[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <3785f913-cffe-4b12-8bed-0586fac16393@intel.com>
Date: Thu, 9 Jan 2025 10:31:35 +0100
From: Przemek Kitszel <przemyslaw.kitszel@...el.com>
To: Hongchen Zhang <zhanghongchen@...ngson.cn>, Michal Schmidt
<mschmidt@...hat.com>, Dave Ertman <david.m.ertman@...el.com>
CC: <netdev@...r.kernel.org>, Jesse Brandeburg <jesse.brandeburg@...el.com>,
Tony Nguyen <anthony.l.nguyen@...el.com>, Daniel Machon
<daniel.machon@...rochip.com>, <intel-wired-lan@...ts.osuosl.org>
Subject: Re: [PATCH net] ice: fix unaligned access in ice_create_lag_recipe
On 1/9/25 02:54, Hongchen Zhang wrote:
> Hi Przemek,
> On 2025/1/8 下午4:59, Przemek Kitszel wrote:
>> On 1/8/25 04:09, Hongchen Zhang wrote:
>
>>> Hi Michal,
>>> On 2024/1/31 pm 7:58, Michal Schmidt wrote:
>>>> new_rcp->recipe_bitmap was written to as if it were an aligned bitmap.
>>>> It is an 8-byte array, but aligned only to 4.
>>>> Use put_unaligned to set its value.
>>>>
>>>> Additionally, values in ice commands are typically in little-endian.
>>>> I assume the recipe bitmap should be too, so use the *_le64 conversion.
>>>> I don't have a big-endian system with ice to test this.
>>>>
>>>> I tested that the driver does not crash when probing on aarch64
>>>> anymore,
>>>> which is good enough for me. I don't know if the LAG feature actually
>>>> works.
>>>>
>>>> This is what the crash looked like without the fix:
>>
>>>> [ 17.599142] Call trace:
>>>> [ 17.599143] ice_create_lag_recipe.constprop.0+0xbc/0x11c [ice]
>>>> [ 17.599172] ice_init_lag+0xcc/0x22c [ice]
>>>> [ 17.599201] ice_init_features+0x160/0x2b4 [ice]
>>>> [ 17.599230] ice_probe+0x2d0/0x30c [ice]
>>>> [ 17.599258] local_pci_probe+0x58/0xb0
>>>> [ 17.599262] work_for_cpu_fn+0x20/0x30
>>
>>> I encountered the same problem on a LoongArch LS3C6000 machine. Can
>>> this patch be merged now?
>>
>> What kernel base do you use?, we have merged the Steven Patches long ago
> My test is based on 6.6.61 which contains Steven's patch:
> 8ec08ba97fab 2024-05-07 ice: Refactor FW data type and fix bitmap
> casting issue [Steven Zou]
>
> It seems that Steven's patch can not solve the unaligned access problem
> caused by new_rcp->recipe_bitmap, So is Michal's patch (may need some
> change in ice_add_sw_recipe()) still needed?
>
thank you, I see now
I agree that ice_aqc_recipe_data_elem::recipe_bitmap[8] should be
changed to __le64, together with updated accesses. Best way to do so
will be as in Steven's patch.
@Michal, will you be OK with us reimplementing this, or you want to
follow up?
Powered by blists - more mailing lists