JP2730209B2 - Input/Output Control Method - Google Patents

Description

[Detailed description of the invention] [Summary] This relates to an I/O control method that shortens the time to stop an I/O device under a faulty channel during the process of resetting the system as a channel fault countermeasure, and for an I/O device having an I/O path other than the path connected to the faulty channel, the channel fault countermeasure processing unit and I/O request processing unit recognize the faulty I/O path, and notify the I/O processor, which is the hardware that controls the channel, not to select the faulty I/O path, thereby allowing the I/O request to be executed even during channel fault countermeasures, with the objective of preventing the system from being stopped due to a channel fault. The method comprises a plurality of I/O devices, a plurality of I/O paths provided for each of the plurality of I/O devices, and channels on the I/O paths that control the I/O devices, and the I/O path selection, which selects which channel to use, is performed by the I/O processor, which is the hardware that controls the channel.
In a computer system operated by a processor, the system has a channel fault detection unit which detects channel faults, a channel fault countermeasure processing unit which analyzes channel faults and performs countermeasure processing, an I/O request processing unit which executes I/O requests, an I/O request issuance permission mask which is set to on for I/O paths which can accept issuance of I/O requests and to off for I/O paths which cannot accept them, and a failure occurrence mask which is set to on for I/O paths which have a fault and to off for others, as information on the I/O device, and when a channel fault occurs, the channel fault detection unit detects the fault and notifies the channel fault countermeasure processing unit of the fault, the channel fault countermeasure processing unit performs analysis processing of the channel fault and detects the faulty I/O path, and the I/O request processing unit accepts an I/O request from a user program and sets an I/O path from the device information storage unit which has the I/O request issuance permission mask set to on and the failure occurrence mask set to off, and
The I/O processor is configured to include a data processing device.

[Industrial application field]

The present invention relates to an input/output control method for shortening the time for stopping an input/output device under a faulty channel during the process of resetting the system as a countermeasure against a channel fault.

In recent years, the increasing speed of CPUs in computer systems has led to demands for faster I/O devices. In addition, the demand for highly reliable systems has led to demands for early detection of faulty devices and early recovery by the system.

PRIOR ART

For this reason, systems have been provided that offload the path selection of I/O devices to the hardware, freeing software from the automatic I/O path selection function and I/O request queue management function, thereby reducing CPU overhead.
Since there was no method for selecting a path when a channel fault occurred, it was necessary to stop all I/O devices under the faulty channel until the channel fault countermeasure processing was completed.

Figure 2 shows the hardware configuration of the input/output control.

In the figure, 84, 85, 86, and 87 are channels, 90, 91, and 92 are I/O devices, and 93 is I/O device 90 connected to channel 85.
95 is an input/output path of the I/O device 90 connected to channel 86, and 96 is an input/output path of the I/O device 90 connected to channel 87. 88 is an input/output path of the I/O device 90 connected to channel 8
4, 85 is an input/output control device, and 89 is channel 8
6, 87. 79 is a data processing device that controls the overall processing. 80 is a channel fault countermeasure processing unit, 81 is an input/output request processing unit, and 82 is a
is the device information storage unit. 83 is an I/O processor, which is hardware that selects the I/O path. An I/O path is a path taken when performing I/O, and when a channel is involved, it is called a channel path. The computer system in this diagram is a system in which the paths between channels and I/O devices are dynamically connected and reconnected. In this system, in order to make efficient use of channel resources, the connection to the channel is cut off while a disk or other device is performing a mechanical operation, allowing the channel to perform other tasks. Furthermore, when a disk or other device is reconnected to a channel, a usable channel path is selected from among all the channel paths connected to the system that initiated the I/O operation. This is called the dynamic reassociation function (DP
This is called the R function.

Fig. 6 is a functional block diagram of the prior art. First, input/output processing during normal operation will be explained with reference to the hardware configuration in Fig. 2.

