When the save folder did not exist, and we could not create
it, we would handle up to one Biobuf worth of message, and
then fail, due to a failed tee. The sequence of events leading
up to this was:
openfolder() -> error
tee(0, fd, -1) -> wait for read
write(0, data) ->
write(fd, data) -> ok
write(-1, data) -> error, tee terminates
write(0, attachment) -> error
This change prevents us from writing to a closed fd, and
therefore from erroring out when sending.
We also warn the user.
---
To: 9front@9front.org
Date: Sun, 07 Feb 2021 14:56:39 +0100
From: kvik@a-b.xyz
Subject: Re: [9front] transient dns errors cause smtp failure
Reply-To: 9front@9front.org
I think I found a reason for DNS failing on known good domains.
/sys/src/cmd/ndb/dns.h:156,157
/* tune; was 60*1000; keep it short */
Maxreqtm= 8*1000, /* max. ms to process a request */
So, 8 seconds is how much the resolver will bother with a request it
has been handed, before dropping it on the floor with little
explanation.
It seems quite possible that this is too short a timeout on a machine
during a spam queue run, which predictably stresses the compute and
network resources.
In turn, negative response caching might explain why a particular
unlucky domain would basically stop receiveing any mail for a while.
I'm dying to know if bumping this limit would clear up the queue of
such DNS errors.
---
[narrator: it did.]
* Add the %ll length modifier,
* Convert nil to "<nil>" under %s (not in APE),
* Cast void* to uintptr under %p,
* Use "0x" hex prefix under %p,
* Fix manual page mentions of %P to %p,
* Fix empty result for fp conversions,
* Fix zero padding of left-aligned fp conversions,
* Remove deprecated #pragma ref uses.
Most of these were introduced in APE prior to 9front.
I've omitted the %z conversion specifier since Plan 9 code
rarely uses the usize type. This may need to be added later
for the benefit of native ports of alien code.
I'm not sure if this LFSR is the same one used by the hardware or is
arbitrary, but it matches the noise sequence used by all other snes
emulators I looked at.
s1 and s2 should store the last and next to last output, but were set
in the wrong order, causing them both to be the last output. This
breaks filters 2 and 3, which both utilize s2.
despite the kernel never doing any efi runtime service calls,
overriding the runtime service regions makes some machines
lock up. so consider them reserved.
the boot service regions should also, in theory, be usable
by the os, but linux says otherwise...
fseeko returns 0 on success, not the new stream position.
This allows flacenc to update the streaminfo block when it is finished
(for example to set the number of samples and checksum).
There may be two iso endpoints with the same ID if it is asynchronous
or adaptive (one for data, one for feedback), and rw iso endpoints are
unusable (error out with "iso i/o is half-duplex").
There may be two iso endpoints with the same ID if it is asynchronous
or adaptive (one for data, one for feedback), and rw iso endpoints are
unusable (error out with "iso i/o is half-duplex").
The value of `k` in dtoa() is an estimate of
floor(log10(d)), where `d` is the number being
converted. The code was asserting that 'k' was
less than 100, but the full range of a double
goes to 10^308 or so.
This means that the majority of the range of
a double would trigger that assert.
validateattachment has no business with the mime boundary; it is not
part of the attachment itself.
Also, it causes the boundary to be dropped in the message output from
upas/vf, effectively dropping the following attachment (though the
content is still present after the last boundary of the wrapped first
attachment part).
Consider the following sequence of events:
1. upas/vf is run on a message containing two attachments.
2. The first attachment does not have a known extension, so is saved
to a temporary file *including* the following mime boundary.
3. This file is opened as p->tmpbuf, which is used for subsequent
reads until switching back to stdin.
4. The attachment fails validateattachment, so upas/vf wraps it in a
multipart with a warning message.
5. problemchild() calls passbody(p, 0), which copies from p->tmpbuf
until it hits the outer boundary line, which it excludes, seeks
back one line, then returns the outer multipart.
6. problemchild() then writes its own boundary, and then copies one
line from *stdin* to stdout, expecting the outer boundary.
However, this boundary was already read from stdin in 2, so it ends
up reading the first line of the subsequent part instead.
