linuxkpi: chase spinlock_t in DECLARE_WAIT_QUEUE_HEAD

[ekhramtsov]

Remove member m to unbreak consumers, e.g. drm-kmod 5.10-lts branch.

[...]
/tmp/drm-kmod/drivers/gpu/drm/ttm/ttm_module.c:38:8: error: no member named 'm' in 'struct mtx'
   38 | static DECLARE_WAIT_QUEUE_HEAD(exit_q);
      |        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/tmp/src/sys/compat/linuxkpi/common/include/linux/wait.h:113:33: note: expanded from macro 'DECLARE_WAIT_QUEUE_HEAD'
  113 |         MTX_SYSINIT(name, &(name).lock.m, spin_lock_name("wqhead"), MTX_DEF)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/tmp/src/sys/sys/mutex.h:528:4: note: expanded from macro 'MTX_SYSINIT'
  528 |                 (mtx),                                                  \
      |                  ^~~
/tmp/drm-kmod/drivers/gpu/drm/ttm/ttm_module.c:38:8: error: no member named 'm' in 'struct mtx'
   38 | static DECLARE_WAIT_QUEUE_HEAD(exit_q);
      |        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/tmp/src/sys/compat/linuxkpi/common/include/linux/wait.h:113:33: note: expanded from macro 'DECLARE_WAIT_QUEUE_HEAD'
  113 |         MTX_SYSINIT(name, &(name).lock.m, spin_lock_name("wqhead"), MTX_DEF)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[...]

Fixes: ae38a1a ("linuxkpi: spinlock: Simplify code")
MFC after: ?

[evadot]

This isn't correct, this will break all other drm-kmod version. Also 5.10 isn't supported for main and stable/14, please use drm-515-kmod or drm-61-kmod.

[evadot]

Wait sorry I'm misread the patch, it's correct.

[bsdimp]

Merged 84f4633

[ekhramtsov]

stable/14 is still affected and expected EoL of stable/14 is November 30, 2028.

Also according to FreshPorts, graphics/drm-510-kmod is available for i386 unlike graphics/drm-515-kmod, also freebsd/freebsd-ports@5eb743945303.

[ekhramtsov]

stable/14 code slush is expected on October 18 and releng/14.2 branch is expected on November 1, see schedule.

No one has yet volunteered to fix drm-kmod 6.1 for non-amd64 as far as can the package section of FreshPorts tell.

Isn't it early for i386 to lose modesetting X.Org given that stable/15 is likely to lose agp(4) thus drm-kmod could become useless on AGP-era machines and x11-drivers being dropped, thus causing revert to VESA which may be inadequate possible due to loss of resolution and performance?

drm-kmod starting with tag drm_v5.11_0 (bisected to) also includes the new ttmv3 allocator which uses phys contig allocations but lacks bus_dma(9) for DMA allocations, resulting in stalls due to page reclaim taking long with enough free memory fragmentation, which makes a modern desktop that uses DRM drivers with TTM unusable after a long uptime or several poudriere runs. Until bus_dma(9) for ttmv3 is implemented only 5.10 remains viable as a workaround and users are unlikely to use -CURRENT just for 5.10.

@evadot, what is your stance on drm-kmod 5.10 for FreeBSD 14?