++++++++++++++++++- drivers/net/ethernet/sfc/siena/selftest.c | 21 ++++++++++++++++++++- 3 files changed, 60 insertions(+), 3 deletions(-)drivers/net/ethernet/sfc/falcon/selftest.c +20
-1@@ -40,7 +40,26 @@*/ struct ef4_loopback_payload {struct ethhdr header; -struct iphdr ip; +struct { +#if defined(__LITTLE_ENDIAN_BITFIELD) +__u8 ihl:4 , + version:4 ; +#elif defined (__BIG_ENDIAN_BITFIELD) +__u8 version:4 , + ihl:4 ; +#else +#error "Please fix <asm/byteorder.h>" +#endif +__u8 tos; +__be16 tot_len; +__be16 id; +__be16 frag_off; +__u8 ttl; +__u8 protocol; +__sum16 check; +__be32 saddr; +__be32 daddr; + } __packed ip; /* unaligned struct iphdr */struct udphdr udp;__be16 iteration;char msg[64 ];drivers/net/ethernet/sfc/selftest.c +20
-1@@ -43,7 +43,26 @@*/ struct efx_loopback_payload {struct ethhdr header; -struct iphdr ip; +struct { +#if defined(__LITTLE_ENDIAN_BITFIELD) +__u8 ihl:4 , + version:4 ; +#elif defined (__BIG_ENDIAN_BITFIELD) +__u8 version:4 , + ihl:4 ; +#else +#error "Please fix <asm/byteorder.h>" +#endif +__u8 tos; +__be16 tot_len; +__be16 id; +__be16 frag_off; +__u8 ttl; +__u8 protocol; +__sum16 check; +__be32 saddr; +__be32 daddr; + } __packed ip; /* unaligned struct iphdr */struct udphdr udp;__be16 iteration;char msg[64 ];drivers/net/ethernet/sfc/siena/selftest.c +20
-1@@ -43,7 +43,26 @@*/ struct efx_loopback_payload {struct ethhdr header; -struct iphdr ip; +struct { +#if defined(__LITTLE_ENDIAN_BITFIELD) +__u8 ihl:4 , + version:4 ; +#elif defined (__BIG_ENDIAN_BITFIELD) +__u8 version:4 , + ihl:4 ; +#else +#error "Please fix <asm/byteorder.h>" +#endif +__u8 tos; +__be16 tot_len; +__be16 id; +__be16 frag_off; +__u8 ttl; +__u8 protocol; +__sum16 check; +__be32 saddr; +__be32 daddr; + } __packed ip; /* unaligned struct iphdr */struct udphdr udp;__be16 iteration;char msg[64 ];
Re: [PATCH 3/3] sfc: selftest: fix struct packing
In reply to Edward CreeQuoted 23 lines · Edward Cree
Quoted 6 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
Three of the sfc drivers define a packed
loopback_payloadstructure with an ethernet header followed by an IP header. However, the kernel definition of iphdr specifies that this is 4-byte aligned, causing a W=1 warning:Quoted 7 lines
net/ethernet/sfc/siena/selftest.c:46:15: error: field ip within 'structefx_loopback_payload' is less aligned than 'struct iphdr' and is usually due to 'structefx_loopback_payload' being packed, which can lead to unaligned accesses [-Werror,-Wunaligned-access]struct iphdr ip;As the iphdr packing is not easily changed without breaking other code, change the three structures to use a local definition instead.
Signed-off-by Arnd Bergmann <[email protected]>Duplicating the definition isn't the prettiest thing in the world; it'd
do for a quick fix if needed but I assume W=1 warnings aren't blocking anyone, so maybe defer this one for now and I'll follow up soon with a rewrite that fixes this more cleanly? My idea is to drop the__packedfrom the containing struct, makeefx_begin_loopback()copy the layers separately, andefx_loopback_rx_packet()similarly do something less direct than casting the packet data to the struct.
Quoted 2 lines
But I don't insist on it; if you want this fix in immediately then I'm
okay with that too.
It's not urgent, if you can come up with a nicer solution, that is probably better than applying my patch now. I have a patch [1] that addresses -Wunaligned-access for all x86 and arm randconfig builds, and this currently affects 23 drivers. Most of the changes don't have changelog texts yet, and some need a more detailed analysis to come up with a correct patch. I'd probably aim for linux-6.6 or later to get them all done, at which point we could move the warning from W=1 to the default set.
Arnd
This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Quoted 10 lines · On Mon, 19 Jun 2023 11:12:09 +0200 you wrote:
From: Arnd Bergmann <[email protected]>
The driver now fails to link when
CONFIG_INETis disabled, so add an explicitKconfigdependency:ld.lld: error: undefined symbol:
ip_route_output_flowQuoted 2 lines
Quoted 2 lines
Quoted 2 lines
referenced by
tc_encap_actions.c
drivers/net/ethernet/sfc/tc_encap_actions.o:(efx_tc_flower_create_encap_md) in archive vmlinux.a[...]
Here is the summary with links:
- [1/3,v2] sfc: add CONFIG_INET dependency for TC offload
https://git.kernel.org/netdev/net-next/c/40cba83370c2 - [2/3,v2] sfc: fix uninitialized variable use https://git.kernel.org/netdev/net-next/c/f61d2d5cf142
- [3/3] sfc: selftest: fix struct packing
(no matching commit)
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html
RE: [PATCH 3/3] sfc: selftest: fix struct packing
In reply to Edward Cree...
Quoted 7 lines
Duplicating the definition isn't the prettiest thing in the world; it'd
do for a quick fix if needed but I assume W=1 warnings aren't blocking anyone, so maybe defer this one for now and I'll follow up soon with a rewrite that fixes this more cleanly? My idea is to drop the__packedfrom the containing struct, makeefx_begin_loopback()copy the layers separately, andefx_loopback_rx_packet()similarly do something less direct than casting the packet data to the struct.
Maybe you can get away with adding a 16bit pad before the ethernet header so that the IP header is actually aligned.
(Then fight all the stuff that stops you doing a memcpy() that runs into a second field of a structure.)
Failing that maybe a single shared copy of the misaligned IP header.
I also suspect you could just add __packed to the two 32bit address fields.
That would generate better code on systems that care.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
Edward Cree
maintainerRe: [PATCH 3/3] sfc: selftest: fix struct packing
Quoted 2 lines · David Laight
Maybe you can get away with adding a 16bit pad before the ethernet header so that the IP header is actually aligned.
That's what I ended up doing, because my original idea was
overcomplicated and turned out super ugly.
See https://lore.kernel.org/netdev/[email protected]/T/
Quoted 2 lines
(Then fight all the stuff that stops you doing a
memcpy()that runs into a second field of a structure.)
Yeah, I don't know how you're meant to annotate that stuff. I guess I'll have to wait until Kees shouts at me and tells
me what to do :S
-ed