(C)Copyright 1994-1998 SysKonnect, a business unit of Schneider & Koch & Co. Datensysteme GmbH. All rights reserved ================================================================================ Name : devices.pci.48110040.rte.README Readme File for SK-NET FDDI-FP Network Device Driver (NDD) Subject: SK-NET FDDI-FP, PCI NDD Driver for AIX for IBM PowerPC Serie Version/ Date/ Author of this file: 1.49 / 98/10/08 / A. Fischer Version/ Date/ Author of the referred subject: v1.21b02 / 98/10/06 / A.Fischer This file contains (1) overview (2) installation instructions for the PCI device driver on AIX 4.1 (3) reference list of additional device configuration parameters (4) known limitaions (5) bugs fixed (6) problems fixed ================================================================================ (1) OVERVIEW ============ The network device driver (NDD) supports max. 32 SK FDDI PCI network adapters. Requirements: ------------- The SK-NET FDDI-FP network device driver is designed for network access on AIX 4.1 based PowerPC machines with PCI bus. For proper driver access the corresponding FDDI protocol interface must be installed for TCP/IP configuration. Required software for network driver only: devices.mca.8ef4.com - Software package from AIX installation CD Required Software for network driver's diagnostic package only: bos.diag.rte 184.108.40.206 - Diagnostic software package from AIX installation CD devices.pci.48110040.rte 220.127.116.11 - Network driver package If you select the complete NDD software package both the network driver and the diagnostic package will be installed. (2) INSTALL THE DRIVER ON AIX 4.1 ================================= For the following actions you should login as root on your console: * Software Installation - Installation from console * Removing Software Package - Deinstallation from console * Configuration of the NDD * Copyright messages Software Installation --------------------- You have probably done these steps below: Run the `smit` program and step through the following menus: "Software Installation and Maintenance" "Install and Update Software" "Install/Update Selectable Software (Custom Install)" "Install Additional Device Software" Some additional input is required when the following text appears: "INPUT device / directory for software [ ]" Press F4 and select the device /dev/fd0 (Diskette Driver) "SOFTWARE to install [ ]" Press F4 to list the software package on your installation diskette. Then move the cursor to the desired item and press F7 <RETURN>: "x.x.x.x devices.pci.48110040" Installation from console: You can also install the network driver package from the system console with the `installp` command as follows: Installation from floppy: installp -d /dev/fd0 devices.pci.48110040 Installation from HD : installp -d <image-file> devices.pci.48110040 Typically the <image-file> is a disk dump file with the extension .dd . You can create a 3,5" install diskette from this file: dd if=<image-file> of=/dev/rfd0 bs=32k . Removing Software Package ------------------------- Before you deinstall the NDD package from your hard disk you must remove the network driver from the kernel: rmdev -l fi0 Remove FDDI protocol from network driver rmdev -l fddi0 -d Remove the network device driver To remove the software package step through the following smit menus: "Software Installation and Maintenance" "Maintain Installed Software" "Remove Software Products" Press F4 to list the installed software packages. Use the search key "/" to get to the network device software package. Select the driver package `devices.pci.48110040.driver` with F7 and return to the previous menu. Set the field `PREVIEW only` to `no` to remove the software package .... Then hit the RETURN key to remove the software package. Deinstallation from console: You can remove the driver package with the `installp` comand as follows. Deinstallation from HD : installp -u devices.pci.48110040 You can also invoke the `smit` command for removing the driver software. Please refer to the following section. Configuration of the NDD ------------------------ You can configure up to 32 network devices of the same type. Normally the number of devices is restricted due to the limited number of system resources available. You need not care for system configuration para- meters e.g. interrupt or IO space. This is automatically done when you reboot the computer or if you execute the `smit` command: "Devices" "Configure Devices Added After IPL" The configuration manager on your computer detects the network adapter automatically in the PCI configuration space. It creates the network devices for each existing adapter. If all network adapters have been configured the devices are set to the state `Available`. Check the FDDI network driver state by entering this command: lsdev -C | grep fddi You can also configure some parameters that can be changed by the user. These are parameters for configuring the MAC address, software queue or the SBA/ESS (Synchronous Bandwidth Allocator/End Station Support). Find a detailed description below. To get to the driver's configuration menu step through these menus: "Devices" "Communication" "SK-NET FDDI-FP Adapter" "Adapter" "Change / Show Characteristics of an SK-NET FDDI-FP Adapter" and select a network device from