JPH08147199A - Process execution information collection device - Google Patents

Abstract

(57) [Summary] [Purpose] Regarding the process execution information collection device that collects and outputs process execution information, the number of executions and the execution time of dynamic library functions are required without re-translation / recombination of measurement targets. The objective is to be able to obtain only such information and to reduce the performance degradation when measurement is not performed. [Configuration] When collecting execution information by determining the notification information from the kernel 45 in the dynamic linker 10,
Execution information collection part 22 by mapping the information collection library 20
To be started every predetermined time. In addition, the entry information of the link table 32 used when linking the dynamic libraries is rewritten so as to call the function execution trace unit 23, and when the dynamic library function is called for the second time or later, it is passed through the function execution trace unit 23. To call the function.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a process execution information collecting device which collects and outputs process execution information such as the number of executions and execution time of a dynamic library function.

In recent years, performance problems have arisen in various programs, and there is a demand to collect as much information as possible to find out the cause. In particular, since it is difficult to obtain execution information about dynamic library functions, information about dynamic library functions is used as one piece of information for easier program analysis in order to improve software quality and performance. It is useful to get.

[0003]

2. Description of the Related Art The prior art is as follows. When collecting information about library functions, for example, in UNIX operating system, it can be collected by using UNIX command such as prof. In that case, as shown in FIG. 11, measurement is performed in the load module. Must be embedded in the instruction for. Therefore, it is necessary to re-translate (compile) / re-join (link) the measurement target program by specifying the option for collecting function information.

When the execution information of the library function is collected by the above method, since the information of the dynamic library cannot be collected, the program must be retranslated / recombined using the static library corresponding to the dynamic library. I had to do it.

In addition, in the related art, generally, the information until one process is started and ended is always collected. Another way to collect information about library functions without retranslation is to set a breakpoint at the entry point of the function in the debugger, run the process, and collect the information in the debugger. In this case, it is necessary to set the break point, and the break point
When I got to the point, I needed to perform the operation again.

In this case, since only the information of each process can be obtained, it is necessary to output the information of each process to the screen or the recording medium and then merge the function information of each process. . In addition, the output of each process to a recording medium or screen imposes a load on each process itself.

In other conventional techniques, there is a method of analyzing the instruction code and calculating the execution time based on the method.

[0008]

[Problems to be Solved by the Invention] For example, the above-mentioned prof
The following problems have occurred when using commands.

When there is a source program file, it is necessary to retranslate the target source program file for acquiring the function execution time. Also, if the source program file does not exist, it was necessary to give up acquiring the execution time information of the function.

Although it can be used for a main program that calls a dynamic library by dynamic linking, the execution time information of the dynamic library function cannot be collected, so in practice, there are many problems in which part of the program. This sometimes caused a misunderstanding about whether it was hanging.

In order to obtain the execution time information for the dynamic library function, it is necessary to prepare a static library corresponding to the dynamic library and rejoin them, which takes time and effort. . In addition, when there is no static library corresponding to the dynamic library, the acquisition of the execution time information of the library function had to be abandoned.

[0012] Since it is not possible to obtain only the information of the requested portion and only the information of the entire one process is obtained, it takes a lot of time to collect the information in the case of a process with a long execution time, and it usually stops. In the case of a process that continues to operate without any action, the function execution time information could not be acquired.

Further, when collecting information about a library function using a debugger, it is necessary to check the entrance of the function, set a break point, and re-execute when the break point is reached. There was a problem. In particular, it was difficult to obtain execution information during operation.

When attempting to collect information on processes that are being activated at the same time, each process is output to a screen or recording medium, which causes a problem that a load is applied to each process. In addition, there is a problem that it is necessary to merge the output information of individual processes.

In the method of analyzing the instruction code and calculating the execution time, it takes a long time to collect and output the information. The present invention solves the above-mentioned problems, and can acquire only the information of the requested part of the execution information of the function of the execution module and the dynamic library function without performing the retranslation / reconnection, and the measurement can be performed. It is an object of the present invention to provide a means for minimizing performance deterioration even when the same module is used when not performing.

