diff mbox series

[net-next,1/3] sfc: use padding to fix alignment in loopback test

Message ID 441f4c197394bdeddb47aefa0bc854ee1960df27.1687427930.git.ecree.xilinx@gmail.com (mailing list archive)
State Superseded
Delegated to: Netdev Maintainers
Headers show
Series sfc: fix unaligned access in loopback selftests | expand

Checks

Context Check Description
netdev/series_format success Posting correctly formatted
netdev/tree_selection success Clearly marked for net-next
netdev/fixes_present success Fixes tag not required for -next series
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 8 this patch: 8
netdev/cc_maintainers success CCed 8 of 8 maintainers
netdev/build_clang success Errors and warnings before: 8 this patch: 8
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/deprecated_api success None detected
netdev/check_selftest success No net selftest shell script
netdev/verify_fixes success No Fixes tag
netdev/build_allmodconfig_warn success Errors and warnings before: 8 this patch: 8
netdev/checkpatch fail ERROR: code indent should use tabs where possible WARNING: please, no space before tabs WARNING: quoted string split across lines
netdev/kdoc success Errors and warnings before: 0 this patch: 0
netdev/source_inline success Was 0 now: 0

Commit Message

edward.cree@amd.com June 22, 2023, 10:18 a.m. UTC 
From: Edward Cree <ecree.xilinx@gmail.com>

Add two bytes of padding to the start of struct efx_loopback_payload,
 which are not sent on the wire.  This ensures the 'ip' member is
 4-byte aligned, preventing the following W=1 warning:
net/ethernet/sfc/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;

Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
---
 drivers/net/ethernet/sfc/selftest.c | 45 +++++++++++++++++------------
 1 file changed, 27 insertions(+), 18 deletions(-)

Comments

Arnd Bergmann June 22, 2023, 11:38 a.m. UTC | #1 
On Thu, Jun 22, 2023, at 12:18, edward.cree@amd.com wrote:
>  drivers/net/ethernet/sfc/selftest.c | 45 +++++++++++++++++------------
>  1 file changed, 27 insertions(+), 18 deletions(-)
>
> diff --git a/drivers/net/ethernet/sfc/selftest.c 
> b/drivers/net/ethernet/sfc/selftest.c
> index 3c5227afd497..74b42efe7267 100644
> --- a/drivers/net/ethernet/sfc/selftest.c
> +++ b/drivers/net/ethernet/sfc/selftest.c
> @@ -42,12 +42,16 @@
>   * Falcon only performs RSS on TCP/UDP packets.
>   */
>  struct efx_loopback_payload {
> +	char pad[2]; /* Ensures ip is 4-byte aligned */
>  	struct ethhdr header;
>  	struct iphdr ip;
>  	struct udphdr udp;
>  	__be16 iteration;
>  	char msg[64];
>  } __packed;

There should probably be an extra __aligned(4) after the __packed,
so that the compiler knows the start of the structure is aligned,
otherwise (depending on the architecture and compiler), any access
to a member can still turn into multiple single-byte accesses.

I would also expect to still see a warning without that extra
attribute. Aside from this, the patch looks good to me.

     Arnd
Edward Cree June 22, 2023, 12:02 p.m. UTC | #2 
On 22/06/2023 12:38, Arnd Bergmann wrote:
> On Thu, Jun 22, 2023, at 12:18, edward.cree@amd.com wrote:
>>  drivers/net/ethernet/sfc/selftest.c | 45 +++++++++++++++++------------
>>  1 file changed, 27 insertions(+), 18 deletions(-)
>>
>> diff --git a/drivers/net/ethernet/sfc/selftest.c 
>> b/drivers/net/ethernet/sfc/selftest.c
>> index 3c5227afd497..74b42efe7267 100644
>> --- a/drivers/net/ethernet/sfc/selftest.c
>> +++ b/drivers/net/ethernet/sfc/selftest.c
>> @@ -42,12 +42,16 @@
>>   * Falcon only performs RSS on TCP/UDP packets.
>>   */
>>  struct efx_loopback_payload {
>> +	char pad[2]; /* Ensures ip is 4-byte aligned */
>>  	struct ethhdr header;
>>  	struct iphdr ip;
>>  	struct udphdr udp;
>>  	__be16 iteration;
>>  	char msg[64];
>>  } __packed;
> 
> There should probably be an extra __aligned(4) after the __packed,
> so that the compiler knows the start of the structure is aligned,
> otherwise (depending on the architecture and compiler), any access
> to a member can still turn into multiple single-byte accesses.

Ok, will add that in v2.