the list of available devices "fddi0 Available 04-0X SK-NET FDDI-FP Network Adapter" "fddi1 Available 04-0X SK-NET FDDI-FP Network Adapter" .... A configuration window appears.Change the configuration parameters that you want to change for the active device and press RETURN. The network device will be updated with the new configuration parameter.Please note that some configuration becomes active if the network device is started again. To do this you should detach any existing FDDI protocol from the network device and start the protocol again. Please note that the net- work driver attaches to the FDDI ring if the device is opened by a user (network protocol). The driver detaches a device from the FDDI ring on the last close. Copyright messages: ------------------- The NDD does not display any boot message on screen.When installing the driver package the following copyright messages will be displayed: "Copyright (C) 1995, 1996, 1997, 1998 SysKonnect a business unit of Schneider & Koch & Co. Datensysteme GmbH. All Rights Reserved." "Synchronous Bandwidth Allocator, Copyright International Business Machines Corporation, 1993 - 1998. All rights reserved. (3) ADDITIONAL DEVICE CONFIGURATION PARAMETERS ============================================== "USER Data" User data to be passed to the SMT for management information base (MIB). "PMF password" Parameter Management Frame (PMF) for the adapter. A nonzero PMF password enables autorization checking of the adapter for remote change requests. A PMF password of 0 disables the authorization checking of the adapter. The PMF password consists of 16 hexadecimal digits. Default value is 0. "Transmit Queue Size" Number of tx - frames that can be queued by the driver software queue in both the synchronous and the aysnchronous transmit queue when the ring is under heavy load. Valid values: 3-250. Default value : 30. "MAX T-REQ" The parameter provides local write access to the 2s complement TReq value registered as fddiMAC 51 in the SMT Standard.The value for TREQ is speci- fied in milliseconds. The valid range reaches from 5 ms to 165 ms. If any value is specified which is outside the limits, the upper/lower limit will be used. If no value is specified the default value for TREQ (165 ms) will be used. "TVX lower bound" This parameter provides local write access to the TVX attribute registered as fddiMAC 54 in the SMT Standard.This attribute provides local control of the recovery time from transient ring errors. The value for TVX is speci- fied in microseconds. The valid range reaches from 2500 us to 10000 us. If a value is specified outside the limits, the upper/lower limit is used. If no value is specified the default value for TVX (2700 us) is used. "Enable Alternate MAC / SMT Address" If the attribute is set to yes,the network node address will be set to the alternate address instead of the manufacturer provided unique address. If the attribute is set to no, this has no effect. Valid values: yes, no. Default value is no. "Alternate MAC/SMT Address" If the `Enable Alternate Mac / SMT Address` attribute is set to yes, this address describes the MAC address that is used as alternate network address. A valid value is 12 hexadecimal digit. Default value is none. "SBA_Payload" The parameter defines the requested synchronous bandwidth for manual static allocations. The valid range is from 0-1562 bytes per 125 micro seconds.For example, if the required bandwidth is 1 MBit/s (125,000 Bytes/s) the value of the payload is 125,000 * 125E-6 = 15.625 round up to 16. If a value is specified outside the limits the upper/lower limit is used. The default value is zero - no synchronous bandwidth is used. The user is required to define the amount of bandwidth to be able to send synchronous frames.If the parameter SbaPayLoad is not specified the keyword SbaOverHead, MaxTNeg, MinSegmentSize and SbaCategory have no effect. The end station supports either the static allocation model, where the re- quested payload is specified by the parameter SbaPayload or by the dynamic allocation model, where the required synchronous bandwidth is allocated by the multimedia application. If a multimedia application is used that can dynamically allocate the bandwidth, the parameter SbaPayLoad should not be specified. "SBA_Overhead" This parameter defines the requested overhead for static allocations. The valid range reaches from 0 to 5000 bytes. If a value is specified which is outside the limits, the upper/lower limit is taken. The default value is 50 bytes. This parameter has only an effect if the parameter SbaPayLoad is specified. "Max_TNEG" This parameter defines the maximum token rotation delay that is acceptable to the application(s) using synchronous bandwidth. The valid range reaches from 5 ms to 165 ms. If a value is specified outside the limits, the upper/ lower limit will be taken. The default value is 25 ms.This parameter has an effect, if the parameter SbaPayLoad is specified. "Min_Segm_Size" This parameter defines the minimum synchronous segmentation size. The value corresponds to the amount of bytes to be transmitted on a per token opportu- nity. The valid range reaches from 1 to 4478 bytes. If a value is specified which is outside the limits,the upper/lower limit is used.The parameter has only an effect if the