In the figure, 101 is a channel fault detection unit that detects the occurrence of a channel fault. 102 is a fixed fault detection unit using an input/output statistical method that mainly detects soft faults.
is an I/O request, which is an I/O request of a user program. 104 is a channel fault countermeasure processing unit, 105 is an I/O request processing unit, 106 is a device information storage unit, and 107 is an IO
P (I/O processor). 108 is the Q for general I/O requests.
It is a user interface that temporarily suspends I/O requests.
Reference numeral 9 denotes an active I/O request recovery unit, which stops active I/O requests and places them in a queue. Reference numeral 110 denotes a reserve path switching unit, which transfers I/O information to another I/O path.
Reference numeral 112 denotes a system reset issuing unit, which resets all I/O path information under the faulty channel. Reference numeral 112 denotes a system reset analysis unit, which analyzes the results of the system reset. Reference numeral 113 denotes a path group re-establishment unit, which returns the faulty I/O path to a healthy path group. Reference numeral 114 denotes a general I/O request quiesce release unit, which releases the quiesce of I/O requests under the faulty channel. Reference numeral 115 denotes a restart request unit, which resumes I/O requests in the waiting queue. Reference numeral 116 denotes an I/O request acceptance unit, which accepts I/O requests from user programs. Reference numeral 117 denotes a CCW creation unit, which creates a command to make an I/O request to the channel. Reference numeral 118 denotes an I/O request issuance request mask determination unit, which makes an I/O request using the above-mentioned path group as a parameter. Reference numeral 121 denotes an I/O request waiting management unit, which makes the I/O request wait if there is a preceding request.
Reference numeral 120 denotes a status analysis unit, which analyzes the results of input/output execution. Reference numeral 122 denotes a unit for notifying the request source, which notifies the user program of the execution result. Reference numeral 123 denotes an input/output request retry unit, which retries the execution if the execution result is abnormal. Reference numeral 124 denotes a subchannel number, which stores the number of the input/output device. Reference numeral 125 denotes quiesce control information, which is sent to the input/output request quiesce unit 108.
This contains information about I/O requests that have been quiesced.
is reserved information, which is information about input/output that a certain input/output device has. 128 is an input/output request mask, which stores input/output paths that are available for input/output. 129
is path group information, which stores the path group when making an I/O request to the IOP 107. 131 is an activation path selection unit, which selects a path from the path group for which input/output is actually performed. 132 is an I/O activation unit, which makes an I/O request to a channel. 133 is a dynamic path reconnection processing unit, which performs the DPR function.

In Fig. 6, some processing is the same as that in Fig. 2, so it will be described below. The channel fault countermeasure processing unit 104 in Fig. 6 and the channel fault countermeasure processing unit 80 in Fig. 2 are the same. In this respect, 104 and 80 are represented as being the same. Similarly, 105 and 79, 106 and 82, and 107 and 83 are the same.

In the above configuration, when a normal I/O request is received, the I/O request processing unit 105 accepts the I/O request. The CCW creation unit 117 adds a system CCW to the CCW created by the user,
An I/O request issuance request mask decision unit 118 sets an I/O request issuance request mask (LPM), and an I/O request issuing unit 119 issues an I/O request (SSCH command).
In the case of an I/O request to device 90, there are four paths (93, 94, 95, 96) that can issue an I/O request to the device, and all of these I/O paths are online if there are no abnormalities. In this case, the I/O request mask 128 for device 90 in device information storage unit 106 has the mask bits corresponding to the above four paths turned on. In addition, in a system in which the path selection of an I/O device is offloaded to the hardware side, the I/O device normally holds a path group. The above four paths are the path group of the device, and the dynamic path reassociation function (DPR function) allows any path in this group to be used for reassociation. This path group information is stored in the I/O control device 88 on the hardware side.

The I/O request processing unit 105 checks whether there is a preceding request to the I/O device 90, and if an I/O request is already being executed, the I/O request queue management unit 121 makes the request queued. If there is no preceding request, the I/O request issue request mask is set and an I/O request is issued.

When the IOP accepts this I/O request, it selects the startup path for the I/O device. The IOP knows which channel the I/O device is connected to, and starts the channels connected to the requested I/O device in order. If the channel is currently executing an I/O request for another I/O device, the channel is considered busy. In this case, the software is not notified, and the IOP selects the next channel within itself and attempts to start the device again.

