dont kill the calling process when demand load fails if fixfault()
is called from devproc. this happens when you delete the binary
of a running process and try to debug the process accessing uncached
pages thru /proc/$pid/mem file.
fixes to procctlmemio():
- fix missed unlock as txt2data() can error
- make sure the segment isnt freed by taking a reference (under p->seglock)
- access the page with segment locked (see comment)
- get rid of the segment stealer lock
other stuff:
- move txt2data() and data2txt() to segment.c
- add procpagecount() function
- make return type mcounseg() to ulong
handle reads and writes with 9pqueue(2) so they can
be flushed and wont hang the filesystem. this also
lets us get rid of the timeouts.
ftdi is still full of braindamage that should be
rewritten, but i dont have a device to test.
instead of naming devices by ther dynamically assigned device address,
we hash device uniqueue fields from the device descriptor and produce
a 5 digit hex string that will identify the device across machines.
when there is a collision (less than 1% chance with 100 devices),
usbd will append the device address to the name to make it uniqueue
for this machine.
the hname is passed to drivers in the devid argument, which now has
the form addr:hname, where the colon and hname can be omited (for backwards
compatibility).
when the new behaviour isnt desired, nousbhname= environment variable
can be defined giving the old behaviour.
pipeline = 1 with a dovecot imap server causes FETCH and OK responses
get interleaved so some message bodies accidentally get merged together.
disabling it will make fetching mail over imap slower, but it works.
make the Page stucture less than half its original size by getting rid of
the Lock and the lru.
The Lock was required to coordinate the unchaining of pages that where
both cached and on the lru freelist.
now pages have a single next pointer that is used for palloc.head
freelist xor for page cache hash chains in Image.pghash[].
cached pages are not on the freelist anymore, but will be reclaimed
from images by the pager when the freelist runs out of pages.
each Image has its own 512 hash chains for cached page lookup. That is
2MB worth of pages and there should be no collisions for most text images.
page reclaiming can be done without holding palloc.lock as the Image is
the owner of the page hash chains protected by the Image's lock.
reclaiming Image structures can be done quickly by only reclaiming pages from
inactive images, that is images which are not currently in use by segments.
the Ref structure has no Lock anymore. Only a single long that is atomically
incremented or decremnted using cmpswap().
there are various other changes as a consequence code. and lots of pikeshedding,
sorry.
the code was correct. erealloc9p() terminates the process
on error, but the code was handling realloc() error explicitely
and responded the request with Enomem error.
the vmware svga video card emulated by qemu (qemu -vga vmware) complains and eventually causes a panic if the rectangles aren't clipped.
messages like the following can be observed from qemu before the kernel panics:
vmsvga_update_rect: update h was < 0 (-20000)
vmsvga_update_rect: update height too large y: 10000, h: 0
vmsvga_update_rect: update w was < 0 (-20000)
vmsvga_update_rect: update width too large x: 10000, w: 0
i could only reproduce this in qemu 2.0.50 on the master branch, when using the ui and had selected 'Zoom To Fit' from the View menu.
fix bug introduced by amd64 support:
forgot to update ring index i on receive. surprisingly
this was working until there where more than one packet
to process. sorry.
ilock the controller while processing rings. this should
be fixed and use kprocs instead.
we have to make sure the *swap address* doesnt go away,
after putting the swap address in the segment pte.
after we unlock the segment, the process could be
killed or fault which would cause the swap address to
be freed *before* we write the page to disk when it
pulls the page from the cache and putswap() swap pte.
keeping a reference to the page is no good. we have
to hold on the swap address. this also has the advantage
that we can now test if the swap address is still
referenced and can avoid writing to disk.
as with the Block refcount changes, _xinc() and _xdec() arent
used anymore, so remove them.
architecure can still define ainc()/adec() when it needs them.
change Proc.nlocks from Ref to int and just use normal increment and decrelemt
as done in erik quanstros 9atom.