[0016]

FIG. 1 is a diagram showing a configuration example of the present invention. In the figure, 1 is a process to be measured, 10 is a dynamic linker improved for measurement, 11 is a measurement target determination unit that determines the measurement target, and 12 is a link table (hereinafter referred to as PLT (Procedure Linkage Table)). ), The PLT entry rewriting unit, 20 is an information collecting library, 21 is an initial processing unit that performs initialization at the time of measurement, 22 is an execution information collecting unit that is activated at regular intervals, and 23 is a dynamic library function. A function execution trace unit that is activated when calling the, a collected information output unit that outputs collected information, a error processing unit that calls the collected information output unit 24 or the shared memory information output unit when an error occurs, and a 30 Main load module to be measured,
31 is an instruction text part, 32 is a link table (PLT) which is used by setting the address information of the control transfer destination at the time of dynamic linking, and 33a and 33b each store one or more functions to be dynamically linked. Dynamic library,
Reference numeral 40 is a measurement monitoring processing unit that receives and processes a command related to measurement, 41 is a flag setting unit that sets a flag indicating whether or not to measure, and 42 is a shared memory that outputs the collected measurement information to a predetermined storage device or output device. An information output unit, 45 is a kernel which is a central part of the operating system, 46 is an in-kernel flag indicating whether or not to measure, 51 is an input device such as a keyboard, 52 is an output device such as a display device, 53 is a disk device, etc. Storage device, 60 is a shared memory shared between processes, 63
Represents in-process memory.

According to the present invention, there is a function of obtaining the value of the program counter at fixed time intervals and collecting execution information of functions (hereinafter referred to as profile function), and the information when the dynamic library function is called. It has a function to collect data (hereinafter referred to as the trace function). These functions are realized by the following means in a normal load module state without modifying the main load module 30 and the dynamic libraries 33a, 33b, ....

First, the profile function will be described. When there is an instruction to execute the profile function by inputting a command from the input device 51, the measurement / monitoring processing unit 40 sets the fact to the in-kernel flag 46 by the flag setting unit 41. The kernel 45 notifies the dynamic linker 10 of the in-kernel flag 46 at the start of execution of the measurement target process 1.
Notify the status of.

When the dynamic linker 10 learns from the notified flag information that there is an instruction to execute the profile function, it maps the information collection library 20 and first calls the initial processing section 21. The initial processing unit 21 makes settings to activate the execution information collection unit 22 at predetermined time intervals. After that, when the execution control is transferred to the main load module 30, the execution information collecting unit 22 is activated at regular intervals by the signal signal of the timer.

The execution information collecting unit 22 includes the main load module 30 or the dynamic library 33a whose execution is suspended,
The value of the program counter such as 33b is sampled, the number of times the address indicated by it is sampled is counted according to the value of the program counter, and the count information is recorded in the in-process memory 63 or the shared memory 60. At the end of the execution of the main load module 30 or the end of the measurement, the collected information output unit 24 causes the main load module 30 and the dynamic libraries 33a, 3 related to the information and the measurement target stored in the in-process memory 63 or the shared memory 60 to be collected.
The head address and the tail address of the instruction text portion of 3b are output to a predetermined storage device 53 or output device 52. Alternatively, when the count information is output to the shared memory 60, the shared memory information output unit 42 provided in the measurement monitoring processing unit 40 stores the measured information in the shared memory 60 for each predetermined amount in the storage device 53 or the output device. Output to 52.

Before or at the start of the processing, the measurement monitoring processing unit 40 sets the information in the shared memory 60 when the information about the measurement time or the process to be measured is specified by the command input from the input device 51. To do. The measurement target determining unit 11 refers to the information and determines whether to perform the process by the initial processing unit 21 or the process by the execution information collecting unit 22.

The trace function is realized as follows. When a trace function execution instruction is given by a command input from the input device 51, the measurement monitoring processing unit 40 sets the fact to the in-kernel flag 46 by the flag setting unit 41.
When the execution of the measurement target process 1 is started, the kernel 45
The state of the in-kernel flag 46 is notified to the dynamic linker 10.

