(I hope.)
My original instinct to make ndis_return_packet() asynchronous was correct.
Making ndis_rxeof() submit packets to the stack asynchronously fixes
one recursive spinlock acquisition, but it's also possible for it to
happen via the ndis_txeof() path too. So:
- In if_ndis.c, revert ndis_rxeof() to its old behavior (and don't bother
putting ndis_rxeof_serial() back since we don't need it anymore).
- In kern_ndis.c, make ndis_return_packet() submit the call to the
MiniportReturnPacket() function to the "ndis swi" thread so that
it always happens in another context no matter who calls it.
workaround was for hardware where the clock was not latched, not for
hardware that was too slow. Also, make variable names more specific for ddb
printing.
While I would have prefered to have a solution that didn't move
knowledge of this into the pci layer. However, this is literally the
only exception that's listed in the PCI standard to the usual way of
decoding BARs. atapci devices in legacy mode now ignore the first 4
bars and hard code the values to the legacy ide values (well, for each
of the controllers that are in legacy mode). The 5th bar is handled
normally.
Remove the zero bar handling. zero bars should be ignored at all
other times, and since we handle that specially, we don't need the
older workaround.
what the ACPI-safe workaround is intended to fix. Requested by phk.
Set the bushandle and tag when attaching the timer, don't do it each time
in read_counter(). Pointed out by bde.
Move test_counter() to the end. Staticize acpi_timer_reg.
Clearly comment the assumptions on the structure of keys (addresses)
and masks, and introduce a macro, LEN(p), to extract the size of these
objects instead of using *(u_char *)p which might be confusing.
Comment the confusion in the types used to pass around pointers
to keys and masks, as a reminder to fix that at some point.
Add a few comments on what some functions do.
Comment a probably inefficient (but still correct) section of code
in rn_walktree_from()
The object code generated after this commit is the same as before.
At some point we should also change same variable identifiers such
as "t, tt, ttt" to fancier names such as "root, left, right" (just
in case someone wants to understand the code!), replace misspelling
of NULL as 0, remove 'register' declarations that make little sense
these days.
like "the foo(4) manual page" to "foo(4)". Uniformized the remaining
instances of "manual page" and "manpage" to "man page". Uniformized
some nearby sentence breaks. Reformatted the whole paragraph containing
these changes only for DUMMYNET.
of you with other cards, please do review and test the drivers for
MP-safety and disable Giant in the interrupt routines when you are
sure of proper functionality.
wireless ever since I added the new spinlock code. Previously, I added
a special ndis_rxeof_serial() function to insure that when we receive
a packet, we never end up calling the MiniportReturnPacket() routine
until after the receive handler has finished. I set things up so that
ndis_rxeof_serial() would only be used for serialized miniports since
they depend on this property. Well, it turns out deserialized miniports
depend on a similar property: you can't let MiniportReturnPacket() be
called from the same context as the receive handler at all. The 2100B
driver happens to use a single spinlock for all of its synchronization,
and it tries to acquire it both while in MiniportHandleInterrupt() and
in MiniportReturnPacket(), so if we call MiniportReturnPacket() from
the MiniportHandleInterrupt() context, we will end up trying to acquire
the spinlock recursively, which you can't do.
To fix this, I made the ndis_rxeof_serial() handler the default. An
alternate solution would be to make ndis_return_packet() submit
the call to MiniportReturnPacket() to the NDIS task queue thread.
I may do that in the future, after I've tested things a bit more.
supported. Symptoms of this bug included unnecessary use of ACPI-safe
and a dmesg that has deltas of about 2^24:
ACPI timer looks BAD min = 2, max = 16777206, width = 16777204
ACPI timer looks BAD min = 2, max = 7, width = 5
ACPI timer looks GOOD min = 4, max = 5, width = 1
ACPI timer looks BAD min = 2, max = 16777206, width = 16777204
ACPI timer looks BAD min = 2, max = 7, width = 5
ACPI timer looks BAD min = 2, max = 16777210, width = 16777208
ACPI timer looks BAD min = 4, max = 16777189, width = 16777185
ACPI timer looks GOOD min = 4, max = 5, width = 1
ACPI timer looks BAD min = 2, max = 7, width = 5
ACPI timer looks BAD min = 4, max = 16777189, width = 16777185
To fix this:
* Use a 32 bit timecounter mask when the timer is 32 bits.
* In test_counter(), use the acpi_TimerDelta function which handles 24/32
bit timers and wraparound.
Miscellaneous fixes:
* Use C99 initializers for timecounter struct.
* Use u_int and uint32_t where appropriate instead of unsigned.
* Remove whitespace-only lines
* Remove the old PIIX4 PCI workaround. The timecounter testing code has
been in use for long enough to prove it's functional.
globally available. acpi_TimerDelta() subtracts two readings from the
ACPI PM timer and returns the difference. It properly distinguishes between
24-bit and 32-bit timers and handles wraparound.
2. Document that this means that kernel modules must be rebuilt.
3. While I'm here, fix my sorting error in callout.h
Requested by: many [1], scottl [2], bde [3]
it checked for rt == NULL after dereferencing the pointer).
We never check for those events elsewhere, so probably these checks
might go away here as well.
Slightly simplify (and document) the logic for memory allocation
in rt_setgate().
The rest is mostly style changes -- replace 0 with NULL where appropriate,
remove the macro SA() that was only used once, remove some useless
debugging code in rt_fixchange, explain some odd-looking casts.
implementation taken directly from OpenBSD.
I've resisted committing this for quite some time because of concern over
TIME_WAIT recycling breakage (sequential allocation ensures that there is a
long time before ports are recycled), but recent testing has shown me that
my fears were unwarranted.
TIME_WAIT recycling cases I was able to generate with http testing tools.
In short, as the old algorithm relied on ticks to create the time offset
component of an ISN, two connections with the exact same host, port pair
that were generated between timer ticks would have the exact same sequence
number. As a result, the second connection would fail to pass the TIME_WAIT
check on the server side, and the SYN would never be acknowledged.
I've "fixed" this by adding random positive increments to the time component
between clock ticks so that ISNs will *always* be increasing, no matter how
quickly the port is recycled.
Except in such contrived benchmarking situations, this problem should never
come up in normal usage... until networks get faster.
No MFC planned, 4.x is missing other optimizations that are needed to even
create the situation in which such quick port recycling will occur.
