Skip to contentSkip to conversation

iio: buffer: Fix demux update

PATCH In review Nuno Sá 3 participants 5 messages Nov 12, 2020 – Nov 28, 2020 IIO

lore
Share & tools
lore

Nuno Sá

author reviewer
#0 lore raw

PATCH 1/1 iio: buffer: Fix demux update

When updating the buffer demux, we will skip a scan element from the device in the case in_ind != out_ind and we enter the while loop. in_ind should only be refreshed with find_next_bit() in the end of the loop.

Fixes 5ada4ea9be16 ("staging:iio: add demux optionally to path from device to buffer")
---
 drivers/iio/industrialio-buffer.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/iio/industrialio-buffer.c +3 -3
index a4f6bb96d4f4..276b609d7917 100644
@@ -865,12 +865,12 @@ static int iio_buffer_update_demux(struct iio_dev *indio_dev,
 				       indio_dev->masklength,
 				       in_ind + 1);
 		while (in_ind != out_ind) {
-			in_ind = find_next_bit(indio_dev->active_scan_mask,
-					       indio_dev->masklength,
-					       in_ind + 1);
 			length = iio_storage_bytes_for_si(indio_dev, in_ind);
 			/* Make sure we are aligned */
 			in_loc = roundup(in_loc, length) + length;
+			in_ind = find_next_bit(indio_dev->active_scan_mask,
+					       indio_dev->masklength,
+					       in_ind + 1);
 		}
 		length = iio_storage_bytes_for_si(indio_dev, in_ind);
 		out_loc = roundup(out_loc, length);
Signature · 1 line
2.29.2
#1 lore raw
Quoted 7 lines · On Thu, 12 Nov 2020 15:43:22 +0100 Nuno Sá <[email protected]> wrote:

When updating the buffer demux, we will skip a scan element from the device in the case in_ind != out_ind and we enter the while loop. in_ind should only be refreshed with find_next_bit() in the end of the loop.

Fixes 5ada4ea9be16 ("staging:iio: add demux optionally to path from device to buffer")

Yikes that's been there a long time.

Could you provide an example of a particular layout and the result of this being wrong?

Thanks,

Jonathan

Quoted 24 lines
---
 drivers/iio/industrialio-buffer.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/iio/industrialio-buffer.c +3 -3
index a4f6bb96d4f4..276b609d7917 100644
@@ -865,12 +865,12 @@ static int iio_buffer_update_demux(struct iio_dev *indio_dev,
 				       indio_dev->masklength,
 				       in_ind + 1);
 		while (in_ind != out_ind) {
-			in_ind = find_next_bit(indio_dev->active_scan_mask,
-					       indio_dev->masklength,
-					       in_ind + 1);
 			length = iio_storage_bytes_for_si(indio_dev, in_ind);
 			/* Make sure we are aligned */
 			in_loc = roundup(in_loc, length) + length;
+			in_ind = find_next_bit(indio_dev->active_scan_mask,
+					       indio_dev->masklength,
+					       in_ind + 1);
 		}
 		length = iio_storage_bytes_for_si(indio_dev, in_ind);
 		out_loc = roundup(out_loc, length);
author reviewer
#2 lore raw
Quoted 21 lines

-----Original Message-----
From: Jonathan Cameron <[email protected]>
Sent: Thursday, November 12, 2020 6:28 PM
To: Sa, Nuno <[email protected]>
Cc: [email protected]; Jonathan Cameron <[email protected]>; Lars-Peter Clausen <[email protected]>; Peter Meerwald-Stadler <[email protected]>
Subject: Re: [PATCH] iio: buffer: Fix demux update

Quoted 1 line · On Thu, 12 Nov 2020 15:43:22 +0100 Nuno Sá <[email protected]> wrote:

When updating the buffer demux, we will skip a scan element from

the

Quoted 2 lines

device in the case in_ind != out_ind and we enter the while loop. in_ind should only be refreshed with find_next_bit() in the end of

the

Quoted 2 lines

loop.

Quoted 9 lines
Quoted 1 line
Fixes 5ada4ea9be16 ("staging:iio: add demux optionally to path from

device to buffer")

Quoted 1 line

Yikes that's been there a long time.

Could you provide an example of a particular layout and the result of this being wrong?

Hi Jonathan,

Let's say:

iio_dev_mask: 0x0111
buffer_mask: 0x0100

We would get out_ind = 2 and in_ind = 0 and enter the loop. In the first iteration we call find_next_bit() before doing the in_ind=0 computation which means we will skip it and go directly to bit 1... And if we continue the path flow, we see that bit 2 will be computed two times, so if we are lucky and scan_index0_len == scan_index2_len this will go unnoticed...

Honestly, I didn't test this but it looks one of those things more or less clear by reading the code or am I missing something here?