It is not clear why we used atomic increment in the fist place as even if we
get preempted by interrupt and scheduled before we write back the incremented
value, it shouldnt be a problem and we'll just continue where we left off as
our process is the only one that can write to it.
Yoann Padioleau found that the Mach pointer Lock.m wasnt maintained
consistently for lock() vs canlock() and ilock(). Fixed.
Use uintptr instead of ulong for maxlockpc, maxilockpc and ilockpc debug variables.
webfs forks the namespace to isolate itself from its mount
point which has the side effect that it captures the mount
of previous instances of webfs mounted on /mnt/web.
explicitely unmount the mountpoint in our namespace copy
to drop the reference.
when there are multiple readers of /dev/usbevent, we have to
serialize the processing to make sure that only one driver
is opening the devices control endpoint at a time.
to do this, we assume the device is busy after reading the
event file until the next read or clunk on the same fid.
to mark a device busy, we set the dev->aux pointer to the
fid processing a event. And the Event structure takes a
reference to the device producing the event.
the problem arised from cdc ethernet and nusb/serial sharing
the same device class, and we need to run the particular driver
to figure out if the device can be used. doing this concurrently
fails because devusb allows only one open per endpoint.
the palloc.pages array takes arround 5% of the upages which
gives us:
16GB = ~0.8GB
32GB = ~1.6GB
64GB = ~3.2GB
we only have 2GB of address space above KZERO so this will not
work for long.
instead, pageinit() was altered to accept a preallocated memory
in palloc.pages. and preallocpages() in pc64/main.c allocates the
in upages memory, mapping it in the VMAP area (which has 512GB).
the drawback is that we cannot poke at Page structures now from
/proc/n/mem as the VMAP area is not accessible from it.
Without an explicit signal for a truncation, copy propagation will
sometimes propagate a 32-bit truncation and end up overwriting uses of
the original 64-bit value.
This was independently discovered and fixed in Go. See:
http://golang.org/issue/1315https://codereview.appspot.com/6002043/
Thanks Charles Forsyth for tips and advice.
as we do system reset and reboot only from boot processor cpu0 now,
theres no need for active.rebooting conditional variable.
mpshutdown() will unconditionally park application processors and
and cpu0 boots the new kernel or calls mpshutdown() causing system
reset.
in vmware, mpshutdown() used to hang in i8042reset() when not
called from the boot processor, so instead of reseting from first
cpu that acquires the shutdown lock, we park all application
processors and let the boot processor do the reset.
procwrite() did truncate the offset to 32bit ulong.
introduce off2addr() function that does the sign
extension hack and use it conststently for Qmem
reads and writes.
for the CMclose procctl, the fd number was not
bounds checked before indexing in the Fgrp.fd
array.
for the CMclosefiles, we looped fd from 0..maxfd-1,
but need to loop from 0..maxfd as maxfd is inclusive.
newns() (called by auth_chuid()) already prepares the
environment variables and puts us in a sane working
directory (as specified by the namespace file).
on amd64, the text segment is aligned and padded to
2MB, but segment granularity is 4K which can give
us page faults that are beyond the highest file
offset. this is perfectly valid, but was not handled
correctly in pio().
use two per process memory slots, one for the
pid and one for the fd instead of a global table
avoiding the case when the table gets full.
instead of calling pread() on the cached fd
(dangerous as it has side effects when the
fd was not closed), we check if the cached fd
is still good using fd2path() when called
the first time in this process.
theres big performance regression with this using
cwfs. cwfs calls time() to update atime on every
read/write which now causes walks on /dev.
reverting to the previous version for now. in the
long run, we'll use new _nsec() syscall but this
has to wait for a later release once new kernels
are established.
trackers do like the new default Mozilla/5.0 (compatible)
user agent. so force useragent to hjdicks and give option
to override it in case trackers get even more clever in
the future.
in dhcpwatch, the sleep time "secs" could become
zero potentially freezing the lease time.
give up when in Sinit state in dhcpquery() as this
is a terminal state.
from the specification:
software may reset the entire HBA by setting GHC.HR to '1'.
