[Bug 282479] LinuxKPI: <linux/lockdep.h> lockdep_is_held() user doesn't build after LLVM 19 import

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=282479

--- Comment #4 from Evgenii Khramtsov <2khramtsov@gmail.com> ---
(In reply to Bjoern A. Zeeb from comment #3)

> Also if you have no DEBUG kernel (or no INVARIANTS) *__m may not be used in a function but only declared, which leads to an error.  Memory serving me right this can happen in iwlwifi.

Comment #1 says that -CURRENT with both GENERIC and GENERIC-NODEBUG kernels
(that include iwlwifi) build fine after removing the __diagused macro (which
expands to the __unused attribute) from the parameter list. I didn't
investigate further.

> It is not clear where your sources are from and whether you are building the modules with the same options as your kernel or with the kernel (given your sources seem to be in /tmp).

Comment #1 says -CURRENT and drm-kmod 5.10-lts branch:

base 8d6feaaaa26f (/usr/src) and
https://github.com/freebsd/drm-kmod/commit/17498c322eea (/tmp/drm-kmod)

cd /tmp/drm-kmod && make -j$(nproc)

> There were other compile-time fixes for drm-kmod committed a few days ago.

"path does not exist" types of warning fixes do not apply to 5.10-lts

> Sorry but 5.10 is not for current.  Assuming amd64 please use 6.1-lts or simply ports they'll do the right thing and 6.1-lts compiles (and wulf tried master he said; see review of the commit).

drm-kmod >=5.11 with the new ttmv3 allocator is unusable on amdgpu as it lacks
bus_dma(9) for DMA allocations, and instead requests 4 MB phys contig
allocations without a scatter gather which become a burden after enough free
phys mem fragmentation (prolonged uptime or poudriere building) see
https://github.com/freebsd/drm-kmod/issues/302#issuecomment-2318230263
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=277476#c0

"The right thing" is subjective, please don't give unsolicited advice.

-- 
You are receiving this mail because:
You are the assignee for the bug.