in the arm __swp() and sparc64 casa() and casax() functions is actually
being used as an input and output and not just the value of the register
that points to the memory location. This was the underlying source of
the mbuf refcount problems on sparc64 a while back. For arm this should be
a nop because __swp() has a constraint to clobber all memory which can
probably be removed now.
Reviewed by: alc, cognet
MFC after: 1 week
variables rather than void * variables. This makes it easier and simpler
to get asm constraints and volatile keywords correct.
MFC after: 3 days
Tested on: i386, alpha, sparc64
Compiled on: ia64, powerpc, amd64
Kernel toolchain busted on: arm
address, writting non-canonical address can cause kernel a panic,
by restricting base values to 0..VM_MAXUSER_ADDRESS, ensuring
only canonical values get written to the registers.
Reviewed by: peter, Josepha Koshy < joseph.koshy at gmail dot com >
Approved by: re (scottl)
- Let creator_bitblt() return ENODEV as it's not implemented (missed
in sys/dev/fb/creator.c rev. 1.6).
- As a speed optimization inline the creator_ras_wait() etc. helper
functions and also cache setting the font increment, font width
and plane mask. [1]
- I got the meaning of V_DISPLAY_BLANK wrong, it's blank like turn
off and not blank like turn on and clear the screen. So move
clearing the screen to creator_clear() were it hopefully belongs.
- Properly implement V_DISPLAY_BLANK, V_DISPLAY_STAND_BY and
V_DISPLAY_SUSPEND. This makes blank_saver.ko and green_saver.ko
work. [1]
- Change the order of operations in creator_fill_rect(), i.e. write
y before x and cy before cx. This fixes drawing the top part of
the border with Elite3D cards when switching from Xorg to a VTY.
- Move setting the chip configuration we use and invalidating the
cache variables to creator_set_mode() and set the V_ADP_MODECHANGE
flag. This causes creator_set_mode() to be called when the X server
shuts down which fixes the screen corruption caused most of the
time by Xorg not restoring the original configuration present at
startup.
Inspired by/based on: Xorg [1]
Approved by: re (scottl)
psm(4), ukbd(4), ums(4) and usb(4) on by default. Modulo some nits with
the most annoying one probably being USB keyboards no longer working at
the OFW boot prompt after halting FreeBSD these drivers work fine on
sparc64 including X and there's nothing left that I'd consider a show-
stopper. I.e. graphical consoles on sun4u machines should either work
out of the box or by plugging in a card that is supported by either
creator(4) or machfb(4). The exception obviously are SBus-only machines
without UPA slots like some Ultra 1 (but which also still lack support
in other areas) and certain Exx0 (but which probably are mainly used
with serial consoles anyway). I'll try to add a cgsix(4) for these later
as Sun CG6 cards are probably the most common SBus framebuffer cards in
sun4u machines. I however don't see much sense in adding drivers for the
dozen of SBus framebuffers that were destined for sparc v8 machines.
The rest of the USB drivers aren't enabled as I'm only aware of ukbd(4)
and ums(4) as well as ohci(4) working with the on-board ALI M5237 and
Sun PCIO-2 controllers. Aue(4) definitely doesn't work on sparc64, yet.
Thanks to:
- Jake for the initial work on syscons(4) on sparc64 and creator(4).
- Marcel for uart(4) and especially for its support for the SCCs which
are only used on sparc64 so far. In various regards it wouldn't have
been possible to enable syscons(4) by default on sparc64, yet, without
uart(4).
- All that tested patches.
Ok'ed by: scottl (RE hat), tmm
vm_page's machine-dependent fields. Use this function in
vm_pageq_add_new_page() so that the vm_page's machine-dependent and
machine-independent fields are initialized at the same time.
Remove code from pmap_init() for initializing the vm_page's
machine-dependent fields.
Remove stale comments from pmap_init().