The dynamic linker 10 includes the dynamic library 33a.
Etc., and when it is known that there is an instruction to execute the trace function from the notified flag information, the information collecting library 20 is also mapped and the address of the function execution trace unit 23 is obtained. After that, when the execution control is transferred to the main load module 30, the dynamic linker 1 for dynamic linking (coupling) is executed at the first call of each function such as the dynamic library 33a.
0 is called. In the dynamic linker 10, the PLT entry rewriting unit 12 rewrites the corresponding entry in the link table 32. Normally, the address information of the called dynamic library function is set in the corresponding entry of the link table 32, but if the execution of the trace function is instructed, the address information of the dynamic library function is set. Instead, the address information of the function execution trace unit 23 is set and the function execution trace unit 23 is called.

The function execution trace unit 23 collects execution information such as the current time, the process ID, and the function name of the dynamic library function in the shared memory 60, and then passes control to the called function of the dynamic library 33a. . Since the corresponding entry in the link table 32 has been rewritten,
Each time the dynamic library function is called, the function execution trace unit 23 is directly called from the function calling source. When the execution of the main load module 30 ends or the measurement ends, the collected information output unit 24
Outputs the information stored in the shared memory 60 to a predetermined storage device 53 or output device 52. Alternatively, the shared memory information output unit 42 provided in the measurement monitoring processing unit 40 outputs the measurement information in the shared memory 60 to the storage device 53 or the output device 52 for each predetermined amount.

Regarding the above processing, the measurement monitoring processing unit 40
When information about a process to be measured, a dynamic library, or a function in the dynamic library is specified by a command input from the input device 51, the information is set in the shared memory 60. Based on the information of the measurement target set in the shared memory 60, the measurement target determination unit 11 sets the P
PL is selected so that the entry rewriting target of the link table by the LT entry rewriting unit 12 is selected, and control is transferred to the function execution tracing unit 23 only for the process, dynamic library, or function in the dynamic library that requires tracing.
The T entry rewriting unit 12 is instructed.

If an error occurs during the execution of the above profile function or trace function, the error processing unit 25
Is called, and the error processing unit 25
Calls the collection information output unit 24.
The collection information output unit 24 outputs the collection information recorded in the shared memory 60 or the in-process memory 63 until an error occurs to the storage device 53 or the output device 52.
Alternatively, when both the profile function and the trace function output information to the shared memory 60, the error processing unit 25
Calls the shared memory information output unit 42, and the shared memory information output unit 42 outputs the measurement information in the shared memory 60 to the storage device 53 or the output device 52.

[0027]

According to the first aspect of the invention, the initial processing unit 21 is called at the head of the dynamic linker 10 and the execution information collecting unit 22 is set to transfer control at regular intervals. The instruction to be executed can be counted by the execution information collecting unit 22 every time. By obtaining information on how many times each address is executed, execution time information for each function can be calculated according to the ratio of the number of executions. Further, since the execution information collection unit 22 also counts the address of the dynamic library, the execution time information of the dynamic library function can be calculated.

In the case of the prior art, in order to write the command for starting the information collection at the head of the command text portion 31, it was necessary to translate / combine the program for measuring the execution time, but in the present invention, It is not necessary to write an instruction for starting information collection to the main load module 30 in advance, and it is not necessary to translate / combine for information collection because it is dealt with in the dynamic linker 10 for measurement.

In the inventions according to claim 2 and claim 3,
Information on whether or not the measurement monitoring processing unit 40 performs measurement is set in the kernel 45, and in the dynamic linker 10, the kernel 45 is set.
Measurement / non-measurement is judged by the notification from, and in the case of non-measurement, only normal processing is performed like the conventional dynamic linker. Therefore, the overhead in the main load module 30 and the dynamic linker 10 is small, and it is possible to make the performance deterioration when measurement is not performed very small.

According to the fourth aspect of the present invention, the shared memory 60 shared by the measurement monitoring processing unit 40 and each process having the dynamic linker 10 for measurement is used. By having the process, the measurement target determination unit 11 called from the initial processing unit 21 and the execution information collection unit 22 of each process determines whether or not each process is a measurement target, and when it is a measurement target, the measurement is performed. However, it is possible to perform processing that does not perform measurement if it is not the measurement target. Further, by providing the shared memory 60 with a flag or the like indicating when to collect the execution information, it is possible to collect the information only when the information is desired to be collected and not to collect the information at other times.