When the I/O processing is completed, an I/O interrupt is notified to the software. The I/O request processing unit retrieves (TSCH command) and analyzes the status to obtain the end state of the I/O request and the device status. If the I/O request is completed normally, it notifies the request source of the normal end. On the other hand, if the I/O request is completed abnormally, it notifies the fixed fault detection unit 102 using the I/O statistics method of this abnormal state, and then
This I/O request is retried (ERP). If the request does not end normally after a certain number of retries, the I/O request source is notified that the I/O process ended abnormally.

Next, the processing performed by the conventional technology when a channel failure occurs will be described.

There are two ways to detect faults in I/O devices: one is a serious channel fault notified by the hardware via a machine check interrupt, and the other is to statistically accumulate I/O errors that occur within the system and identify the location of the fault.

When an I/O machine check interrupt is reported,
The channel fault detection unit 101 issues the STCRW command, collects the status, and analyzes which channel is faulty. In addition, the fixed fault detection unit using the input/output statistics method:
If a particular channel has a high frequency of I/O errors,
The channel is determined to be faulty. In this way, the faulty channel is identified and the channel fault countermeasure processing unit 104 is scheduled. For example, by a machine check interrupt,
Assume that channel 84 is reported as faulty.

The channel fault countermeasure processing unit 104 first detects the faulty channel 8
The quiesce unit 108 performs a quiesce process on the I/O requests so that all the subordinate I/O devices 90, 91, and 92 will not execute any general I/O requests even if they are accepted thereafter, and the requests are made to wait in a queue.

Next, the active I/O request recovery unit 109 issues an HSCH command to the I/O device currently executing the I/O request to stop the I/O request. This is equivalent to a system reset (RC) during data transfer.
This is because when a command such as the HP command is issued, data cannot be guaranteed. Any I/O request that was in progress is forcibly terminated and a retry is attempted in the form of a pseudo I/O error.

In addition, the HSCH command issued by the channel fault countermeasure processing unit 104 is controlled so that it can be executed even when the I/O device is in a state where I/O requests are paused.
The input/output request 104 is level-controlled so that it can be executed.

Furthermore, if the I/O device holds a reservation, the reservation information is transferred to another I/O path by the reservation path switching unit 110 so that the reservation information is not lost due to a system reset. This is done by executing a UR command from another path. Note that if a path group exists, a system reset automatically causes the path for which the system reset was issued to leave the group, and the reservation information is transferred to the remaining path group, so in this case there is no need to issue a UR command.

After the above preparations are completed, the system reset issuing unit 11
A system reset is issued to the faulty channel with a command of 1. This command resets the faulty channel and the channel path routes of all I/O devices connected to the channel.

System reset completion is system reset analysis part 11
A machine check interrupt is issued at 2 to notify the user of this. If the system reset is successful, the faulty channel path that was removed by the issuance of the system reset returns to being a healthy path, so that the path group re-establishment unit 113 issues an ESTABLISH I/O command specifying this faulty path in order to incorporate it back into the path group. In this case, if a path has been specified, the I/O request processing unit 105 adds an SMPR command to the I/O request issuance mask 128 for this path, and to prohibit the dynamic rebinding (DPR) function.

After the path group reestablishment process is complete, general I/O requests are
The quiesce release unit 114 releases the quiesce of I/O requests from all I/O devices under the faulty channel, making them eligible to execute general I/O requests as well. Next, the restart request unit 115 schedules restart processing for all I/O devices and executes the I/O requests in the waiting queue. At this point, the path group has also returned to the state before the fault, so I/O processing is the same as normal.

[Problem to be solved by the invention]

As described above, with conventional technology, if the I/O device under the faulty channel contains something important to the system and frequently accessed, such as a magnetic disk that stores a dataset that is the target of paging, I/O requests to this I/O device have to wait until the countermeasure processing for the faulty channel is completed, resulting in a problem that the system may stop during this time.

