PROBLEM: (81943) (PATCH ID: OSF510-042) ******** This patch is required for support of the SuperDLT1 tape drive. If the patch is not present, opening the tape will fail. PROBLEM: (82100, 82161) (PATCH ID: OSF510-048) ******** This fixes a problem encountered on a heavily loaded HSG80, in which a device may become unavailable to other cluster members if a cluster node crashes at the same time an error occurs on that device. PROBLEM: (81814) (PATCH ID: OSF510-010) ******** This patch prevents panics from occurring if AdvFS detects corruption in the per-fileset frags file and attempts to work around the corruption. The problem would occur only after the following message/event were registered: Warning: AdvFS has detected an inconsistency in a frag metadata file free list. PROBLEM: (81258) (PATCH ID: OSF510-014) ******** AdvFS filesets mounted with the -o noatimes option or the -o ro (readonly) option should not update the file access times as seen using the "ls -lu" command. Prior to this patch, memory mapped files were having their access times updated even if the noatimes or readonly mount options were specified. This patch corrects that problem. PROBLEM: (79769) (PATCH ID: OSF510-015) ******** A kernel memory fault can occur on an smp machine when one thread is extending a clone frags file and another thread does a stat system call on a file with a frag. The stack trace is as follows: 5 panic src/kernel/bsd/subr_prf.c : 804 6 trap src/kernel/arch/alpha/trap.c : 1762 7 _XentMM src/kernel/arch/alpha/locore.s : 1748 8 x_page_to_blk src/kernel/msfs/bs/bs_stg.c : 6191 9 x_page_to_blkmap src/kernel/msfs/bs/bs_stg.c : 5928 10 x_page_mapped src/kernel/msfs/bs/bs_stg.c : 6281 11 bs_frag_has_stg src/kernel/msfs/bs/bs_bitfile_sets.c : 2229 12 msfs_getattr src/kernel/msfs/osf/msfs_vnops.c : 1525 13 vn_stat src/kernel/vfs/vfs_vnops.c : 1277 14 stat1 src/kernel/vfs/vfs_syscalls.c : 3056 15 lstat src/kernel/vfs/vfs_syscalls.c : 3034 16 syscall src/kernel/arch/alpha/syscall_trap.c : 627 17 _Xsyscall src/kernel/arch/alpha/locore.s : 1512 PROBLEM: (81251, 83087) (PATCH ID: OSF510-087) ******** This patch provides an improvement to AdvFS performance when the first bytes of user data (and subsequent storage requests) is written to a domain. PROBLEM: (82689) (PATCH ID: OSF510-060) ******** This patch corrects read-ahead behavior for AdvFS for both local and NFS reads. Read performance is increased by approximately 10% with the addition of this patch. This patch does not include any 'correctness' fixes. This read performance fix can be detected by timing the read requests or by monitoring internal kernel paths. PROBLEM: (#, -----, 81882, BUG) (PATCH ID: OSF510-011) ******** Under a specific set of unlikely circumstances it is possible for revision 4 PCA hardware to falsely report PCI hung bus errors, which will cause a uncorrectable hardware machine check and operating system panic. The following console error messages indicate this problem has occurred. The "PCI bus hung during Fault" text in the error message is the key to detecting this error. Note: the false PCI bus hung error only occurs with revision 4 PCA hardware. iop: ioa_err_sum = 0x100000000 Bit 32: Hose 0 PCA reported an Uncorrectable Error pca: pca_whatami = 0x8000034 Bit 2: ASIC revision Bit 4: Backplane revision Bit 5: Backplane revision Bit 27: Microprocessor present pca: pca_err_sum = 0x8000 Bit 15: PCI bus hung during Fault pca: ne_whatami = 0x100000302 pca: fe_whatami = 0x100000202 pca: pci0_err_sum = 0x10068000 Bit 15: PCI bus hung during Fault Bit 17: Failing PCI command Bit 18: Failing PCI command Bit 28: PCA was PCI initiator when bus was hung pca: pci0_err_addr = 0x2030000 pca: pci1_err_sum = 0x4060000 Bit 17: Failing PCI command Bit 18: Failing PCI command Bit 26: PCI slot that was active during the failure pca: pci1_err_addr = 0xc0011500 Machine Check SYSTEM Fatal Abort Machine check code = 0x100000202 Ibox Status = 0000000000000000 Dcache Status = 0000000000000000 Cbox Address = 0000000000000000 Fill Syndrome 1 = 0000000000000000 Fill Syndrome 0 = 0000000000000000 Cbox Status = 0000000000000000 EV6 captured status of Bcache mode = 0000000000000000 EV6 Exception Address = fffffc000097aff0 EV6 Interrupt Enablement and Current Processor mode = 0000007ee0000000 EV6 Interrupt Summary Register = 0000002000000000 EV6 TBmiss or Fault status = 0000000000000000 EV6 PAL Base Address = 0000000000020000 EV6 Ibox control = fffffffc0c306396 EV6 Ibox Process_context = 0000718000000004 Cpu Fault Summary = 0x1 QBB bit-directed reg. dump QBB 0 CSRs to be logged summary = 0x1100 QBB 1 CSRs to be logged summary = 0x0 QBB 2 CSRs to be logged