> I would also expect to still see a warning without that extra
> attribute. Aside from this, the patch looks good to me.
My compiler (gcc 8.3.1) doesn't reproduce the original warning, so
 I'm flying slightly blind here :(  If you could build-test this on
 your setup with/without the __aligned(4), I'd be very grateful.

-ed
Arnd Bergmann June 22, 2023, 12:19 p.m. UTC | #3 
On Thu, Jun 22, 2023, at 14:02, Edward Cree wrote:
> On 22/06/2023 12:38, Arnd Bergmann wrote:
>> On Thu, Jun 22, 2023, at 12:18, edward.cree@amd.com wrote:
>> 
>> There should probably be an extra __aligned(4) after the __packed,
>> so that the compiler knows the start of the structure is aligned,
>> otherwise (depending on the architecture and compiler), any access
>> to a member can still turn into multiple single-byte accesses.
>
> Ok, will add that in v2.
>
>> I would also expect to still see a warning without that extra
>> attribute. Aside from this, the patch looks good to me.
> My compiler (gcc 8.3.1) doesn't reproduce the original warning, so
>  I'm flying slightly blind here :(  If you could build-test this on
>  your setup with/without the __aligned(4), I'd be very grateful.

I can confirm that your patch addresses the warning for me
with any version of clang I have installed, and that the __aligned(4)
is not needed for the warning, though it is still a good idea for
code generation.

     Arnd
diff mbox series

Patch

diff --git a/drivers/net/ethernet/sfc/selftest.c b/drivers/net/ethernet/sfc/selftest.c
index 3c5227afd497..74b42efe7267 100644
--- a/drivers/net/ethernet/sfc/selftest.c
+++ b/drivers/net/ethernet/sfc/selftest.c
@@ -42,12 +42,16 @@ 
  * Falcon only performs RSS on TCP/UDP packets.
  */
 struct efx_loopback_payload {
+	char pad[2]; /* Ensures ip is 4-byte aligned */
 	struct ethhdr header;
 	struct iphdr ip;
 	struct udphdr udp;
 	__be16 iteration;
 	char msg[64];
 } __packed;
+#define EFX_LOOPBACK_PAYLOAD_LEN	(sizeof(struct efx_loopback_payload) - \
+					 offsetof(struct efx_loopback_payload, \
+					 	  header))
 
 /* Loopback test source MAC address */
 static const u8 payload_source[ETH_ALEN] __aligned(2) = {
@@ -282,7 +286,7 @@  void efx_loopback_rx_packet(struct efx_nic *efx,
 			    const char *buf_ptr, int pkt_len)
 {
 	struct efx_loopback_state *state = efx->loopback_selftest;
-	struct efx_loopback_payload *received;
+	struct efx_loopback_payload received;
 	struct efx_loopback_payload *payload;
 
 	BUG_ON(!buf_ptr);
@@ -293,13 +297,14 @@  void efx_loopback_rx_packet(struct efx_nic *efx,
 
 	payload = &state->payload;
 
-	received = (struct efx_loopback_payload *) buf_ptr;
-	received->ip.saddr = payload->ip.saddr;
+	memcpy(&received.header, buf_ptr,
+	       min_t(int, pkt_len, EFX_LOOPBACK_PAYLOAD_LEN));
+	received.ip.saddr = payload->ip.saddr;
 	if (state->offload_csum)
-		received->ip.check = payload->ip.check;
+		received.ip.check = payload->ip.check;
 
 	/* Check that header exists */
-	if (pkt_len < sizeof(received->header)) {
+	if (pkt_len < sizeof(received.header)) {
 		netif_err(efx, drv, efx->net_dev,
 			  "saw runt RX packet (length %d) in %s loopback "
 			  "test\n", pkt_len, LOOPBACK_MODE(efx));
@@ -307,7 +312,7 @@  void efx_loopback_rx_packet(struct efx_nic *efx,
 	}
 
 	/* Check that the ethernet header exists */
-	if (memcmp(&received->header, &payload->header, ETH_HLEN) != 0) {
+	if (memcmp(&received.header, &payload->header, ETH_HLEN) != 0) {
 		netif_err(efx, drv, efx->net_dev,
 			  "saw non-loopback RX packet in %s loopback test\n",
 			  LOOPBACK_MODE(efx));
@@ -315,16 +320,16 @@  void efx_loopback_rx_packet(struct efx_nic *efx,
 	}
 
