On 29/05/12 05:32, Ben Hutchings wrote:
> On Mon, 2012-05-28 at 18:31 +0100, Torne (Richard Coles) wrote:
>> From: "Torne (Richard Coles)" <to...@google.com>
>>
>> MMC CSD info can specify very large, ridiculous timeouts, big enough to
>> overflow timeout_ns on 32-bit machines. This can result in the card
>> timing out on every operation because the wrapped timeout value is far
>> too small.
>>
>> Fix the overflow by capping the result at 2 seconds.  Cards specifying
>> longer timeouts are almost certainly insane, and host controllers
>> generally cannot support timeouts that long in any case.
>>
>> 2 seconds should be plenty of time for any card to actually function;
>> the timeout calculation code is already using 1 second as a "worst case"
>> timeout for cards running in SPI mode.
> 
> Needs a 'Signed-off-by'.
> 
>> ---
>>  drivers/mmc/core/core.c |   11 ++++++++++-
>>  1 files changed, 10 insertions(+), 1 deletions(-)
>>
>> diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
>> index 0b6141d..3b4a9fc 100644
>> --- a/drivers/mmc/core/core.c
>> +++ b/drivers/mmc/core/core.c
>> @@ -512,7 +512,16 @@ void mmc_set_data_timeout(struct mmc_data *data, const 
>> struct mmc_card *card)
>>      if (data->flags & MMC_DATA_WRITE)
>>              mult <<= card->csd.r2w_factor;
>>  
>> -    data->timeout_ns = card->csd.tacc_ns * mult;
>> +    /*
>> +     * The timeout in nanoseconds may overflow with some cards. Cap it at
>> +     * two seconds both to avoid the overflow and also because host
>> +     * controllers cannot generally generate timeouts that long anyway.
>> +     */
>> +    if (card->csd.tacc_ns <= (2 * NSEC_PER_SEC) / mult)
>> +            data->timeout_ns = card->csd.tacc_ns * mult;
>> +    else
>> +            data->timeout_ns = 2 * NSEC_PER_SEC;
> 
> We clearly need to guard against overflow here, and this is the correct
> way to clamp the multiplication.  I can't speak as to whether 2 seconds
> is the right limit.

The host controllers I have looked at have a limit of around 2.5 seconds.

But why not just use the size of the type as the limit? e.g.

        if (card->csd.tacc_ns <= UINT_MAX / mult)
                data->timeout_ns = card->csd.tacc_ns * mult;
        else
                data->timeout_ns = UINT_MAX;

> 
> Ben.
> 
>>      data->timeout_clks = card->csd.tacc_clks * mult;
>>  
>>      /*
> 

--
To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to