The present invention aims to prevent the system from stopping due to a channel fault by recognizing the faulty I/O path in the channel fault countermeasure processing unit and I/O request processing unit for an I/O device having an I/O path other than the path connected to the faulty channel, and notifying the I/O processor, which is the hardware that controls the channel, not to select the faulty I/O path, thereby allowing the I/O request to be executed even while channel fault countermeasures are being taken.

[Means for Solving the Problems] FIG. 1 is a diagram for explaining the principle of the present invention.

In the figure, 1 is the channel fault countermeasure processing unit, which is channel 7.
The I/O request processor 2 performs an I/O request when a channel fault occurs.
Reference numeral 3 denotes a device information storage unit, which has an I/O request issuance permission mask 4 that displays I/O paths that are permitted and prohibited for issuing I/O requests, and a fault occurrence mask 5 that displays I/O paths where a fault has occurred, and stores information about I/O devices. Reference numeral 6 denotes a data processing device which processes the entire system. Reference numeral 8 denotes an I/O device, and 9 denotes an I/O path, which becomes the route through which channel 7 controls I/O device 8. Reference numeral 10 denotes hardware which selects the I/O path in the I/O processor. Reference numeral 11 denotes a channel fault detection unit which detects channel faults.

In addition, since there are some parts in Fig. 1 that are the same as those in Fig. 2, they will be shown below. The channel fault countermeasure processing unit 1 in Fig. 1 and the channel fault countermeasure processing unit 80 in Fig. 2 are the same. For brevity, 1 and 80 are represented as the same. Similarly, 2 and 81, 3 and 82, 6,
79,10 and 83 are identical.

In the above configuration, when a channel fault occurs, the channel fault detection unit 11 in the data processing device 6 notifies the channel fault countermeasure processing unit 1 of the fault. The channel fault countermeasure processing unit 1 temporarily stops the I/O devices 8 under the faulty channel, stores the faulty I/O path for each I/O device 8 in the fault occurrence mask 5 in the device information storage unit 3, and displays that fault countermeasures are being taken. After this, it collects I/O requests being executed for all I/O devices 8 under the faulty channel. After this, it releases the suspension of the I/O requests and requests the I/O request processing unit 2 to restart the I/O requests. The I/O request processing unit 2 detects paths available for I/O requests and makes I/O requests to the I/O devices using the available paths as parameters. Next, the channel fault countermeasure processing unit 1 issues a system reset. Next, it resets the fault occurrence mask and turns off the display indicating that fault countermeasures are being taken.

[Action]

As described above, in the present invention, the input/output path on which a fault has occurred is stored in the device information storage unit as a fault occurrence mask,
The paths available for input/output are determined based on the input/output request issuance permission mask and the fault occurrence mask, and an input/output request is made when a system reset is issued using this available path as an argument. This makes it possible to execute the input/output request using other input/output paths even while the channel fault countermeasure processing unit is issuing a system reset to the faulty channel and executing countermeasure processing.

[Example]

FIG. 3 shows a functional block diagram of an embodiment of the present invention.
However, the shaded boxes indicate the parts that differ from the conventional technology.

FIG. 3 is almost the same as FIG. 6, so the modifications of the present invention compared to the prior art will be explained.

45 is a section for displaying the fault countermeasure in progress and setting the fault path mask, which is set when a channel fault occurs. 53 is a section for canceling the fault countermeasure in progress and resetting the fault path mask, which is a process performed when the fault is cleared. 56 is an SMPR
A command addition section returns data to the channel that has made an I/O request. 58 is a processing section that performs an exclusive OR with the faulty path mask, and sets paths that are capable of I/O requests during a fault. 71 is a fault countermeasure processing display that indicates that a channel fault has occurred. 72 is a faulty path mask that indicates the path that is experiencing a fault.

