On Tue, Oct 31, 2023 at 12:18:13PM +0200, Laurent Pinchart wrote:
> On Tue, Oct 31, 2023 at 10:14:02AM +0000, Sakari Ailus wrote:
> > On Tue, Oct 31, 2023 at 09:07:06AM +0000, Sakari Ailus wrote:
> > > On Tue, Oct 31, 2023 at 10:53:47AM +0200, Laurent Pinchart wrote:
> > > > On Tue, Oct 31, 2023 at 06:34:39AM +0000, Sakari Ailus wrote:
> > > > > On Tue, Oct 31, 2023 at 01:38:09AM +0200, Laurent Pinchart wrote:
> > > > > > On Mon, Oct 30, 2023 at 10:43:47PM +0000, Sakari Ailus wrote:
> > > > > > > On Mon, Oct 30, 2023 at 01:26:58PM +0100, Tommaso Merciai wrote:
> > > > > > >
> > > > > > > ...
> > > > > > >
> > > > > > > > > > +static int alvium_get_host_supp_csi_lanes(struct alvium_dev *alvium)
> > > > > > > > > > +{
> > > > > > > > > > + u64 val;
> > > > > > > > > > + int ret = 0;
> > > > > > > > > > +
> > > > > > > > > > + alvium_read(alvium, REG_BCRM_CSI2_LANE_COUNT_RW, &val, &ret);
> > > > > > > > >
> > > > > > > > > Missing error checking before the use of the value. The same pattern
> > > > > > > > > remains prevalent throughout the driver.
> > > > > > > > >
> > > > > > > > > I think it'd be easier if you didn't use a temporary variable for reading,
> > > > > > > > > but instead had a register width specific access function. You could even
> > > > > > > > > introduce a helper macro to read this information as I suggested in an
> > > > > > > > > earlier review.
> > > > > > > >
> > > > > > > > oks.
> > > > > > > > We are moving to use the following macros:
> > > > > > > >
> > > > > > > > #define alvium_read_check(alvium, reg, value) \
> > > > > > > > { \
> > > > > > > > int ret = alvium_read(alvium, reg, value, NULL); \
> > > > > > > > if (ret) \
> > > > > > > > return ret; \
> > > > > > > > }
> > > > > > > >
> > > > > > >
> > > > > > > You could do something like (entirely untested):
> > > > > > >
> > > > > > > #define ALVIUM_DECLARE_READ(sign, bits) \
> > > > > > > static int
> > > > > > > alvium_read_ ## sign ## bits(struct alvium_dev *alvium, u32 reg, \
> > > > > > > sign ## bits *val, int *err) \
> > > > > > > { \
> > > > > > > u64 val64; \
> > > > > > > int ret; \
> > > > > > > \
> > > > > > > if (err && *err < 0) \
> > > > > > > return *err; \
> > > > > > > \
> > > > > > > alvium_read(alvium, reg, &val64, &ret); \
> > > > > > > if (ret < 0) { \
> > > > > > > if (err) \
> > > > > > > *err = ret; \
> > > > > > > return ret; \
> > > > > > > } \
> > > > > > > \
> > > > > > > *val = val64; \
> > > > > > > \
> > > > > > > return 0; \
> > > > > > > }
> > > > > > >
> > > > > > > ALVIUM_DECLARE_READ(u, 32);
> > > > > > >
> > > > > > > And then, e.g. instead of (and failing to check ret):
> > > > > > >
> > > > > > > u64 val;
> > > > > > >
> > > > > > > alvium_read(alvium, REG_BCRM_CONTRAST_VALUE_RW, &val, &ret);
> > > > > > > alvium->dft_contrast = val;
> > > > > > >
> > > > > > > you'd have a single call:
> > > > > > >
> > > > > > > alvium_read_u32(alvium, REG_BCRM_CONTRAST_VALUE_RW,
> > > > > > > &alvium->dft_contrast, &ret);
> > > > > > >
> > > > > > > And so on.