Eliminate the Boolean variable pmap_initialized from the alpha, amd64,
i386, and ia64 pmap implementations. Its use is no longer required
because of the above changes and earlier changes that result in physical
memory that is being mapped at initialization time being mapped without
pv entries.
Tested by: cognet, kensmith, marcel
- Implement sampling modes and logging support in hwpmc(4).
- Separate MI and MD parts of hwpmc(4) and allow sharing of
PMC implementations across different architectures.
Add support for P4 (EMT64) style PMCs to the amd64 code.
- New pmcstat(8) options: -E (exit time counts) -W (counts
every context switch), -R (print log file).
- pmc(3) API changes, improve our ability to keep ABI compatibility
in the future. Add more 'alias' names for commonly used events.
- bug fixes & documentation.
mutex instead of a MTX_DEF one in order to defer preemption while
reading the date and time registers. If we don't manage to read them
within the time slot where we are guaranteed that no updates occur we
might actually read them during an update in which case the output is
undefined.
the number of registered adapters instead of determining again whether
stdout is a supported card (and which might have failed to attach and
register).
- Drop creator_set_mode() and move the relevant parts to creator_fill_rect()
and creator_putc() respectively. This is a bit cleaner than having to
make sure that creator_set_mode() was called before creator_fill_rect()
or creator_putc() are used and matches better what Xorg does.
- Fix a bug in the handling of the FBIOSCURSOR IOCTL; the code was meant
to return ENODEV for all invocations expect when used to disable the
cursor and not just when used for enabling the cursor.
- In case the adapter is the OFW stdout move its OFW cursor to the start
of the last line on halt so OFW output doesn't get intermixed with what
FreeBSD left on the screen. With hindsight this is what the faking of a
hardware cursor which was removed in the last revision really was about,
i.e. to keep the OFW updated about the current cursor position. The new
approach however is simpler while producing the same result and doesn't
cause the first letter of the OFW output to be turned into a blank and
a newline.
- Add variable names to the prototypes of creator_cursor_*() which were
added in the last revision and list them alphabetically in order to match
the style of this file.
for the SYS_RES_IOPORT -> SYS_RES_MEMORY transition again. While it
was helpful to not need to change all of the affected drivers in a
single pass together with ebus(4) we probably shouldn't start into
6.0 with such a hack.
This requires some of the modules of affected drivers to be rebuilt,
namely: auxio(4), snd_audiocs(4) and puc(4).
by default, yet.
- Replace "graphics cards" with "framebuffers" in the description
of creator(4) in order to make it uniform with the description of
machfb(4) and the latter occur both on-board and as add-on cards.
- Use register macros instead of magic values in the code. [1]
- Check the return values of OF_getprop() and other stuff that actually
can fail.
- Let the unimplemented video driver methods return ENODEV rather
than 0 so other code isn't tricked into thinking a certain operation
was successfull. In case of e.g. the video driver creator_ioctl()
this caused vidcontrol(1) to return random garbage information.
Remove the TODO macros in the unimplemented video driver methods
which did a printf("%s: unimplemented\n", __func__). Under certain
circumstances these managed to invoke a printf() when a low-level
console device wasn't attached, yet, causing a Fast Data Access MMU
Miss. These macros were only really usefull for development anyway.
- Set the struct video_adapter and struct video_info va_flags and
vi_flags etc. as appropriate.
- In creator_configure() don't rely on hitting the node which is the
chosen console device first when searching the OFW tree for adapters
compatible with this driver. Instead just check whether the chosen
console device is a viable target for this driver. Targets that are
not the console (including additional cards in multi-head configs)
will be attached through creator_upa_attach(). I think this how the
code in creator_configure() was actually meant to work.
Honour the VIO_PROBE_ONLY flag and don't initialise and register the
console device twice when creator_configure() is called a second time
during sc_probe_unit().
Let creator_configure() return the number of the found adapters,
i.e. 1 in case probing succeeds, as it's expected. The return values
of video adapter configure functions however currently aren't checked
so this doesn't make a difference at the moment.
