Patches and Fixes for this Version

JAGad28181

Symptoms: Frames larger than 8220 bytes are discarded when received by the HDLC.FRAME layer (level-1).

Defect/Fix: The ACC HDLC/LAP-B (ABM) User's Guide states that the maximum frame size is 10066 bytes. The limit has been raised to that value.

JAGad54596

Symptoms: When x25init is run against an X.25 link repeatedly, a DMA timeout or firmware error may result. This error could also happen when using other protocols such as HDLC-LAPB or HDLC-LAPD.

Defect/Fix: During the disable processing, a NULL pointer is incorrectly de-referenced by the code in unstdt_build(). A check has been added to avoid this de-reference.

JAGab67173

Symptoms: zconfig() portsc request hangs if a card is not available.

Defect/Fix: During a card startup, if the startup fails, zmon leaves the card in whatever state it was in previously (e.g. being Reset). This allows API requests to be accepted by LDM and DAM to operate on a non-functioning mux. If the API request is a zconfig/zport request, it is never completed and therefore it hangs the calling program. The zmon restart mechanism has been modified to fix this defect.

JAGad03287

Symptoms: Z7340A: zmntr mx display shows incorrect values for Txusd.

Defect/Fix: During the testing cycle of 8-Port PCI software and hardware, the zmntr mx command shows incorrect values for the TXusd field of each port. The value is in bytes, but does not correctly reflect the configured buffer size. It uses a fixed buffer size of 256 bytes, rather than the configured zbufsize.

The number is most noticeably wrong when the buffer size is set to a large value (e.g. zbufsize = 4096). You may need to set Unack-limit to a larger value (e.g. 50000) to see larger values of TXusd.

JAGad05873

Symptoms: zmlog logs a null char in hyphen lines.

Defect/Fix: This problem does not affect the log visually. However, when the log file is being filtered by the 'head' program, the extra null confuses 'head' and the correct line can't be extracted properly. The code has been changed to remove the extra NULL character from hyphen lines.

JAGad12116

Symptoms: On-line addition of card fails when ZCOM is already started.

Defect/Fix: When the ZCOM subsystem is started, zmon builds a table of all installed ACC cards which includes their type and hardware path. This table is used to verify that the configured ACC card is actually present in the system prior to a download. The download is not allowed to proceed if the configuration does not match the physical hardware.

With an on-line card addition after ZCOM is started, zmon always rejects the card download because the new card does not exist in its table. To fix this problem, zmon has been modified to clear the flag indicating the hardware table has been initialized whenever zmon receives a Interface Request Record (IRR) denoting an on-line card addition. This causes zmon to rebuild its hardware table, therefore allowing the download to proceed properly.

JAGad13610

Symptoms: zqmve() gives an error when moving messages between program ZLUs.

Defect/Fix: When the 'MV' command in ZTERM is used, it gives the error:

"Error on ZQMVE: Different nodes not allowed" when moving a message from one program ZLU to another program ZLU. Both are local ZLUs.

JAGad13821

Symptoms: ZTERM's TX behaves inconsistently when out of system buffers.

Defect/Fix: In QIP test LDM043, it uses ZTERM TX to send a large number of messages to exhaust all ZCOM buffers (e.g. TX <ZLU> 1000 030000). In some systems, ZTERM's TX got suspended - this is expected. But in some systems, ZTERM's TX returned with a ZCOM error: "Error on ZSEND: Not enough system free buffers".

JAGad15652

Symptoms: Zmasterd takes up to 7 minutes to startup the ZCOM subsystem with large I/O systems.

Defect/Fix: During ZCOM subsystem startup, ioscan is run to verify that the cards installed in the system match those configured through the TTGEN configuration (.answ) file. Most of the time taken during startup is spent waiting for ioscan to scan the backplane for hardware. The fix is not to scan the physical backplane, but simple to retrieve the hardware information from the kernel that was gathered at system bootup. The code now schedules ioscan with the parameters of "-FkC acc" instead of just "-F".

JAGad15656

Symptoms: 'x25init' command fails with the following message in Nettl log, when 200 SVCs are configured on all the 8 ports.

N2Z: The zconfig() call to create a new L3 ZLU has unexpectedly failed
with error -15. Unable to startup the X.25 line (Mux = 0, Port 7,
Subc 0)!

Defect/Fix: The problem is caused by not having enough ZCOM Terminal entries configured in the .answ file. The fix is to tune these parameters in .answ file and then start-up the ZCOM subsystem.

JAGad21198

Symptoms: System panics with a Data Page Fault.

