Correct the "sub_lsb" shift for both ltc2497 and ltc2499.
An earlier version of the code shifted by 14 but this was a consequence
of reading three bytes into a __be32 buffer and using be32_to_cpu(), so
eight extra bits needed to be skipped. Now we use get_unaligned_be24()
and thus the additional skip is wrong.
Fixes 2187cfe ("drivers: iio: adc: ltc2497: LTC2499 support")
Signed-off-by: Ian Ray <[email protected]>
---
drivers/iio/adc/ltc2497.c | 6 ++----
1 file changed, 2 insertions(+), 4 deletions(-)
diff --git a/drivers/iio/adc/ltc2497.c b/drivers/iio/adc/ltc2497.c
index 17370c5..ec198c6 100644
--- a/drivers/iio/adc/ltc2497.c
+++ b/drivers/iio/adc/ltc2497.c
@@ -28,7 +28,6 @@ struct ltc2497_driverdata {
struct ltc2497core_driverdata common_ddata;
struct i2c_client *client;
u32 recv_size;
- u32 sub_lsb;
/*
* DMA (thus cache coherency maintenance) may require the
* transfer buffers to live in their own cache lines.
@@ -65,10 +64,10 @@ static int ltc2497_result_and_measure(struct ltc2497core_driverdata *ddata,
* equivalent to a sign extension.
*/
if (st->recv_size == 3) {
- *val = (get_unaligned_be24(st->data.d8) >> st->sub_lsb)
+ *val = (get_unaligned_be24(st->data.d8) >> 6)
- BIT(ddata->chip_info->resolution + 1);
} else {
- *val = (be32_to_cpu(st->data.d32) >> st->sub_lsb)
+ *val = (be32_to_cpu(st->data.d32) >> 6)
- BIT(ddata->chip_info->resolution + 1);
}
@@ -122,7 +121,6 @@ static int ltc2497_probe(struct i2c_client *client)
st->common_ddata.chip_info = chip_info;
resolution = chip_info->resolution;
- st->sub_lsb = 31 - (resolution + 1);
st->recv_size = BITS_TO_BYTES(resolution) + 1;
return ltc2497core_probe(dev, indio_dev);
--
2.10.1
On Fri, 27 Jan 2023 14:57:14 +0200
Ian Ray <[email protected]> wrote:
> Correct the "sub_lsb" shift for both ltc2497 and ltc2499.
Hi Iain, Thanks for the patch.
>
> An earlier version of the code shifted by 14 but this was a consequence
> of reading three bytes into a __be32 buffer and using be32_to_cpu(), so
> eight extra bits needed to be skipped. Now we use get_unaligned_be24()
> and thus the additional skip is wrong.
>
> Fixes 2187cfe ("drivers: iio: adc: ltc2497: LTC2499 support")
This isn't in standard Fixes tag format. If nothing else comes
up I'll fix up whilst applying.
> Signed-off-by: Ian Ray <[email protected]>
> ---
> drivers/iio/adc/ltc2497.c | 6 ++----
> 1 file changed, 2 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/iio/adc/ltc2497.c b/drivers/iio/adc/ltc2497.c
> index 17370c5..ec198c6 100644
> --- a/drivers/iio/adc/ltc2497.c
> +++ b/drivers/iio/adc/ltc2497.c
> @@ -28,7 +28,6 @@ struct ltc2497_driverdata {
> struct ltc2497core_driverdata common_ddata;
> struct i2c_client *client;
> u32 recv_size;
> - u32 sub_lsb;
> /*
> * DMA (thus cache coherency maintenance) may require the
> * transfer buffers to live in their own cache lines.
> @@ -65,10 +64,10 @@ static int ltc2497_result_and_measure(struct ltc2497core_driverdata *ddata,
> * equivalent to a sign extension.
