Re: [PATCH] [-next] net: marvell: prestera: Add explicit padding
From: Volodymyr Mytnyk [C] <hidden>
Date: 2021-11-02 11:36:39
Also in:
lkml
Hi All,
For some unknown reason, the bb5dbf2cc64d5cfa ("net: marvell: prestera: add firmware v4.0 support") changes have been merged into net-next w/o review comments addressed and waiting for the final patch set to be uploaded. Any idea why ?
Right now, I'm working on fixing all the issues/comments and rebasing them based on latest net-next master. Also, my changes include those posted in this thread to fix m68k build and comments related to structure pack/align.
Should I rebase my changes based on yours now ? Is it possible to make a relation chain ?
The bb5dbf2cc64d5cfa mail thread discussion (waiting for new v5 patchset to be uploaded) can be found at:
[PATCH net-next v4] net: marvell: prestera: add firmware v4.0 support
https://www.spinics.net/lists/kernel/msg4127689.html
Regards,
Volodymyr
On Tue, Nov 2, 2021 at 9:24 AM Geert Uytterhoeven [off-list ref] wrote:quoted
On m68k: In function ‘prestera_hw_build_tests’, inlined from ‘prestera_hw_switch_init’ at drivers/net/ethernet/marvell/prestera/prestera_hw.c:788:2: ././include/linux/compiler_types.h:335:38: error: call to ‘__compiletime_assert_345’ declared with attribute error: BUILD_BUG_ON failed: sizeof(struct prestera_msg_switch_attr_req) != 16 ... The driver assumes structure members are naturally aligned, but does not add explicit padding, thus breaking architectures where integral values are not always naturally aligned (e.g. on m68k, __alignof(int) is 2, not 4). Fixes: bb5dbf2cc64d5cfa ("net: marvell: prestera: add firmware v4.0 support") Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>Looks good to me, Reviewed-by: Arnd Bergmann <arnd@arndb.de>quoted
Compile-tested only. BTW, I sincerely doubt the use of __packed on structs like: union prestera_msg_switch_param { u8 mac[ETH_ALEN]; __le32 ageing_timeout_ms; } __packed; This struct is only used as a member in another struct, where it is be naturally aligned anyway.Agreed, this __packed attribute is clearly bogus and should be removed. Same for +struct prestera_msg_event_port_param { + union { + struct { + u8 oper; + __le32 mode; + __le32 speed; + u8 duplex; + u8 fc; + u8 fec; + } __packed mac; + struct { + u8 mdix; + __le64 lmode_bmap; + u8 fc; + } __packed phy; + } __packed; +} __packed __aligned(4); This makes no sense at all. I would suggest marking only the individual fields that are misaligned as __packed, but not the structure itself. and then there is this + union { + struct { + u8 admin:1; + u8 fc; + u8 ap_enable; + union { + struct { + __le32 mode; + u8 inband:1; + __le32 speed; + u8 duplex; + u8 fec; + u8 fec_supp; + } __packed reg_mode; + struct { + __le32 mode; + __le32 speed; + u8 fec; + u8 fec_supp; + } __packed ap_modes[PRESTERA_AP_PORT_MAX]; + } __packed; + } __packed mac; + struct { + u8 admin:1; + u8 adv_enable; + __le64 modes; + __le32 mode; + u8 mdix; + } __packed phy; + } __packed link; which puts misaligned bit fields in the middle of a packed structure! Arnd