 	/* Check packet length */
-	if (pkt_len != sizeof(*payload)) {
+	if (pkt_len != EFX_LOOPBACK_PAYLOAD_LEN) {
 		netif_err(efx, drv, efx->net_dev,
 			  "saw incorrect RX packet length %d (wanted %d) in "
-			  "%s loopback test\n", pkt_len, (int)sizeof(*payload),
-			  LOOPBACK_MODE(efx));
+			  "%s loopback test\n", pkt_len,
+			  (int)EFX_LOOPBACK_PAYLOAD_LEN, LOOPBACK_MODE(efx));
 		goto err;
 	}
 
 	/* Check that IP header matches */
-	if (memcmp(&received->ip, &payload->ip, sizeof(payload->ip)) != 0) {
+	if (memcmp(&received.ip, &payload->ip, sizeof(payload->ip)) != 0) {
 		netif_err(efx, drv, efx->net_dev,
 			  "saw corrupted IP header in %s loopback test\n",
 			  LOOPBACK_MODE(efx));
@@ -332,7 +337,7 @@  void efx_loopback_rx_packet(struct efx_nic *efx,
 	}
 
 	/* Check that msg and padding matches */
-	if (memcmp(&received->msg, &payload->msg, sizeof(received->msg)) != 0) {
+	if (memcmp(&received.msg, &payload->msg, sizeof(received.msg)) != 0) {
 		netif_err(efx, drv, efx->net_dev,
 			  "saw corrupted RX packet in %s loopback test\n",
 			  LOOPBACK_MODE(efx));
@@ -340,10 +345,10 @@  void efx_loopback_rx_packet(struct efx_nic *efx,
 	}
 
 	/* Check that iteration matches */
-	if (received->iteration != payload->iteration) {
+	if (received.iteration != payload->iteration) {
 		netif_err(efx, drv, efx->net_dev,
 			  "saw RX packet from iteration %d (wanted %d) in "
-			  "%s loopback test\n", ntohs(received->iteration),
+			  "%s loopback test\n", ntohs(received.iteration),
 			  ntohs(payload->iteration), LOOPBACK_MODE(efx));
 		goto err;
 	}
@@ -363,7 +368,8 @@  void efx_loopback_rx_packet(struct efx_nic *efx,
 			       buf_ptr, pkt_len, 0);
 		netif_err(efx, drv, efx->net_dev, "expected packet:\n");
 		print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 0x10, 1,
-			       &state->payload, sizeof(state->payload), 0);
+			       &state->payload.header, EFX_LOOPBACK_PAYLOAD_LEN,
+			       0);
 	}
 #endif
 	atomic_inc(&state->rx_bad);
@@ -385,14 +391,15 @@  static void efx_iterate_state(struct efx_nic *efx)
 	payload->ip.daddr = htonl(INADDR_LOOPBACK);
 	payload->ip.ihl = 5;
 	payload->ip.check = (__force __sum16) htons(0xdead);
-	payload->ip.tot_len = htons(sizeof(*payload) - sizeof(struct ethhdr));
+	payload->ip.tot_len = htons(sizeof(*payload) -
+				    offsetof(struct efx_loopback_payload, ip));
 	payload->ip.version = IPVERSION;
 	payload->ip.protocol = IPPROTO_UDP;
 
 	/* Initialise udp header */
 	payload->udp.source = 0;
-	payload->udp.len = htons(sizeof(*payload) - sizeof(struct ethhdr) -
-				 sizeof(struct iphdr));
+	payload->udp.len = htons(sizeof(*payload) -
+				 offsetof(struct efx_loopback_payload, udp));
 	payload->udp.check = 0;	/* checksum ignored */
 
 	/* Fill out payload */
@@ -418,7 +425,7 @@  static int efx_begin_loopback(struct efx_tx_queue *tx_queue)
 	for (i = 0; i < state->packet_count; i++) {
 		/* Allocate an skb, holding an extra reference for
 		 * transmit completion counting */
-		skb = alloc_skb(sizeof(state->payload), GFP_KERNEL);
+		skb = alloc_skb(EFX_LOOPBACK_PAYLOAD_LEN, GFP_KERNEL);
 		if (!skb)
 			return -ENOMEM;
 		state->skbs[i] = skb;
@@ -429,6 +436,8 @@  static int efx_begin_loopback(struct efx_tx_queue *tx_queue)
 		payload = skb_put(skb, sizeof(state->payload));
 		memcpy(payload, &state->payload, sizeof(state->payload));
 		payload->ip.saddr = htonl(INADDR_LOOPBACK | (i << 2));
+		/* Strip off the leading padding */
+		skb_pull(skb, offsetof(struct efx_loopback_payload, header));
 
 		/* Ensure everything we've written is visible to the
 		 * interrupt handler. */