To fix this, pass 0 to passbody() in save() to exclude it from the
attachment file and make it available in stdin when expected.
Reading nested subparts of messages into the root
message array allows deeply nested multipart trees
of messages to show correctly in the message view.
H-blank DMA should only transfer 16 bytes per h-blank, rather than
waiting for the first h-blank and then transferring the whole size.
HDMAC should read 0xff when the transfer is finished, and 0 in the
high bit when the transfer is ongoing. Also, if 0 is written in the
high bit, the current transfer should be aborted.
Introduce two flags, DMAREADY and DMAHBLANK rather than special
constants 1 and -1. If dma is non-zero, there is an ongoing DMA. If
DMAREADY is set, the next chunk is ready to transfer.
Reference: https://gbdev.io/pandocs/#ff55-hdma5-cgb-mode-only-new-dma-length-mode-start
Tested with pokemon crystal.
What was happening is that when the game was loading N background tiles
into vram (each 16 bytes, so one per h-blank), it did something like
this:
- start an hdma transfer for N+1 tiles
- after the Nth tile is transferred, it would read HDMA5, clear the
high bit, then write it back to abort the transfer.
games/gb would instead transfer all N+1 tiles at once, overwriting one
extra tile with whatever was 1 past the end of the source array, and
then would interpret the cancel request as the start of a new transfer
of 16 bytes, which would copy an additional tile past the end. The end
result is that every transfer would end up copying N+2 tiles instead
of just N, overwriting certain tiles with whatever was after the end
of the source data.
According to [0], input clock 0 should divide by 1024=2¹⁰, not 2¹².
This caused audio to run at quarter-speed in one game I tried.
[0] https://gbdev.io/pandocs/#ff07-tac-timer-control-r-w
Tested with zelda: oracle of seasons, and dr. mario
---
upas/vf was converted to use tmdate, but the formatter was never
installed. This caused it to send attachments to validateattachment
with header `From virusfilter %τ%`, which always failed since upas/fs
would just skip over the message.
On the pi400, the xhci reset firmware mailbox request
assumes that the pci windows match the ones specified
in the device tree. The inbound window (pcidmawin)
also varies now depending on the amount of memory
installed.
It is all pretty ridiculous, as the firmware could as
well just read the pci controllers hardware register
to determine the window configuration and the os could
keep a nice simple 1:1 mapping (with pci dma addresses
== physical addresses).
The old parser code was rubbish and only worked for trivial
expressions. The new code properly handles complex expressions,
including short circuit evaluation.
As such, the BUGS section has been removed from the test(1) man page.
The description of an unimplemented feature has also been removed.
Despite pervious efforts, mk clean still doesn't remove libcommon.a*
files from cmd/upas/common/. To fix this, let's tell cmd/mklib to do
the job instead.
Runq spawns a number of processes, and wait()s for them
in 2 different places. Because of the way that the exit
handling is done, the wait can get the wrong message.
It turns out that only one place in the code needs to
wait for the child, and in all other cases, it's just
muddling the problem.
This change adds the RFNOWAIT call to all the processes
we don't need to wait for, so that the places that do
need wait will always get the correct child.
Screenlock should use libdraw(2) to init the display
and create the window, instead of looking at the screen
file directly. Also, to prevent new windows from popping
up over screenlock, bring it to the top periodically.
on arm32, we can do one of 4 shifts
by a constant:
reg<<(0..31)
reg>>(1..32)
((u32int)reg)>>(1..32)
reg ROT (0..31)
There's no way to encode a 0 bit right
shift, so when encoding reg>>0, flip
it to the equivalent nop reg<<0, which
can be encoded.
This prevents an incorrect warning for a comparison such as `0 < x`,
where x is an unsigned type. Previously, this would get normalized as
`x >= 0` rather than `x > 0` when checking the comparison.
With ntlm auth, we were trying to set 0 bytes of
the auth struct to its size. The args were clearly
swapped. Fix it.
While we're here, remove some dead code.
When running a mail queue, it's useful to run it with limited
parallelism. This helps mailing lists process messages in a
reasonable time.
At the same time, we can remove the load balancing from runq,
since the kinds of systems that this matters on no longer
exist, and running multiple queues at once can be better
done through xargs.