summary = 0x0 QBB 3 CSRs to be logged summary = 0x0 QBB 4 CSRs to be logged summary = 0x0 QBB 5 CSRs to be logged summary = 0x0 QBB 6 CSRs to be logged summary = 0x0 QBB 7 CSRs to be logged summary = 0x0 panic (cpu 0): System Uncorrectable Machine Check PROBLEM: (82320) (PATCH ID: OSF510-032) ******** This patches fixes a kernel memory fault which can occur during scheduler load balancing on a NUMA system. The problem can be identified by a stack trace which will start with a syscall to routine numa_move_current_thread and end with the memory fault in routine u_map_is_locked (~line 2770). PROBLEM: (82179) (PATCH ID: OSF510-006) ******** This patch fixes a panic that occurs in madvise() when called with MADV_DONTNEED when running in lockmode 4. The panic string would be "lock_done: lock not currently owned" PROBLEM: (81478) (PATCH ID: OSF510-007) ******** This patch fixes a performance problem when HPTC programs on NUMA machines migrate threads betwen QBBs. PROBLEM: (82131) (PATCH ID: OSF510-008) ******** This patch fixes a Kernel Memory Fault which can happen when all of physical memory is in use. This problem can be identified by the stacktrace of the offending thread, which will show that the illegal memory reference occurred in vm_map_entry_split(). This will probably appear as line 2706 of vm_map.c. PROBLEM: (82198, 82050) (PATCH ID: OSF510-049) ******** This patch fixes two problems. - A problem seen in a cluster when one member whose boot partition is on a device whose scsi wid changes while the node is down. - This also corrects a failure that is seen as a user_cmd timeout. "Cannot invoke dsfmgr from kernel: run /sbin/dn_setup -boot to synchronize dsfmgr with kernel" PROBLEM: (82124) (PATCH ID: OSF510-030) ******** The patch fixes a kernel memory fault when accessing a shared text segment after or during scheduler load balancing on a NUMA system. There are two different stack traces associated with the failure. The most common trace starts with routine vm_fault and ends with a memory fault in routine pmap_seg_enter (~line 6059). The second trace begins with routine numa_move_current_thread and ends with a memory fault in routine pmap_collect (~line 4094). PROBLEM: (MGO85573B, N/A) (PATCH ID: OSF510-012) ******** This patch fixes a bug such that when 'fuser -k' is issued on a dismounted NFS mount point in which some process is running, a hang will occur. A new flag, -p, has been introduced. When the -p flag is used with the -c flag, as in 'fuser -c -p -k /mnt', the parent directories are ignored and the processes are gracefully killed. PROBLEM: (82347, SSRT0700U) (PATCH ID: OSF510-023) ******** 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. PROBLEM: (81933, 82226, 82278) (PATCH ID: OSF510-047) ******** This patch improves webserver performance, fixes an IPV6 related crash and a hang in soclose(). PROBLEM: (80225, 81884, 82170, 82273, 82309) (PATCH ID: OSF510-039) ******** This patch for Compaq POSIX Threads Library fixes the following problems: 1. A threaded applications may not recover the use of a CPU following an offline/online procedure. 2. User threads blocked in the kernel against a specific VP may never unblock. 3. Problems with libpthread tracing for Totalview debugger 4. User threads are not bound to the correct RAD upon returning from pthread_rad_attach. PROBLEM: (82700) (PATCH ID: OSF510-059) ******** Prior to this patch, /sbin/dd was dependent on the threads package. This dependency caused problems when installing patches that affected the thread package. So this patch eliminates the dependency by making /sbin/dd non-threaded. PROBLEM: (82705) (PATCH ID: OSF510-065) ******** This patch fixes SPECweb99 httpd hangs in umc_get_page() routine waiting for the page. The stack on a hung process is: crash> set fffffc00324a0380 CONTEXT: CURRENT PID: 780 COMMAND: "zeus.web" THREAD: fffffc00324a0380 CPU: 3 RAD: 0 EVENT: fffffc000a27fa80 STATE: WAIT crash> st CPU: 3 PID: 780 THREAD: fffffc00324a0380 COMMAND: "zeus.web" 1: umc_get_page+356: thread_block() 2: sosend+1172: umc_get_page(0xfffffc004f0dae00, 0xa0c58000, 0xfffffc000 07b01e8) 3: rwuio+212: sosend(0xfffffc00f1ac6700, 0x0, 0xfffffe0689d2f7a0, 0x0, 0x0, 0xfffffe0689d2f690) 4: writev+236: rwuio(???, ???, 0xfffffe0689d2f8a8, 0xfffffe0689d2f7a0, 0x1) 5: syscall+592: writev(0xfffffc000c846cc0, 0xfffffe0689d2f8b8, 0xfffffe0 689d2f8a8) 6: _Xsyscall+92: syscall(0x0) PROBLEM: (82585) (PATCH ID: OSF510-073) ******** This patch includes performance fixes for systems doing raw I/O, raw async I/O, and systems with large disk farms (high disk count). PROBLEM: (QAR82396) (PATCH ID: OSF510-084) ******** This patch enables the "getconf VENDOR_ABBREV" command to return the correct abbreviated vendor name, for example, "COMPAQ" instead of "Compaq Computer Corporation" in a Tru64 UNIX system. PROBLEM: (none) (PATCH ID: OSF510-063) ******** This patch provides the device driver for a new graphics card. PROBLEM: (HPAQ418B4) (PATCH ID: OSF510-053) ******** This patch fixes a problem where some network based multimedia applications will cause a kernel memory fault when exiting. A sample stack trace is as follows: 1 panic(s = 0xfffffc00007d10a0 = "kernel memory fault") 2 trap 3 _XentMM 4 igmp_leavegroup 5 in_delmulti 6 ip_freemoptions 7 in_pcbfree 8 in_pcbdetach 9 udp_usrreq 10 soclose 11 soo_close 12 closef 13 close_ufe 14 exit 15 rexit 16 syscall 17 _Xsyscall PROBLEM: (none) (PATCH ID: OSF510-050) ******** This patch provides support for the DEGPA-TA (1000BaseT) Gigabit Ethernet device. PROBLEM: (BCGMA09HM) (PATCH ID: OSF510-064) ******** This patch fixes a potential deadlock on systems using shared memory segments and granularity hints. This can occur when allocating a gh region larger then the available free memory. Example stack traces for deadlocked threads from a forced crash: 0 thread_block 1 gh_collect 2 gh_free_list 3 gh_free_granhint 4 u_shm_oop_deallocate 5 vm_map_entry_delete 6 u_map_delete 0 thread_block 1 gh_collect 2 gh_free_list 3 alloc_gh_phys 4 nshmget 5 syscall 6 _Xsyscall PROBLEM: (82227) (PATCH ID: OSF510-035) ******** This patch improves UDP performance by removing an unneeded lock from the UDP output path. PROBLEM: (BCSM412FX) (PATCH ID: OSF510-062) ******** This patch fixes a panic in in_pcbfree() when NFS is implemented over TCP. If an NFS server stops servicing requests to clients and the server is rebooted to clear the NFS problem, the clients panic in in_pcbfree(). A typical client stack trace is: panic in_pcbfree tcp_input ipintr PROBLEM: (BCGMA0M0S) (PATCH ID: OSF510-089) ******** This patch fixes a lock contention for multiple writers which would use 100% of CPU time. This problem has been seen when running Oracle database doing 'Table Creates'. Tru64 UNIX continues to add UBC buffers which causes their free memory to go from 6 gb to 200 mb, and does not release them back. The Oracle database has to be shut down to recover UBC buffers. PROBLEM: (81889, n/a) (PATCH ID: OSF510-095) ******** This patch resolves hang-like behavior when LSM volumes are used to create AdvFS domain volumes. With some systems the default preferred IO byte transfer size for read and write requests is initially set too high. Run the "chvol" command on the domain volumes and see if the rblks or wblks field is very large. These are the sizes of the read and write IO transfer size that AdvFS uses to issue reads or writes, respectively. In some cases, the transfer size converted to bytes may be as large as 1GB or 2GB's. The patch will cause AdvFS during domain activation to check if an AdvFS volume's underlying driver is LSM. If the default "preferred" IO byte transfer size is 64MB's or larger for LSM drivers, then AdvFS will instead use a lower size of 256KB's (512 blocks). If a larger value is desirable, then you can still use "chvol" to change the transfer size for reads or writes. "chvol -l" command will display the preferred, maximum and minimum transfer sizes. The patch also eliminates confusing warning messages about exceeding the preferred read or write transfer size seen when the IO transfer size has been set higher than the preferred transfer size default. PROBLEM: (none) (PATCH ID: OSF510-094) ******** This patch fixes periodic slowdowns seen on large systems that are consuming large amounts of memory due to file I/O. These changes make the reclaiming of memory in use for file buffers more efficient. There is also a fix for a lock timeout seen on the vdIoLock because of a large number of buffers on the smoothsync queues. PROBLEM: (83490) (PATCH ID: OSF510-101) ******** This patch fixes a race condition introduced in v51supportos-94-beltz and should be applied with that patch. PROBLEM: (HPAQ30LKL) (PATCH ID: OSF510-097) ******** This patch fixes inaccuracy problems when using setrlimit/getrlimit with a threaded application. PROBLEM: (81007, 82961) (PATCH ID: OSF510-119) ******** This patch addresses multiple issues for the KZPCC family of RAID Array 2000 (RA2000) controllers. - Errors seen when concurrent opens are issued to separate logical partitions on the same logical device. - Change to the preferred chunk size from 16 KB to 64 KB which may increase data transfer rates. PROBLEM: (BCGM919P3, EVT376821B, EVT376821, 82934) (PATCH ID: OSF510-110) ******** This patch fixes a hang seen while running