On Thu, 22 Jan 2026 09:51:02 +0100, Aleksandr Loktionov wrote:
> Add a u32 buffer array member to union ixgbe_hic_hdr2 and use it
> directly instead of casting the union address to u32 pointer. This
> avoids potential strict aliasing violations and makes the code more
> explicit about the buffer usage.
>
> The ixgbe_host_interface_command function expects a void* buffer, so
> providing a proper u32 array member in the union is the correct
> approach rather than relying on pointer casting. This eliminates the
> type punning issue where we were casting the union pointer to u32*.
>
> By using buffer.buf instead of &buffer, we pass the address of the
> u32 array directly, which is semantically correct and avoids any
> potential undefined behavior from strict aliasing rule violations.
This commit message is unnecessarily verbose, looks like AI-generated.
The kernel is built with -fno-strict-aliasing, so it's okay to not
follow the rule.
What you're fixing is likely an alignment issue. (see below)
>
> Fixes: 49425dfc7451 ("ixgbe: Add support for x550em_a 10G MAC type")
> Signed-off-by: Aleksandr Loktionov <[email protected]>
> Reviewed-by: Jedrzej Jagielski <[email protected]>
> ---
> drivers/net/ethernet/intel/ixgbe/ixgbe_type.h | 1 +
> drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c | 2 +-
> 2 files changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_type.h
> b/drivers/net/ethernet/intel/ixgbe/ixgbe_type.h
> index 61f2ef6..eb5bf3b 100644
> --- a/drivers/net/ethernet/intel/ixgbe/ixgbe_type.h
> +++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_type.h
> @@ -2798,6 +2798,7 @@ struct ixgbe_hic_hdr2_rsp {
> };
>
> union ixgbe_hic_hdr2 {
> + u32 buf[1];
The alignment of this union was 1 byte. By adding a u32 member, you're
effectively making it align to u32 (4 bytes).
> struct ixgbe_hic_hdr2_req req;
> struct ixgbe_hic_hdr2_rsp rsp;
> };
> diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
> b/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
> index 76d2fa3..4a0ccbf 100644
> --- a/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
> +++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_x550.c
> @@ -1228,7 +1228,7 @@ static int ixgbe_update_flash_X550(struct ixgbe_hw *hw)
> buffer.req.buf_lenl = FW_SHADOW_RAM_DUMP_LEN;
> buffer.req.checksum = FW_DEFAULT_CHECKSUM;
>
> - status = ixgbe_host_interface_command(hw, &buffer, sizeof(buffer),
> + status = ixgbe_host_interface_command(hw, buffer.buf, sizeof(buffer),
> IXGBE_HI_COMMAND_TIMEOUT, false);
`buffer` is a local variable allocated on stack, and the compiler did
not guarantee its alignment. As ixgbe_host_interface_command() casts
`buffer` to a u32 array, this may cause an unaligned-access exception
on some arch.
For your reference, I addressed a similar issue previously:
https://lore.kernel.org/all/[email protected]/
Please update your message, and try not to use completely-AIGC phrases.
> return status;
> }
> --
> 2.52.0
>
>
Regards,
Qingfang