Querying battery (or temperature) using ACPI takes quite some
resources, which makes the battery discharge faster. It doesn't make
much sense to have it queried as often either. So, when using ACPI:
1) set battery query period to 10s minimum
2) set temperature query period to 5s minimum
When _startbuf is invoked, it would crash on the second invocation
if creating a mux segment failed. This is because the first attempt
would assign the return value -1 to the global mux variable, and
the second attempt would notice that the global mux was not nil,
and would attempt to use it.
This change only assigns to the global variable if the allocation
of the segment was a success.
While we're here, we should also check the return of the rfork call.
When invoking with dd with an invalid size suffix, we
silently accept the suffix. This can lead to confusion,
because lines like:
dd -bs 1K
dd -bs 1m
will silently copy in 1-byte increments. This has caught
people by surprise. While we're at it, megabytes are
convenient, so let's have them too.
Passwd used to produce a very confusing error
about DES not being enabled whenever the password
was mistyped. This happened because we attempted
to guess what authentication method to use, and
preseneted the error from the wrong one on failure.
This puts the legacy mode behind a flag, so that
we don't even try the old method unless it's
explicitly requested.
This adds the new function pointer PCArch.clockinit(),
which is a timer dependent initialization routine.
It also takes over the job of guesscpuhz(). This way, the
architecture ident code can switch between different
timers (i8253, HPET and XEN timer).
Revert the change, as it causes system lockups on bootup
on some systems with USB OHCI controllers, suspected to be
caused by BIOS/SMM accessing the device as BIOS handover
has not been executed yet.
We might bring that back when the problem has is better
understood.
when loading large binaries such as netsurf, with many
symbols, our hash table fills up with collisions and
loading the symbol table gets very slow. Bumping it up
drops the time to lstk() in acid on netsurf from 4 minutes
to 8 seconds.
Call exits(0) instead of returning from main. Also call sysfatal if
writing of image data fails. Previously, qr(1) would exit with
default non-nil status "main" unconditionally as a result of returning
from main.
/$objtype/include/ape/math.h contained an almost
identical copy of math.h for each architecture.
The only difference between them architectures
was that some had an incorrect version of isinf
defined.
This change picks one of the versions of math.h
with a correct definition, moves it to /sys/include,
and removes the redundant versions.
Tilting allows using left/right rotated or invetrted display orientation.
This can be changed at runtime such as: echo tilt right > /dev/vgactl
This removes the old panning and vga overlays as they are only implemented
with some ancient vga controllers.
The idea is to avoid the magic files that contain
per process information in devcons when possible.
It will make it easier to deprecate them in the future.
Previously, mmurelease() was always called with
palloc spinlock held.
This is unneccesary for some mmurelease()
implementations as they wont release pages
to the palloc pool.
This change removes pagechainhead() and
pagechaindone() and replaces them with just
freepages() call, which aquires the palloc
lock internally as needed.
freepages() avoids holding the palloc lock
while walking the linked list of pages,
avoding some lock contention.
On 12/18/20, Jacob Moody wrote:
> Hello,
>
> I recently ran in to some issues with pointing an unbound server towards a
> 9front dns server as its upstream.
> The parsing seemed to fail when ndb/dns received a DNSKEY RR from it's own
> upstream source on behalf of unbound.
> This patch catches and stores the DNSKEY from the upstream server to prevent
> this.
we might as well handle the per process cycle
counter in the portable part instead of duplicating the code
in every arch and have inconsistent implementations.
we now have a portable kenter() and kexit() function,
that is ment to be used in trap/syscall from user,
which updates the counters.
some kernels missed initializing Mach.cyclefreq.
Provide a central function to change the user id
of the calling process.
This is mostly used by programs to become the none
user, followed by a call to newns().
vt sets several environment variables ($TERM, $COLS, $LINES)
after exiting. This change rforks the environment so that this
detritus doesn't get left behind.
Using strlen in strndup will walk past the first
n bytes up to the terminator, which may not be
present. This is not what we want.