> */
> if (st->recv_size == 3) {
> - *val = (get_unaligned_be24(st->data.d8) >> st->sub_lsb)
> + *val = (get_unaligned_be24(st->data.d8) >> 6)
> - BIT(ddata->chip_info->resolution + 1);
> } else {
> - *val = (be32_to_cpu(st->data.d32) >> st->sub_lsb)
> + *val = (be32_to_cpu(st->data.d32) >> 6)
> - BIT(ddata->chip_info->resolution + 1);
> }
>
> @@ -122,7 +121,6 @@ static int ltc2497_probe(struct i2c_client *client)
> st->common_ddata.chip_info = chip_info;
>
> resolution = chip_info->resolution;
> - st->sub_lsb = 31 - (resolution + 1);
> st->recv_size = BITS_TO_BYTES(resolution) + 1;
>
> return ltc2497core_probe(dev, indio_dev);
On Fri, 27 Jan 2023 17:47:56 +0000
Jonathan Cameron <[email protected]> wrote:
> On Fri, 27 Jan 2023 14:57:14 +0200
> Ian Ray <[email protected]> wrote:
>
> > Correct the "sub_lsb" shift for both ltc2497 and ltc2499.
I dropped the ltc2499 mention here as that one seems to be unaffected
by this patch (beyond as a simplification)
>
> Hi Iain, Thanks for the patch.
>
> >
> > An earlier version of the code shifted by 14 but this was a consequence
> > of reading three bytes into a __be32 buffer and using be32_to_cpu(), so
> > eight extra bits needed to be skipped. Now we use get_unaligned_be24()
> > and thus the additional skip is wrong.
> >
> > Fixes 2187cfe ("drivers: iio: adc: ltc2497: LTC2499 support")
>
> This isn't in standard Fixes tag format. If nothing else comes
> up I'll fix up whilst applying.
>
> > Signed-off-by: Ian Ray <[email protected]>
Fix looks good to me. (I noted my reasoning below on basis someone might
find it useful)
Applied to the fixes-togreg branch of iio.git with fixes tag fixed and
marked for stable.
Please take a look at:
https://git.kernel.org/pub/scm/linux/kernel/git/jic23/iio.git/commit/?h=fixes-togreg&id=6327a930ab7bfa1ab33bcdffd5f5f4b1e7131504
to check I didn't mess anything up.
Thanks,
Jonathan
> > ---
> > drivers/iio/adc/ltc2497.c | 6 ++----
> > 1 file changed, 2 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/iio/adc/ltc2497.c b/drivers/iio/adc/ltc2497.c
> > index 17370c5..ec198c6 100644
> > --- a/drivers/iio/adc/ltc2497.c
> > +++ b/drivers/iio/adc/ltc2497.c
> > @@ -28,7 +28,6 @@ struct ltc2497_driverdata {
> > struct ltc2497core_driverdata common_ddata;
> > struct i2c_client *client;
> > u32 recv_size;
> > - u32 sub_lsb;
> > /*
> > * DMA (thus cache coherency maintenance) may require the
> > * transfer buffers to live in their own cache lines.
> > @@ -65,10 +64,10 @@ static int ltc2497_result_and_measure(struct ltc2497core_driverdata *ddata,
> > * equivalent to a sign extension.
> > */
> > if (st->recv_size == 3) {
> > - *val = (get_unaligned_be24(st->data.d8) >> st->sub_lsb)
> > + *val = (get_unaligned_be24(st->data.d8) >> 6)
Prior to the buggy patch.
1) Read into top 3 bytes of BE32 which corresponds to 8 bits high.
2) Shift by 14.
Before this patch
1) Read only 3 bytes
2) shift by (31 - (resolution + 1) where resolution is 16 so shift by 14.
Indeed bug.
After this patch.
1) Read 3 bytes
2) shift by 6
Matches original behaviour - fix is correct.
> > - BIT(ddata->chip_info->resolution + 1);
> > } else {
> > - *val = (be32_to_cpu(st->data.d32) >> st->sub_lsb)
> > + *val = (be32_to_cpu(st->data.d32) >> 6)
Prior to this patch
sub_lsb = 31 - (resolution + 1) where in this case resolution is 24
31 - 25 = 6 so this change is a noop change, so this is good.
> > - BIT(ddata->chip_info->resolution + 1);
> > }
> >
> > @@ -122,7 +121,6 @@ static int ltc2497_probe(struct i2c_client *client)
> > st->common_ddata.chip_info = chip_info;
> >
> > resolution = chip_info->resolution;
> > - st->sub_lsb = 31 - (resolution + 1);
> > st->recv_size = BITS_TO_BYTES(resolution) + 1;
> >
> > return ltc2497core_probe(dev, indio_dev);
>
On Sat, Jan 28, 2023 at 06:00:32PM +0000, Jonathan Cameron wrote:
>
> On Fri, 27 Jan 2023 17:47:56 +0000
> Jonathan Cameron <[email protected]> wrote:
>
> > On Fri, 27 Jan 2023 14:57:14 +0200
> > Ian Ray <[email protected]> wrote:
> >
> > > Correct the "sub_lsb" shift for both ltc2497 and ltc2499.
