DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org D7B88224E0
Subject: Re: [REGRESSION] brcmfmac: NULL pointer deference starting
 next-20181107
To:     Jon Hunter <jonathanh@nvidia.com>,
        Arend van Spriel <arend.vanspriel@broadcom.com>,
        Kalle Valo <kvalo@codeaurora.org>,
        linux-tegra <linux-tegra@vger.kernel.org>,
        linux-wireless@vger.kernel.org,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Ard Biesheuvel <ard.biesheuvel@linaro.org>
References: <dbc5e8f9-6f14-77d2-be2a-f0738e13b773@nvidia.com>
 <9f72ac4f-a83a-7af7-3c26-b1ced6d98653@broadcom.com>
 <9e0a7997-4c14-a3a1-c935-d674270533aa@redhat.com>
 <000bfd39-10cf-661c-9512-0410782168dc@nvidia.com>
From:   Hans de Goede <hdegoede@redhat.com>
Message-ID: <b33e91b9-e375-5169-dda6-dc664700eefc@redhat.com>
Date:   Tue, 13 Nov 2018 14:32:54 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.3.0
MIME-Version: 1.0
In-Reply-To: <000bfd39-10cf-661c-9512-0410782168dc@nvidia.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Sender: linux-wireless-owner@vger.kernel.org
Precedence: bulk

HI,

On 13-11-18 14:21, Jon Hunter wrote:
> 
> On 13/11/2018 10:40, Hans de Goede wrote:
>> Hi,
>>
>> On 13-11-18 11:24, Arend van Spriel wrote:
>>> + Ard as this involves EFI.
>>>
>>> On 11/12/2018 2:24 PM, Jon Hunter wrote:
>>>> Hi Hans, Kalle,
>>>>
>>>> Starting with next-20181107 I am seeing the following NULL pointer
>>>> deference on Tegra (note the firmware is missing on this board) ...
>>>>
>>>> [   14.072883] brcmfmac: brcmf_fw_alloc_request: using
>>>> brcm/brcmfmac4329-sdio for chip BCM4329/3
>>>>
>>>> [   14.130287] brcmfmac mmc1:0001:1: Direct firmware load for
>>>> brcm/brcmfmac4329-sdio.nvidia,cardhu-a04.txt failed with error -2
>>>>
>>>> [   14.156283] brcmfmac mmc1:0001:1: Direct firmware load for
>>>> brcm/brcmfmac4329-sdio.txt failed with error -2
>>>>
>>>> [   14.177769] Unable to handle kernel NULL pointer dereference at
>>>> virtual address 00000008
>>>>
>>>> [   14.197303] pgd = 60bfa5f1
>>>>
>>>> [   14.211842] [00000008] *pgd=00000000
>>>>
>>>> [   14.227373] Internal error: Oops: 5 [#1] SMP ARM
>>>>
>>>> [   14.244244] Modules linked in: brcmfmac sha256_generic sha256_arm
>>>> snd cfg80211 brcmutil soundcore snd_soc_tegra30_ahub tegra_wdt
>>>>
>>>> [   14.269109] CPU: 1 PID: 114 Comm: kworker/1:2 Not tainted
>>>> 4.20.0-rc1-next-20181107-gd881de3 #1
>>>>
>>>> [   14.269114] Hardware name: NVIDIA Tegra SoC (Flattened Device Tree)
>>>>
>>>> [   14.269154] Workqueue: events request_firmware_work_func
>>>>
>>>> [   14.269177] PC is at efivar_entry_size+0x28/0x90
>>>>
>>>> [   14.269362] LR is at brcmf_fw_complete_request+0x3f8/0x8d4 [brcmfmac]
>>>>
>>>> [   14.269369] pc : [<c0c40718>]    lr : [<bf2a3ef4>]    psr: a00d0113
>>>>
>>>> [   14.269374] sp : ede7fe28  ip : ee983410  fp : c1787f30
>>>>
>>>> [   14.269378] r10: 00000000  r9 : 00000000  r8 : bf2b2258
>>>>
>>>> [   14.269384] r7 : ee983000  r6 : c1604c48  r5 : ede7fe88  r4 :
>>>> edf337c0
>>>>
>>>> [   14.269389] r3 : 00000000  r2 : 00000000  r1 : ede7fe88  r0 :
>>>> c17712c8
>>>>
>>>
>>> Hi Jon,
>>>
>>> I tried building drivers/firmware/efi/vars.c using tegra_defconfig.
>>> Had to enable CONFIG_EFI. So the null pointer access is a 0x00000008
>>> so I looked at the disassembly below:
>>>
>>> int efivar_entry_size(struct efivar_entry *entry, unsigned long *size)
>>> {
>>>        310:       e1a05001        mov     r5, r1
>>>           const struct efivar_operations *ops = __efivars->ops;
>>> ==>  314:       e5936008        ldr     r6, [r3, #8]
>>>
>>> So I think __efivars is NULL on your platform. It is private to the
>>> source file. Not sure how the driver should deal with this. Maybe use
>>> efi_enabled() but not sure what feature to use. My best bet would be
>>> EFI_RUNTIME_SERVICES.
>>
>> Ah right, thank you for catching this I had looking into this
>> on my TODO list, but you beat me to it.
>>
>> IMHO the best fix here would be to modify efivar_entry_size(),
>> adding:
>>
>>      if (!ops)
>>          return -ENOENT;
>>
>> Which makes it return the same error as when we do have efivar
>> support but the requested variable is not found.
> 
> So the above did not work. I see a patch from Arend and I will give this
> a try.

Ah right, looking at Arend's patch my little snippet got the test wrong.

Hopefully Arend's patch will fix things.

Regards,

Hans