- In creator_upa_attach() don't rely on probing and attaching the
adapter which is the console first, in case there are multiple
adpaters and one of them is the console this could lead into using
the video adapter unit 0 twice.
- Make the check for DACs with inverted cursor control a bit more
precise and actually honour that information when turning the cursor
on or off. Add a helper function creator_cursor_enable() for this
in order to keep code duplication low. [1]
- Don't bother with faking a hardware cursor in case a device is the
console. Apparently this was meant to start kernel output right after
where the firmware left. In general this isn't worth the fuzz and
also had no real effect as creator_set_mode() did clear the screen
in any case, not just in case a device was not the console.
- Implement creator_fill_rect() and use it to actually blank the
display in creator_blank_display() when the mode is V_DISPLAY_BLANK,
moving blanking the display out of creator_set_mode(). Use it also
to implement creator_set_border() so the border can be re-drawn
when switching to a VTY from X, exiting X, etc. (which leaves us
with a black border most of the time).
- Implement the video driver creator_ioctl(), moving the implementation
of the IOCTL interface from the fbN CDEV version of creator_ioctl()
into the video driver version and use the latter to implement the
former. Use fb_commonioctl() to handle most of the FBIO IOCTLs.
This gives programs like vidcontrol(1) which use the video driver
creator_ioctl() a chance of working.
Implement turning off the cursor via the FBIOSCURSOR IOCTL, which
Xorg uses to in order to inform the OS that it's taking over the
cursor. In creator_putm() check whether the cursor is enabled and
(re-)install it if necessary, moving installing the cursor out of
creator_init() and into a helper function creator_cursor_install().
This fixes the missing mouse pointer when switching to a VTY from X,
exiting X, etc.
- Some clean-up (remove unused/useless code, etc.).
o sparc64/creator/creator_upa.c / sparc64/sparc64/sc_machdep.c:
- Attach syscons(4) as an own pseudo-device on the nexus rather than
directly in creator_upa_attach(), similiar to attaching syscons(4)
as a pseudo-device on isa(4) on other archs. This makes it a whole
lot easier to do the right thing in multi-head configs, especially
with different types of graphics adapters. [2]
- Set SC_AUTODETECT_KBD by default so USB keyboards work out of the
box. [2]
Based on/obtained from: Xorg 'ffb' driver [1]
Based on/obtained from: FreeBSD/powerpc [2]
Use bus_generic_probe() and add a bus_add_child() interface method to
allow device drivers to use the identify method to add themselves if
need be (e.g. syscons(4)).
- Use FBSDID.
consist of the expected number of address and size cells (we can't use
dynamic arrays here because at the point in the boot process when this
code is used malloc() doesn't work, yet). This fixes a Fast Data Access
MMU Miss when uart(4) (erroneously) calls OF_decode_addr() to decode
the address of PS/2 keyboards. PS/2 keyboards use a different and also
undocumented scheme at the first parent node than mapping at 'ranges'
properties. It's however not worth implementing that other scheme and
actually also fits atkbdc(4) better to just start at the first parent
node of PS/2 keyboards which is the 8042 controller (I have atkbdc(4)
working that way).
- Use FBSDID.
MFC after: 1 month
- Add locking.
- Account for if the MC146818_NO_CENT_ADJUST flag is set we don't need
to check wheter year < POSIX_BASE_YEAR.
- Add some comments about mapping the day of week from the range the
generic clock code uses to the range the chip uses and which I meant
to add in the initial version.
- Minor clean-up, use __func__ instead of hardcoded function names in
error strings.
o in the rtc(4) front-end additionally:
- Don't leak resources in case mc146818_attach() fails.
- Account for ebus(4) defaulting to SYS_RES_MEMORY for the memory
resources since ebus.c rev. 1.22.
- Add support for storing the century in MK48TXX_WDAY_CB on MK48Txx with
extended registers when the MK48TXX_NO_CENT_ADJUST flag is set (and which
is termed somewhat confusing as it actually means don't manually adjust
the century in the driver).