Therefore, for each process, it is possible to collect information about only the requested process and only the requested part. In the invention according to claim 5, for function tracing, the entry of the linking table 32 is rewritten so that control is passed to the function execution tracing section 23, and the function execution tracing section 23 immediately before entering the dynamic library function. Since the function information (function name, process name, process ID, library name, time, etc.) is acquired, it is necessary at any time without retranslating / rejoining the source programs such as the main load module 30 and the dynamic library 33a. Information can be obtained.

In the inventions according to claims 6 and 7,
Information on whether or not the measurement monitoring processing unit 40 performs measurement is set in the kernel 45, and in the dynamic linker 10, the kernel 45 is set.
Measurement / non-measurement is judged by the notification from, and in the case of non-measurement, only normal processing is performed like the conventional dynamic linker. Therefore, the overhead in the main load module 30 and the dynamic linker 10 is small, and it is possible to make the performance deterioration when measurement is not performed very small.

According to the eighth aspect of the present invention, the shared memory 60 shared by the measurement monitoring processing unit 40 and each process having the dynamic linker 10 for measurement is used, and the information of the measurement target is stored in the shared memory 60. By having it, the measurement / non-measurement is determined accordingly, and the content rewritten by the PLT entry rewriting unit 12 in the dynamic linker 10 is used as the address information of the function execution trace unit 23 in the case of the measurement target, and is not measured. Since the address information of the dynamic library function is used for, the call of the dynamic library function when it is not the measurement target can be made the same normal call interface as before, and the overhead when calling the dynamic library function that is not measured. Can be eliminated.

According to the ninth aspect of the invention, when an error occurs during information collection, the error processing section 25 is called, and the error processing section 25 calls the collected information output section 24 to collect the information collected up to that point. Output. Therefore, even if an error occurs during information collection, the information collected until then can be output without losing the information until the error occurs.

In the tenth aspect of the invention, the execution information collecting unit 22 or the function execution tracing unit 23 records the collected information in the shared memory 60, and the shared memory information output unit 42
Since it is output to the storage device 53 or the output device 52 for each predetermined amount, the overhead at the time of collecting information can be reduced, and particularly when collecting measurement information of a plurality of processes, a large load is placed on each process. It is possible to collect necessary information without spending time.

[0036]

An embodiment of the present invention will be described below with reference to the drawings. [First Embodiment] FIG. 2 is a block diagram of an embodiment of the present invention.

In FIG. 2, the same reference numerals as those in FIG.
Corresponding to the one shown in (1), 34 is a register for the kernel 45 to notify the dynamic linker 10 of flag information (measurement / non-measurement) regarding measurement, and 61 is shared by the measurement target process 1 and the measurement monitoring processing unit 40. Shared memory for setting measurement object, 62 is a shared memory for recording collected information for recording collected information, 70 is a collected information file for storing information output by the collected information output unit 24, 71 is a symbol of the main load module Tables 72a and 72b represent symbol tables of the dynamic libraries 33a and 33b, respectively, and 80 represents an information display processing unit for editing and outputting the information in the collection information file 70.

When a measurement start (or measurement preparation) command for a trace or a profile is input, the flag setting unit 41 of the measurement monitoring processing unit 40 sets the in-kernel flag 46 to ON (measurement). Further, when the measurement end command is input, the flag setting unit 41 returns the in-kernel flag 46 to OFF (non-measurement).

When the process 1 is started, the kernel 45 discriminates the in-kernel flag 46, and the dynamic linker 10
Notify the measurement / non-measurement of execution information to. This notification includes
An unused register 34, which is predetermined as an interface with the dynamic linker 10, is used.

Further, the measurement monitoring processing unit 40 measures the process such as process name, process ID, or user ID and user name in the measurement target setting shared memory 61 based on the information specified by the measurement instruction command. Set the judgment conditions that identify whether or not to perform, and the flag that indicates whether or not to perform measurement at each time point. For example, if the command specifies that measurement should be started after 10 minutes, the flag setting process of whether to perform measurement at each time point is 1
The processing is such that the measurement start flag is set in the measurement target setting shared memory 61 when 0 minutes has elapsed.