Defect/Fix: The problem happened because a NULL pointer was referenced. To fix this problem, a check has been added if the pointer is NULL.

JAGad21596

Symptoms: ZTERM mistakenly takes msg request code 11 (ZCOM_MRQCODE_DEL) as "port configuration" to report zport return status (by x25stat API). The correct request code should be 14 (ZCOM_MRQCODE_PORT).

Defect/Fix: While doing QIP API047 test, in an erroneous ZTERM PT command (ZPORT configuration), the returned status was incorrectly reported. E.g., ZPORT status 11 is reported as "Invalid subchannel number" (which is the status for terminal configuration request). ZPORT status 11 should be "Mode/baud rate incompatibility". zterm uses an incorrect request code in calling x25stat().

JAGad21615

Symptoms: System panics with Spinlock Deadlock when running reliability tests.

Defect/Fix: When the system was running out of memory, it would cause nli2zcom to drop a data packet and issue a reset request on the VC. This particular test caused many resets and reset confirms to be issued during heavy data transfer loads. The nli2zcom driver showed that a reset request (or response) was being sent to the DAM at the same time that an inbound event was arriving from the DAM on the same card. The DAM held the IFT lock and passed the event to the nli2zcom driver which found the N2Z lock already held. The N2Z driver locked its spinlock and passed a request to the DAM which found the IFT lock already held. Thus a classic spinlock deadlock occurred.

The fix is to have the nli2zcom driver drop its lock when issuing a request to the DAM and then reacquire its lock upon return from the DAM. This was done for both the outbound Reset Request and Reset Response functions.

JAGad24633

Symptoms: zconfig PORTSC enable/disable subchannel has insufficient validation.

Defect/Fix: The zconfig PORTSC for enabling/disabling subchannels did not check for valid card type and valid subchannel numbers.

The zconfig call to 2-port and 8-port cards returns ok as long as the port number is in range. The zconfig call to 4-port card accepts all subchannel numbers, including 40.

The LDM should reject the subchannel configuration (for 2,8-port cards) and the illegal subchannel number for 4-port card with ZCOM error: ZESUBCH (-49, Illegal subchannel number). All the above problems have been corrected.

JAGad31008

Symptoms: MPU activity field in the zmntr mx command shows incorrect value.

Defect/Fix: The zmntr utility can be used to check the status of an ACC MUX. The command zmntr mx also shows the utilization of the processor on the ACC MUX. With the development of the TRAU/GPRS protocol, the customer runs some tests, especially load tests. But the MPU activity with the use of the TRAU/GPRS protocol stays always at 0%.

The problem occurred because the $STAT message in the DAM was given lower priority compared to messages in Express Queue, High Priority Queue and Low Priority queue. Because of this, the status always got updated after processing messages in the queues. Hence the % activity always showed 0% because the CPU already finished processing messages and is now idle. The code change has been done such that the $STAT message is given higher priority than the messages in the Express Queue, High Priority, and Low Priority Queue.

JAGad40862

Symptoms: The x25stat command needs enhancment to show "VC Up Time" for all active virtual circuits.

Defect/Fix: Provided -u option in the x25stat command so that the user can request the "VC Up Time" for all active virtual circuits. A new ioctl has been added for this purpose. This option must be used in conjunction with the -v option of the x25stat command.

JAGad56962

Symptoms: ACC firmware does not free buffers when the terminal is deactivated and disabled.

Defect/Fix: When messages are sent down to an HDLC-LAPB terminal after enabling it, but before the link is established, those messages are never completed or flushed unless the link comes up. This leads to a variety of problems, including card buffer limits being reached.

The Level 2 state machine has been adjusted to make sure messages are flushed after the first attempt at link establishment. After the T1 x N2 SABM attempt, all messages are flushed and no further messages are accepted by the protocol as the link down status is set.

JAGad66065

Symptoms: zmasterd stop causes system panic.

Defect/Fix: The customer is using ACC/ISDN cards with OTS and FTAM software for collecting information from phone pbx. During the night transfers, the card is intermittently hung and transfer is stalled. When trying to stop the card by the MC/SG scripts, the stopping of zmasterd caused system panic.

The panic is due to accessing a null pointer and trying to do bcopy of a value accessed by that null pointer. Now the defect has been fixed by checking if the pda->resp_qdb is NULL. Only if it is not NULL, bcopy is done.

JAGad66909

Symptoms: File /dev/n is getting created when executing zmasterd cold *.answ.