collect and the vdump utility. This patch prevents the hang in tok_wait from occurring. PROBLEM: (83842) (PATCH ID: OSF510-124) ******** This patch prevents stat(), lstat(), fstat(), statfs(), fstatfs(), getmntinfo(), and getfsstat() from returning EOVERFLOW errors for programs compiled on Tru64 UNIX v4.0* or earlier. A program that is compiled on v4.0* or earlier and was linked static (cc -non_shared) and calls the getcwd() or statvfs() functions can fail. Any program that was compiled on v4.0* or earlier that calls statfs(), fstatfs(), getmntinfo(), or getfsstat() could fail, such as the Oracle 8.1.7 install. PROBLEM: (MGO10736A, EVT0396650, TPOB36405, BCSM10NZK) (PATCH ID: OSF510-175) ******** This patch fixes a problem where threads can hang in x_load_inmem_xtnt_map() when called from x_page_to_blkmap(). A typical hung thread will have the following calls at the top of its stack trace: 0 thread_block 1 lock_read 2 x_load_inmem_xtnt_map 3 x_page_to_blkmap 4 x_page_to_iolist 5 blkmap PROBLEM: (82597) (PATCH ID: OSF510-078) ******** This patch fixes a kernel memory fault when writing to /proc. This issue can be seen as a kernel memory fault or possibly other failures in u_anon_rss_enforce. This issue is predominently seen as a KMF, the condition only occurs when anon_rss_enforce is set to 2 (HARD LIMIT). A typical stack trace is as follows: ... 4 panic src/kernel/bsd/subr_prf.c : 1339 5 trap src/kernel/arch/alpha/trap.c : 2262 6 _XentMM src/kernel/arch/alpha/locore.s : 2126 7 u_anon_rss_enforce src/kernel/vm/u_mape_anon.c : 4666 8 u_anon_fault src/kernel/vm/u_mape_anon.c : 1277 9 u_map_copy_overwrite src/kernel/vm/vm_umap.c : 2084 10 procfs_write src/kernel/procfs/procfs_vnops.c : 2222 11 vn_write src/kernel/vfs/vfs_vnops.c : 1372 12 rwuio src/kernel/bsd/sys_generic.c : 2257 13 write src/kernel/bsd/sys_generic.c : 2179 14 syscall src/kernel/arch/alpha/syscall_trap.c : 719 15 _Xsyscall src/kernel/arch/alpha/locore.s : 1796 Typical reproduction of this issue can be seen by doing the following: 1) Add "anon_rss_enforce=2" to the vm section of /etc/sysconfigtab 2) Reboot with new setting 3) Run dbx on any a.out file "$ dbx a.out" PROBLEM: (VNO65701A) (PATCH ID: OSF510-159) ******** This patch fixes an issue with lightweight wiring of pages and shared memory regions. This problem can occur when running Oracle and may cause ORA-01034: Oracle not available" errors when attempting to connect to Oracle databases. PROBLEM: (83520, 83540, 83568, 83645, 83725, 83862) (PATCH ID: OSF510-196) ******** This patch fixes a system panic when the system has at least one AdvFS domain and the system is configured for lockmode=4 kernel lock statistics collection. The system panic message will be: panic (cpu 0): lock_read: hierarchy violation The crash-data file will show (typical example): lock_read: hierarchy violation pc of caller: 0xffffffff00183ccc lock address: 0xfffffc000757eb30 lock info addr: 0xfffffc0000f582a0 lock class name: bfAccessT.putpage_lk class already locked: vdT.mcell_lk current spl level: 0 with a stack trace that always includes "msfs_putpage": 4 panic(s = (unallocated - symbol optimized away)) ["../../../../src/kernel/bsd/subr_prf.c"] 5 lock_fault(lp = (unallocated - symbol optimized away), ["../../../../src/kernel/kern/lock.c"] 6 lock_read(l = (unallocated - symbol optimized away)) ["../../../../src/kernel/kern/lock.c"] 7 msfs_putpage(0xfffffc00039af100, ...) ["../../../../src/kernel/msfs/osf/msfs_misc.c"] 8 ubc_page_alloc(0x0, ...) ["../../../../src/kernel/vfs/vfs_ubc.c"] In all cases the "lock class name" is "bfAccessT.putpage_lk" but the "class already locked" may be other locks such as: class already locked: vdT.del_list_lk class already locked: vdT.stgMap_lk class already locked: bfAccessT.xtntMap_lk PROBLEM: (83930) (PATCH ID: OSF510-107) ******** This change corrects some I/O rate fluctuations and thread unresponsiveness that had been seen when vm free pages dropped to a low level and used pages were being recycled. PROBLEM: (84601) (PATCH ID: OSF510-126) ******** If Oracle customers are running one of the affected Oracle configurations, Oracle may have already detected an inconsistency in the database and reported errors similar to the following in the alert log and trace file: ORA-01578: ORACLE data block corrupted (file # 1, block # 100) ORA-01119: data file 1: '/scratch/820/qa/dbs/t_db1.f' or ORA-00368: checksum error in redo block ORA-00354: Log corruption near block #231 Oracle customers that have run the dbverify (dbv) utility may have encountered an error message similar to the following: *** Corrupt block relative dba: 0x0040900b (file 0, block 36875) Bad