keyword SbaPayLoad is specified. "SBA_Category" This parameter defines the session ID of the SBA_Category for the static allocation.The valid ranges reaches from 0 to 65535.If a value is specified which lies outside the limits, the upper / lower limit will be taken. This parameter has only an effect, if the keyword SbaPayLoad is specified. "Sync_Tx_Mode" This keyword defines the synchronous transmission mode.The default value is 'SPLIT' where only packets identified as synchronous packets shall be trans- mitted by the synchronous queue.The other value is 'ALL' that means that all LLC packets received from upper layers are transmitted via the synchronous queue. This keyword has only an effect, if the end station support was able to get the required synchronous bandwidth from the SBA. "SBA_Command" The SbaCommand specifies a SBA local action to start / stop the SBA of the network device. Valid values: start, stop. Default value : stop. "SBA_Available" The parameter SbaAvailable specifies the maximum synchronous bandwidth in percent available for the primary path. If the value is outside the valid range, the upper/lower limit will be taken. Valid values: 0 .. 100. Default value is 50. (4) KNOWN LIMITATIONS ===================== No limitations are known. (5) KNOWN BUGS ============== (6) PROBLEMS FIXED ================== Problems fixed in rel: v1.02b01: -------------------------------- The DAS adapters did not insert into a dual ring. This problem is fixed. A lack of performance may be caused on the send side due to a problem when scheduling the transmit queue. The performance problem when sending with max. speed has been fixed in this driver release. A strange behaviour of the LEDs was detected when loading or unloading the network driver. This problem is fixed. The error message "ST3L: parity error in receive queue 2" has been printed out although this error did not occur. This problem had been fixed. This message was caused due to checking the wrong bit. The problem is fixed. Problems fixed in rel: v1.02b03: -------------------------------- NFS accesses are very slow with the PCI FDDI board if there is no traffic. Since NFS keeps on waiting for tx-mbuf to be freed immediately, the driver checks the outgoing mbufs for an optional free - function. If such a free- function exists (like in NFS mbufs) a tx-complete interrupt is generated after the mbuf has been sent. This prevents NFS from running slowly. When running the driver on multiprocessor machines an invalid IO address may be used for IO access that causes one processor reading invalid data and blocking the system. This bug has been fixed in this release. When queueing outgoing frames with a very high rate (more than 70% of the tx frames) frames may be lost when the driver was not able to clear the tx queue. When running netperf - tests on TCP/IP with the RFC1323 option set to one a lack of performance was detected.In this mode additional space in the tx ring is needed for sending higher fragmented frames. A higher number of tx fragments in the transmit path prevents the driver from queueing too many frames. Problems fixed in rel: v1.02b04: -------------------------------- Allocation of RX mbufs has been changed. The driver allocates now receive buffers with two 4K mbufs instead of a single mbuf with more than 4K size. This was neccessary due to an 'out of mbuf' error that was caused by the driver's preallocation function for receive buffers. System resources exhausted due to the large amount of receive buffer storage of the driver. Problems fixed in rel: v1.02b05: -------------------------------- Allocation of RX mbufs has been made configurable.Configuration parameters for allocation of both frames with two 4K mbufs and with single mbufs with more than 4K size have been added to the smit configuration files. Now the two configuration parameters `Extended receive mbuf size` and `Receive frame count` can be used to specify the type of allocation and the number of max. receive buffers (frames) in the receive queue. Support of VPD has been implemented in the driver. Error logging has been implemented in the driver. Problems fixed in rel: v1.02b06 ------------------------------- The following bug in the SMT has been fixed: After a ring op changes or a driver reset it could happen, that the FORMAC and the ASIC became asynchronous.As a consequence the following may happen - PCI bus violations, the machine my hang up. - The FORMAC may send frames which will never stripped by any station after a ring operational change. The bug has been fixd in this driver release release. Problems fixed in rel: v1.02b07 ------------------------------- The RX/TX descripors have been placed into physical contiguous memory. If a large number of descriptors are used a page size of 4K memory will be exceeded that may not be physically contguous. Problems fixed in rel: v1.02b08 ------------------------------- In the previous release the driver did not correctly set the multicast RX- mode automatically if the multicast address table was full.The problem has been fixed in this release. Independent of setting the multicast RX - mode by I/O control the driver enables the promisuous mode automatically if the multicast address table is full and disables the promisuous mode again if the address table