While we're here, do some cleanups.
hget supports adding custom headers with -r;
it makes sense for hpost to do the same, both
because custom headers are more likely necessary
with POSTs, and for consistency.
the tulip driver is used in microsofts hypver-v
as the legacy ethernet adapter for pxe booting.
to make the driver work on pc64, we need to
store the Block* pointers in a separate array
instead of stuffing them into buffer address 2
of the hardware descriptor.
also, enable the driver in the pc64 kernel.
The initial protocol handling in exportfs for
cpu and import services is a huge mess.
Saparate the code out into its own program with
its own oexportfs(4) manpage.
The OCEXEC flag used to be maintained per channel,
making it shared between all the file desciptors.
This has a unexpected side effects with regard to
channel passing drivers such as devdup (/fd),
devsrv (/srv) and devshr (/shr).
For example, opening a /srv file with OCEXEC
makes it impossible to be remounted by exportfs
as it internally does a exec() to mount and
re-export it. There is no way to reset the flag.
This change makes the OCEXEC flag per file descriptor,
so a open with the OCEXEC flag only affects the fd
group of the calling process, and not the channel
itself.
On rfork(RFFDG), the per file descriptor flags get
copied.
On dup(), the per file descriptor flags are reset.
The second modification is that /fd, /srv and /shr
should reject the ORCLOSE flag, as the files that
are returned have already been opend.
enable pci busmaster before set the fis-receive-enable
bit in the port command register.
not doing so triggers a crash in qemu like:
address_space_unmap: Assertion `mr != NULL' failed.
as qemu tries to process the dma command list as soon
as we set that flag and busmaster dma needs to be enabled
at this point.
Bhyve returns 0 in MTRRCap register, so we
can use that instead on relying on cpuid only
to see if MTRR's are supported.
That way we can get rid of the sanity check
in memory.c.
Opening a /srv file sets the close-on-exec flag on the
shared channel breaking the exportfs openmount() hack.
The devsrv tries to prevent posting a channel with the
close-on-exec or remove-on-close flags. but nothing
currently prevents this poisoning on open.
Until this gets fixed in eigther exportfs or devsrv,
i'll back out the changes that could have potential side
effects like this.
On AMD64, CR0/CR4 are 64-bit registers, with
the upper half reserved. So use uintptr type
to store the register values to get 32 bit on 386
and 64 bit on AMD64.
The -v flag now does not create a new rio window,
while -w flag does (restores the old behaviour).
This allows vmx to run under vncs and is in general
mode aligned to other emulators and programs.
Removes the 128 kB limit for files making up the database.
We used to skip over and complain about files that exceeded
the limit, forcing the user to generate hash files.
This caused things to inexplicably stop working after a file
hit the hidden limit, which is unreasonable behaviour considering
that libndb happily, albeit slowly, works with bigger files.
The previous resize optimization now means that the wfill()
is skipped on resize for libdraw programs.
So do it once /dev/mouse is closed and the window processes
the Refresh message.
"" looks for patterns in the form 'prompt;' or 'prompt%',
and gets confused when proof emits 'illegal;'. This change
replaces the ';' with a ':', which both matches other
conventional error outputs and prevents "" from getting
confused.
Initially the code tried to guess the date format. This
turned out to be a bit too magical, so the feature was
removed, but the manpage still documented the nonfeature.
As long as the client as the mouse file open
and maintains reading the winname file of the window
after a resize we will avoid drawing the text frame
on a resize as it will be overdrawn by the client.
This reduces flicker on resize somewhat for slow systems.
This implements proper intrdisable() support for all
interrupt controllers.
For enable, (*arch->intrassign)(Vctl*) fills in the
Vctl.enable and Vctl.disable pointers with the
appropriate routines and returns the assigned
vector number.
Once the Vctl struct has been linked to its vector
chain, Vctl.enable(Vctl*, shared) gets called with a
flag if the vector has been already enabled (shared).
This order is important here as enabling the interrupt
on the controller before we have linked the chain can
cause spurious interrupts, expecially on mp system
where the interrupt can target a different cpu than
the caller of intrenable().
The intrdisable() case is the other way around.
We first disable the interrupt on the controller
and after that unlink the Vctl from the chain.
