On 2/12/19 06:25, Kalle Valo wrote:
Pablo Greco <pgreco@xxxxxxxxxxxxxxxxx> writes:
Some compilers (tested with 4.8.5 from CentOS 7) fail properly process
FIELD_GET inside an inline function, which ends up in a BUILD_BUG_ON.
Convert inline function to a macro.
Fixes commit bf92e7685100 ("mt76: mt7615: add support for per-chain
signal strength reporting")
Reported in https://lkml.org/lkml/2019/9/21/146
Reported-by: kbuild test robot <lkp@xxxxxxxxx>
Signed-off-by: Pablo Greco <pgreco@xxxxxxxxxxxxxxxxx>
---
drivers/net/wireless/mediatek/mt76/mt7615/mac.c | 5 +----
1 file changed, 1 insertion(+), 4 deletions(-)
diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/mac.c b/drivers/net/wireless/mediatek/mt76/mt7615/mac.c
index c77adc5d2552..77e395ca2c6a 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/mac.c
@@ -13,10 +13,7 @@
#include "../dma.h"
#include "mac.h"
-static inline s8 to_rssi(u32 field, u32 rxv)
-{
- return (FIELD_GET(field, rxv) - 220) / 2;
-}
+#define to_rssi(field, rxv) ((FIELD_GET(field, rxv) - 220) / 2)
What about u32_get_bits() instead of FIELD_GET(), would that work? I
guess chances for that is slim, but it's always a shame to convert a
function to a macro so we should try other methods first.
Anything that doesn't check field at build time should work, but between
losing a check, or turning an inline into a macro, I'd rather use the macro.
Or even better if we could fix FIELD_GET() to work with older compilers.
The problem is not FIELD_GET itself, is that the compiler is trying to
use "field" as a variable, instead as the macro expansion of GENMASK, as
if the function wasn't inline.
In the linked page you can see this message
BUILD_BUG_ON failed: (((field) + (1ULL << (__builtin_ffsll(field) - 1)))
& (((field) + (1ULL << (__builtin_ffsll(field) - 1))) - 1)) != 0
_compiletime_assert(condition, msg, __compiletime_assert_, __LINE__)
which is is not right, because "field" should never be used for that check.
Pablo.