contains less multicast addresses than the address table can hold ( assumed that the multicast receive mode is not enabled by I/O - control permanently). Problems fixed in rel: v1.03b01: -------------------------------- Previous PCI drivers that include SMT software releases 2.6.2 and before had the following SMT bug: A dual homing problem has been detected in DAS and DAC: If you plug port B and then port A and then unplug port B again the ring would come up after 50 seconds. The bug has been fixed in this new release. Now the ring comes up immediately.The network driver contains SMT software release 2.6.3 that complies to the SMT version 7.3. A bug in the receive part of the driver has been fixed, that causes a loss of performance only: on heavy receive load the receive descriptor ring was completely cleared, that delayed the next receive interrupt.An internal rx counter was not reset on adapter restart. As a consequence the rx-ring was not refilled correctly. Basically the problem does not happen if the driver was not restarted or was reloaded. The problem occurs the first time after the driver was stopped (but not unloaded) and restarted and if system mbufs become low. Error Log Entry: LABEL: SKFX_HWM_SWE IDENTIFIER: A06C8DD0 FAILURE REASON: HWM: Out of RxD condition detected The field "Max Packets on S/W Transmit Queue:" was set to the max. transmit queue length. It should be a 'high water' mark of the maximum number of the packets ever queued up to the driver. In the new driver release it reflects the high water mark of frames that have been queued after the last reset of FDDI statistic counters. This field can be reset by 'fddistat -r' command. Problems fixed in rel: v1.04b01: -------------------------------- A bug was fixed that could cause a driver crash or at least invalid data: When the driver requeues a receive descriptor because the descriptor ring of the receive side was empty an incorrect physical address may passed to receive descriptor ring. If the length of the next receive frame exceeds a page size of 4K the system could crash.Typically this happens only on very strong receive load or if the driver receives an SMT - frame followed by a any rame with a length of more than 4K. This does not happen if the driver configuration parameter "Extended receive mbuf size" was set to "yes" (no default value). Internal NDD statistics will be set by the driver correctly. The former driver release did not set the NDD statistic variables.NDD statistics can be displayed with the 'ndb' command. The receive address table returned from the NDD_MIB_ADDR - IO control was incorrect. This has been corrected. The IO controls NDD_MIB_QUERY and NDD_MIB_GET were not implemented in the former driver release but the driver returned zero indicating valid data have been copied into the caller's receive buffer. Now both IO controls are implemented and will return correct data. Problems fixed in rel: v1.04b02 ------------------------------- Additional trace IDs have been added for the trace groups 'open / close / config / control and error conditions. This allows a trace separated from the RX and TX path (that will normally produce very much output). Problems fixed in rel: v1.04b03 ------------------------------- Statistic counters for in/out errors and purged frames were not correctly updated. This has been fixed in this latest release. Problems fixed in rel: v1.04b04 ------------------------------- The IOCTRL command 'NDD_MIB_ADDR' returned an invalid mac address. Now it returns the physical address format of the mac address. The NDD_RUNNING flag is set in the ndd_flags field before the driver is ready to transmit packets. The NDD_RUNNING flag remains set in the field after the cable is unplugged.Now the flag is set when the NDD is attached to the ring and is removed when the NDD is detached from the ring. Not all trace points are formated in the trace format file.Additional IDs have been added to the trace file for the rx- function: mac_drv_rx_init. The command fddistat showed an invalid statistic value for the Connection Policy Violation. This is now set to zero because there is no appropriate SMT variable available for this statistic counter. A bug in the PowerManagement handler function was fixed that lets the NDD crash on SMP environment only (invalid lock access). Problems fixed in rel: v1.04b05 / v1.04 --------------------------------------- A bug in the PowerManagement handler functions was fixed that caused the driver not to attach to the ring again after coming up from hibernation. IOCTLs have been added to the ADD FILTER command for receiving SMT frames from the network driver's SMT. The network driver description has been changed for the lsdev -C command. Problems fixed in rel: v1.05b01 / v1.05 --------------------------------------- The error log message identifies now the adapter number that displayed the error e.g: fddi0, fddi1 ..... In error logs of previous driver releases it could happen that an error text was truncated.The maximum length of error message field 'Detail_Data' is now 128 characters instead of 64 characters before. If the driver is busy and the customer tries to unload the driver it does no more generate an error log entry (rmdev -l fddiX). In previous driver releases no error message description was added