When software sets the GHC.HR bit to '1', the HBA shall perform
an internal reset action. The bit shall be cleared to '0'
by the HBA when the reset is complete.
sites like google return the wrong characterset when
they do not recognize the user-agent. so setting default
user agent to something thats likely to pass these
idiotic browser tests.
> warning: a.c:9 useless or misleading comparison: UINT < 0
the error can be observed by compiling the following code
with warnings enabled:
#include <u.h>
#include <libc.h>
uint r;
void
main(int argc, char *argv[])
{
int r;
if(r < 0){
exits(0);
}
}
the offending code in the compiler is:
- if(l->op == ONAME && l->sym->type){
- lt = l->sym->type;
- if(lt->etype == TARRAY)
- lt = lt->link;
- }
compiler handles scope by overwritin and reverting
symbols while parsing. in the ccom phase, the nodes symbol
(n->sym) is not in the right scope and we wrongly think r
is uint instead of int.
it is not clear to me what this code tried to accomplish in
the first place nor could anyone answer me this question.
the risk is small as this change doesnt affect the compiled
program, only the warning, so removing the offending code.
the malloc pool allocator is limited in its allocation
size. as almost all data structures in cwfs are never
freed, use brk() in ialloc() instead of mallocalign().
this means memory returned by ialloc() cannot be freed!
to make sure we do not call free by accident, remove
the #define malloc(n) ialloc(n, 0) macro and use ialloc()
directly as in the original code to show the intend
of permanent allocations.
this was a big mistake. we should never attempt to
timeout or retry a scsi command from the controller
driver because theres no way to tell how long a
command would take or if a command has side effects
when being retried.
number of bank slots in Conf.mem[4] was too small
for kenjis machine, set it to maximum 16 (the
size of the RAM map in pc64/memory.c).
also increasing the UPA memory map to 64. the
e820 map on my x200s has 31 entries and many
holes. this gets rid of the "mapfree: ... losing"
messages on boot.
make mntflushfree() return the original rpc and do the
botched clunk check on the original instead of the
current rpc.
so if we get a botched flush of a clunk, we abandon the
fid of the channel as well.
if theres an error transmitting a Tclunk or Tremove request,
we cannot assume the fid to be clunked. in case this was
a transient error, reusing the fid on further requests
will fail.
as a work arround, we zero the channels fid and allocate
a new fid before the chan is reused.
this is not correct as we essentially leak the fid
on the fileserver, but we will still be able to use
the mount.
the shift instructions does not change the zero flag
when the shift count is 0, so we cannot remove the
compare instruction in this case.
this fixes oggdec under 386.
ftrvxmtrx repots devices that use the endpoint number for
input and output of different types like:
nusb/ether: parsedesc endpoint 5[7] 07 05 81 03 08 00 09 # ep1 in intr
nusb/ether: parsedesc endpoint 5[7] 07 05 82 02 00 02 00
nusb/ether: parsedesc endpoint 5[7] 07 05 01 02 00 02 00 # ep1 out bulk
the previous change tried to work arround this but had the
concequence that only the lastly defined endpoint was
usable.
this change addresses the issue by allowing up to 32 endpoints
per device (16 output + 16 input endpoints) in devusb. the
hci driver will ignore the 4th bit and will only use the
lower 4 bits as endpoint address when talking to the usb
device.
when we encounter a conflict, we map the input endpoint
to the upper id range 16..31 and the output endpoint
to id 0..15 so two distinct endpoints are created.
nusb code assumes endpoint numbers are unique. It's true in general
case, but it becomes false once the direction bit is ignored. The
commit adds a check so that two endpoints of different types are not
merged into one with Eboth direction. It does overwrite endpoint
though, so it shouldn't be considered as a full fix.
the 802.11 spec only specifies the msb of the rate for
Beacon, Probe Response, Association Response, Reassociation Response,
Mesh Peering Open, and Mesh Peering Confirm management frames
...