- Add the MI part of interfacing the watchdog functionality of MK48Txx with
extended registers with watchdog(9). This is inspired by the SunOS/Solaris
drivers for the 'eeprom' devices also having watchdog support. I actually
expected this to work out of the box on Sun Exx00 machines with 'eeprom'
devices which have a 'watchdog-enable' property. On terminal count of the
the watchdog timer however only the MK48TXX_FLAGS_WDF bit rises but the
reset signal and the interrupt respectively (depending on whether the
MK48TXX_WDOG_WDS bit of the chip and the MK48TXX_WDOG_ENABLE_WDS flag
of the driver respectively is set) goes nowhere. Apparently passing the
reset signal on to the WDR line of the CPUs has to be enabled somewhere
else but we don't have documentation for the Exx00 specific controllers.
I decided to commit this nevertheless so it can be enabled in the eeprom(4)
front-end later in e.g. 6.0-STABLE without breaking the API. Besides the
Exx00 the watchdog part of the MK48Txx should also work on E250 and E450.
Possibly also without extra fiddling on these machines but I haven't
found someone willing to give it a try on such a machine so far.
- Use uintXX_t instead of u_intXX_t, use __func__ instead of hardcoded
function names in error strings.
eeprom_ebus_attach() and eeprom_sbus_attach() into eeprom_attach()
respectively. Since the introduction of the ofw_bus interface some
time ago and now that ebus(4) also uses SYS_RES_MEMORY for the
memory resources since ebus.c rev. 1.22 there is no longer a
need to have separate front-ends for ebus(4), fhc(4) and sbus(4).
- Fail gracefully instead of panicing when the model can't be
determined.
- Don't leak resources when mk48txx_attach() fails.
- Use FBSDID.
compatibility with ISA devices while in fact all known EBus devices
actually use memory space turned out to be not a good idea as so far
there is only the 'rtc' device known to show up either on an EBus or
ISA bus but not on any of the other busses used on sparc64. However
there are quite a couple of them that show up on either EBus, FireHose
or SBus. In order to save extra code in the respective drivers switch
ebus(4) to actually use SYS_RES_MEMORY for the memory resources of
its children. At least for transition still accept SYS_RES_IOPORT
and silently change it to SYS_RES_MEMORY. [1]
- In ebus_probe() use ofw_bus_get_name() instead of re-implementing it
via ofw_bus_get_node() and OF_getprop().
- Remove some unused variables.
- Use FBSDID.
Discussed with: tmm (some time ago)
the iteration variable as the RID when adding the respective resource
to the child via bus_set_resource(). In case a device has both I/O
and memory resources this generates gaps in the newbus resources of
the child, e.g. its first memory resource might end up as RID 1.
To solve this mimic resource_list_add_next() via resource_list_find()
and bus_set_resource(); we can't just use resource_list_add_next()
here as this would circumvent the limit checks in isa_set_resource()
of the common ISA code.
This however is more or less a theoretical problem so far as all known
ISA devices on sparc64 soley use I/O space.
- Just use bus_generic_rl_release_resource() for isa_release_resource()
instead of re-implementing the former.
- Improve some comments to better reflect reality, minor clean-up and
simplifications, return NULL instead of 0 were appropriate.
front-end and the LSI64854 and NCR53C9x code in case one of these
functions fails. Add detach functions to these parts and make esp(4)
detachable.
- Revert rev. 1.7 of esp_sbus.c, since rev. 1.34 of sbus.c the clockfreq
IVAR defaults to the per-child values.
- Merge ncr53c9x.c rev. 1.111 from NetBSD (partial):
On reset, clear state flags and the msgout queue.
In NetBSD code to notify the upper layer (i.e. CAM in FreeBSD) on reset
was also added with this revision. This is believed to be not necessary
in FreeBSD and was not merged.
