don't increment scatterlist length twice
this occurs as sg_dma_len() returns the length member of struct scatterlist
where as on x86 linux it returns a dma_length member of the struct
Problem reported by Ryan Fahy in FreeBSD drm-kmod PR 468.
Avoids a 'Data modified on freelist' panic on boot when using discrete
Intel cards (DG2). DG2 has other issues, so remains disabled for now.
revert scatterlist dma_length changes
deraadt reports this caused 'flip_done timed out', 'commit wait timed out'
and a black screen when starting X on meteor lake
cpp: no longer default to -traditional
calendar(1) and xrdb(1) that use cpp as a generic macro processor
already call "/usr/libexec/cpp -traditional".
ok jsg@ deraadt@
sys/vfs: discard buffers after vclean error
vclean() ignored vinvalbuf(V_SAVE) failures and continued into
VOP_RECLAIM(). If dirty buffer writeback failed, the vnode could be
reclaimed and later reused while old buffers still referenced the same
struct vnode identity.
Report the failed flush, then force buffer invalidation before reclaim
proceeds. vclean() has no failure channel; discarding dirty buffers may
lose data, but reusing the vnode with stale buffers is worse, and a
panic would risk unrelated dirty state.
OK claudio@
create_priority_list: add bounds check for priority_len
Overflow should not be possible since we deduplicate the options
but an explicit check won't hurt. Also move the dedup check for
options 121 and 245 right before they are stored to match the other
checks. OK deraadt@ florian@
sndiod: Add the server.thru control to midithru/N ports
It defines if programs connected to the midithru/N port can see each
other. It defaults to 1, which is the current behavior.
sndiod: Use pointers to midithru structures instead of indices
This makes handling of the midithru structure more similar to others,
making the code more readable. No behavior change.
sndiod: Implement control of midithru ports with sndioctl
For now there's a single new server.port control that specifies which
hardware MIDI ports programs will see. It's the equivalent of what's
the server.device control is for audio.
SSL_alert_desc_string: update manpage after change to return "!!" always
Also change some totally unreadable roff macros to slightly less unreadable
mdoc markup.
The mwx_delete_key() callback from net80211 can't use a task.
While adding a key must be done via task to be able to sleep, the
delete key operation must operate in fire and forget mode because
defering the work results in a use-after-free (or at least bzero).
Because of this split the functions to update and remove a key
and use the no wait msg sending function in the delete case.
This fixes a panic on interface down.
