This patch corrects the following: - Fixes a problem in which a recursive panic occurs during certain lockmode violations. - Fixes a problem with the CPU auto_action console environment variable. If the auto_action console environment variable is set to BOOT or RESTART, when the CPU is to be stopped, the processor immediately boots and the user cannot observe that the CPU had halted. - Fixes a problem that occurs on systems with more than 256 MB of memory in which the write performance of the dumpsys file is very slow. - Corrects a potential problem in the handling of a ieee_get_state_at_signal(3) C-library call. - Fixes a pmap_enter() race condition that can lead to a kernel-stack-not-valid halt, which subsequently causes a "tb_shoot ack timeout" panic. Also fixes crash-dump code to properly dump PMR data from upper memory regions on systems with memory holes. - Corrects a simple lock timeout problem in several vm_page routines. - Fixes a problem where a file system is busy when trying to unmount it. - Corrects a problem where we need to hold file references on raw device accesses. - Fixes a problem when a processor is commanded to stop during a heavy load but does not actually halt. - Fixes a problem when a setuid program is exec'ed, and the error message "privileges disabled because of outstanding IPC access to task" is issued. - Corrects a potential boot panic problem by limiting the size of the bufcache. - This patch fixes the following two problems that occur on an NFS file server using a Network Appliance server: o New files may not be listed in directory reads. For example, when the ls command is used not all the files may be listed. o When a directory listing is requested from a Network Appliance server, more data than was requested may be returned and the extra data is lost by the Digital UNIX client. The problem can be seen by doing using the ls command; not all the files on the server are listed. - A flaw in NFS client operation can result in a KMF panic. - Fixes two separate problems which are: 1) This patch fixes a problem where a user application may receive an EIO error back from an fsync() call. 2) A message size of zero passed to msgsnd() can result in a kernel memory fault panic. - Fixes two separate problems which are: 1) This corrects a performance problem with posix timers. 2) This patch avoids a "kernel memory fault" panic from sigsgdisp(). The problem has only been seen when shutting down an Oracle database. - Corrects performance related problems. A system running low on free memory or showing heavy pagein activity will need this patch. - Corrects a small accounting problem where the measured time for a process was an integral rather than mean value. - Fixes several problems in AdvFS. - Includes several corrections for AdvFS, ufs and vfs. - This patch must be installed to allow the X.25 WAN layered product to run on DIGITAL UNIX V4.0E. - Corrects a system performance problem on a freshly booted AlphaServer 4000 and similar systems. - Fixes a routing corruption that could been seen as a kernel memory fault or a corruption within the 128 byte kernel memory bucket. - Corrects a problem where incorrect NFS client locking caused a KFM panic. - Fixes a problem where NFS clients may hang in the uninterruptable state. - Fixes a problem in which the msleep() syscall was not marked as a cancellation point. This problem occurred only if your C programs were complied using the -pthread switch. - Fixes a problem that can cause an NFS client application to hang, or causes a "lock already owned by thread" panic when lockmode=4. - Fixes a problem with rpc where heavy traffic (such as mail over NFS) may cause performance problems. - Fixes a problem with poor performance of NFS/UDP over a GigaBit Ethernet network interface (DEGPA). - Fixes a rare problem in the NFS subsystem in which a "kernel memory fault" panic will be seen originating from the free() routine either when the NFS server daemon (nfsd) is processing a request or is shut down. - Fixes problems with sqrt instruction emulation and a rare rounding problem with denormalized values. - Fixes a problem where the system crashes with the following error message: lw_remove: light weight wiring(s) found - Fixes a problem that causes a panic to occur at boot time if the system has a large boot time memory requirement and only a portion of memory has been tested. - Fixes a dealock that can occur when a thread is in sigwaitprim(), and a second signal in the sigwait set is being delievered. An example stack from of the sigwait thread is: simple_lock_time_violation() mpsleep() sigwaitprim() syscall() _Xsyscall() And the delievering threads stack would be: psignal_internal() kill() syscall() _Xsyscall() - Fixes a problem in the subsystem configuration bootstrap process where a status variable might not get initialized correctly and therefore subsequent entries from /etc/sysconfigtab might not get set correctly. - Corrects a "simple_lock: time limit exceeded" panic in softclock_scan(). - Fixes a panic which has the following error message: simple_lock: time limit exceeded - Fixes an LMF license problem where LMF fails to decrement the OSF-USR license when xdm is selected as the default window manager. - Fixes problems with sqrt instruction emulation when operating on IEEE Signaling NaNs or other IEEE special values. - Fixes a problem where a network hang can occur because the network isr thread is not running at realtime priority. As a result, some of the realtime daemons or user programs may preempt the netisr thread and hang network activity. - A potential security vulnerability has been discovered, where under certain circumstances, system integrity may be compromised. This may be in the form of improper file or privilege management. Compaq has corrected this potential vulnerability. - Fixes resolves a performance problem associated with page coloring for real time applications (rt_preempt_opt=1). There is a new tuneable, "vm_page_color_private", which is modified in the "vm" section of the sysconfigtab file. The default value for this variable is "0", so to enable this feature the variable must be set to "1". - Corrects a problem where a signal is delievered, but not responded to by the target process. - Fixes a problem in which sem_wait() does not get interrupted by a SIGCHLD signal if the program is compiled/linked -pthread. - Changes the behavior of mlockall() to make a process unswappable. - Fixes a panic of "get_color_bucket: empty buckets" when the sysconfig attribute "private-cache-percent" is non-zero. - Fixes a prior regression that prevented full crash dumps from working properly. - Fixes several problems in the kernel. 1) A panic with the message vm_unwire: page is not wired 2) A panic with the message kernel_object_bad: bad operation 3) A system hang due to deadlock between the swapin thread and ps both accessing the same task. - Provides functionality to allow the detection of unlinked referenced files. This is to assist customers in recovering disk space. - Fixes a problem in which the kernel misses profiling samples for multithreaded applications that are implemented to use system contention scope (SCS) threads. - In multithreaded programs, the fork() system call was failing to preserve the floating point state. - Fixes a system hang caused by an infinite loop with out-of-band networking data. - Fixes a problem where NFS does not update mtime and atime for special files and named pipes. - Fixes a kernel problem, where proper locking/reference count management was not being performed. This could result in a "lock-terminate: lock held" system panic. - Fixes a panic with the following panic string: "pgmv: session leader attempted setpgrp" - Fixes an kernel memory fault caused by a streams SMP race condition. - Fixes a problem that occurs on AdvFS systems. The system will panic with the following error message: malloc_overflow: guard space corruption - Fixes a problem where several processes accessing the same AdvFS file can hang in ubc_lookup(). - Fixes an kernel memory fault caused by a streams SMP race condition. - Fixes a problem where a system panic will occur when accessing an ISO9660 format CDROM. - Fixes the following problems: o A kernel memory fault system panic in routine spec_reclaim. o When executing the "file" command against a lat (BSD) special device, the "file" process will hang. o On multiCPU systems, hangs can occur in the revoke system call when multiple threads attempt to call "revoke" at the same time. - Fixes a problem where process accounting data was not written to the accounting file when it was on an NFS-mounted file system. This problem occurred on Dataless Management Services (DMS) client systems. - When a thread issues a resume request against another thread before it has actually suspended, the target thread is incorrectly suspended. At times it may suspend indefinitely. - nxm_get_state() system call did not return a failure status if a target thread is not blocked at the time the call is executed. - Provides platform support for software interleaving on the AlphaServer 4000 family of platforms. - Support for two new DMA driver function prototypes that were added in V4.0F; dma_get_num_val_sgentrys() and dma_get_sgentry_reset(). These functions provide source and binary compatibility across Y2K-compliant versions of Digital/Tru64-UNIX. - Fixes