- Nuno Sá

Quoted 4 lines

Thanks,

Jonathan

maintainer
#3 lore raw
Quoted 37 lines · Nuno
Quoted 20 lines

-----Original Message-----
From: Jonathan Cameron <[email protected]>
Sent: Thursday, November 12, 2020 6:28 PM
To: Sa, Nuno <[email protected]>
Cc: [email protected]; Jonathan Cameron <[email protected]>; Lars-Peter Clausen <[email protected]>; Peter Meerwald-Stadler <[email protected]>
Subject: Re: [PATCH] iio: buffer: Fix demux update

Quoted 1 line · On Thu, 12 Nov 2020 15:43:22 +0100 Nuno Sá <[email protected]> wrote:

When updating the buffer demux, we will skip a scan element from

the

Quoted 2 lines

device in the case in_ind != out_ind and we enter the while loop. in_ind should only be refreshed with find_next_bit() in the end of

the

Quoted 2 lines

loop.

Quoted 9 lines
Quoted 1 line
Fixes 5ada4ea9be16 ("staging:iio: add demux optionally to path from

device to buffer")

Quoted 1 line

Yikes that's been there a long time.

Could you provide an example of a particular layout and the result of this being wrong?

Hi Jonathan,

Let's say:

iio_dev_mask: 0x0111
buffer_mask: 0x0100

Quoted 8 lines

We would get out_ind = 2 and in_ind = 0 and enter the loop. In the first iteration we call find_next_bit() before doing the in_ind=0 computation which means we will skip it and go directly to bit 1... And if we continue the path flow, we see that bit 2 will be computed two times, so if we are lucky and scan_index0_len == scan_index2_len this will go unnoticed...

Honestly, I didn't test this but it looks one of those things more or less clear by reading the code or am I missing something here?

Mostly I was wondering why it hadn't bitten us before. I think you've identified why with your "if we are lucky and scan_index0_len == scan_index2_len" then this will go unnoticed.

It's very rare (though not unheard of) for a device to have it's main channels of different widths (timestamp doesn't matter for this as it is always at the end). The demux also only kicks in if we have a restricted channel mask (or are using a kfifo and a buffer_cb which is rather rare). I suspect we have few if any devices that actually run into this problem.

I guess I originally tested this code with devices I had at the time and none of them would have tripped this.

Anyhow, whilst I agree with your analysis I'd like to leave this on list for perhaps another week before applying it on the basis I'm paranoid and would ideally like a few more eyes on this.

Good spot!

Jonathan

Quoted 7 lines

- Nuno Sá

Quoted 4 lines

Thanks,

Jonathan

maintainer
#4 lore raw
Quoted 59 lines · On Sat, 14 Nov 2020 16:26:38 +0000 Jonathan Cameron <[email protected]> wrote:
Quoted 37 lines · Nuno
Quoted 20 lines

-----Original Message-----
From: Jonathan Cameron <[email protected]>
Sent: Thursday, November 12, 2020 6:28 PM
To: Sa, Nuno <[email protected]>
Cc: [email protected]; Jonathan Cameron <[email protected]>; Lars-Peter Clausen <[email protected]>; Peter Meerwald-Stadler <[email protected]>
Subject: Re: [PATCH] iio: buffer: Fix demux update

Quoted 1 line · On Thu, 12 Nov 2020 15:43:22 +0100 Nuno Sá <[email protected]> wrote:

When updating the buffer demux, we will skip a scan element from

the

Quoted 2 lines

device in the case in_ind != out_ind and we enter the while loop. in_ind should only be refreshed with find_next_bit() in the end of

the

Quoted 2 lines

loop.

Quoted 9 lines
Quoted 1 line
Fixes 5ada4ea9be16 ("staging:iio: add demux optionally to path from

device to buffer")

Quoted 1 line

Yikes that's been there a long time.

Could you provide an example of a particular layout and the result of this being wrong?

Hi Jonathan,

Let's say:

iio_dev_mask: 0x0111
buffer_mask: 0x0100

Quoted 8 lines

We would get out_ind = 2 and in_ind = 0 and enter the loop. In the first iteration we call find_next_bit() before doing the in_ind=0 computation which means we will skip it and go directly to bit 1... And if we continue the path flow, we see that bit 2 will be computed two times, so if we are lucky and scan_index0_len == scan_index2_len this will go unnoticed...

Honestly, I didn't test this but it looks one of those things more or less clear by reading the code or am I missing something here?

Mostly I was wondering why it hadn't bitten us before. I think you've identified why with your "if we are lucky and scan_index0_len == scan_index2_len" then this will go unnoticed.

It's very rare (though not unheard of) for a device to have it's main channels of different widths (timestamp doesn't matter for this as it is always at the end). The demux also only kicks in if we have a restricted channel mask (or are using a kfifo and a buffer_cb which is rather rare). I suspect we have few if any devices that actually run into this problem.

Quoted 7 lines

I guess I originally tested this code with devices I had at the time and none of them would have tripped this.

Anyhow, whilst I agree with your analysis I'd like to leave this on list for perhaps another week before applying it on the basis I'm paranoid and would ideally like a few more eyes on this.

Ah well. I'll take silence as meaning either everyone is happy or no one else is going to read it ;)

Applied with a bit more text in the description to highlight that, whilst a bug, it's actually not a common situation.

Given timing I've applied this to the togreg branch of iio.git ready for the next merge window.

thanks,

Jonathan

Quoted 12 lines

Good spot!

Jonathan

Quoted 7 lines

- Nuno Sá

Quoted 4 lines

Thanks,

Jonathan