Re: [dpdk-dev] [PATCH v1 6/6] net: fix rte_ether conflicts with libc

[Olivier MATZ <olivier.matz@6wind.com>]

Hi Adrien,

On Thu, Dec 21, 2017 at 02:00:06PM +0100, Adrien Mazarguil wrote:
> Applications can't combine either net/ethernet.h or netinet/ether.h
> together with rte_ether.h due to the redefinition of struct ether_addr and
> various macros by the latter.
> 
> This patch adapts rte_ether.h to rely on system definitions while
> maintaining DPDK additions.
> 
> An unforeseen consequence of involving more system header files compilation
> issues with some base drivers (i40e, ixgbe) defining their own conflicting
> types (e.g. __le64). This is addressed by explicitly including rte_ether.h
> where missing to ensure system definitions always come first.
> 
> Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
> Cc: Olivier Matz <olivier.matz@6wind.com>
> Cc: Bruce Richardson <bruce.richardson@intel.com>

[...]

> --- a/drivers/net/qede/base/bcm_osal.h
> +++ b/drivers/net/qede/base/bcm_osal.h
> @@ -334,7 +334,6 @@ u32 qede_find_first_zero_bit(unsigned long *, u32);
>  	qede_find_first_zero_bit(bitmap, length)
>  
>  #define OSAL_BUILD_BUG_ON(cond)		nothing
> -#define ETH_ALEN			ETHER_ADDR_LEN
>  
>  #define OSAL_BITMAP_WEIGHT(bitmap, count) 0
>  

Not sure we can update code in a 'base' driver as easily.
It should be checked how it can be done with the qede maintainer.


> diff --git a/lib/librte_net/rte_ether.h b/lib/librte_net/rte_ether.h
> index 06d7b486c..19e62ea89 100644
> --- a/lib/librte_net/rte_ether.h
> +++ b/lib/librte_net/rte_ether.h
> @@ -44,6 +44,7 @@
>  extern "C" {
>  #endif
>  
> +#include <net/ethernet.h>
>  #include <stdint.h>
>  #include <stdio.h>
>  
> @@ -52,15 +53,7 @@ extern "C" {
>  #include <rte_mbuf.h>
>  #include <rte_byteorder.h>
>  
> -#define ETHER_ADDR_LEN  6 /**< Length of Ethernet address. */
> -#define ETHER_TYPE_LEN  2 /**< Length of Ethernet type field. */
> -#define ETHER_CRC_LEN   4 /**< Length of Ethernet CRC. */
> -#define ETHER_HDR_LEN   \
> -	(ETHER_ADDR_LEN * 2 + ETHER_TYPE_LEN) /**< Length of Ethernet header. */
> -#define ETHER_MIN_LEN   64    /**< Minimum frame len, including CRC. */
> -#define ETHER_MAX_LEN   1518  /**< Maximum frame len, including CRC. */
> -#define ETHER_MTU       \
> -	(ETHER_MAX_LEN - ETHER_HDR_LEN - ETHER_CRC_LEN) /**< Ethernet MTU. */
> +#define ETHER_MTU ETHERMTU /**< Deprecated, defined for compatibility. */
>  
>  #define ETHER_MAX_VLAN_FRAME_LEN \
>  	(ETHER_MAX_LEN + 4) /**< Maximum VLAN frame length, including CRC. */
> @@ -72,8 +65,11 @@ extern "C" {
>  
>  #define ETHER_MIN_MTU 68 /**< Minimum MTU for IPv4 packets, see RFC 791. */
>  
> +#ifdef __DOXYGEN__
> +
>  /**
> - * Ethernet address:
> + * Ethernet address.
> + *
>   * A universally administered address is uniquely assigned to a device by its
>   * manufacturer. The first three octets (in transmission order) contain the
>   * Organizationally Unique Identifier (OUI). The following three (MAC-48 and
> @@ -82,11 +78,25 @@ extern "C" {
>   * A locally administered address is assigned to a device by a network
>   * administrator and does not contain OUIs.
>   * See http://standards.ieee.org/regauth/groupmac/tutorial.html
> + *
> + * This structure is defined system-wide by "net/ethernet.h", however since
> + * the name of its data field is OS-dependent, a macro named "addr_bytes" is
> + * defined as an alias for the convenience of DPDK applications.
> + *
> + * The following definition is only for documentation purposes.
>   */
>  struct ether_addr {
>  	uint8_t addr_bytes[ETHER_ADDR_LEN]; /**< Addr bytes in tx order */
>  } __attribute__((__packed__));
>  
> +#endif /* __DOXYGEN__ */
> +
> +#if defined(__FreeBSD__)
> +#define addr_bytes octet
> +#else
> +#define addr_bytes ether_addr_octet
> +#endif
> +
>  #define ETHER_LOCAL_ADMIN_ADDR 0x02 /**< Locally assigned Eth. address. */
>  #define ETHER_GROUP_ADDR       0x01 /**< Multicast or broadcast Eth. address. */

This kind of #define looks a bit dangerous to me: it can trigger
strange bugs because it will replace all occurences of addr_bytes
after this header is included.

Wouldn't it be a good opportunity to think about adding the rte_ prefix
to all variables/functions of rte_ether.h?