FreeBSD 5.4 Open Issues
Open Issues
This is the beginning of a list of open issues that need to be worked on
or resolved for FreeBSD 5.4. If you have any updates for this list, please
e-mail re@FreeBSD.org. The feature list for 5.4 is still a work in
progress, so items may be added, removed, or modified before we're done.
Show stopper defects for 5.4-RELEASE
Issue |
Status |
Responsible |
Description |
Required features for 5.4-RELEASE
Issue |
Status |
Responsible |
Description |
Desired features for 5.4-RELEASE
Issue |
Status |
Responsible |
Description |
KAME IPSEC without the Giant Lock |
In progress |
George V. Neville-Neil |
Currently, the KAME IPSEC implementation contains inadequate locking
to operate without the Giant lock over the network stack, forcing
kernels compiled with the KAME IPSEC implementation (not FAST_IPSEC)
to run the network stack with the Giant lock, reducing parallelism,
increasing lock contention, and increasing latency by preventing
preemption. For 5.4-RELEASE, it is desirable to complete the locking
work for KAME IPSEC so that it can run without the Giant lock. |
All Network Interface Drivers MPSAFE |
In progress |
-- |
Currently, some network interface drivers are not safe without the
Giant lock due to missing synchronization. These drivers are protected
by running non-INTR_MPSAFE and with the IFF_NEEDSGIANT flag set, which
cause interrupt threads to acquire the Giant lock before executing the
driver's interrupt handler, and to perform if_start (interface
transmit start) asynchronously once the Giant lock can be acquired.
This results in these drivers performing less well due to increased
lock contention, decreased ability to preempt, and latency associated
with asynchronous launching of latency-critical events. For 5.4, all
network drivers should be able to operate without the Giant lock. |
NetIPX without the Giant Lock |
In progress |
Robert Watson |
Currently, the IPX/SPX (netipx) implementation contains inadequate
locking to operate without the Giant lock over the network stack,
forcing kernels compiled with IPX support to run the network stack
with the Giant lock, reducing parallelism, increasing lock contention,
and increasing latency by preventing preemption. For 5.4-RELEASE, it
is desirable to complete the locking work for IPX so that it can run
without the Giant lock. |
Fix regression in file(1) |
-- |
-- |
The new version of file(1) does not cross-build properly. If
sparc64 executables are built on i386 the compiled magic numbers
file that gets installed is wrong and file(1) doesn't work
properly. |
SIGABRT under load |
In progress |
Alan L. Cox |
Under very high load (Kris sees this on the package builders) a
limitation of 16 sumultaneous exec's may be hit in exec_map().
A workaround has been implemented with a fix planned for 5.5. |
DDB_UNATTENDED may not work |
Not done |
-- |
There are reports of problems getting dumps after panics. |
fix for ttywakeup panic |
In progress |
Doug White |
There appears to be a race condition opening/closing ttys. |
Documentation items that must be resolved for 5.4
Issue |
Status |
Responsible |
Description |
Testing focuses for 5.4-RELEASE
Issue |
Status |
Responsible |
Description |
File descriptor locking |
Done |
Jeff Roberson |
File descriptor locking needs to be merged back from HEAD. |
IPFilter mpsafe fixes |
Done |
Robert Watson |
IPFilter is currently not mpsafe but doesn't run with Giant. |
Update sysinstall to handle separate packages CD |
Done |
John Baldwin |
Avoid the need for a disc1-gnome/disc1-kde as was done for 4.11 |
busdma fixes |
Done |
Scott Long |
More work is needed on busdma to fix bounce buffer problems. |
AMR driver MPSAFE |
Done |
Scott Long |
Making the AMR driver MPSAFE will greatly increase its performance. |
CARP support |
Done |
Gleb Smirnoff |
Common Address Redundancy Protocol allows multiple hosts on the same
local network to share a set of IP addresses. |
kstack overflows in softupdates |
Done |
-- |
There had been reports of kernel stack overflows in softupdates code
that caused filesystem problems on reboot. This was fixed during the
early stage of code freeze but should be tested. |
ATAPI CDROMs in PIO mode during install |
Done |
Ken Smith |
People are having difficulties with various ATAPI CDROMs on
various architectures. We need to either do all installs
in PIO mode or provide a boot menu method of selecting it. |
truss not working |
Done |
Jeff Roberson |
See PR kern/78664. |
Deadlock under heavy interrupt load on MP Opteron systems |
Done |
Doug White |
A deadlock related to delivering IPIs on AMD 64-bit processors has
been identified. |
SCHED_ULE update |
Done |
Jeff Roberson |
Many improvements have been made to the ULE scheduler in 6-CURRENT.
These should be merged back to 5.4. The merging is done but ULE is
still known to cause panics for some people, especially on SMP
systems. Try it with extreme caution. |