The MSB of each Supported Rate octet in other
management frame types is ignored by receiving STAs.
this should make no difference but on some netgear ap's not
setting this bit seems to ignore these data rates.
driver sets wifi->rates array to tell wifi layer what
rates it supports. when we receive beacon, we determine
the minimum and maximum data rates and set wn->minrate
and wn->maxrate to point to the entries in wifi->rates.
it is the responsibility of the driver to use this
information on transmit.
to run aux/wpa at boot, we need factotum to be running. tho
factotum was started only after the network was configured.
what we do now is start factotum early, not fetching keys
from secstore. once network is available and the auth server
is known, we fetch keys from secstore using auth/secstore in
bootrc.
to pass the authserver for p9 authentication to factotum, we
write it in /net/ndb and the special _authdial() in factotum
will picks it up.
as we are using auth/secstore binary in any case, we remove
the duplicated secstore code from factotum and make it just
exec auth/secstore to fetch the keys on startup (unless -n
or -S is specified).
load x-stretched scanline and use image replication bit to let
devdraw do the y-stretching. this reduces slow RGB15 -> display
conversions as devdraw caches small numbers of converted source
scanlines (one in hour case). setting pixels should also be a
bit faster. (only 3 writes instead of 9 for x3 scaling)
remove unused memsetb() routine.
replace foo ? 1 : 0 with foo != 0
avoid double calculation of rgb components in readbyte()
handle byte aligned color components in writebyte()
which lets us avoid the read-modify-write and the shifting.
surprisingly, the branches in the loop are way less important
than avoiding the memory access.
this change makes ganes/snes playable at -3 scaling.
used to set audio device frequency thru /dev/volume tho
only ac97 driver supports this. as a quick work arround,
upsample the 32000 hz audio signal to 44100 hz (without
any interpolation).
move the sample buffer room check from audiosample() into
dspstep() so that when the buffer is full (shouldnt happen),
we wont advance dspstate so samples will not get dropped.
we allow protocol path to begin with # for dial, so should
allow this for announce as well. this is primarily usefull
when booting the fileserver to listen on alternate ip stack.
according to erik, virtualbox puts the source overrides
before the ioapic entries so the addirq() call fails
as no ioapics have been declared yet. use a second pass
over the table after we processed the apic entries.
the original format for addresses was %8lux which was changed
to %p for amd64. this broke linuxemu which assumes fixed format
in the segment file. as a compromize we change it to %8p and
amd64 port of linuxemu will hopefully use a more robust parser :)
quote handling was broken with 21-bit runes. nextrec()
returned quoted rune as long rune | (Runemax+1) to escape
it.
with 16-bit runes, storing that long into 16-bit Rune
would automatically remove the escaping, but with 21-bit
runes, Rune is uint32 so the escaping would remain. we
now use (Runemask+1) instead, and mask the escaping off
explicitely when storing back to Rune.
add 0xffff to tab1 as range 0xffff-0x10ffff has 4 byte utf-8 sequence.
use Runemax (0x10ffff) instead of Runemask (0x1fffff) to denote
the last valid rune for inverted [^] match as Runemask is out of the
valid rune space.
when the previous instruction sets the zero flag,
we can remove the CMPL/CMPQ instruction.
this removes compares for zero/non zero tests only.
it only looks at the previous non-nop instruction
to see if it sets our compare value register.
we have to wait for the pcmconv process to exit before
exiting yourselfs because otherwise pcmconv could
keep /dev/audio open and prevent further reopens for
a short period of time.
the rule that was used to copy header files from ../pc
accidently overwrote dat.h when ../pc/dat.h was updated
because it matched on all *.h files that was also found
in ../pc directory. change to exact match on $PCHEADERS
to prevent this.