On a multiprocessor, the xfree() of the Vctl struct
is delayed to avoid freeing it while it is still
in use by another cpu.
The xen port now also uses pc/irq.c which has been
made generic enougth to handle xen's irq scheme.
Also, archgeneric is now a separate file to avoid
pulling in dependencies from the 8259 interrupt
controller code.
Plan 9 memcpy(2) uses the same implementation as memmove(2) to handle
overlapping ranges. Hovewer, the MIX MOVE instruction, as described
in TAOCP, specifically does not do this. It copies words one at a
time starting from the lowest address.
This change also expands the address validation to check that all
addresses within the source and destination ranges are valid before
proceeding.
Fixes 3 issues in our upas mkfiles:
- mk/mkfile and send/mkfile were rebuilding
only the rfc822.tab.$O, even though the
header also needed to be rebuilt.
- CLEANFILES had a pattern that would not
get expanded.
- Third, ../upas/mkfile was being included
in the wrong place and making the wrong
rule default.
currently the EFI loader's behavior is to search all disks in a
firmware-defined order. we search the list returned by the firmware
in reverse order in the hopes of searching the first 9FAT instead of
the ESP, but this results in unintuitive behavior when there are
multiple FAT partitions (possibly in multiple disks), such as loading
a plan9.ini and kernel from a different disk than the one you executed
the EFI loader from.
to resolve this, we change the EFI loader to instead prefer read
plan9.ini and the kernel from the same disk as the EFI loader was read
from, and then fall back to the old behavior, since the old behavior
is relied on by current installations.
dc crashes because a Blk* sometimes ends getting double freed.
To make it crash, any of these lines will do:
(each line is a separate input to dc):
1 sa 2 :a le d sa v :a
1 sa 2 :a le d sa :a
1 sa 2 :a le d sa c
Fix by assigning p to sptr->val before EMTPY causes a jump.
Additionally, dcgetwd() can return 0. all other uses check for
0 ptr; Also fix a buffer overflow.
It appears that our IDT overlaps with the data structures
passed from grub in multiboot load.
So defer setup of the interrupt table after the multiboot
parameters have been processed.
The driver used to register the interrupt handler just
after reset, tho the Ctlr struct, including the buffer
descriptor arrays where only allocated on attach.
This moves most of the reset/init out of pnp
function and into attach. This also means we can
error out and even retry on the next attach.
The logic of the reseter kproc has been changed:
now it is only started once the first initialization
completely succeeded. This avoids the strange qlock
passing.
Implement a shutdown function so the device gets
halted for /dev/reboot.
Assume 64 bit physical addresses for dma.
Check that pci bar0 is actually I/O.
the hid 1.11 specification says that for hid devices which arent in
the boot subclass (subclass 1), it is only optional to support the set
protocol command. for my devices, trying to set protocol results in a
stall error and unusable devices.
fixes my Tex Shinobi keyboard and Playstation 4 controller.
The new MTRR code handles overlapping ranges
and supports AMD specific TOM2 MSR.
The format in /dev/archctl now only shows
the effective cache ranges only, without
exposing the low level registers.
Before the "native" awk work, a call to the fflush function resulted
in one or more calls to the APE fflush(2).
Calling fflush on a stream open for reading has different behavior
based on the environment: within APE, it's a no-op¹; on OpenBSD, it's
an error²; in musl, it depends on whether or not the underlying file
descriptor is seekable³; etc. I'm sure glibc is subtly different.
Now that awk uses libbio, things are different: calling Bflush(2) on a
file open for reading simply discards any data in the buffer. This
explains why we're seeing truncated input. When awk attempts to read
in the next record, there's nothing in the buffer and no more data to
read so it gets EOF and exits normally. Note that this behavior is not
documented in bio(2). It was added in the second edition but I haven't
figured out why or what depends on it.
The simple fix is to have awk only call Bflush on files that were
opened for writing. You could argue that this is the only correct
behavior according to the awk(1) manual and it is, in fact, how GNU
awk behaves⁴.
