netdev - Re: [PATCH v5 13/15] octeontx2-af: Add support for CGX link management

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAK8P3a2zm-kxXqC-fN9MPYH1aiWiUtt32m=xYrDTfse=0+aPgw@mail.gmail.com>
Date:   Mon, 1 Oct 2018 14:46:49 +0200
From:   Arnd Bergmann <arnd@...db.de>
To:     Sunil Kovvuri <sunil.kovvuri@...il.com>
Cc:     Networking <netdev@...r.kernel.org>,
        David Miller <davem@...emloft.net>, linux-soc@...r.kernel.org,
        lcherian@...vell.com, nmani@...vell.com, sgoutham@...vell.com
Subject: Re: [PATCH v5 13/15] octeontx2-af: Add support for CGX link management

On Mon, Oct 1, 2018 at 1:37 PM <sunil.kovvuri@...il.com> wrote:

> + * Not applicable for commands :
> + * CGX_CMD_LINK_BRING_UP/DOWN/CGX_EVT_LINK_CHANGE
> + * check struct cgx_lnk_sts comments
> + */
> +struct cgx_err_sts_s {
> +#if defined(__BIG_ENDIAN_BITFIELD)
> +       uint64_t reserved2:35;
> +       uint64_t type:10;
> +       uint64_t reserved1:9;
> +#else
> +       uint64_t reserved1:9;
> +       uint64_t type:10;               /* cgx_error_type */
> +       uint64_t reserved2:35;
> +#endif
> +};
> +
> +/* Response to cmd ID as CGX_CMD_GET_FW_VER with cmd status as
> + * CGX_STAT_SUCCESS
> + */
> +struct cgx_ver_s {
> +#if defined(__BIG_ENDIAN_BITFIELD)
> +       uint64_t reserved2:47;
> +       uint64_t minor_ver:4;
> +       uint64_t major_ver:4;
> +       uint64_t reserved1:9;
> +#else
> +       uint64_t reserved1:9;
> +       uint64_t major_ver:4;
> +       uint64_t minor_ver:4;
> +       uint64_t reserved2:47;
> +#endif
> +};

This is not exactly what I meant in my comment. I think this may work
correctly on all architectures that we support in Linux, but my point
was that the layout of bit fields within a structure is completely up
to the implementation, so it's really better to just avoid it entirely.

Note that the data is being read with a call like

ereg.val = cgx_read(cgx, lmac->lmac_id,  CGX_EVENT_REG);

which in turn is implemented using readq(), and that already
performs a byte swap in order to do the right thing on big-endian
architectures. The non-reserved bitfields in the two structures
above span byte boundaries, so I can't even wrap my head around
where the respective bits are supposed to be for all combinations
of __BIG_ENDIAN/__LITTLE_ENDIAN and __BIG_ENDIAN_BITFIELD/
__LITTLE_ENDIAN_BITFIELD.

      Arnd