This makes ncr53c9x.c to be in sync with NetBSD up to rev. 1.114.
- Conditionalize the LSI64854 support on sbus(4) only instead of sbus(4)
and esp(4) as it's also required for the 'dma', 'espdma' and 'ledma'
busses/devices as well as the 'SUNW,bpp' device (printer port) which
all hang off of sbus(4).
- Add a driver for the 'dma', 'espdma' and 'ledma' (pseudo-)busses/
devices. These busses and devices actually represent the LSI64854 DMA
engines for the ESP SCSI and LANCE Ethernet controllers found on the
SBus of Ultra 1 and SBus add-on cards. With 'espdma' and 'ledma' the
'esp' and 'le' devices hang off of the respective DMA bus instead of
directly from the SBus. The 'dma' devices are either also used in this
manner or on some add-on cards also as a companion device to an 'esp'
device which also hangs off directly from the SBus. With the latter
variant it's a bit tricky to glue the DMA engine to the core logic of
the respective 'esp' device. With rev. 1.35 of sbus.c we are however
guaranteed that such a 'dma' device is probed before the respective
'esp' device which simplifies things a lot. [1]
- In the esp(4) SBus front-end read the part-unique ID code of Fast-SCSI
capable chips the right way. This fixes erroneously detecting some
chips as FAS366 when in fact they are not. Add explicit checks for the
FAS100A, FAS216 and FAS236 variants instead treating all of these as
ESP200. That way we can correctly set the respective Fast-SCSI config
bits instead of driving them out of specs. This includes adding the
FAS100A and FAS236 variants to the NCR53C9x core code. We probably
still subsume some chip variants as ESP200 while in fact they are
another variant which however shouldn't really matter as this will
only happen when these chips are driven at 25MHz or less which implies
not being able to run Fast-SCSI. [3]
- Add a workaround to the NCR53C9x interrupt handler which ignores the
stray interrupt generated by FAS100A when doing path inquiry during
boot and which otherwiese would trigger a panic.
- Add support for the 'esp' devices hanging off of a 'dma' or 'espdma'
busses or which are companions of 'dma' devices to esp(4). In case of
the variants that hang off of a DMA device this is a bit hackish as
esp(4) then directly uses the softc of the respective parent to talk
to the DMA engine. It might make sense to add an interface for this
in order to implement this in a cleaner way however it's not yet clear
how the requirements for the LANCE Ethernet controllers are and the
hack works for now. [2]
This effectively adds support for the onboard SCSI controller in
Ultra 1 as well as most of the ESP-based SBus add-on cards to esp(4).
With this the code for supporting the Performance Technologies SBS430
SBus SCSI add-on cards is also largely in place the remaining bits
were however omitted as it's unclear from the NetBSD how to couple
the DMA engine and the core logic together for these cards.
Obtained from: OpenBSD [1]
Obtained from: NetBSD [2]
Clue from: BSD/OS [3]
Reviewed by: scottl (earlier version)
Tested with: FSBE/S add-on card (FAS236), SSHA add-on card (ESP100A),
Ultra 1 (onboard FAS100A), Ultra 2 (onboard FAS366)
device and which also applies to the children. This is very usefull for
drivers for the various subordinate busses so they don't need to fiddle
with the OFW node of their parent themselves. As SBus busses hang of the
nexus and we don't use the ofw_bus interface for nexus devices, yet, this
would also require special knowledge about this in the drivers for the
SBus children which these shouldn't need to have.
This includes switching to use an unshifted IGN in the sc_ign member of
the sbus(4) softc internally.
- For SBus child devices where there are variants that are actually split
split into two SBus devices (as opposed to the first half of the device
being a SBus device and the second half hanging off of the first one)
like 'auxio' and 'SUNW,fdtwo' or 'dma' and 'esp' probe the SBus device
which is a prerequisite to the driver attaching to the second one with
a lower order. This saves us from dealing with different probe orders
in the respective device drivers which generally is more hackish.