> > > > > > >
> > > > > > > You can drop sign if you don't need signed reads but some of the struct
> > > > > > > fields you're writing something appear to be signed.
> > > > > > >
> > > > > > > It'd be good to check the register size matches with the size of *val, too.
> > > > > > > Maybe something like:
> > > > > > >
> > > > > > > WARN_ON((CCI_REG ## bits(0) && CCI_REG_WIDTH_MASK) >> CCI_REG_WIDTH_SHIFT
> > > > > > > != sizeof(sign ## bits));
> > > > > >
> > > > > > I think this could actually be automated, and implemented in v4l2-cci.
> > > > > > Something like the following:
> > > > > >
> > > > > > diff --git a/drivers/media/v4l2-core/v4l2-cci.c b/drivers/media/v4l2-core/v4l2-cci.c
> > > > > > index bc2dbec019b0..27f1eaa7777d 100644
> > > > > > --- a/drivers/media/v4l2-core/v4l2-cci.c
> > > > > > +++ b/drivers/media/v4l2-core/v4l2-cci.c
> > > > > > @@ -16,7 +16,7 @@
> > > > > >
> > > > > > #include <media/v4l2-cci.h>
> > > > > >
> > > > > > -int cci_read(struct regmap *map, u32 reg, u64 *val, int *err)
> > > > > > +int __cci_read(struct regmap *map, u32 reg, void *val, int *err)
> > > > > > {
> > > > > > unsigned int len;
> > > > > > u8 buf[8];
> > > > > > @@ -37,19 +37,19 @@ int cci_read(struct regmap *map, u32 reg, u64 *val, int *err)
> > > > > >
> > > > > > switch (len) {
> > > > > > case 1:
> > > > > > - *val = buf[0];
> > > > > > + *(u8 *)val = buf[0];
> > > > > > break;
> > > > > > case 2:
> > > > > > - *val = get_unaligned_be16(buf);
> > > > > > + *(u16 *)val = get_unaligned_be16(buf);
> > > > > > break;
> > > > > > case 3:
> > > > > > - *val = get_unaligned_be24(buf);
> > > > > > + *(u32 *)val = get_unaligned_be24(buf);
> > > > > > break;
> > > > > > case 4:
> > > > > > - *val = get_unaligned_be32(buf);
> > > > > > + *(u32 *)val = get_unaligned_be32(buf);
> > > > > > break;
> > > > > > case 8:
> > > > > > - *val = get_unaligned_be64(buf);
> > > > > > + *(u64 *)val = get_unaligned_be64(buf);
> > > > > > break;
> > > > > > default:
> > > > > > dev_err(regmap_get_device(map), "Error invalid reg-width %u for reg 0x%04x\n",
> > > > > > @@ -64,7 +64,7 @@ int cci_read(struct regmap *map, u32 reg, u64 *val, int *err)
> > > > > >
> > > > > > return ret;
> > > > > > }
> > > > > > -EXPORT_SYMBOL_GPL(cci_read);
> > > > > > +EXPORT_SYMBOL_GPL(__cci_read);
> > > > > >
> > > > > > int cci_write(struct regmap *map, u32 reg, u64 val, int *err)
> > > > > > {
> > > > > > @@ -119,7 +119,7 @@ int cci_update_bits(struct regmap *map, u32 reg, u64 mask, u64 val, int *err)
> > > > > > u64 readval;
> > > > > > int ret;
> > > > > >
> > > > > > - ret = cci_read(map, reg, &readval, err);
> > > > > > + ret = __cci_read(map, reg, &readval, err);
> > > > > > if (ret)
> > > > > > return ret;
> > > > > >
> > > > > > diff --git a/include/media/v4l2-cci.h b/include/media/v4l2-cci.h
> > > > > > index 0f6803e4b17e..31223ce8d741 100644
> > > > > > --- a/include/media/v4l2-cci.h
> > > > > > +++ b/include/media/v4l2-cci.h
> > > > > > @@ -7,6 +7,9 @@
> > > > > > #ifndef _V4L2_CCI_H
> > > > > > #define _V4L2_CCI_H
> > > > > >
> > > > > > +#include <linux/bitfield.h>
> > > > > > +#include <linux/build_bug.h>
> > > > > > +#include <linux/log2.h>
> > > > > > #include <linux/types.h>
