Re: [PATCH 1/3] iio: buffer: remove usage of list iterator variable for list_for_each_entry_continue_reverse()

[Jakob Koschel]

> On 1. Apr 2022, at 14:40, Sa, Nuno <Nuno.Sa@xxxxxxxxxx> wrote:
> 
> Hi Jakob,
> 
>> -----Original Message-----
>> From: Jakob Koschel <jakobkoschel@xxxxxxxxx>
>> Sent: Friday, April 1, 2022 1:07 AM
>> To: Jonathan Cameron <jic23@xxxxxxxxxx>
>> Cc: Lars-Peter Clausen <lars@xxxxxxxxxx>; Dan Carpenter
>> <dan.carpenter@xxxxxxxxxx>; Jakob Koschel
>> <jakobkoschel@xxxxxxxxx>; linux-iio@xxxxxxxxxxxxxxx; linux-
>> kernel@xxxxxxxxxxxxxxx; Mike Rapoport <rppt@xxxxxxxxxx>; Brian
>> Johannesmeyer <bjohannesmeyer@xxxxxxxxx>; Cristiano Giuffrida
>> <c.giuffrida@xxxxx>; Bos, H.J. <h.j.bos@xxxxx>
>> Subject: [PATCH 1/3] iio: buffer: remove usage of list iterator variable
>> for list_for_each_entry_continue_reverse()
>> 
>> [External]
>> 
>> In preparation to limit the scope of the list iterator variable to the
>> list traversal loop, use a dedicated pointer to iterate through the
>> list [1].
>> 
>> Since that variable should not be used past the loop iteration, a
>> separate variable is used to 'remember the current location within the
>> loop'.
>> 
>> To either continue iterating from that position or start a new
>> iteration (if the previous iteration was complete) list_prepare_entry()
>> is used.
>> 
>> Link: https://urldefense.com/v3/__https://lore.kernel.org/all/CAHk-
>> =wgRr_D8CB-D9Kg-
>> c=EHreAsk5SqXPwr9Y7k9sA6cWXJ6w@xxxxxxxxxxxxxx/__;!!A3Ni8CS0y
>> 2Y!q8llw5UCaMIsAU7tPtPDhwVor0wy032I7FJHv0VxBZksNuRJF04HjWe
>> 0XYG7OQ$  [1]
>> Signed-off-by: Jakob Koschel <jakobkoschel@xxxxxxxxx>
>> ---
>> drivers/iio/industrialio-buffer.c | 7 +++++--
>> 1 file changed, 5 insertions(+), 2 deletions(-)
>> 
>> diff --git a/drivers/iio/industrialio-buffer.c b/drivers/iio/industrialio-
>> buffer.c
>> index 208b5193c621..151a77c2affd 100644
>> --- a/drivers/iio/industrialio-buffer.c
>> +++ b/drivers/iio/industrialio-buffer.c
>> @@ -1059,7 +1059,7 @@ static int iio_enable_buffers(struct iio_dev
>> *indio_dev,
>> 	struct iio_device_config *config)
>> {
>> 	struct iio_dev_opaque *iio_dev_opaque =
>> to_iio_dev_opaque(indio_dev);
>> -	struct iio_buffer *buffer;
>> +	struct iio_buffer *buffer, *tmp = NULL;
>> 	int ret;
>> 
>> 	indio_dev->active_scan_mask = config->scan_mask;
>> @@ -1097,8 +1097,10 @@ static int iio_enable_buffers(struct iio_dev
>> *indio_dev,
>> 
>> 	list_for_each_entry(buffer, &iio_dev_opaque->buffer_list,
>> buffer_list) {
>> 		ret = iio_buffer_enable(buffer, indio_dev);
>> -		if (ret)
>> +		if (ret) {
>> +			tmp = buffer;
>> 			goto err_disable_buffers;
>> +		}
>> 	}
>> 
>> 	if (indio_dev->currentmode == INDIO_BUFFER_TRIGGERED) {
>> @@ -1125,6 +1127,7 @@ static int iio_enable_buffers(struct iio_dev
>> *indio_dev,
>> 					     indio_dev->pollfunc);
>> 	}
>> err_disable_buffers:
>> +	buffer = list_prepare_entry(tmp, &iio_dev_opaque-
>>> buffer_list, buffer_list);
> 
> Ok, it's Friday so I might be seeing ghosts... But looking at 'list_prepare_entry()'...
> If tmp != NULL, then all goes well but if tmp == NULL, then we get
> 
> list_entry(&iio_dev_opaque->buffer_list, struct iio_buffer, buffer_list) which
> would require 'struct iio_dev_opaque'. It looks like like 'list_prepare_entry()'
> assumes that pos and head are of the same type... 
> 
> Am I missing something?

The list iterators are weird in this perspective...

If you look at the original code, list_for_each_entry_continue_reverse() is called on 'buffer'.

'buffer' would be a valid struct element of &iio_dev_opaque->buffer_list if the break is hit,
but if no break is hit in the earlier list_for_each_entry() buffer is not a valid entry.

Before the terminating condition of list_for_each_entry() is met, it essentially does:

	buffer = list_entry(&iio_dev_opaque->buffer_list, typeof(*buffer), buffer_list);

the buffer returned here is not a valid pointer to struct however.
But since list_for_each_entry_continue_reverse() immediately calls list_prev_entry(buffer, buffer_list)
on it you end up with the last entry of the list again and start iterating with that one.

It's a very weird design choice but since list_for_each_entry_continue_reverse() expects
a pointer to the element struct and not the list_head struct, you need to pass in this 'bogus'
pointer if you want it to start on the head element.


Keep in mind that the code here is just a more explicit version of this 'type confusion' whereas
with the original code it was just hidden within the list_for_each_entry() macro and far less obvious.
The functionality is exactly the same.

PS: list_prepare_entry() was made for this use case, so basically it is expected to craft this bogus pointer from
the head if pos == NULL.

> 
> - Nuno Sá

Thanks,
Jakob