header found during dbv: Data in bad block - type: 27 format: 2 rdba: 0x0040900d last change scn: 0x0000.0001349a seq: 0x2 flg: 0x04 consistency value in tail: 0x349a1b02 check value in block header: 0xa377, computed block checksum: 0x0 spare1: 0x0, spare2: 0x0, spare3: 0x0 *** PROBLEM: (85230, 85345) (PATCH ID: OSF510-182) ******** This patch addresses two types of system crashes: - Crash caused by VM hash corruption, kernel memory fault. A typical stack trace would be as follows: 0 stop_secondary_cpu src/kernel/arch/alpha/cpu.c : 1326 1 panic src/kernel/bsd/subr_prf.c : 1296 2 event_timeout src/kernel/arch/alpha/cpu.c : 2186 3 printf src/kernel/bsd/subr_prf.c : 981 4 panic src/kernel/bsd/subr_prf.c : 1353 5 trap src/kernel/arch/alpha/trap.c : 2266 6 _XentMM src/kernel/arch/alpha/locore.s : 2143 7 ubc_wire src/kernel/vfs/vfs_ubc.c : 5372 8 u_vp_oop_pagecontrol src/kernel/vm/u_mape_vp.c : 1219 9 u_anon_faultpage src/kernel/vm/u_mape_anon.c : 2046 10 u_anon_fault src/kernel/vm/u_mape_anon.c : 1410 11 u_anon_lockop src/kernel/vm/u_mape_anon.c : 3053 12 u_map_lockvas src/kernel/vm/vm_umap.c : 1526 13 memlk src/kernel/bsd/kern_mman.c : 2309 14 syscall src/kernel/arch/alpha/syscall_trap.c : 722 15 _Xsyscall src/kernel/arch/alpha/locore.s : 1814 - Crash caused by lock hierarchy problem violation. A typical stack trace would be as follows: 4 panic() 5 simple_lock_fault() 6 mcs_lock_hierarchy_violation() 7 free() 8 _ms_free() 9 msfs_fs_cleanup() 10 vop_page_lookup() 11 ubc_page_alloc() 12 ubc_lookup() 13 bs_pinpg_one_int() 14 bs_pinpg_clone() 15 bs_pinpg() 16 fs_write() 17 msfs_write() 18 vn_write() 19 rwuio() 20 write() 21 syscall() 22 _Xsyscall() PROBLEM: (85404) (PATCH ID: OSF510-201) ******** This patch fixes a problem with the driver for Gigabit Ethernet adapters (DEGPA-FA and DEGPA-TA) which prevented its use in a NetRAIN (Redundant Array of Independent Network Adapters) set. In order to add an interface to a NetRAIN set, the interface must not be busy. However, even when the Gigabit Ethernet interface ("alt") is not explicitly started by the system administrator or by a startup script, sometimes the interface will be started automatically as a result of activity on the network. If an attempt is then made to add this interface to a NetRAIN set, NetRAIN will report that the device is busy and thus cannot be used. This patch corrects the Gigabit Ethernet driver so that it will not start unless explicity told to do so by either the system administrator or by a startup script. PROBLEM: (83661, 85312, STL111443) (PATCH ID: OSF510-213) ******** This patch fixes a problem where the setgid bit of a directory was not being set when created, if its parent directory has the setgid bit set. PROBLEM: (84908, 84911) (PATCH ID: OSF510-168) ******** This patch fixes issues with memory allocation attributes. PROBLEM: (83270) (PATCH ID: OSF510-212) ******** The context switch code would only clear the floating point enable bit when context switching away from a thread which was not terminated. If the terminated thread had enabled the floating point unit, the newly context-switched-to thread would be using its floating point state instead of its saved state. PROBLEM: (85447, 85450, MGO07072A, NWO42386A) (PATCH ID: OSF510-211) ******** This patch fixes several virtual memory algorithms related to the allocation and freeing of pages within the kernel. PROBLEM: (83169, 83608, 83745) (PATCH ID: OSF510-111) ******** These changes address panics which can occur if a signal is sent to a multi-threaded task in which one or more threads are calling exit() or exec(). The panic is invariably a kernel memory fault whose faulting virtual address is usually 0x50. The following panic message is typical: trap: invalid memory read access from kernel mode faulting virtual address: 0x0000000000000050 pc of faulting instruction: 0xfffffc0000283bc0 ra contents at time of fault: 0xfffffc0000283bb4 sp contents at time of fault: 0xfffffe0450d77360 PROBLEM: (83309, BCGMA1FFX) (PATCH ID: OSF510-184) ******** This patch fixes the corruption of the CAM hardware database when using hwmgr. This typically can result in a kernel memory fault when the database is being written to disk after a hwmgr operation. > 0 stop_secondary_cpu src/kernel/arch/alpha/cpu.c : 1003 1 panic src/kernel/bsd/subr_prf.c : 1222 2 event_timeout src/kernel/arch/alpha/cpu.c : 1668 3 xcpu_puts src/kernel/bsd/subr_prf.c : 1357 4 printf src/kernel/bsd/subr_prf.c : 892 5 panic src/kernel/bsd/subr_prf.c : 1287 6 afault_trap (source file cannot be determined) 7 _XentUna src/kernel/arch/alpha/locore.s : 2337 8 kds_offset_next src/kernel/io/common/kds.c : 4674 9 