- Remove a stale comment about the 'specials' array above the attaching
of the child devices. This is a remnant of the NetBSD/sparc origin of
this code. There the 'specials' array is also used to probe certain
devices which are prerequisites to others first. Why NetBSD soley
relies on the devices having the expected order in the OFW tree on
sparc64 isn't clear to me, as far as I can tell OFW doesn't guaranteed
such things.
in other codes. Add cpu_set_user_tls, use it to tweak user register
and setup user TLS. I ever wanted to merge it into cpu_set_kse_upcall,
but since cpu_set_kse_upcall is also used by M:N threads which may
not need this feature, so I wrote a separated cpu_set_user_tls.
into _bus.h to help with name space polution from including all of bus.h.
In a few days, I'll commit changes to the MI code to take advantage of thse
sepration (after I've made sure that these changes don't break anything in
the main tree, I've tested in my trees, but you never know...).
Suggested by: bde (in 2002 or 2003 I think)
Reviewed in principle by: jhb
- Merge lsi64854.c rev. 1.25 from NetBSD: nuke trailing whitespace.
- Update NetBSD RCS IDs according to what was actually already merged.
- Remove dv_name from the lsi64854_softc and use device_printf() instead.
- Use __func__ instead of hardcoded function names in error messages.
- Use ulmin() instead of min() for comparing the DMA sizes as the values
involved actually are represented by 64bit unsigned instead of 32bit
unsigned. As far as I can't tell this doesn't make a difference in
practice though.
- Some style(9) fixes (mainly indentation).
- Remove unnecessary braces.
at their old location in sys/dev/esp after they were repo-copied to
sys/sparc64/sbus at rev. 1.1:
sys/dev/esp/lsi64854.c rev. 1.2
sys/dev/esp/lsi64854var.h rev. 1.2
Add some style(9) touch ups; style(9) states that new code should follow
these conventions and, well, this is a new driver.
Tested on: i386, sparc64
Reviewed by: scottl
with the attaching of the children done in the bus attach function like
it's supposed to be.
- In the bus probe nomatch function print the resources of the children
like it's done in the other sparc64 specific bus drivers.
- For the clock frequency IVAR use the per-child values and fall back to
the bus default in case a child doesn't have the respective property
instead of always using the bus default so a child driver doesn't need
to obtain the per-child value itself (see also the commit message of
sys/dev/esp/esp_sbus.c rev. 1.7).
- Add support for pass-through allocations. The comment preceding
sbus_alloc_resource() wasn't quite correct, we need to support pass-
through allocations for the 'espdma' and 'ledma' (pseudo-)busses which
hang off of the SBus in Ultra 1 machines. There can also be actual
bridges like the SBus-to-PCMCIA bridge on the SBus and the XBox (SBus
extension box) probably also involves one.
- Use auto-generated typedefs for the prototypes of the device interface
functions.
- Style(9) fixes (mainly don't use function calls in initializers).
- Use __func__ instead of hardcoded function names in error messages.
- Try to make error messages sound uniform.
- Try to keep the code within 80 columns.
- Correct some typos.
- Correct some function declarations to match their prototypes.
- Remove unused headers, macros and variables.
- Remove a bzero() superfluous due to allocating with M_ZERO.
- Use FBSDID.
Don't use atomic ops to increment interrupt stats.
On sparc64 this reduces delay until tick interrupts are service by 1/10th
on average. In turn this reduces the clock drift caused by these delays
so there's less drift which has to be compensated in tick_hardclock().
This includes switching from atomically incrementing the global cnt.v_intr
to the asm equivalent of PCPU_LAZY_INC(cnt.v_intr) in exception.S
- Correct some comments to match the registers actually used.
- Correct some format specifiers, interrupt levels passed in are u_int.
- Use FBSDID.
Ok'ed by: jhb
- Fix NULL pointer dereferences caused when an ithread or a handler is
NULL which happens when a stray interrupt triggers after the respective
device interrupt was torn down.