> > > > > >
> > > > > > struct i2c_client;
> > > > > > @@ -39,6 +42,8 @@ struct cci_reg_sequence {
> > > > > > #define CCI_REG32(x) ((4 << CCI_REG_WIDTH_SHIFT) | (x))
> > > > > > #define CCI_REG64(x) ((8 << CCI_REG_WIDTH_SHIFT) | (x))
> > > > > >
> > > > > > +int __cci_read(struct regmap *map, u32 reg, void *val, int *err);
> > > > > > +
> > > > > > /**
> > > > > > * cci_read() - Read a value from a single CCI register
> > > > > > *
> > > > > > @@ -48,9 +53,17 @@ struct cci_reg_sequence {
> > > > > > * @err: Optional pointer to store errors, if a previous error is set
> > > > > > * then the read will be skipped
> > > > > > *
> > > > > > + * The type of the @val pointer must match the size of the register being read.
> > > > > > + * Mismatches will result in compile-time errors.
> > > > > > + *
> > > > > > * Return: %0 on success or a negative error code on failure.
> > > > > > */
> > > > > > -int cci_read(struct regmap *map, u32 reg, u64 *val, int *err);
> > > > > > +#define cci_read(map, reg, val, err) ({ \
> > > > > > + u32 __reg = (reg); \
> > > > > > + u32 __size = FIELD_GET(CCI_REG_WIDTH_MASK, __reg); \
> > > > > > + BUILD_BUG_ON(sizeof(*(val)) != roundup_pow_of_two(__size)); \
> > > > > > + __cci_read(map, __reg, (void *)(val), err); \
> > > > > > +})
> > > > > >
> > > > > > /**
> > > > > > * cci_write() - Write a value to a single CCI register
> > > > > >
> > > > > > The change to cci_update_bits() is obviously wrong, I've hacked that to
> > > > > > compile-test the rest with the drivers using cci_read(), and I get nice
> > > > > > build-time errors due to usage of the wrong type :-)
> > > > > >
> > > > > > Is this something that would be considered ? Bonus points to anyone who
> > > > > > would fix cci_update_bits() :-)
> > > > >
> > > > > I like the idea of moving this to v4l2-cci.
> > > > >
> > > > > I'd prefer _Generic() based solution as we'd have exact types there instead
> > > > > of just size. E.g. with the above code, reading a value to a long variable
> > > > > would work on some archs but fail on others.
> > > >
> > > > Doesn't _Generic() treat compatible types identically ?
> > >
> > > Ah, it does, indeed. So that doesn't solve the long problem.
> > >
> > > I guess the code will be more compact with just void *, on the expense of
> > > (some) type checking.
> > >
> > > I'm fine with either.
> >
> > You can't check the variable size matches the register width by using a
> > void pointer. That'd be a source for possibly difficult to debug problems:
> > passing a pointer to u8 variable while reading a 64-bit register overwrites
> > seven bytes on the stack outside the variable itself.
> >
> > Let's use _Generic().
>
> I still don't see how that will help :-) The prototype of the
> __cci_read() function is unrelated to whether or not _Generic() gets
> used. The cci_read() macro will cause a *build* time error if the value
> pointer doesn't match the register size. The __cci_read() function must
> not be used directly by drivers.
Ah, right, this part of the check is indeed in the macro. That doesn't
catch floats but those are unlikely be a practical issue anyway.
The cast to void * is redundant in the macro.
--
Sakari Ailus
