adrian [Sat, 14 Jul 2012 02:52:48 +0000 (02:52 +0000)]
Change the RX EDMA path to first complete the FIFO, then re-populate it
with fresh descriptors, before handling the frames.
Wrap it all in the RX locks.
Since the FIFO is very shallow (16 for HP, 128 for LP) it needs to be
drained and replenished very quickly. Ideally, I'll eventually move this
RX FIFO drain/fill into the interrupt handler, only deferring the actual
frame completion.
adrian [Sat, 14 Jul 2012 02:07:51 +0000 (02:07 +0000)]
Fix EDMA RX to actually work without panicing the machine.
I was setting up the RX EDMA buffer to be 4096 bytes rather than the
RX data buffer portion. The hardware was likely getting very confused
and DMAing descriptor portions into places it shouldn't, leading to
memory corruption and occasional panics.
Whilst here, don't bother allocating descriptors for the RX EDMA case.
We don't use those descriptors. Instead, just allocate ath_buf entries.
sh: Add tests where "export" does not parse differently.
It is planned to expand variable assignments as assignments (no word
splitting, different tilde expansion) when they follow a "declaration
utility" (export, readonly or local). However, a quoted character cannot be
part of a "name" so things like \v=~ are not assignments, and the existing
behaviour applies.
dim [Fri, 13 Jul 2012 21:48:01 +0000 (21:48 +0000)]
Pull in r159895 from upstream clang trunk:
When marking virtual functions as used for a class' vtable, mark all functions
which will appear in the vtable as used, not just those ones which were
declared within the class itself. Fixes an issue reported as comment#3 in
PR12763 -- we sometimes assert in codegen if we try to emit a reference to a
function declaration which we've not marked as referenced. This also matches
gcc's observed behavior.
This should fix clang assertions when building certain components of the
LibreOffice port.
Make the interval timings for EVFILT_TIMER more accurate. tvtohz() always
adds an extra tick to account for the current partial clock tick. However,
that is not appropriate for a repeating timer when the exact tvtohz() value
should be used for subsequent intervals. Fix repeating callouts for
EVFILT_TIMER by subtracting 1 tick from the tvtohz() result similar to the
fix used in realitexpire() for interval timers.
While here, update a few comments to note that if the EVFILT_TIMER code
were to move out of kern_event.c, it should move to kern_time.c (where the
interval timer code it mimics lives) rather than kern_timeout.c.
The etcupdate utility is a tool for managing updates to files that are
not updated as part of `make installworld' such as files in /etc. It
manages updates by doing a three-way merge of changes made to these files
against the local versions. It is also designed to minimize the amount
of user intervention with the goal of simplifying upgrades for clusters
of machines.
The primary difference from mergemaster is that etcupdate requires less
manual work. The primary difference from etcmerge is that etcupdate
updates files in-place similar to mergemaster rather than building a
separate /etc tree.
1796 "ZFS HOLD" should not be used when doing "ZFS SEND" froma read-only pool
2871 support for __ZFS_POOL_RESTRICT used by ZFS test suite
2903 zfs destroy -d does not work
2957 zfs destroy -R/r sometimes fails when removing defer-destroyed snapshot
acpi_cpu: separate a notion of current deepest allowed+available Cx level
... from a user-set persistent limit on the said level.
Allow to set the user-imposed limit below current deepest available level
as the available levels may be dynamically changed by ACPI platform
in both directions.
Allow "Cmax" as an input value for cx_lowest sysctls to mean that there
is not limit and OS can use all available C-states.
Retire global cpu_cx_count as it no longer serves any meaningful
purpose.
For our at91rm9200 boards, register which subtype of SoC is on the
board. We'll use this later to control the differences between these
two variants' pins.
Complete the transition away from newbus to populate the children to
the linker set of CPU modules. The newbus method, although clever,
had many flaws: it didn't really support multiple SoC, many of the
comments about order were just wrong, and it did a few things far too
late to be useful. delay and cpu_reset now work much earlier in the
boot process.
Export the interrupt status vector via soc_data. Set the interrupt
priorities in the AIC in the atmelarm driver before attaching the
children. Delete redunant copies of the code.
Re-merge a couple of changes from NetBSD's libedit.
bin/sh has been taught about el_gets setting the count to -1
on error, so now we can partially revert r238173 to reduce
differences with NetBSD's implementation.
Unfortunately NetBSD's libedit handling of EINTR (Revision
1.44 in read.c + SIGWINCH fixes) still causes trouble in
bin/sh and other utilities and has to be avoided.
Make the SoC stuff a little more modular, and start to move away from
having the CPU device that's a child of atmelarm that does stuff.
o Create a linker_set for the support fucntions for the SoCs.
o Rename soc_data to soc_info.
o Move the delay and reset function pointers to new soc_data struct
o Create elements for all known SoCs
o Add lookup of the SoC we found, and print a warning if it isn't one
we know about.