In addition, since there are some parts in Fig. 3 that are the same as those in Fig. 6, they will be shown below. For example, the channel fault detection unit 37 in Fig. 3 is the same as the channel fault detection unit 101 in Fig. 6. For brevity, 37 and 101 are expressed as being the same. Similarly, 38 and 102, 39 and 103, 4
4 and 108, 46 and 109, 47 and 110, 48 and 114, 49 and 115, 50 and 111, 51
and 112, 52, 113, 54, 116, 55, 117, 57, 118, 59, 119, 61,
121, 60 and 120, 62 and 122, 63 and 123, 64 and 124, 65 and 125, 66 and 12
6, 68 and 128, 69 and 129, 73 and 131, 74 and 132, 75 and 133 are identical. As a whole function, 40 and 104, 41 and 105, 42 and 106, 43
and 107 correspond to each other as having the same function.

Here, the following will mainly explain the parts that have been improved by the present invention, with reference to the hardware configuration diagram of FIG.

In addition, in Fig. 3, there are some processes that are the same as those in Fig. 2, so they will be described below to avoid confusion.
are the same. For brevity, we write 40 and 80 as being the same. Similarly, 41 and 81, 42 and 82, 43 and 83 are the same.

In the configuration of FIG. 3, the channel fault countermeasure processing unit 40 first puts all the I/O devices 90, 91, and 92 under the faulty channel 84 in a queue so that they do not execute any general I/O requests even if they are accepted thereafter, and then executes the general I/O request quiesce unit 44.
Then, the I/O request quiescing process is performed. Next, the fault countermeasure in progress indication and fault path mask setting unit 45 sets the fault countermeasure in progress indication and fault path mask (42, 71, 72) for all devices under the faulty channel. Next, the active I/O request recovery unit 46 stops I/O requests for the currently active I/O device. The active I/O request is forcibly terminated and awaits retry in the form of a pseudo I/O error. Next, the reserve path switching unit 47 moves the reserve information 66 held by the I/O device to another I/O path.

Once the above preparations are complete, it becomes possible to execute I/O requests from online paths other than the faulty channel path. This is because the scope of effect of a system reset is the faulty channel and the channel path paths of all subordinate I/O devices connected to the channel. Now consider executing an I/O request to an I/O device 90 under the faulty channel using a path 95 that uses a channel other than the faulty channel while a system reset is being issued to the faulty channel 84. The general I/O request quiesce release unit 48 releases the quiesce of I/O requests from all I/O devices under the faulty channel,
General I/O requests are also made executable, and restart processing is scheduled for all I/O devices by the restart request unit 49. As a result, I/O requests on the waiting queue are executed.

In the I/O request processing unit 41, if the I/O device is undergoing fault countermeasure processing, an SMPR command is added by an SMPR command adding unit 56 to prohibit dynamic reconnection. Next, a processing unit 58 that performs exclusive OR with the fault occurrence path mask performs exclusive OR with the I/O request mask that is currently considered to be capable of issuing I/O requests (at this point, the I/O request mask 68 of the I/O device 90 has the bits corresponding to all four I/O paths 93, 94, 95, and 96 turned on) and the fault occurrence path mask set by the channel fault countermeasure processing unit 40, so as not to select the faulty channel path.

In this way, the I/O device 90 requests an I/O request from the IOP with the I/O issue request mask (the bits corresponding to the I/O paths 93, 95, and 96 are on). The IOP selects one of these three paths and starts it.
Since the SMPR command is still added when reconnecting, the reconnection is performed using the same path that was started. If an I/O request was made using three paths and the SMPR command was not added, the I/O controller 88 would still believe that the path group was made up of four paths, so
There is a possibility that the failed channel path 94 may be rejoined. In this case, if a system reset hits that path, the completion interrupt for this I/O request will be lost.

In this way, while all devices under the faulty channel are allowed to execute general I/O requests, the system reset issuing unit 50 asynchronously issues a system reset to the faulty channel. If the system reset is successful, the faulty channel path that was removed by issuing the system reset in the path group re-establishment unit 52 returns to a normal path.
In order to incorporate the failed path 94 back into the path group, an ESTABLISH I/O is issued specifying this failed path 94. After the path group has been re-established, the display indicating that a fault countermeasure is being processed is cleared in the fault countermeasure in progress release and failed path mask reset unit 53, and the failed path mask is reset. After this, the I/O request processing unit
Since 41 is not undergoing fault recovery processing, normal input/output processing will be carried out.

