A new commit in LLVM causes an error on the use of 'long double' when
'-mno-x87' is used, which the kernel does through an alias,
'-mno-80387' (see the LLVM commit below for more details around why it
does this).
drivers/power/reset/ltc2952-poweroff.c:162:28: error: expression requires 'long double' type support, but target 'x86_64-unknown-linux-gnu' does not support it
data->wde_interval = 300L * 1E6L;
^
drivers/power/reset/ltc2952-poweroff.c:162:21: error: expression requires 'long double' type support, but target 'x86_64-unknown-linux-gnu' does not support it
data->wde_interval = 300L * 1E6L;
^
drivers/power/reset/ltc2952-poweroff.c:163:41: error: expression requires 'long double' type support, but target 'x86_64-unknown-linux-gnu' does not support it
data->trigger_delay = ktime_set(2, 500L*1E6L);
^
3 errors generated.
This happens due to the use of a 'long double' literal. The 'E6' part of
'1E6L' causes the literal to be a 'double' then the 'L' suffix promotes
it to 'long double'.
There is no visible reason for floating point values in this driver, as
the values are only assigned to integer types. Use NSEC_PER_MSEC, which
is the same integer value as '1E6L', to avoid changing functionality but
fix the error.
Fixes: 6647156c00cc ("power: reset: add LTC2952 poweroff driver")
Link: https://github.com/ClangBuiltLinux/linux/issues/1497
Link: https://github.com/llvm/llvm-project/commit/a8083d42b1c346e21623a1d36d1f0cadd7801d83
Signed-off-by: Nathan Chancellor <[email protected]>
Reviewed-by: Nick Desaulniers <[email protected]>
---
v1 -> v2: https://lore.kernel.org/r/[email protected]/
* A separate review pointed out that NSEC_PER_MSEC is a better choice
than USEC_PER_SEC because ktime_t is nanoseconds and the few functions
that take these values work in nanoseconds. The value is the same but
the documentation is better.
* Pick up Nick's review tag.
drivers/power/reset/ltc2952-poweroff.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/power/reset/ltc2952-poweroff.c b/drivers/power/reset/ltc2952-poweroff.c
index fbb344353fe4..65d9528cc989 100644
--- a/drivers/power/reset/ltc2952-poweroff.c
+++ b/drivers/power/reset/ltc2952-poweroff.c
@@ -159,8 +159,8 @@ static void ltc2952_poweroff_kill(void)
static void ltc2952_poweroff_default(struct ltc2952_poweroff *data)
{
- data->wde_interval = 300L * 1E6L;
- data->trigger_delay = ktime_set(2, 500L*1E6L);
+ data->wde_interval = 300L * NSEC_PER_MSEC;
+ data->trigger_delay = ktime_set(2, 500L * NSEC_PER_MSEC);
hrtimer_init(&data->timer_trigger, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
data->timer_trigger.function = ltc2952_poweroff_timer_trigger;
base-commit: d4439a1189f93d0ac1eaf0197db8e6b3e197d5c7
--
2.34.0.rc0
Hi,
On Fri, Nov 05, 2021 at 08:20:50AM -0700, Nathan Chancellor wrote:
> A new commit in LLVM causes an error on the use of 'long double' when
> '-mno-x87' is used, which the kernel does through an alias,
> '-mno-80387' (see the LLVM commit below for more details around why it
> does this).
>
> drivers/power/reset/ltc2952-poweroff.c:162:28: error: expression requires 'long double' type support, but target 'x86_64-unknown-linux-gnu' does not support it
> data->wde_interval = 300L * 1E6L;
> ^
> drivers/power/reset/ltc2952-poweroff.c:162:21: error: expression requires 'long double' type support, but target 'x86_64-unknown-linux-gnu' does not support it
> data->wde_interval = 300L * 1E6L;
> ^
> drivers/power/reset/ltc2952-poweroff.c:163:41: error: expression requires 'long double' type support, but target 'x86_64-unknown-linux-gnu' does not support it
> data->trigger_delay = ktime_set(2, 500L*1E6L);
> ^
> 3 errors generated.
>
> This happens due to the use of a 'long double' literal. The 'E6' part of
> '1E6L' causes the literal to be a 'double' then the 'L' suffix promotes
> it to 'long double'.
>
> There is no visible reason for floating point values in this driver, as
> the values are only assigned to integer types. Use NSEC_PER_MSEC, which
> is the same integer value as '1E6L', to avoid changing functionality but
> fix the error.
>
> Fixes: 6647156c00cc ("power: reset: add LTC2952 poweroff driver")
> Link: https://github.com/ClangBuiltLinux/linux/issues/1497
> Link: https://github.com/llvm/llvm-project/commit/a8083d42b1c346e21623a1d36d1f0cadd7801d83
> Signed-off-by: Nathan Chancellor <[email protected]>
> Reviewed-by: Nick Desaulniers <[email protected]>
> ---
Thanks, queued to power-supply's fixes branch.
-- Sebastian
>
> v1 -> v2: https://lore.kernel.org/r/[email protected]/
>
> * A separate review pointed out that NSEC_PER_MSEC is a better choice
> than USEC_PER_SEC because ktime_t is nanoseconds and the few functions
> that take these values work in nanoseconds. The value is the same but
> the documentation is better.
>
> * Pick up Nick's review tag.