When an MFI command fails, the driver needs to set bio->bio_resid so that
the upper levels notice. Otherwise we see commands silently failing leading
to data corruption. This mirrors dadone()
Note that the I/O systems in FreeBSD and Solaris/Illumos are sufficiently
different that there is not a 1:1 mapping from scripts that work
with one to the other.
MFC after: 1 month
PCI:
- Properly handle interrupt fallback from MSIX to MSI to legacy.
The host may not have sufficient resources to support MSIX,
so we must be able to fallback to legacy interrupts.
- Add interface to get the (sub) vendor and device IDs.
- Rename flags to VTPCI_FLAG_* like other VirtIO drivers.
Block:
- No longer allocate vtblk_requests from separate UMA zone.
malloc(9) from M_DEVBUF is sufficient. Assert segment counts
at allocation.
- More verbose error and debug messages.
Network:
- Remove stray write once variable.
Virtqueue:
- Shuffle code around in preparation of converting the mb()s to
the appropriate atomic(9) operations.
- Only walk the descriptor chain when freeing if INVARIANTS is
defined since the result is only KASSERT()ed.
Go ahead and disable the interrupts for the DBGU the boot loader may
have left enabled after we detect the CPU, and remove the multiplely
copied code from the SoC modules.
Move the ffclock symbols from FBSD_1.2 to FBSD_1.3 where they should have been
put initially. They were added to head during development of 10-CURRENT, not
9-CURRENT.
adrian [Tue, 10 Jul 2012 07:45:47 +0000 (07:45 +0000)]
Add some debugging and comments about what's going on when reinitialising
the FIFO.
I still see some corner cases where no RX occurs when it should be
occuring. It's quite possible that there's a subtle race condition
somewhere; or maybe I'm not programming the RX queues right.
There's also no locking here yet, so any reset/configuration path
state change (ie, enabling/disabling receive from the ioctl, net80211
taskqueue, etc) could quite possibly confuse things.
adrian [Tue, 10 Jul 2012 06:05:42 +0000 (06:05 +0000)]
Add/fix EDMA RX behaviour.
* For now, kickpcu should hopefully just do nothing - the PCU doesn't need
'kicking' for Osprey and later NICs. The PCU will just restart once
the next FIFO entry is pushed in.
* Teach "proc" about "dosched", so it can be used to just flush the
FIFO contents without adding new FIFO entries.
* .. and now, implement the RX "flush" routine.
* Re-initialise the FIFO contents if the FIFO is empty (the DP is NULL.)
When PCU RX is disabled (ie, writing RX_D to the RX configuration
register) then the FIFO will be completely emptied. If the software FIFO
is full, then no further descriptors are pushed into the FIFO and
things stall.
This all requires much, much more thorough stress testing.
Always clear p_xthread if current thread no longer needs it, in theory, if
debugger exited without calling ptrace(PT_DETACH), there is a time window
that the p_xthread may be pointing to non-existing thread, in practical,
this is not a problem because child process soon will be killed by parent
process.
- Remove the unused and not completed write support for NTFS.
- Fix a bug where vfs_mountedfrom() is called also when the filesystem
is not mounted successfully.
adrian [Mon, 9 Jul 2012 23:58:22 +0000 (23:58 +0000)]
Add some AR9300 HAL descriptor definition changes.
* Add a couple of RX errors;
* Add the spectral scan PHY error code;
* extend the RX flags to be a 16 bit field, rather than an 8 bit field;
* Add a new RX flag.
Fix a bug in code that calculates the number of the first interrupt
vector for a port. This affected the gigabit ports of T422 cards (the
ones with 2x10G ports and 2x1G ports).
Add a clts() wrapper around the 'clts' instruction to <machine/cpufunc.h>
on x86 and use that to implement stop_emulating() in the fpu/npx code.
Reimplement start_emulating() in the non-XEN case by using load_cr0() and
rcr0() instead of the 'lmsw' and 'smsw' instructions. Intel explicitly
discourages the use of 'lmsw' and 'smsw' on 80386 and later processors in
the description of these instructions in Volume 2 of the ADM.
Partially revert r217515 so that the mem_range_softc variable is always
present on x86 kernels. This fixes the build of kernels that include
'device acpi' but do not include 'device mem'.
In epair_clone_destroy(), when destroying the second half, we have to
switch to its vnet before calling ether_ifdetach(). Otherwise if the
second half resides in a different vnet, if_detach() silently fails
leaving a stale pointer in V_ifnet list, and the system crashes trying
to access this pointer later.
Another solution could be not to allow to destroy epair unless both
ends are in the home vnet.