to the error description data base (errpt -w E) for the driver specific message IDs 8500 - 8509.In this new release an appropriate error description is added to the data base file. Instead of the message ID the errorlog output shows now the error description. A bug was fixed that could cause a system crash if a not existing device was initialized. Problems fixed in rel: v1.06b01 ------------------------------- A bug was fixed that could cause a system crash under certain circumstances: When the driver has to requeue outgoing frames due to heavy netload and the cable connected to the corresponding adapter was unpluged and pluged again the system could crash. The problem does not occur,if the cable of an other station was unpluged to generate the ring operational change. Problems fixed in rel: v1.06b02 ------------------------------- Another bug was fixed that let the outgoing traffic stuck after he cable of the corresponding adapter was unpluged and pluged during heavy netload. Any applications e.g. TCP/IP or ping would no more operate correctly though the operating system seems to work fine. The problem could be resolved if the network driver was restarted again. Under certain circumstances the machine could also hang up. On SMP machines tx - frames may be lost if the network driver has to queue outgoing frames due to heavy network load and any existing frames in the tx queue are cleared during a long inoperational time (> 1 sec) of the network adapter.The order of queued frames may be corrupted due to a SMP sync. bug. A system crash or driver hang up was not yet detected concerning this bug. Problems fixed in rel: v1.06b03 / v1.06 --------------------------------------- A new SMT release 2.6.4 has been implemented in this driver release with the following changes: - all stations: SUCCESS was returned on an PMF SET frame when setting the mib.a[PATH0].fddiPATHSbaOverhead , but the parameter cannot be set when mib.a[PATH0].fddiPATHSbaPayload is zero. Now 'not authorizeed' is returned in this situation. - all stations with SBA:DENIED was returned on an PMF SET when setting the mib.a[PATH0].fddiPATHSbaAvailable and the SBA is not started ; 'not authorizeed' should be returned. - all stations: mib.fddiSMTStationStatus was not set correctly. - all stations/concentrator: entering RM4:NON_OP_DUP via RM(34a) , RM(34b) or RM(34c) the station should disconnect from the ring. In the old SMT version it was working, but it immediately reconnected to the ring. Now it will not reconnect. An errorlog is generated. - all stations/concentrator:the Link Error rate was compute wrongly (about factor 12 to high) - all stations/concentrator:the PLCs interrupt mask was not set correctly. The PL_SELF_TEST bit was not set. The incorrect interrupt mask caused a missing interrupt in the TRACE state and an unnecessary wait of 7 sec. Problems fixed in rel: v1.07b01 ------------------------------- A new SMT release 2.6.5 has been implemented in this driver release with the following changes: - all stations : Setting SBA parameter results in success or `OutOfRange` error. - all station concentrator : duplicate address is now detected correctly. The Duplicate Claim with T-Bid != T-Req is signaled directly to RMT. - PCI driver:the PLC signaling starting has changed for the internal port. It is now exactly according to the PLCS description. - SN3 drivers : sometimes the PLL logic got locked up (under strange vol- tage and temperature conditions). In that case the whole card must be reset. The reset is handled by the individual driver. For that purpose a function called drv_reset_indication is introduced. Also an Error log is made in that situation. - all drivers/concentrator: increment fddimibEBError_Ct if neccessary. Problems fixed in rel: v1.07b02 ------------------------------- - The following bug was fixed: Under stress the TX descriptor ring of the driver may run out of free TX descriptors to be used for TX frames.If this happens the driver needs to clear fragments from the TX descriptor ring,that have been already sent. Clearing the TX descriptor ring may conflict with a send operation when a new transmit descriptor list for a TX frame is generated. In this case the TX fragment list was corrupted and did not send the correct data out to the network. As a consequence protocols like TCP / IP receive invalid data frames that seem to have an invalid packet size set in the protocol header.TCP/IP performs retransmission and will get retransmission errors. As a result a bad performance may be achieved. - The following bug was fixed: If the driver was just loaded but not started and was unloaded again the driver did not free receive mbuf resources again.The bug could cause the system running out of mbufs if the command is executed very frequently : mkdev -l fddiX or chdev -l fddi0 -a MBUF_type=yes rmdev -l fddiX chdev -l fddi0 -a MBUF_type=no mkdev -l fddiX chdev -l fddi0 -a MBUF_type=no rmdev -l fddiX chdev -l fddi0 -a MBUF_type=yes ..... ..... A new SMT release 2.6.6 has been implemented in this driver release with the following changes: - FDDI PCI driver: Workaround for SN3 temperature / low voltage problem has changed. The counter of EBUF errors increased. - all stations and concentrators: EBErrors can now be received by SMTLOOK. Problems fixed in rel: v1.07b03 ------------------------------- - The following bug was fixed: When setting a MAC address at driver initialization a canonical address was expected by the driver. As a consequence the ndb and lscfg commands displayed a different MAC address than passed to the driver. The address format that is passed to the driver is also physical format as returned from the driver in the NDD - interface structure and therefore needs to be converted into canonical form.The bug has been fixed in this release. Problems fixed in rel: v1.07b04 ------------------------------- - The following bug was fixed for multiprocessor machines only: In a stress situation it happens that tx frames have to be queued in the transmit queue because the receive path (interrupt was busy). Sometimes the tx - scheduler function was not enabled for clearing the tx - queue. This problem caused performance leaks that can be detected with netperf tool. The bug is fixed in this new release. Problems fixed in rel: v1.07b05 ------------------------------- A new SMT release 2.6.7 has been implemented in this driver release with the following changes: -Setting the TReq parameter was not handled correctly in the ISOLATED ring state.This caused situation where the SBA could not be started correctly. Problems fixed in rel: v1.07b06 ------------------------------- - The following bug was fixed: If the driver was loaded and opened and was closed again the driver did not free receive mbuf resources again. This bug could cause the system running out of mbufs if the command is executed very frequently. Problems fixed in rel: v1.07b07 ------------------------------- The driver release number has been added to the driver's object file that can be requested with the following command: what /etc/drivers/<driver-object> A release string is displayed for a beta release e.g: SK-NET FDDI-PCI rel. v1.07b07 96/11/15 (C) SK" ; or for a final release e.g: SK-NET FDDI-PCI rel. v1.07 96/11/15 (C) SK" ; Problems fixed in rel: v1.08b01 ------------------------------- A bug was fixed for multiprocessor machines concerning the following bug fix that has been done already ( for single processor machines only ) in release: v1.07b01 - SN3 drivers : sometimes the PLL logic got locked up (under strange vol- tage and temperature conditions). In that case the whole card must be reset. The reset is handled by the individual driver. For that purpose a function called drv_reset_indication is introduced. Also an Error log is made in that situation. The workaround for the problem described above did not work correctly on SMP machines.A system lock may occur while recovering from the PLL error condition due to a conflict with the TX path. Now the problem is fixed. Problems fixed in rel: v1.08b02 ------------------------------- A bug was fixed concerning the `PLL logic` problem described above for previous driver release. If the driver needs to recover from the `PLL logic hangup condition` it causes the consumtion of system mbufs. As a consequence the system runs out mbuf resources over a long time. The system locks up. Sometimes the error log entry may occur: HWM: Out of RxD condition detected Problems fixed in rel: v1.08b03 ------------------------------- The following bugs have been fixed: If the tx-ring runs out of space during the tx queue is cleared the tx- queue looses mbufs only if the tx - queue is partially cleared when the `tx ring is full` condition is given. As a consequence the system could run out of system resources after a long time. The problem occurs only if there is at least a network load of more that 90% over a long time. A spurious interrupt may occur after the driver has been closed and the adapter`s interrupt has been disabled. In this case the driver`s inter- rupt handler returned an invalid return code, indicating that there was no valid interrupt. The system displays a SYSINTR error log message. Problems fixed in rel: v1.08b04 ------------------------------- The following bug has been fixed for MP machines: The driver may loose some tx - mbufs while queueing frames due to heavy network load. The lack is caused by a missing SMP lock. Problems fixed in rel: v1.08b05 ------------------------------- The following bug has been fixed for MP machines: On heavy network load the tx-queue may be overload on a tx packet storm that caused the driver to exceed the max. number of packets in the tx - queue. As a consequence to much mbuf resources are held by the driver . The protocol stack may hang due to a resource allocation failure . This problem does not occur on single processor machines. Normally the mbuf resources are returned to the system's mbuf pool when the driver is exiting the stress situation. The bug is fixed in this new release. Spurious interrupts may still occur when the driver recoveres from the PLL elasticity event ( see also rel. v1.07b01 ) when there is heavy tx- load e.g. a UDP tx storm. This problem is also fixed. Problems fixed in rel: v1.08b06 ------------------------------- The following bug has been fixed: SBA/ESS did not work correctly . The allocation of synchronous bandwith failed. It was impossible to use synchronous bandwith between two nodes that requested synchronous bandwidth from an SBA. Now the bug is fixed. Problems fixed in rel: v1.09b01 ------------------------------- The following bug has been fixed: If there is transient interrupt source that does not persist to be read again by the second level interrupt handler , this can cause the driver to disable interrupts without enabling them again. As a consequence the driver seems to hang in the receive side . In this case the led does not indicate a 'ring operational change' when disconnecting the driver from the FDDI ring (green led is still on). Problems fixed in rel: v1.09b04 ------------------------------- The following has been changed: The ODM data base entry for the interrupt level has been changed as follows (IBM/ID:214907): default value == "" (empty) value range == "" (empty) This allows the system to use a wider range of interrupt levels for the PCI driver. To avoid a possible deadlock situation in the PM handler the MP- lock in the the PM handler function has been removed except for critical code as stopping and starting the network driver (IBM/ID:218111). The following bug has been fixed: The driver may let the system crash due to an invalid mbuf pointer that is used in the `nd_trace` kernel interface function. This may happen if data is sent via NFS.The problem (IBM/ID:217487) is fixed in this driver release. The driver may let the system crash during an internal data copy of SMT data, that have been received from the network addressed to the driver's local SMT. This problem is caused by a ring-op, when the driver receives large SMT info frames, or after loading and connecting the driver to the network The problem (IBM/ID:217732) is fixed in this driver release. Problems fixed in rel: v1.10b01 ------------------------------- The following bugs have been fixed: The machine initialization by BIOS or Firmware of the PCI command register was overwritten since the command register was not read back before is was configured. The bug has been fixed in this release. There was an uninitialized local variable that is used to update the m_flag of an mbuf that could cause kernel crashes. This function is called only if a driver user has requested SMT frames to be returned from the driver . The problem has been triggered by a new version of the snmpd command in AIX 4.2. The problem does not occur in normal operation with TCP/IP of NFS. The bug has been fixed in this release. When executing a `sna -display` command SNA gets a timout error if only SNA is attached to the driver but no other protocol stack e.g. FDDI.The timeout was cuased by a wrong signaling of the driver's LIMBO and RUNNING status to the upper protocol with both the ndd interface structure and the nd_status interface function. The incorrect signaling is shown in picture (I). The correction is shown in picture (II): L=Driver is loaded O=First open of network driver (NDD_open) C=(Re-)Connect to FDDI ring D=Detach from FDDI ring *=nd_status - messages I) SK FDDI driver - status signaling (rel 1.09b04) ----------------------------------------------- L O C D C __________________________________ 1 (set) NDD_UP ___|.........................................0 ___________ ________ 1 (set) NDD_RUNNING __________| |________|...............0 ________ __________ 1 (set) NDD_LIMBO ___| |_________| |_______........0 *NDD_CONNECTED *NDD_LIMBO_EXIT X X *NDD_LIMBO_ENTER X II) SK FDDI driver - status signaling (rel 1.10b01) ----------------------------------------------- L O C D C __________________________________ 1 (set) NDD_UP ___|.........................................0 ___________ ________ 1 (set) NDD_RUNNING __________| |________|...............0 __________ 1 (set) NDD_LIMBO ____________________| |_______........0 *NDD_CONNECTED X *NDD_LIMBO_EXIT X *NDD_LIMBO_ENTER X Problems fixed in rel: v1.11b01 ------------------------------- 64 Bit enhancement: The 64bit enhancement does not apply to the ability of addressing physical 64 bit wide addresses by the NIC. There is still no such enhancement in the driver. Furthermore this means only that the driver may also be loaded to a 64 bit hardware platform. Problems fixed in rel: v1.12b01 ------------------------------- In case of a 'bus list overflow' error a frame's fragmentation of physical addresses may exceed internal storage capacity of the TX descriptor ring.In this case the frame was discarded . The driver has been enhanced to handle this exception by allocating a new mesage block and copying data into this new data block before executing the DMA operation . If this operation also fails the driver will generate a 'bus list overflow'. This is not a bug fix but may solve TX problems especially if a protocol sends frames that exist of several mbufs that do not fit temporarily into the TX descriptor ring. Changes for 64 Bit enhancement: The DMA_MAXMIN_512K flag passed to the D_MAP_INIT kernel function has been changed to the value DMA_MAXMIN_2M. Also the DMA_BYPASS flag that is passed to the D_MAP_LIST function has been changed to DMA_READ. Problems fixed in rel: v1.12b02 ------------------------------- The following bug has been fixed: After returning from completed DMA operation when unmapping physical DMA addresses an invalid pointer was passed to the kernel function D_UNMAP_LIST. This occures also if the ring is down and the protocol layer tries to send a frame to the network. The bug may cause a system crash.A