>
> I dropped the ltc2499 mention here as that one seems to be unaffected
> by this patch (beyond as a simplification)
Makes sense.
>
> >
> > Hi Iain, Thanks for the patch.
> >
> > >
> > > An earlier version of the code shifted by 14 but this was a consequence
> > > of reading three bytes into a __be32 buffer and using be32_to_cpu(), so
> > > eight extra bits needed to be skipped. Now we use get_unaligned_be24()
> > > and thus the additional skip is wrong.
> > >
> > > Fixes 2187cfe ("drivers: iio: adc: ltc2497: LTC2499 support")
> >
> > This isn't in standard Fixes tag format. If nothing else comes
> > up I'll fix up whilst applying.
> >
> > > Signed-off-by: Ian Ray <[email protected]>
>
> Fix looks good to me. (I noted my reasoning below on basis someone might
> find it useful)
>
> Applied to the fixes-togreg branch of iio.git with fixes tag fixed and
> marked for stable.
>
> Please take a look at:
> https://git.kernel.org/pub/scm/linux/kernel/git/jic23/iio.git/commit/?h=fixes-togreg&id=6327a930ab7bfa1ab33bcdffd5f5f4b1e7131504
> to check I didn't mess anything up.
lgtm
Many thanks!
>
> Thanks,
>
> Jonathan
>
>
> > > ---
> > > drivers/iio/adc/ltc2497.c | 6 ++----
> > > 1 file changed, 2 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/drivers/iio/adc/ltc2497.c b/drivers/iio/adc/ltc2497.c
> > > index 17370c5..ec198c6 100644
> > > --- a/drivers/iio/adc/ltc2497.c
> > > +++ b/drivers/iio/adc/ltc2497.c
> > > @@ -28,7 +28,6 @@ struct ltc2497_driverdata {
> > > struct ltc2497core_driverdata common_ddata;
> > > struct i2c_client *client;
> > > u32 recv_size;
> > > - u32 sub_lsb;
> > > /*
> > > * DMA (thus cache coherency maintenance) may require the
> > > * transfer buffers to live in their own cache lines.
> > > @@ -65,10 +64,10 @@ static int ltc2497_result_and_measure(struct ltc2497core_driverdata *ddata,
> > > * equivalent to a sign extension.
> > > */
> > > if (st->recv_size == 3) {
> > > - *val = (get_unaligned_be24(st->data.d8) >> st->sub_lsb)
> > > + *val = (get_unaligned_be24(st->data.d8) >> 6)
> Prior to the buggy patch.
>
> 1) Read into top 3 bytes of BE32 which corresponds to 8 bits high.
> 2) Shift by 14.
>
> Before this patch
> 1) Read only 3 bytes
> 2) shift by (31 - (resolution + 1) where resolution is 16 so shift by 14.
> Indeed bug.
>
> After this patch.
> 1) Read 3 bytes
> 2) shift by 6
>
> Matches original behaviour - fix is correct.
>
>
> > > - BIT(ddata->chip_info->resolution + 1);
> > > } else {
> > > - *val = (be32_to_cpu(st->data.d32) >> st->sub_lsb)
> > > + *val = (be32_to_cpu(st->data.d32) >> 6)
>
> Prior to this patch
> sub_lsb = 31 - (resolution + 1) where in this case resolution is 24
> 31 - 25 = 6 so this change is a noop change, so this is good.
>
> > > - BIT(ddata->chip_info->resolution + 1);
> > > }
> > >
> > > @@ -122,7 +121,6 @@ static int ltc2497_probe(struct i2c_client *client)
> > > st->common_ddata.chip_info = chip_info;
> > >
> > > resolution = chip_info->resolution;
> > > - st->sub_lsb = 31 - (resolution + 1);
> > > st->recv_size = BITS_TO_BYTES(resolution) + 1;
> > >
> > > return ltc2497core_probe(dev, indio_dev);
> >
>
>