kds_offset_next src/kernel/io/common/kds.c : 4695 10 kds_offset_next src/kernel/io/common/kds.c : 4695 11 kds_offset_lock src/kernel/io/common/kds.c : 4818 12 camdb_write_status src/kernel/io/cam/camdb.c : 4113 13 camdb_write_asynchronous src/kernel/io/cam/camdb.c : 4164 PROBLEM: (GB_G00942) (PATCH ID: OSF510-188) ******** This patch corrects an AdvFS panic which can occur during a rmfset operation. The panic string: "rbf_delete_int: can't find bf attributes" PROBLEM: (78804) (PATCH ID: OSF510-099) ******** This patch fixes an issue with some remote ioctls for tape/changer drivers not working in a cluster. PROBLEM: (82879, 81918, 84896) (PATCH ID: OSF510-149) ******** This patch fixes a panic which comes from a page fault on a user buffer while we are already holding the write lock. A representative stack trace for the original hung thread looks like: 0 thread_block 1 lock_wait 2 lock_read 3 msfs_getpage 4 u_vp_fault 5 u_map_fault 6 vm_fault 7 trap 8 _XentMM 9 old_bcopy 10 copyin 11 uiomove 12 fs_write 13 vn_write 14 rwuio 15 write 16 syscall 17 _Xsyscall PROBLEM: (117-1-16330) (PATCH ID: OSF510-206) ******** This problem occurs when using the user level context switching (ucontext) routines and switching a thread's stack. The result is that the pthread mutex API routines can return EINVAL because we were unable to validate the mutex. This problem has been corrected by this patch. PROBLEM: (117-2-203) (PATCH ID: OSF510-136) ******** This modification fixes a problem in which the system panicked with a kernel memory fault while the class scheduler was being configured. PROBLEM: (85651) (PATCH ID: OSF510-209) ******** This patch fixes cluster hangs where I/O stops, and a hwmgr -view -clu command does not return. However, the systems will respond to pings. This is caused by the ubc_memory_purge in routine cfs_putpage being blocked when doing FSOP_PUTPAGE. An example stack trace is as follows: THREAD: fffffc01ffdb1500 1: sleep_prim+616: thread_block() 2: mpsleep+28: sleep_prim() 3: icscli_wait+284: mpsleep() 4: icstnc_cli_rcall+452: icscli_wait() 5: tnc_rcall+208: icstnc_cli_rcall() 6: rcfscall_service+220: tnc_rcall() 7: rcfscall+44: rcfscall_service() 8: cfscall_fsync+344: rcfscall() 9: cfs_putpage+760: cfscall_fsync() 10: ubc_memory_purge+2192: cfs_putpage() 11: ubc_purge_thread+180: ubc_memory_purge() crash> last -s | grep fffffc01ffdb1500 00:16:51 4 1048576 fffffc01ffdb1500 [ubc_purge_thread] This thread has been blocked for 16+sec. PROBLEM: (84103, 82583, 81968) (PATCH ID: OSF510-140) ******** This patch fixes the following system panics: - A "simple_lock: lock already owned by cpu" panic when anon_rss_enforce is non-zero and lockmode is set to 4. This remove occurs when a process, whose RSS (resident set size; the number of pages a process can have in memory) limit is exceeded, tries to expand its heap. An example stack trace is as follows: 1 panic() 2 simple_lock_fault() 3 simple_lock_state_violation() 4 pmap_clear_reference() 5 u_anon_rss_purge() 6 u_anon_rss_enforce() 7 u_anon_fault() 8 u_map_fault() 9 vm_fault() 10 trap() - A " panic: vm_page_activate: already active" panic that can occur on a system during memory shortages An example stack trace is as follows: 14 panic() 15 vm_page_activate() 16 vm_pageout_scan() 17 vm_pageout() - A "mcs_lock: no queue entries available" panic that can occur on a GS160 system. This is caused by an abandoned page mistakenly being reclaimed off the the 0/O hash. The page is then removed off a UBC free list where two stale page pointers were connected, hereby connecting the ACTIVE and INACTIVE list. When attempting to deactivate pages (move them from the ACTIVE queue to the INACTIVE queue) we encounter an INACTIVE page, which causes an inadvertant failure to unlock the page. Continued attempts to deactivate INACTIVE pages results in the lock queue being filled. This can also cause a "kernel memory fault" panic. An example stack trace is as follows: 1 panic() 2 event_timeout() 3 printf () 4 panic() 5 simple_lock_fault() 6 mcs_lock_entry_violation() 7 vm_pagelru_approximation() 8 vm_pageout() PROBLEM: (82138, 82611, 83447) (PATCH ID: OSF510-117) ******** This patch fixes a problem in which a heavy load placed on an HSG80 can disable the device. PROBLEM: (ALC-2-076) (PATCH ID: OSF510-192) ******** This patch fixes a timing window where flushing data to disk can be incomplete when a system is going down. Note this can only occur if all of these conditions are true: o More than one thread calls the reboot() system call without first going through shutdown, /sbin/reboot, or /sbin/halt (note the operating system itself does not do this, it would have