Next, FIG. 4 shows a detailed diagram of the device information storage unit.

In FIG. 4, 151 is a device information storage unit for one I/O device.
Reference numeral 52 denotes the fault countermeasure processing display section; when this is on, fault countermeasures are being performed, and when it is off, fault countermeasures are not being performed. Reference numeral 153 denotes an I/O request issuance permission mask, which indicates which channel paths are permitted to issue I/O requests and which are not. A path with a mask value of 1 indicates that I/O requests can be issued, and a path with a mask value of 0 indicates that I/O requests cannot be issued. Reference numeral 154 denotes a fault occurrence mask, which indicates a channel path in which a fault has occurred. Reference numeral 155 denotes mask A, which is an example of an I/O request issuance permission mask. Reference numeral 156 denotes mask B, which is an example of a fault occurrence mask. By taking the exclusive OR of mask A and mask B, it is possible to control so that I/O requests are not issued from the faulty channel path.

Finally, the difference between the processing of the present invention and the processing of the prior art will be shown by a time chart.

FIG. 5 is a time chart of the fault processing.
The upper part of the figure is a time chart of a fault processing according to the prior art, and the lower part is a time chart of a fault processing according to the present invention.

The fault processing diagram in Fig. 5 shows the operation of channels 84 and 85 in the configuration diagram in Fig. 2. The time chart shows the processing from when a fault occurs in channel 84 to when the fault is restored. Details of each processing step have already been explained, so they will be omitted here. As shown in the diagram, in the prior art,
Processing is suspended from when a fault occurs in channel 84 until the fault is repaired and restarted, but in the present invention, normal processing can be resumed once the input/output processing unit issues an I/O request and a healthy channel 85 starts input/output processing. In other words, in the present invention, input/output processing can be resumed significantly faster than in the prior art.

[Effects of the Invention]

As described above, according to the present invention, even if a channel failure occurs, it is possible to execute an I/O request to an I/O device under the faulty channel while the faulty channel is being restored, which greatly contributes to speeding up I/O processing in the system.

[Brief description of the drawings]

Fig. 1 is an explanatory diagram of the principle of the present invention, Fig. 2 is a hardware configuration diagram of I/O control, Fig. 3 is a functional block diagram of one embodiment of the present invention, Fig. 4 is a detailed diagram of the device information storage unit, Fig. 5 is a time chart diagram of fault processing, and Fig. 6 is a functional block diagram of the prior art. In Fig. 1, 1 is a channel fault countermeasure processing unit, 2 is an I/O request processing unit, 3 is a device information storage unit, 4 is an I/O request issuance permission mask, 5 is a fault occurrence mask, 6 is a data processing device, 7 is a channel, 8 is an I/O device, 9 is an I/O path, and 10 is an I/O processor (IOP).
11 is a channel fault detection unit.

Claims (1)

(57) [Claims] [Claim 1] In a computer system having an I/O request processing unit (2) that executes I/O requests, a plurality of I/O devices (8), a channel (7) that controls the I/O devices (8), an I/O path (9) provided for each of the plurality of I/O devices (8) and the channel (7), a device information storage unit (3) that stores paths that are available and unavailable for the I/O requests, and an I/O processor (10) that selects which I/O path to use for an I/O request, a channel fault detection unit (11) that detects faults in the channel
and a channel fault countermeasure processing unit (1) that analyzes a channel fault and processes the countermeasure. When an input/output fault occurs, the channel fault detection unit (1)
an input/output control system in which a channel fault countermeasure processing unit (1) detects a fault and notifies the channel fault countermeasure processing unit (1) of the fault, the channel fault countermeasure processing unit (1) analyzes the channel fault and stores the failed input/output path in the device information storage unit (3), the input/output request processing unit (2) accepts an input/output request from a user program and continues input/output processing by referring to the device information storage unit (3), and at the same time, the channel fault countermeasure processing unit (1) interrupts the path in which the fault has occurred and cancels the input/output command on the failed path, executes restart request processing, stores the recovered path in the device information storage unit (3), and continues the input/output processing command.