Defect/Fix: Instead of redirecting the output of ttgen to /dev/null it has been redirected to /dev/n ull. One extra space caused the problem. Hence /dev/n was getting created. This extra space has been removed and now the output will be redirected to /dev/null.

JAGad67062

Symptoms: Kernel build fails because the path of spinlock.h mentioned in zcomsys.h is incorrect.

Defect/Fix: During kernel build, bx25 header file (/usr/conf/acc/csihdw.h) includes zcomsys.h from /usr/conf/acc directory. zcomsys.h includes spinlock.h as follows:

#ifdef _KERNEL
#include <sys/spinlock.h>
#endif

This makes the compiler get spinlock.h file from /usr/include/sys/spinlock.h. /usr/include/sys becomes the relative base directory (not /usr/conf/acc). /usr/include/sys/spinlock.h has following lines:

#ifdef _KERNEL_BUILD
#include "../h/types.h"
...
#endif

The kernel build fails here since it is looking for /usr/include/sys/../h/spinlock.h which does not exist. The exact error indicated by the compiler is:

cpp: "/usr/include/sys/spinlock.h", line 15: error 4036: Can't open
include file '../h/types.h'.

The fix is to change zcomsys.h to:

#include "../h/spinlock.h"

instead of

#include <sys/spinlock.h>

JAGad67065

Symptoms: The zconfig.3x man-page contained incorrect information about the DSC feature for the subchannels.

Defect/Fix: The man-page has been modified so that it contains the updated information for supporting the DSC feature for subchannels.

JAGad67069

Symptoms: When an x25stat is done, the Max Frame size that is displayed is shown in bits instead of bytes.

Defect/Fix: n2z_upper.c has been changed such that the max frame size is displayed in bytes instead of bits.

JAGad83768

Symptoms: Add support for ACC PCI cards on Superdome systems.

Defect/Fix: When running ACC 3.30 with Z7340A cards on a Superdome system, zmasterd cold start fails, and the cards do not start up. zmon complains that there is no ACC card in the slot configured. This is because an extra "0" was added to the I/O path on Superdome systems. For example, if an N-class system had an I/O path of 16/4/1/0, on a Superdome the same path would look like "16/4/1/0/0". The problem is in the function nacc1_get_interface_info() which looks at the path entries starting with the rightmost element and working left. On a superdome system, this yielded a bus address of 4:1 and a slot address of 0 instead of 16:4 and slot address 1.

The address parsing has been changed to start with the leftmost element and then work right for the PCI bus. This should result in consistent bus and slot addresses regardless of the target system that supports the PCI bus.

JAGad87791

Symptoms: ACC card description in ioscan output is insufficient.

Defect/Fix: ACC card description in "ioscan -fC acc" output needs to be more informative. The code has been changed so that "ioscan -fC acc" output displays the card number and the description. For instance, the "Description" field of "ioscan -fC acc" output is as follows:

Description
==============
Z7340A PCI bridge
Z7340A 8-port serial ACC

JAGad94415

Symptoms: ACC 3.xx sends the calling address in call accept packet.

Defect/Fix: ACC 3.xx has a non NULL calling address in the call accept packet, which is refused by some X.25 implementations. This problem has been fixed by providing a tunable parameter, which can be set using /opt/acc/bin/n2z_cntrl. By default this flag is set to 0. So, the calling address in call accept packet is 0 by default. If set to 1, the call address packet contains the calling address.

JAGad96420

Symptoms: ioctls in n2z_cntrl fails with "Bad address" error.

Defect/Fix: When trying to set the tunable values using n2z_cntrl, it fails with "Bad address" error. For example, n2z_cntrl -d0 will display "n2z_cntrl: Bad address". The reason is that the ioctls were called directly instead of using I_STR as the second parameter of ioctl and strioctl structure as the third parameter. This has been fixed as follows:

Instead of calling ioctl as follows:

ioctl(ssmfd, N2Z_CTS_CD_FLAG, &flag)

ioctl is now be called by filling the strioctl structure and then calling:

ioctl(ssmfd, I_STR,&strioctl)

JAGad71268

Symptoms: The processing in ZCOM API zconfig terminal clear/delete is incomplete. If this feature is used, it may result in a memory leak in the ZCOM buffer pool and/or a system panic.

Defect/Fix: A new queue "pendg_txreq" was added to the PTT in order to support the DAM's $DATA processing. This queue is used as a temporary holding place for the selected tx requests while building a $DATA backplane transaction. However, in zconfig terminal CLEAR & DELETE, this new queue is not checked or cleared at all. This means the DAM may be actively using this queue while the LDM is clearing or deleting the associated terminal.