to be an application program which is calling reboot()). o O_SYNC is not in use. o AdvFS data logging is not in use. PROBLEM: (84151) (PATCH ID: OSF510-163) ******** The compiler can generate code that tries to place the pages of an ALLOCATABLE (heap) array, before the pages are actually allocated. The placement is done with calls to nmadvise. This can cause the system to crash. PROBLEM: (82489, SSRT0676U) (PATCH ID: OSF510-155) ******** 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. SYNOPSIS mountd [-d] [-i] [-n] [-s] [-r] [-R] [exportsfile] FLAGS ... -r Have mountd listen for requests on a reserved port. This is the default behavior. -R mountd may listen on an unreserved port. PROBLEM: (BCGMC0X62) (PATCH ID: OSF510-194) ******** This patch eliminates a kernel memory fault in AdvFS that has the following footprint: crash> tf > 0 boot src/kernel/arch/alpha/machdep.c : 2774 1 panic src/kernel/bsd/subr_prf.c : 1334 2 trap src/kernel/arch/alpha/trap.c : 2259 3 _XentMM src/kernel/arch/alpha/locore.s : 2115 4 bs_derefpg src/kernel/msfs/bs/bs_buffer2.c : 3564 5 get_quota_blks_used src/kernel/msfs/fs/fs_quota.c : 2972 6 quota_activate src/kernel/msfs/fs/fs_quota.c : 3391 7 advfs_mountfs src/kernel/msfs/osf/msfs_vfsops.c : 3255 8 msfs_mount src/kernel/msfs/osf/msfs_vfsops.c : 2882 9 local_mount src/kernel/vfs/vfs_syscalls.c : 1453 10 mount1 src/kernel/vfs/vfs_syscalls.c : 1611 11 syscall src/kernel/arch/alpha/syscall_trap.c : 713 12 _Xsyscall src/kernel/arch/alpha/locore.s : 1785 (dbx) p kernel_memory_fault_data struct { fault_va = 0xc0 fault_pc = 0xffffffff0010e900 fault_ra = 0xffffffff00166258 fault_sp = 0xfffffe0e25db72c0 crash> 0xffffffff0010e900/i (dbx) 0xffffffff0010e900/i [bs_derefpg:3568, 0xffffffff0010e900] ldq t0, 192(s0) Here's where we panic in bs_derefpg(): 3567 /* Update statistics counters. */ 3568 if (bp->bfAccess->dataSafety == BFD_FTX_AGENT ) { PROBLEM: (83207, 83660, 83372, 83770, 83668, 76843, 83207, 81784, 76147) (PATCH ID: OSF510-122) ******** This patch fixes multiple problems with SCSI tape handling including improvements to backup procedures, SCSI passthrough, an increase to the local IO size for transfers, a fix for a system crash that can occur during a bus reset and a fix for a panic with the following panic string: "PWS_CCB_QUE_REMOVE: CCB NOT ON ANY LIST" PROBLEM: (BCPMB0MC4, 117-2-306, 84332, 82443) (PATCH ID: OSF510-157) ******** This patch fixes a system hang caused by netisr queue corruption due to a race condition that is primarily encountered by third party drivers and layered products that call schednetisr_nospl(). This patch also fixes a lockmode 4 panic in netisr_del_rad where netisr_del_rad attempted to release a lock it did not hold. PROBLEM: (84435, 84148, 84634, 82059) (PATCH ID: OSF510-134) ******** mixed speed cpu system from hanging a test suite Fixes a performance problem with 100% cpu usage nmadvise placed memory pages sometimes ignored. nmadvise performance enhancements. PROBLEM: (83859, 83567, BCGMA0MX9) (PATCH ID: OSF510-129) ******** This patch fixes a problem seen when trying to write out coredumps to fibrechannel devices. Previously, the fibrechannel-specific code would make calls to non high IPL-safe routines. This patch fixes the fibre code to only call routines which will not lower the IPL. This was more likely to occur on active systems than inactive. PROBLEM: (83399) (PATCH ID: OSF510-181) ******** This patch fixes a kernel memory fault and invalid memory ifetch panic which can occur in AlphaServer SC systems running Quadrics' RMS software. A typical panic follows, where the faulting virtual address and faulting pc are both zero: trap: invalid memory ifetch access from kernel mode faulting virtual address: 0x0000000000000000 pc of faulting instruction: 0x0000000000000000 ra contents at time of fault: 0xfffffc00005bde78 sp contents at time of fault: 0xfffffe05449bf760 panic (cpu 3): kernel memory fault PROBLEM: (83943) (PATCH ID: OSF510-109) ******** This patch fixes a bug in the POSIX Threads Library for Tru64 UNIX V5.1 that would result in a DECthreads Bugcheck and process termination. Threaded applications might encounter this problem when pthread_kill() is used on a thread that is marked as blocked in the kernel. This problem is identified by a stack trace similar to the following, where the __errBugcheck frame occurs right after the _schReadyForInterrupt frame: 0 __nxm_thread_kill(0x3ff800e3834, 0x20002225038, 0x3ff805ce968, ... 1 __sigAbortProcess(0x20, 0x0, 0x0, 0x0, 0x20002227600) [0x3ff805b752c] ... 2 __errBugcheck(0x3ff805934b8, 0x3, 0x20002227600, 0x20001bf7cc2, ... 3 __schReadyForInterrupt(0x1e, 0x20002227600, 0x3ffc01b5050, 0x0, ... 