system crash has not been detected yet but is likely especially on heavy network load. The configuration utility that is used for loading and initializing the driver has been changed. Some printf messages have been removed. Problems fixed in rel: v1.12b03 ------------------------------- The following bug has been fixed: Due to a bug in the driver' s configuration part configuration of more than 10 devices was not possible since the network device driver did not handle the logical device name fddi10 (and higher) correctly. The bug is now fixed in this release. Problems fixed in rel: v1.20b02 ------------------------------- The following bug has been fixed: In previous driver releases the 'Purged Frame' - error counter was not set to zero when the fddi statictic was reset with the command: 'fddistat -r'. The bug is now fixed and the statistic counter will be cleard correctly. New features of rel: v1.20b02 ----------------------------- SubVendor/SubDevice ID support has been added to the driver's installation part. Packaging has been changed and supports now multiple vendor / device ids that will be recognized automatically if the hardware is present . For a detailed list of supported products you may run the 'smit list software' command that shows the contents of the installation media. Since AIX 4.3 recognizes PCI SubVendor/Device IDs for autoconfiguration a product specific hardware description will be installed to the object data base depending on the NIC' s SubVendor/Device ID provided the installation media lists the corresponding type of network adapter. Please note : there is only one diagnostic tool on the installation media. If the diagnose is not automatically installed by executing 'smit cfgmgr' command you should install it manually from the list of software that is shown by the `smit list software` command sequence . Normally the diagnose is only installed automatically if the installed NIC identifies the ` base vendor/device ID` of the diagnose software that is normally 0x48110040 for SK network cards . The DIAGNOSE can be run on any 32-bit NIC listed on the installation media ( THE DIAGNOSE IS NOT VALID FOR NICs THAT SUPPORT 64BIT WIDE PCI SLOTS ) . This sample shows the contents of an AIX installation media from SK: devices.pci ALL @devices.pci _all_filesets SysKonnect FDDI PCI Adapter @devices.pci.48110040.rte SysKonnect FDDI PCI Adapter Diagnostics @devices.pci.48110040.diag SysKonnect FDDI PCI Adapter SK-5521 @devices.pci.48112155.rte SysKonnect FDDI PCI Adapter SK-5522 @devices.pci.48112255.rte SysKonnect FDDI PCI Adapter SK-5541 @devices.pci.48114155.rte SysKonnect FDDI PCI Adapter SK-5543 @devices.pci.48114355.rte SysKonnect FDDI PCI Adapter SK-5544 @devices.pci.48114455.rte SysKonnect FDDI PCI Adapter SK-5821 @devices.pci.48112158.rte SysKonnect FDDI PCI Adapter SK-5822 @devices.pci.48112258.rte SysKonnect FDDI PCI Adapter SK-5841 @devices.pci.48114158.rte SysKonnect FDDI PCI Adapter SK-5843 @devices.pci.48114358.rte SysKonnect FDDI PCI Adapter SK-5844 @devices.pci.48114458.rte If a NIC is installed with a SubVendor/Device ID different from 48110040 (that represents the base product) only the rte part is installed but not the diagnose.Since the first entry 'devices.pci.48110040.rte' of the list of supported products is a prerequisite for all other products it will be installed automatically if the hardware matches any of the other network products. The following object database configuration parameters have been added: - 'Enable Ring State Log' The driver configuration parameter allows the network administrator to log FDDI ring state transitions.If enabled this new driver feature displays an INFO message in the system log whenever the FDDI ring state changed. It allows the network administrator to keep track of unexpected events in the FDDI network e.g. the failure of a network component. - 'PCI Cache Line Size' This configuration parameter overwrites the system PCI configuration value for PCI Cache Line Size.The default value is `ignore` that does not modify the PCI configuration register . If set to any value of the selection list the driver changes the PCI configuration register to one of the following values : 2,4,8,16,32,64 or 128 (not valid for 32 bit PCI Nic) . - 'PCI Latency Timer' This configuration parameter specifies the max. time the PCI adapter can continue with bus master transfers after the system arbiter has removed the GNT signal. The time is specified in units of PCI bus clocks. Default value is -1 that does not change the PCI configuration value . The valid range of values is: -1 - 255 ." - 'PCI Memory Write and Invalidate Cycle' The 'Memory Write and Invalidate Cycle' configuration parameter is used to overwrite the system PCI configuration value.The default value is 'ignore' that does not modify the PCI configuration register. If set to `enable` or `disable` bit 4 of the command register of the PCI configuration space is set to the corresponding bit value 1 or 0 (MWIEN-bit). Problems fixed in rel: v1.21b01 ------------------------------- A new SMT software release 2.8.2 has been included.This contains a bug fix regarding dual homing: the dual homing operation mode on multiple adapters did not work correctly since multiple NICs affected each other and could cause link state failures.