Re: [PATCH 1/2] iio: light: adux1020: fix interrupt clear value on SMBus-only adapters
From: Jonathan Cameron
Date: Sat Nov 14 2026 - 19:00:41 EST
On Tue, 10 Nov 2026 22:17:44 +0000
Raffaele Tranfaglia <raffaele.tranfaglia@xxxxxxxxx> wrote:
> On Tue, 10 Nov 2026 at 10:17, Andy Shevchenko
> <andriy.shevchenko@xxxxxxxxx> wrote:
> > Can you give a Kunit test case? Because regmap specifies the val_bits == 16.
> > See also [3].
>
> I read your analysis in [3] and I agree with it for the bus it
> describes. regmap has two write paths though, selected in
> __regmap_init() by what the bus provides [4]:
>
> - a bus with raw .write/.read (regmap_i2c, used when the adapter has
> I2C_FUNC_I2C) goes through _regmap_bus_raw_write(), which uses
> format_val, so regmap_format_16_be() drops everything above 16 bits,
> as you describe;
>
> - a bus with only .reg_write/.reg_read (regmap_smbus_word_swapped, used
> when the adapter only has I2C_FUNC_SMBUS_WORD_DATA) skips the format
> initialization and goes through _regmap_bus_reg_write() [5], which
> passes val to bus->reg_write() unmodified.
> regmap_smbus_word_write_swapped() then returns -EINVAL for
> val > 0xffff [6].
>
> i2c-stub is SMBus only, so adux1020 takes the second path there.
>
> The existing regmap KUnit buses match these two paths (regmap-ram is
> reg_write only, regmap-raw-ram is formatted), so I added a test case for
> each, writing 0xffffff0f to a regmap with val_bits = 16 and expecting
> 0xff0f on the bus:
>
> # raw_write_wider_than_val_bits: pass:10 fail:0 skip:0 total:10
> ok 37 raw_write_wider_than_val_bits
>
> # write_wider_than_val_bits: EXPECTATION FAILED at
> drivers/base/regmap/regmap-kunit.c:342
> Expected 0xff0f == data->vals[param->from_reg], but
> 0xff0f == 65295 (0xff0f)
> data->vals[param->from_reg] == 4294967055 (0xffffff0f)
> # write_wider_than_val_bits: pass:0 fail:10 skip:0 total:10
> not ok 2 write_wider_than_val_bits
>
> I can send the test to Mark as an RFC, to ask whether regmap_write()
> should mask the value to val_bits for reg_write buses, or reject it
> consistently.
Indeed, please do that as otherwise we are going to see repeats
of the subtle issue you found here.
>
> [4] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/base/regmap/regmap.c?h=v7.4-rc1#n963
> [5] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/base/regmap/regmap.c?h=v7.4-rc1#n2153
> [6] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/base/regmap/regmap-i2c.c?h=v7.4-rc1#n127
>
> > > However, even if regmap changed, the driver would still be passing a
> > > 32-bit value for a 16-bit register.
> >
> > Is it a problem?
>
> Today it is on SMBus-only adapters, as above. If regmap ends up masking
> the value, patch 1 could still be used as the fix for stable releases
> since, from my understanding, small self-contained fixes are preferred
> for backporting.
I'd definitely want to know the longer term consistency fix is making
progress, but sure. Add a little more info to the patch description
and I don't mind pickup up the fix.
>
> Best regards,
> Raffaele