>
> drivers/power/reset/ltc2952-poweroff.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/power/reset/ltc2952-poweroff.c b/drivers/power/reset/ltc2952-poweroff.c
> index fbb344353fe4..65d9528cc989 100644
> --- a/drivers/power/reset/ltc2952-poweroff.c
> +++ b/drivers/power/reset/ltc2952-poweroff.c
> @@ -159,8 +159,8 @@ static void ltc2952_poweroff_kill(void)
>
> static void ltc2952_poweroff_default(struct ltc2952_poweroff *data)
> {
> - data->wde_interval = 300L * 1E6L;
> - data->trigger_delay = ktime_set(2, 500L*1E6L);
> + data->wde_interval = 300L * NSEC_PER_MSEC;
> + data->trigger_delay = ktime_set(2, 500L * NSEC_PER_MSEC);
>
> hrtimer_init(&data->timer_trigger, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
> data->timer_trigger.function = ltc2952_poweroff_timer_trigger;
>
> base-commit: d4439a1189f93d0ac1eaf0197db8e6b3e197d5c7
> --
> 2.34.0.rc0
>
Hi Sebastian,
On Tue, Nov 16, 2021 at 03:26:14PM +0100, Sebastian Reichel wrote:
> Hi,
>
> On Fri, Nov 05, 2021 at 08:20:50AM -0700, Nathan Chancellor wrote:
> > A new commit in LLVM causes an error on the use of 'long double' when
> > '-mno-x87' is used, which the kernel does through an alias,
> > '-mno-80387' (see the LLVM commit below for more details around why it
> > does this).
> >
> > drivers/power/reset/ltc2952-poweroff.c:162:28: error: expression requires 'long double' type support, but target 'x86_64-unknown-linux-gnu' does not support it
> > data->wde_interval = 300L * 1E6L;
> > ^
> > drivers/power/reset/ltc2952-poweroff.c:162:21: error: expression requires 'long double' type support, but target 'x86_64-unknown-linux-gnu' does not support it
> > data->wde_interval = 300L * 1E6L;
> > ^
> > drivers/power/reset/ltc2952-poweroff.c:163:41: error: expression requires 'long double' type support, but target 'x86_64-unknown-linux-gnu' does not support it
> > data->trigger_delay = ktime_set(2, 500L*1E6L);
> > ^
> > 3 errors generated.
> >
> > This happens due to the use of a 'long double' literal. The 'E6' part of
> > '1E6L' causes the literal to be a 'double' then the 'L' suffix promotes
> > it to 'long double'.
> >
> > There is no visible reason for floating point values in this driver, as
> > the values are only assigned to integer types. Use NSEC_PER_MSEC, which
> > is the same integer value as '1E6L', to avoid changing functionality but
> > fix the error.
> >
> > Fixes: 6647156c00cc ("power: reset: add LTC2952 poweroff driver")
> > Link: https://github.com/ClangBuiltLinux/linux/issues/1497
> > Link: https://github.com/llvm/llvm-project/commit/a8083d42b1c346e21623a1d36d1f0cadd7801d83
> > Signed-off-by: Nathan Chancellor <[email protected]>
> > Reviewed-by: Nick Desaulniers <[email protected]>
> > ---
>
> Thanks, queued to power-supply's fixes branch.
Is there a timeline for getting this to Linus, since it has been a month
and a half now since this was applied? It breaks several x86_64 configs
with tip of tree clang, which harms our testing of the upcoming release,
and other people are now tripping over this and reporting it on the LLVM
issue tracker:
https://github.com/llvm/llvm-project/issues/52924
Also, I noticed that your fixes branch is not flowing into linux-next,
meaning this is not fixed there either. Would it be possible for you to
ask Stephen Rothwell to add it?
Cheers,
Nathan
> >
> > v1 -> v2: https://lore.kernel.org/r/[email protected]/
> >
> > * A separate review pointed out that NSEC_PER_MSEC is a better choice
> > than USEC_PER_SEC because ktime_t is nanoseconds and the few functions
> > that take these values work in nanoseconds. The value is the same but
> > the documentation is better.
> >
> > * Pick up Nick's review tag.
> >
> > drivers/power/reset/ltc2952-poweroff.c | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/power/reset/ltc2952-poweroff.c b/drivers/power/reset/ltc2952-poweroff.c
> > index fbb344353fe4..65d9528cc989 100644
> > --- a/drivers/power/reset/ltc2952-poweroff.c
> > +++ b/drivers/power/reset/ltc2952-poweroff.c
> > @@ -159,8 +159,8 @@ static void ltc2952_poweroff_kill(void)
> >
> > static void ltc2952_poweroff_default(struct ltc2952_poweroff *data)
> > {
> > - data->wde_interval = 300L * 1E6L;
> > - data->trigger_delay = ktime_set(2, 500L*1E6L);
> > + data->wde_interval = 300L * NSEC_PER_MSEC;
> > + data->trigger_delay = ktime_set(2, 500L * NSEC_PER_MSEC);
> >
> > hrtimer_init(&data->timer_trigger, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
> > data->timer_trigger.function = ltc2952_poweroff_timer_trigger;
> >
> > base-commit: d4439a1189f93d0ac1eaf0197db8e6b3e197d5c7
> > --
> > 2.34.0.rc0
> >