Unsure if already aware, but was investigating a build failure with 545.29.06 and the upcoming gcc-14 which enables -Werror=incompatible-pointer-types by default (regardless of CFLAGS), failing with:
kernel/nvidia-drm/nvidia-drm-gem.c:115:16: error: initialization of 'int (*)(struct drm_gem_object *, struct iosys_map *)' from incompatible pointer type 'void * (*)(struct drm_gem_object *)' [-Wincompatible-pointer-types]
115 | .vmap = nv_drm_gem_prime_vmap,
| ^~~~~~~~~~~~~~~~~~~~~
And upon closer look the issue is not there, but rather in conftest.sh’s:
CODE="
#include <drm/drm_gem.h>
int conftest_drm_gem_object_vmap_has_map_arg(
struct drm_gem_object *obj, struct dma_buf_map *map) {
return obj->funcs->vmap(obj, map);
}"
compile_check_conftest "$CODE" "NV_DRM_GEM_OBJECT_VMAP_HAS_MAP_ARG" "" "types"
…resulting in (if remove the /dev/null):
conftest4394.c: In function 'conftest_drm_gem_object_vmap_has_map_arg':
conftest4394.c:24:46: error: passing argument 2 of 'obj->funcs->vmap' from incompatible pointer type [-Wincompatible-pointer-types]
24 | return obj->funcs->vmap(obj, map);
| ^~~
| |
| struct dma_buf_map *
conftest4394.c:24:46: note: expected 'struct iosys_map *' but argument is of type 'struct dma_buf_map *'
…and in kernel/nvidia-drm/nvidia-drm-gem.c there’s a comment which explains what’s happening:
/*
* The 'dma_buf_map' structure is renamed to 'iosys_map' by the commit
* 7938f4218168 ("dma-buf-map: Rename to iosys-map").
*/
#if defined(NV_LINUX_IOSYS_MAP_H_PRESENT)
typedef struct iosys_map nv_sysio_map_t;
#else
typedef struct dma_buf_map nv_sysio_map_t;
#endif
conftest.sh does not know about this rename. I’ve built this against the 6.6.11 linux kernel but should reproducible with any >5.18 kernel that has this rename.
Still some time before gcc-14 releases and spreads to distros, but would be good to get this fixed before it becomes a problem. This affects other driver branches too (at least 535.146.02, have not tried older ones).
There could potentially be other tests that need reviewing with gcc-14 (considering the >/dev/null 2>&1 silencing), albeit it’s the only one that resulted in a straight up build failure for me.
As a quickfix, I’ve done this locally (fortunately seems to know about this define during the test):
--- a/kernel/conftest.sh
+++ b/kernel/conftest.sh
@@ -5071,6 +5071,11 @@
CODE="
#include <drm/drm_gem.h>
+ #if defined(NV_LINUX_IOSYS_MAP_H_PRESENT)
+ typedef struct iosys_map nv_sysio_map_t;
+ #else
+ typedef struct dma_buf_map nv_sysio_map_t;
+ #endif
int conftest_drm_gem_object_vmap_has_map_arg(
- struct drm_gem_object *obj, struct dma_buf_map *map) {
+ struct drm_gem_object *obj, nv_sysio_map_t *map) {
return obj->funcs->vmap(obj, map);
}"