1. /sys/src/ape/lib/ap/stdio/fflush.c
2. https://cvsweb.openbsd.org/src/lib/libc/stdio/fflush.c?rev=1.9
3. https://git.musl-libc.org/cgit/musl/tree/src/stdio/fflush.c
4. https://git.savannah.gnu.org/cgit/gawk.git/tree/io.c#n1492
Changeset 50ad211fb12f broke the libcommon rule in
mkupas. Deleting the 'mk clean' in the recipe fixes
this.
Cleanup includes deleting UPDATE vars from all mkfiles,
reorganization of vars in TARG,LIB,OFILE,HFILE order,
and deletion of extra vars used for UPDATE.
loading the interrupt vector table early allows
us to handle traps during bootup before mmuinit()
which gives better diagnostics for debugging.
we also can handle general protection fault on
rdmsr() and wrmsr() which helps during
cpuidentify() and archinit() when probing for
cpu features.
before removing the double map at 0, load our
initial gdt pointer with its new KZERO based
virtual address.
this is prerequisite for handling traps early during
bootup before mmuinit() loads the final gdt.
The fsdestroyfid() is called regardless if the open succeeded
or failed. This causes erroneous videoclose() when opening
the frame or video file while the camera is active.
When using /dev/reboot, the MSI vecor might have already
been setup causing interrupts to fire on the designated
cpu while we send the commands to the card.
reseting irbsts bits in hdacmd() only works
while interrupts are disabled during hdareset().
once interrupts are enabled we need to reset the
irbsts bits in the interrupt handler or else the
interrupt never clears and locks up the system.
Upas/marshal -F was broken with the '-8' command, and silly
without it: It used aliases passed on the command line, so
the destination address was ignored with -8 was passed.
In addition, it would create a new mailbox for any aliases
being sent to, instead of putting them all in one location.
The new -S option is similar to -F, but specifies where the
message should go.
The change 3306:c5cf77167bfe made the code reuse MTRR slots
of the default memory type.
But this did not take overlapping ranges into account!
If two or more variable-range MTRRs overlap, the following rules apply:
a. If the memory types are identical, then that memory type is used.
b. If at least one of the memory types is UC, then UC memory type is used.
c. If at least of of the memory types is WT. and the only other memory type
is WB, then th WT memory type is used.
d. If the combination of memory types is not listed above,
then the memory type used in undefined.
It so happend that on a Dell Latitude E7450 that the BIOS defines
the default type as UC. and the first slot defines a 16GB range
of type WB. Then the rest of the ranges mark the PCI space back
as UC, but overlapping the first WB range! This works because
of rule (B) above.
When trying to make the framebuffer write-combining, we would
falsely reuse one of the UC sub-ranges and making the UC memory
into WB as a side effect.
Thanks to Fulton for his patience and providing debug logs and
doing experiments for us to narrow the problem down.
With some newer UEFI firmware, not all pci bars get
programmed and we have to assign them ourselfs.
This was already done for memory bars. This change
adds the same for i/o port space, by providing a
ioreservewin() function which can be used to allocate
port space within the parent pci-pci bridge window.
Also, the pci code now allocates the pci config
space i/o ports 0xCF8/0xCFC so userspace needs to
use devpnp to access pci config space now. (see
latest realemu change).
Also, this moves the ioalloc()/iofree() code out
of devarch into port/iomap.c as it can be shared
with the ppc mtx kernel.
libcommon.a$O doesn't end with a .a, so mk
doesn't know how to look inside it in order
to check if the files are up to date.
This means that when 'mk clean' is run,
libcommon.a$O looks up to date:
% mk clean
...
% mk
mk: 'default' is up to date
Deleting the library works around this problem.
When $wsys doesn't exist (eg, drawterm -G, or
rcpu from a text console), the profile would
create an empty $wsys variable, and sessions
started in this environment would fail with a
null list in concatenation.
This change tests if /mnt/term/env/wsys exists
before assigning it.
This prevents VESA bios from accessing the pci
CONFIG_ADDRESS/CONFIG_DATA registers (0xCF8/0xCFC)
directly to access pci config space.
This makes sure the access to pci config space is
properly serialized by the kernel.
Our qsort has an optimization to recurse on one
half of the array, and do a tail call on the other
half. Unfortunately, the condition deciding which
half of the array to recurse on was wrong, so we
were recursing on the larger half of the array and
iterating on the smaller half.
