hppa: Limit __HAVE_ATOMIC64_OPS to userland.
And revert changes to include 64-bit atomic r/m/w operations in the
rump build, since they're not available in the hard kernel anyway.
The 64-bit CAS logic relies on 64-bit floating-point load/store (plus
RAS or hash-locking), so it doesn't work in the kernel.
Further followup to fix the hppa build after:
PR kern/60729: incosistent st_ino from the first stat(2) on a socket
ipsec: Fix fencepost error in policy parsing.
If there are no addresses appended, the input buffer length may
rightly match sizeof(*xisr) exactly.
Should fix some of the IPsec tests I just broke!
PR kern/60669: netipsec key_sp2msg buffer overrun
Pull up following revision(s) (requested by tls in ticket #483):
sys/dev/pci/if_ixl.c: revision 1.104
sys/dev/pci/if_ixl.c: revision 1.105
Be a little more conservative with DMA sync for the ATQ, though what was there already was probably OK.
Remove pointless and confusing check against NULL.
Pull up following revision(s) (requested by tls in ticket #481):
sys/dev/pci/if_ixl.c: revision 1.102
sys/dev/pci/if_ixl.c: revision 1.103
Fix link state handling in ixl driver (PR 60736):
1. The admin queue ("ATQ") polling could skip completions if there was
more than one pending command. Force ixl_atq_poll() to go in order
and simplify by looping around the post_locked and done_locked functions.
This makes ixl_atq_poll_done() superfluous, so remove that.
2. Add missing "goto out" in ixl_get_link_status() so we return EBUSY
instead of falling through to cv_timedwait(). This is much
less likely with fix 1 above but worth fixing anyway.
3. Mostly undo 1.92 - don't block waiting on link in ixl_init(), instead,
fire off a link status check and ignore the result. Meanwhile, in
ixl_init_locked(), enable the hardware link status interrupt.
[7 lines not shown]
Pull up following revision(s) (requested by jdc in ticket #480):
sys/dev/i2c/dbcool.c: revision 1.70
sys/dev/i2c/dbcool_var.h: revision 1.19
sys/dev/i2c/dbcool.c: revision 1.69
Also match "i2c-adt7475" (Sun Fire V245).
Add fan speed dividers using properties.
Pull up following revision(s) (requested by jdc in ticket #479):
share/man/man4/bq4802rtc.4: revision 1.1
sys/arch/sparc64/conf/files.sparc64: revision 1.171
sys/dev/ic/bq4802reg.h: revision 1.3
distrib/sets/lists/man/mi: revision 1.1836
distrib/sets/lists/manhtml/mi: revision 1.54
sys/arch/sparc64/dev/bq4802_ebus.c: revision 1.3
share/man/man4/Makefile: revision 1.753
Add a note about the validity of the battery-valid flag.
Add a sysctl to control the clock oscillator.
Only write to the chip at attach time if the configuration needs changing.
Add a manual page for bq4802rtc(4) (Texas Instruments bq4802Y/bq4802LY).
Add build glue (Makefile and set lists) for bq4802rtc.4 manual page.
[2 lines not shown]
Pull up following revision(s) (requested by jdc in ticket #478):
sys/dev/i2c/adt7462reg.h: revision 1.2
sys/dev/i2c/adt7462reg.h: revision 1.3
sys/dev/i2c/adt7462.c: revision 1.2
sys/dev/i2c/adt7462.c: revision 1.3
Correct the logic for tach/fan measurement.
Add tach pulse entries whilst here.
Use the pin configuration registers to discover monitored fans.
Don't display the fan fault sensor if there are no monitored fans.
Add some GPIO definitions. Tidy up unused definitions.
Add temperature offsets using properties.
Also check the tach. enable register when checking for connected fans.
Pull up following revision(s) (requested by jdc in ticket #476):
sys/dev/pci/radeonfb.c: revision 1.123
Also display the total graphics memory if it's more than 64MB. E.g.:
radeonfb0: 128 MB, 64 MB aperture at 0x10000000, 64 KB registers at 0x00100000
Pull up following revision(s) (requested by jdc in ticket #475):
sys/arch/sparc64/sparc64/ofw_patch.c: revision 1.12
sys/arch/sparc64/sparc64/ofw_patch.c: revision 1.13
sys/arch/sparc64/sparc64/ofw_patch.c: revision 1.14
sys/arch/sparc64/sparc64/ofw_patch.c: revision 1.15
sys/arch/sparc64/sparc64/ofw_patch.c: revision 1.16
Add some i2c fixes for V245 and a dsrtc fix also needed on V240.
Remove the ds1307 from the E250 patches - we can't reliably read from it.
Add environmental sensors and GPIO properties for Fire V245.
Make the V245 "disks present" ordered as 0123 instead of 3210.
Add information for sensors on V215/V245.
Add patches for incorrect values on SB2500-S dbcool and V215/V245 adt7462sm.
[5 lines not shown]
amd64: Make XSAVE area overlap mcontext_t on signal delivery.
Additionally, if the stack pointer is already such that pushing a
struct sigframe_siginfo would make it congruent to -8 modulo the
stack alignment, as the signal handler expects (-8 for the return
address on an aligned stack frame), don't re-align it again -- it's
just fine as is.
Previously, for NetBSD 11.0, we would copy out an XSAVE area
separately from ucontext_t, and copy out the FXSAVE content twice in
two different places on the stack -- one in the ucontext_t for
compatibility so older applications can still read x87/SSE registers,
and one in the XSAVE area:
+-----------------------+ \
| (other XSAVE stuff) | |
| --------------------- | |
| Hi16_ZMM | |
| --------------------- | |
[90 lines not shown]
bind - libdns - match format to arg (for clang)
Since the format took the trouble to use %hu I assumed that
an unsigned short was intended to be printed (sprintf) so converted
the arg from unsigned to unsigned short, rather than just
adopting clang's suggestion of changing the format to %u.
This might fix the clang builds.
Get rid of the wait in ixl_get_link_status, get rid of IXL_LINK_FLAG_WAITDONE;
fix another instance of the cv_timedwait() pattern noticed by riastradh@ in
PR kern/60736 . The only remaining change from revision 1.92 is the addition
of sc_link_state_done_task which avoids a lock inversion.
Fix link state handling in ixl driver (PR 60736):
1. The admin queue ("ATQ") polling could skip completions if there was
more than one pending command. Force ixl_atq_poll() to go in order
and simplify by looping around the post_locked and done_locked functions.
This makes ixl_atq_poll_done() superfluous, so remove that.
2. Add missing "goto out" in ixl_get_link_status() so we return EBUSY
instead of falling through to cv_timedwait(). This is much
less likely with fix 1 above but worth fixing anyway.
3. Mostly undo 1.92 - don't block waiting on link in ixl_init(), instead,
fire off a link status check and ignore the result. Meanwhile, in
ixl_init_locked(), enable the hardware link status interrupt.
Addresses the source of the multiple pending completions that triggered
problem 1 while still ensuring we get link status at initialization.
crypto(4): Crank down max number of ops to something reasonable.
Largest driver I found, qat(4), has at most 16384 sessions, so 64k
oughta be enough for anyone.
PR kern/60281: crypto(4): bugs in reference counting and test