4 pthread_kill(0x3ffbfdf43d8, 0x3ffbfdf4388, 0xe, 0x177db030, ... ... PROBLEM: (83392) (PATCH ID: OSF510-180) ******** This patch corrects the behavior of the FIONBIO, FIOASYNC, and FIONREAD ioctl's in a cluster environment. These commands would fail, returning ENOTTY when they should have succeeded. PROBLEM: (BCGMA0VHK) (PATCH ID: OSF510-092) ******** This patch fixes a problem in which the system call fcntl(fd, F_DUPFD, 15) fails with "too many files" even after fd limits have been increased. PROBLEM: (HPAQB1GH8, BCGMC0SKX, 84064, 84731) (PATCH ID: OSF510-167) ******** This patch corrects two problems with the scheduler. - Enables NUMA load balancing in other processor sets then the default processor set (pset 0). - Enables the processor to do load balancing for multi-threaded applications. PROBLEM: (ALC-2-076) (PATCH ID: OSF510-158) ******** This patch provides support for activating temporary atomic write data logging on all files within the mount point. The data logging is turned off when the fileset is unmounted. PROBLEM: (78704) (PATCH ID: OSF510-179) ******** This patch fixes a hang in the ufs filesystem. PROBLEM: (BCPMA1QJ4, 83367) (PATCH ID: OSF510-178) ******** This patch fixes kernel build failures due to an undefined ss_sched function. PROBLEM: (81620, n/a) (PATCH ID: OSF510-068) ******** Execution of an interpreter (e.g., file beginning with #!/bin/csh) can fail if the total number of characters in environment variables is approximately one 8K or other page size multiple. It is unusual to use so much environment variable space, and unlikely for the size of the space to be very close to a multiple of the page size. If the problem does occur, and the execution of an interpreter file will fail with "file not found". For example: /updmnt/isl/updeng: /updmnt/sbin/it.d/bin/install_ap: not found PROBLEM: (83442, 85273, 84454) (PATCH ID: OSF510-199) ******** This patch provides full KZPCC support in version 2.0 of the i2o block storage driver. RESTRICTION: For TCR-V5.1 installations, KZPCC support is restricted to data-only service; devices on the KZPCC controller cannot be used for system or boot partitions in a cluster. Additionally, this patch fixes the problem where extraneous console messages will appear when hardware is added or deleted. PROBLEM: (SOO01415A, NWO41533A, SOO15249B, SOO09843A, BCGMA0M0S, n/a) (PATCH ID: OSF510-156) ******** This patch fixes a performance problem with v5.1 where threads doing large IO transfers could spend excess time in ubc_page_alloc(). PROBLEM: (84826) (PATCH ID: OSF510-169) ******** This patch fixes nmadvise with a modification to VM to allow migration of shared memory PROBLEM: (82423) (PATCH ID: OSF510-162) ******** This patch provides functionality to support EMC storage boxes that support Persistent Reserves (SCSI command set) as defined by the final SCSI specification. PROBLEM: (83129, 84172) (PATCH ID: OSF510-200) ******** On wildfires with PCI adapters of mixed revision, the PCA registers are set up incorrectly. This leads to kernel memory faults of variable characteristics. While this is not a configuration which is sold, it may exist in the field, and will certainly exist in-house. PCA and their revisions may be identified by their model numbers: PCA4 (B4171-AC) PCA3 (B4171-AA E02 or B4171-AB - doesn't matter) PROBLEM: (84104, 86180) (PATCH ID: OSF510-224) ******** PROBLEM: This patch fixes two issues seen on the GS320/160/80 AlphaServers: - A "u_anon_free: page_busy" system panic when using System V shared memory locked by a single process. This is seen when attempting to delete a user process' map. An example stack trace is as follows: 4: panic() 5: u_anon_free() 6: u_anon_oop_deallocate() 7: vm_map_entry_delete() 8: u_map_delete() 9: vm_map_exit() Examination of the suspect page revelas a negative wire count. px ((vm_page *))->pg_wire_count 0xffff - Failures ranging from unitialized simple lock panics, kernel memory fault panics, and process hangs, on GS320/160/80 systems configured with at least one memory less quad. The "simple_lock: uninitialized lock" or "kernel memory fault" panics originating in get_free_ptepage. These panics are a result of either attempting to obtain a free_page_table lock (fp) using an ID of a non-existent MAD, or attempting to dereference the MAD associated with the non-existent MAD. The process hangs are caused when the hanging process thread_blocks waiting for a ptpage request to be filled. The request will never be serviced because it has been made to a non-existent MAD. Identification of this hang condition can be done as follows: crash> *thread.wait_mesg
= "ptpage"