- Remove the critical section around INTR_FAST handlers which actually
was a nested critical section. Both tl0_intr() and tl1_intr() already
enter a critical section for calling intr_execute_handlers().
MFC after: 3 days
call tick_stop() again after tick_init() as tick interrupts already
have been disabled as part of tick_init().
- In spinlock_enter() replace the magic value for PIL TICK with the
respective macro.
- Use FBSDID.
SpitFire erratum #54) which can cause writes to the TICK_CMPR register
to fail. This seems to fix the dying clocks problem reported by jhb@
and kris@. [1]
- In tick_start() don't reset the tick counter of the boot processor to
zero. It's initially reset in _start() and afterwards but _before_
tick_start() is called on the BSP the APs synchronise with the tick
counter of the BSP in mp_startup(). Resetting the tick counter of the
BSP in tick_start() probably also was the cause of problems seen when
using the CPU tick counter as timecounter on SMP machines.
Not resetting the tick counter of the BSP in mp_startup() makes the
tick counters and tick interrupts between the BSP and APs be pretty
much in sync as it's supposed to be. This also means there's no longer
a real reason to have separate tick_start() and tick_start_ap() so
merge them and zap tick_start_ap(). This is also a first step in
simplifying the interface to the tick counters in preparation to use
alternate clock hardware where available.
- Switch to the algorithm used on FreeBSD/ia64 for updating the tick
interrupt register and which compensates the clock drift caused by
varying delays between when the tick interrupts actually trigger and
when they are serviced. Not compensating the clock drift mainly hurts
interactive performance especially when using WITNESS. [2]
For further information about the algorithm also see the commit log
of sys/ia64/ia64/interrupt.c rev. 1.38.
On sparc64 the sysctls for monitoring the behaviour of the tick
interrupts are machdep.tick.adjust_edges, machdep.tick.adjust_excess,
machdep.tick.adjust_missed and machdep.tick.adjust_ticks.
- In tick_init() just use tick_stop() for stopping the tick interrupts
until a proper handler is set up later. This also stops the system
tick interrupt on USIII systems earlier.
- In tick_start() check for a rough upper limit of HZ.
- Some minor changes, e.g. use FBSDID, remove unused headers, etc.
Info obtained from: Linux [1]
Ok'ed by: marcel [2]
Additional testing by: kris (earlier version of the workaround), jhb
X-MFC after: 3 days [1]
disabling interrupts before updating the saved pil in the thread. If we
save the value first then it can be clobbered if an interrupt comes in
and the interrupt handler tries to acquire a spin lock.
Submitted by: marius
critical_enter() and critical_exit() are now solely a mechanism for
deferring kernel preemptions. They no longer have any affect on
interrupts. This means that standalone critical sections are now very
cheap as they are simply unlocked integer increments and decrements for the
common case.
Spin mutexes now use a separate KPI implemented in MD code: spinlock_enter()
and spinlock_exit(). This KPI is responsible for providing whatever MD
guarantees are needed to ensure that a thread holding a spin lock won't
be preempted by any other code that will try to lock the same lock. For
now all archs continue to block interrupts in a "spinlock section" as they
did formerly in all critical sections. Note that I've also taken this
opportunity to push a few things into MD code rather than MI. For example,
critical_fork_exit() no longer exists. Instead, MD code ensures that new
threads have the correct state when they are created. Also, we no longer
try to fixup the idlethreads for APs in MI code. Instead, each arch sets
the initial curthread and adjusts the state of the idle thread it borrows
in order to perform the initial context switch.
This change is largely a big NOP, but the cleaner separation it provides
will allow for more efficient alternative locking schemes in other parts
of the kernel (bare critical sections rather than per-CPU spin mutexes
for per-CPU data for example).
Reviewed by: grehan, cognet, arch@, others
Tested on: i386, alpha, sparc64, powerpc, arm, possibly more
