RDMA/hfi1: use a struct group to avoid warning

[PATCH] RDMA/hfi1: use a struct group to avoid warning

Posted by Arnd Bergmann 10 months, 1 week ago

From: Arnd Bergmann <arnd@arndb.de>

On gcc-11 and earlier, the driver sometimes produces a warning
for memset:

In file included from include/linux/string.h:392,
                 from drivers/infiniband/hw/hfi1/mad.c:6:
In function 'fortify_memset_chk',
    inlined from '__subn_get_opa_hfi1_cong_log' at drivers/infiniband/hw/hfi1/mad.c:3873:2,
    inlined from 'subn_get_opa_sma' at drivers/infiniband/hw/hfi1/mad.c:4114:9:
include/linux/fortify-string.h:480:4: error: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Werror]
    __write_overflow_field(p_size_field, size);
    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This seems to be a false positive, and I found no nice way to rewrite
the code to avoid the warning, but adding a a struct group works.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/infiniband/hw/hfi1/hfi.h | 6 ++++--
 drivers/infiniband/hw/hfi1/mad.c | 4 ++--
 2 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index cb630551cf1a..fca37eb167cc 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -883,8 +883,10 @@ struct hfi1_pportdata {
 	 * cc_log_lock protects all congestion log related data
 	 */
 	spinlock_t cc_log_lock ____cacheline_aligned_in_smp;
-	u8 threshold_cong_event_map[OPA_MAX_SLS / 8];
-	u16 threshold_event_counter;
+	struct_group (zero_event_map,
+		u8 threshold_cong_event_map[OPA_MAX_SLS / 8];
+		u16 threshold_event_counter;
+	);
 	struct opa_hfi1_cong_log_event_internal cc_events[OPA_CONG_LOG_ELEMS];
 	int cc_log_idx; /* index for logging events */
 	int cc_mad_idx; /* index for reporting events */
diff --git a/drivers/infiniband/hw/hfi1/mad.c b/drivers/infiniband/hw/hfi1/mad.c
index b39f63ce6dfc..0dea8d01e868 100644
--- a/drivers/infiniband/hw/hfi1/mad.c
+++ b/drivers/infiniband/hw/hfi1/mad.c
@@ -3870,8 +3870,8 @@ static int __subn_get_opa_hfi1_cong_log(struct opa_smp *smp, u32 am,
 	 * Reset threshold_cong_event_map, and threshold_event_counter
 	 * to 0 when log is read.
 	 */
-	memset(ppd->threshold_cong_event_map, 0x0,
-	       sizeof(ppd->threshold_cong_event_map));
+	memset(&ppd->zero_event_map, 0x0,
+	       sizeof(ppd->zero_event_map));
 	ppd->threshold_event_counter = 0;
 
 	spin_unlock_irq(&ppd->cc_log_lock);
-- 
2.39.5

Re: [PATCH] RDMA/hfi1: use a struct group to avoid warning

Posted by Jason Gunthorpe 10 months ago

On Thu, Apr 03, 2025 at 04:47:53PM +0200, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> On gcc-11 and earlier, the driver sometimes produces a warning
> for memset:
> 
> In file included from include/linux/string.h:392,
>                  from drivers/infiniband/hw/hfi1/mad.c:6:
> In function 'fortify_memset_chk',
>     inlined from '__subn_get_opa_hfi1_cong_log' at drivers/infiniband/hw/hfi1/mad.c:3873:2,
>     inlined from 'subn_get_opa_sma' at drivers/infiniband/hw/hfi1/mad.c:4114:9:
> include/linux/fortify-string.h:480:4: error: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Werror]
>     __write_overflow_field(p_size_field, size);
>     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> 
> This seems to be a false positive, and I found no nice way to rewrite
> the code to avoid the warning, but adding a a struct group works.

Er.. so do we really want to fix it or just ignore this on gcc-11? Or
is there really a compile bug here and it is mis-generating the code?

The unneeded struct group seems ugly to me?

Jason

Re: [PATCH] RDMA/hfi1: use a struct group to avoid warning

Posted by Arnd Bergmann 10 months ago

On Mon, Apr 7, 2025, at 20:27, Jason Gunthorpe wrote:
> On Thu, Apr 03, 2025 at 04:47:53PM +0200, Arnd Bergmann wrote:
>> From: Arnd Bergmann <arnd@arndb.de>
>> 
>> On gcc-11 and earlier, the driver sometimes produces a warning
>> for memset:
>> 
>> In file included from include/linux/string.h:392,
>>                  from drivers/infiniband/hw/hfi1/mad.c:6:
>> In function 'fortify_memset_chk',
>>     inlined from '__subn_get_opa_hfi1_cong_log' at drivers/infiniband/hw/hfi1/mad.c:3873:2,
>>     inlined from 'subn_get_opa_sma' at drivers/infiniband/hw/hfi1/mad.c:4114:9:
>> include/linux/fortify-string.h:480:4: error: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Werror]
>>     __write_overflow_field(p_size_field, size);
>>     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> 
>> This seems to be a false positive, and I found no nice way to rewrite
>> the code to avoid the warning, but adding a a struct group works.
>
> Er.. so do we really want to fix it or just ignore this on gcc-11? Or
> is there really a compile bug here and it is mis-generating the code?
>
> The unneeded struct group seems ugly to me?

Having a clean build would be nice though. Do you think a patch
that just turns off the warning locally would be better?

      Arnd

Re: [PATCH] RDMA/hfi1: use a struct group to avoid warning

Posted by Leon Romanovsky 10 months ago

On Tue, Apr 08, 2025 at 02:40:38PM +0200, Arnd Bergmann wrote:
> On Mon, Apr 7, 2025, at 20:27, Jason Gunthorpe wrote:
> > On Thu, Apr 03, 2025 at 04:47:53PM +0200, Arnd Bergmann wrote:
> >> From: Arnd Bergmann <arnd@arndb.de>
> >> 
> >> On gcc-11 and earlier, the driver sometimes produces a warning
> >> for memset:
> >> 
> >> In file included from include/linux/string.h:392,
> >>                  from drivers/infiniband/hw/hfi1/mad.c:6:
> >> In function 'fortify_memset_chk',
> >>     inlined from '__subn_get_opa_hfi1_cong_log' at drivers/infiniband/hw/hfi1/mad.c:3873:2,
> >>     inlined from 'subn_get_opa_sma' at drivers/infiniband/hw/hfi1/mad.c:4114:9:
> >> include/linux/fortify-string.h:480:4: error: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Werror]
> >>     __write_overflow_field(p_size_field, size);
> >>     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> >> 
> >> This seems to be a false positive, and I found no nice way to rewrite
> >> the code to avoid the warning, but adding a a struct group works.
> >
> > Er.. so do we really want to fix it or just ignore this on gcc-11? Or
> > is there really a compile bug here and it is mis-generating the code?
> >
> > The unneeded struct group seems ugly to me?
> 
> Having a clean build would be nice though. Do you think a patch
> that just turns off the warning locally would be better?

I don't think so, as you will need to disable warning for specific
compiler, which won't be nice.

My preference is to have a fix.

Thanks

> 
>       Arnd