This meant that if we picked the partition poorly,
we were pessimizing our stack usage instead of
optimizing it.
This change reduces our stack usage from O(n)
to O(log(n)) for poorly chosen pivots.
vt chording behaves slightly differently from other
applications: a chord must be fully released before
the next chord can be applied. This makes any change
in chord apply the action.
the driver was not using irb interrupts
and was just polling the irb write pointer
to wait for command completion.
this is not supported by qemu.
qemu requires the use of irb interrupt handshake
and it refuses to accept the next command until we
acknowledge the irb interrupt.
supplying a non-ip address in ADD_EDGE crashes the unix tincd.
the reason was that we where misreporting ADD_EDGE messages;
ignoring the information from our peers; and always supplying
the Address string from our configuration instead of the
connections ip address.
now we just report the edge information as is.
the access functions for pci config space in config mode #1
used to set bit 0 in the register offset if the access was
to a device on any bus different from 0.
it is completely unclear why this was done and i can't find
any documentation on this.
but for sure, this breaks all pci config spacess access to
pci devices behind a bridge on qemu. with -trace pci* it
was discovered that all config space register offsets on
devies behind pci brige where off by one.
on real hardware, setting bit 0 in the offset doesnt appear
to be an issue.
thanks mischief for reporting and providing a qemu demo
configuration to reproduce the problem.
On my 6235 card, if we calibrate the crystal
immediately after disabling wimax, the the
firmware gets unhappy. A short nap before
sending the command prevents the command from
timing out.
When moving messages between folders, mbappend,
deliver, and nedmail were trying to parse the
timestamp ouut of the message. They were doing
it incorrectly, trying to include the user name
as part of the date format.
Change to pass just the date to the date parser.
ori and echoline are reporting regression on some 6000 cards;
which sometimes time out on crystal calibration command;
which is expected by the driver. but the new code used
to force a device reset on any command timeout.
reverting to old behaviour until for now until we have
a chance investigating.
We used to assume a 1:1 pairing of processors to submit queues.
With recent machines, we now got more cpu cores than what some
nvme drives support so we need to distribute the queues across
these cpu's which requires locking on command submission.
There is a feature get/set command to probe the number of submit
and completion queues, but we decided to just handling
submission queue create command error gracefully as it is simpler
and has less chance of regression with existing setups.
Thanks to mischief for investigating and writing the code.
I have the problem that i need to delegate a subdomain
to another name server that is confused about its own zone
(and its own name) returning unusable ns records.
With this, one can make up a nameserver entry in ndb that
is authoritative and owned by us for that nameserver,
and then put it in the soa=delegated ns entry.
This promotes the ns record in the soa=delegated to
Authoritative, which avoids overriding the ns rr's from
the confused server for the delegated zone.
we used to only allocate ports from 6881 to 6890,
which limits the maximum of parallel torrents to 9.
this change make it go up to 9000, which gives us
at best 2120 ports, which is overkill but ports might
be randomly occupied by other connections.
the new date format introduced by the previous commit;
using numeric timezone offsets; needs one character more,
so increase the date format buffer to 31 characters.
the whole idea of a ucallocb() is bad, as even access to the
metadata header would be in uncached memory. also, it tuns out
that it was never used by anyone.
the 9000 series uses a new receive descriptor format
wich appears to reqire 4k aligned buffers. the old
format "halfworks" and just makes the firmware not
respond to any commands after the enable paging command.
the smartfifo command appears to causes problems.
but apparently not issuing it at all seems to work
fine on both the 8265 and 9260. so removing the code
for now.
issuing the bindingquota command before associated
makes association impossible. but enabling afterwards
works fine. (tested in 8265 and 9260).
the prph access functions now mask the address with
0xfffff. it is unclear why linux and openbsd drivers
specify addresses beyond that in ther register constants.
the timeevent change is interesting. the timeevent
needs to be restarted when it has stoped to make sure
probing/association packets are sent during the evnet.
Complete the conversion of upas to remove ctime,
use the new date library, and print time zones
in +hhmm format, instead of NNN format.
This may affect code that expects specific names
for timezones. Fix that code.