The Proper codes are added in the LDM to ensure this queue is also checked and cleared.

JAGad69876

Symptoms: Transmit messages stop after ACC runs out of transmit buffers.

Defect/Fix: When an ACC card is heavily using its internal buffers, the DAM can, in rare situations, be notified that the card is out of buffers for this subchannel or port even though the driver does not have a single pending write request on the card. This is an issue because the write completion is what normally restarts the outbound data flow when the driver goes into outbound flow control (e.g., the card says its out of buffers). In this special case, the driver keeps track of the subchannel(s) which are in this state and starts a timer to restart the outbound I/O since there are no write completions to start the outbound I/O.

After analysis of the problem, it appears that the subchannel is not being placed back onto the transmit linked list when the timer pops.

When the condition arises where the cards rejects an write request on a subchannel and the driver does not have any pending write requests on that subchannel, the driver starts a timer. When the timer pops, the driver finds all subchannels in this state and attempts to restart any pending writes.

The problem is that the driver is not placing the subchannel back into the linked lists of subchannels with a pending write request. Code has been added to insert the subchannel back into the transmit linked list.

JAGad97577

Symptoms: When there is a lot of opening and closing of X.25 sockets and the ACC kernel limit n2z_max_devs is reached, X.25 socket creation with a socket() call returns invalid errno 65535(-1).

The following message is logged in nettl log:

Network NS_LS_N2Z Error 2001, pid 486745

N2Z: Too many streams used! Increase the size of the kernel tunable n2z_max_devs in /stand/system and rebuild your kernel.

Defect/Fix: This problem has been noticed with PVCs and while detaching the PVC in N2Z_F_pvc_detach_up() routine, the VC Stream q_ptr field was made NULL. However in VC Stream close routine N2Z_F0_close(), if q_ptr is NULL, then the further closing and clean-up of the VC stream was not complete, and hence VC Streams were not returned to the free pool. This caused the driver to run out of available VC Streams and the socket() call returns errno. 65535 after the kernel n2z_max_devs is reached.

A new 'lhvcp' flag mask has been introduced. This flag is set during PVC detach processing to indicate that VC Stream close is still pending and thereby avoiding a need to set the VC Stream q_ptr to NULL. A code change has been made to check for this flag and to reset the flag during 'lhvcp' allocation.

JAGad98715

Symptoms: Panic with Data Page Fault with the following stack trace:

panic+0x6c
report_trap_or_int_and_panic+0x94
interrupt+0x208
$ihndlr_rtn+0x0
nacc1_cmplt_read+0x38
nacc2_complete_req+0x34a8
nacc2_end_io+0x580
nacc2_isr+0x126c
sapic_interrupt+0x2c
...

Defect/Fix: When stressing the system by running the tests on 16 cards simultaneously on a HP rp8400 server, we see the above panic. The panic is because, in nacc1_cmplt_read, the DMA receive queue is found to be NULL. Dereferencing this NULL pointer caused the above sustem panic. The fix is to check if qhead is NULL. If it is NULL, we log a trace message and return.

JAGad67070

Symptoms: System panics with the Spinlock timeout failure.

Defect/Fix: The Spinlock timeout failure seems to be happening because the function N2z_Disable_ZLUs() calls the zcntl() holding the SPINLOCK(glock). Because there are no buffers available, the Zc_gosleep() is eventually called which in turn calls sleep(). This is causing the spinlock timeout failure to occur because the spinlock should not be held when the sleep() is called. To fix this, the spinlock is released before calling the zcntl() in function N2z_Disable_ZLUs().

JAGae04681

Symptoms: This is an enhancement to diagnostic capabilities of the ACC Firmware. Tracing the ACC firmware to determine events leading up to and causes of problems has been difficult to do without specialized equipment. A better ACC firmware tracing mechanism was needed.

Defect/Fix: A more advanced ACC firmware tracing environment was developed to enable on-demand, and dynamic tracing of the ACC firmware, without the need forextra equipment, such as a logic analyzer. This environment also enabled the amount of trace points within the firmware code to be substantially increased - since the new mechanism has a very low impact on the ACC card's performance. The new trace mechanism allows the amount of active tracing to be varied and tracing of individual modules and protocols be selectively enabled and disabled dynamically. The mechanism allows tracing to be captured from a running ACC card.

Notes for usage: When gathering ACC card firmware trace for any further problem analysis, it is recommended to always use the raw option of 'fwtrace' utility as shown below.

$ /opt/acc/bin/fwtrace
> dump -raw > /tmp/trace.out