a kernel memory fault from socket code. The kernel memory fault results from failing to get a lock on a list of threads which have requested resources on a socket. - Fixes a system hang in which there is a large number of pending ioctl's on the streams queue. - Fixes a problem that causes system panics when thread_swappable is called with the current_thread as the target thread, when the thread is about to be swapped out. - Fixes a problem with the map entry indexing scheme that results in the following panic: "pmap_release_page: page not found" - Fixes a restart detection problem with the proplistd daemon. Prior to this fix, when mounting a relocated ASE NFS service with property lists, clients did not detect that the proplistd RPC port number had changed. Clients continued to use the proplistd RPC port number of the old ASE cluster member. - Fixes a problem in which certain invalid kernel address ranges may get ignored.This can result in invalid kernel memory accesses to be left unnoticed. - Fixes a problem that causes the Tru64 UNIX Version 5.0 update install procedure to exit with core dumps and /sbin/loader failures on a system. - Fixes a problem with CDFS. Data corruption occurs when reading beyond the end of a partition. - Fixes a problem in the module core() that can cause a panic with the message: vrele: bad ref count. - Fixes 2 separate problems: 1) Panic in the kernel with the following error message: simple_lock: time limit exceeded 2) A panic occurs when booting kernel interactively and setting the memlimit. The panic error message is as follows: kernel memory fault - Fixes a UFS file system panic with the following error message: malloc_check_checksum: memory pool corruption - Improves UFS performance by no longer prematurely writing single pages. - Fixes system crashes seen on ASE or Cluster systems when changing the network interfaces. The stack is not informative and the panic may be "trap: illegal instruction," or "kernel memory fault." - Fixes a problem where process accounting data was not written to the accounting file when the accounting file was on an NFS-mounted file system. - Applications using fcntl() system calls may appear to hang. - Fixes "simple_lock: time limit exceeded" panics. - Patch for a Compaq AlphaServer DS20. A fix has been made to the handling of power supply, temperature, and fan events. - Fixes an nfs/ufs/vm deadlock. While serving a client, the system running ASE/DT as an NFS server can hang with deadlock. - Fixes a problem in which the system may panic with the error message "kernel memory fault". - Fixes two problems; fork can fail to obtain swap space and the resource limitation on core files doesn't work as documented. - Fixes a system hang condition. All NFS related services may deadlock. - Fixes a problem where the system can panic with the console message, "bs_bf_htop: invalid handle\n N1 = 0". - Fixes the following problems: o A system panic, with panic string "simple_lock_terminate: lock busy". o A system panic, with panic string "lock_terminate: lock held". o Removes the restriction of the maximum number of threads allowed system-wide. - Fixes a problem with kdbx. A core file created by kdbx was left in the root directory when recovering from a system crash. - Fixes a problem in mountd. The NFS server allows read/write access to clients not on the exports list and other clients to be incorrectly denied access. - Fixes a problem in mountd where multiple "A" records in DNS database for an NFS client will not always succeed. - Adds automatic detection of a cdfs file system for the mount(8) command. - Fixes a problem with the mount command where it sometimes kills other processes. - Fixes a problem in which mount would incorrectly fall back to Version 2 after certain errors had been encountered using Version 3. - Fixes a system "pause" seen when doing a lot of I/O to UFS filesystems. - Fixes a problem in which the "vmstat -M" command incorrectly matches bucket numbers and bucket indices. - Fixes the database application core dumps when using truss/trace tools by remembering that COW has been setup on a shared pte and process it correctly when a subsequent write access is made to the page. - Fixes a problem with NFS locking when lockmode is set to 4. The system can panic with "lock_write: hierarchy violation" or "lock_write: simple lock owned". - Fixes a problem in which system may memory fault if TCR/ASE server no longer had access to the cdrom device. - Fixes a data corruption problem that can occur when mapping to private regions. - Improves nfs performance by avoiding an undesired cache invalidation that can have a big impact on NFS performance with huge files. - Forces