Re: [PATCH] gpio: davinci: Fix potential buffer overflow

[Alexander Lobakin <aleksander.lobakin@intel.com>]

From: Aleksandr Mishin <amishin@t-argos.ru>
Date: Thu, 28 Mar 2024 19:23:56 +0300

> 
> 
> 28.03.2024 18:27, Alexander Lobakin пишет:
>> From: Aleksandr Mishin <amishin@t-argos.ru>
>> Date: Thu, 28 Mar 2024 12:10:21 +0300
>>
>>> In davinci_gpio_probe() accessing an element of array 'chips->regs'
>>> of size 5 and
>>> array 'offset_array' of size 5 can lead to a buffer overflow, since
>>> the index
>>> 'bank' can have an out of range value 63.
>>> Fix this bug by limiting top index value.
>>>
>>> Found by Linux Verification Center (linuxtesting.org) with SVACE.
>>>
>>> Fixes: c809e37a3b5a ("gpio: davinci: Allocate the correct amount of
>>> memory for controller")
>>> Signed-off-by: Aleksandr Mishin <amishin@t-argos.ru>
>>> ---
>>>   drivers/gpio/gpio-davinci.c | 3 +++
>>>   1 file changed, 3 insertions(+)
>>>
>>> diff --git a/drivers/gpio/gpio-davinci.c b/drivers/gpio/gpio-davinci.c
>>> index bb499e362912..b65df1f2b83f 100644
>>> --- a/drivers/gpio/gpio-davinci.c
>>> +++ b/drivers/gpio/gpio-davinci.c
>>> @@ -257,6 +257,9 @@ static int davinci_gpio_probe(struct
>>> platform_device *pdev)
>>>       spin_lock_init(&chips->lock);
>>>         nbank = DIV_ROUND_UP(ngpio, 32);
>>> +    if (nbank > MAX_REGS_BANKS || nbank > 5) {
>>> +        nbank = MAX_REGS_BANKS < 5 ? MAX_REGS_BANKS : 5;
>>> +    }
>>
>> Static analysis warnings make no sense until you provide a reliable way
>> to trigger the problem on real systems.
>>
>>>       for (bank = 0; bank < nbank; bank++)
>>>           chips->regs[bank] = gpio_base + offset_array[bank];
>>>   
>>
>> Thanks,
>> Olek
>>
> 
> I can only see the code at this time. And I see the following:
> 1. In some configurations CONFIG_ARCH_NR_GPIO value is 2048. So nbank
> value can be 64.
> 2. Previously, a patch was proposed that removes restrictions on the
> number of GPIOs
> (https://lore.kernel.org/all/cb540a9d73cad36d288664f8b275c6308a4a3168.1662116601.git.christophe.leroy@csgroup.eu/).
> 

But no real hardware / device tree which declares such huge number of
GPIOs, right?

CONFIG_ARCH_NR_GPIO is architecture-specific. Davinci is a platform, not
an architecture. If no Davinci board can have a number that would
overflow, the fix doesn't make sense.

Thanks,
Olek