the kernel malloc code to print failure messages only when kmem_debug is enabled, and also corrects some other minor kmem_debug problems. - Fixes a problem where AS1200 systems with more than 3 pairs of memory displays the following warning message on the console during boot: "pmap_get_align: Unaligned memory hole found... Please reset the system to clear any previous memlimit" - Fixes a kernel memory fault caused when a network application walked an inpq array. - Fixes system crashes seen on ASE or Cluster systems when changing the network interfaces. The stack is not informative and the panic may be "trap: illegal instruction," or "kernel memory fault." - Fixes a problem in which signals can be lost in multithreaded applications. - Fixes two problems which are: 1) Locking/unlocking of a segmented shared memory region causes the system to panic with the following message: (cpu 0): vm_page_ssm_unwire. 2) The segmented shared memory does not handle simultaneous shared memory locks and unlock causing the system to panic with the following message: ssm_anon_free: wired/held/busy page - Fixes a problem with soclose() that caused permanent looping on exit while aborting pending connections at a TCP/IP listener socket. - Fixes a problem that only occurs if realtime preemption is enabled and SMP test suites are run. - Fixes a problem with virtual memory. When running the Oracle database, Oracle can not detach from a shared memory segment. - Fixes single-step support in a debugger, such as Ladebug, for instructions that trap or fault. - Fixes an incorrect calculation for memory-usage-by-type when kmem_debug is set. - Fixes a simple_lock: hierarchy violation in sigq_abort() when lockmode is set to 4. - Fixes a system panic on multi-process systems (approximately 12 cpus) with large memory (128GB). The system can panic with panic: lock time on vm_page_free_lock - fixes a problem where the system can panic with the panic string "secsize_resid < d_reclen" when accessing a defective cdrom. - Fixes a problem where partitioned turbolasers return incorrect cpu data for cpus that are not in the partition. - Fixes a problem where ubc_msync() may not flush out all the pages in the requested range. - Fixes var adm messages from truncation on larger configurations by raising the default size (4096) of 'msgbuf_size' to 8192. - Fixes a problem where systems with the BL13 patch kit installed would run low on kernel memory after process accounting had been running for a while. - Fixes the AdvFS race between extending and reading subextent maps and avoids the panic occured in advfs bs_get_bf_xtnt_map code (qar 61985). This patch also fixes the race with removing storage causing kernel memory fault (qar 66890). - Fixes a problem that could result in a incorrect scheduling of threads when they were dispatched from the idle state. - Fixes a problem with making a msfs_putpage() call. The length argument may get its upper bits truncated which will result in an incorrect length calculation. - Fixes a panic in AdvFS which can have the following error messages: panic (cpu 1): bs_cow_pg: pin clone err -OR- panic (cpu 1): bs_cow_pg: cannot get blkMap - Fixes a kernel memory fault caused by a mishandling of multicast addresses on the FDDI interface. - A correction to the port allocation code now prevents a user from allocating port zero if ipport_userreserved is set to 65535. - Fixes a problem where Digital UNIX systems may panic with one of the following strings while attempting to unlock a socket. "Unaligned kernel space access from kernel mode" "kernel memory fault" - Fixes a problem where resets are seen due to proxy packets getting into the ipintrcacheaddr cache. - Fix for AOL systems running Inktomi code. It provides enabling hooks for Inktomi caching server code. - Fixes a kernel memory fault and an SMP race condition with the AltaVista Firewall 98 server on a multi-cpu system. - Fixes a problem when a default IP address and a cluster virtual IP address get interchanged after a network restart. The default interface address is used by all outgoing traffic and the alias address is only usable for the incoming packets. - Fixes a problem on systems running screend where TCP/IP fragments are being dropped. - Fixes a problem where RCP commands issued from a Sun Solaris system to Compaq Tru64 UNIX may sometimes fail incorrectly with the error message "Connection reset by peer." - Fixes a TCP performance problem if the TCP window scale option is turned off when using the HIPPI interface. - Fixes a system panic: "tcphdr too big". - Consists of changes necessary for AV firewall 98 to pass ICSA certification. - This patch involves virtual mac addressing.