When process 1 is first started, control is passed to the modified dynamic linker 10 for measurement. The dynamic linker 10 determines the register 34 at the beginning of process 1. When the register 34 is set so as not to perform the measurement, the normal processing (when the measurement is not performed) is performed thereafter. This normal process is exactly the same as the conventional dynamic linker. Further, when the measurement is set to perform measurement, the dynamic linker 10 maps the information collection library 20, and when there is a measurement instruction regarding the profile, calls the initial processing unit 21.

The initial processing section 21 first calls the measurement object determination section 11 to determine whether or not the own process 1 is a measurement object. In the case of a process that is not the measurement target, control is returned to the dynamic linker 10 without doing anything, and the process of the normal process (process when the measurement is not performed) is performed thereafter.

If it is a measurement target, then the execution information collecting unit 22 is set as an exit function of the signal, and the timer is set so that the signal is periodically generated. Control then returns to the dynamic linker 10.

A signal specified by the initial processing unit 21 is issued to the process at regular intervals, and control is transferred to the execution information collecting unit 22. Here, first, the measurement target determination unit 11 determines whether or not to perform the measurement at that time, and if the measurement is performed, the execution information collection unit 22
Return the address of the instruction text part of the main load module or dynamic library and increment the counter that indicates how many times the address is called.

On the other hand, when the dynamic library function is called in the instruction text portion 31 of the process 1, the control is passed to the corresponding entry of the link table (PLT) 32. If the call to the dynamic library function is the first time, the dynamic linker 10 is operated by the control transfer method similar to the conventional one.
Control is transferred to. The dynamic linker 10 performs the following processing by the PLT entry rewriting unit 12.

When the information collecting library 20 is mapped by the trace instruction, first, the measuring object judging section 1
1, the measurement target determining unit 11 refers to the measurement target setting shared memory 61, and determines whether a process, a library, or a function is a measurement target. If it is not the measurement target, the PLT entry rewriting unit 12 rewrites the normal PLT 32.

In the case of the measurement object, the PLT entry rewriting unit 12 sets the entry of the PLT 32 to the function execution trace unit so that the control is directly passed to the function execution trace unit 23 when the dynamic library function is called for the second time and thereafter. Rewrite so that the address becomes 23. afterwards,
The function execution trace unit 23 is called, the execution information of the function is recorded in the collection information recording shared memory 62, and control is transferred to the original dynamic library function. In the processing of the function execution trace unit 23, the element number of the output destination of the record is obtained from the management information of the collected information recording shared memory 62, the element of the array is advanced by one, and the element number of the management information is updated. Execution information such as the called function name, the called process name, the process ID, and the time information is stored as log information in the array of element numbers acquired in step 1.

If an error does not occur in the execution information collection unit 22 or the function execution trace unit 23 and the processing of the process is completed, the termination processing is finally performed by the dynamic linker 10. Here, the collection information output unit 24 is used. , Counter information corresponding to the return address from the execution information collection unit 22 to the instruction text portion, address information regarding the execution module or dynamic library for which information was collected (the start address and the final address of each text portion), and The function execution information collected by the function execution trace unit 23 is output to the collection information file 70.

When an error occurs during the process of collecting information in the execution information collection unit 22 or the function execution trace unit 23, the error processing unit 25 is called to perform necessary error processing, and then the collected information is output. The count information corresponding to each address collected up to that time by calling the section 24, the address information corresponding to the file of the execution module (main load module and dynamic library) (start address and end address of each text part), and , Outputs trace information related to function execution.

Further, in the processing for the measurement end instruction,
Set it so that it does not generate the signal that was being generated at regular intervals, and then perform normal process operations.

FIG. 3 is a diagram for explaining a method of collecting execution information by the execution information collecting unit and its output. If the profile information measurement is set, the dynamic linker 10
Secures an information collection area 100 as shown in FIG. The information collection area 100 has a count area (initial value is all 0) for each address corresponding to the address of the instruction text part of the main load module 30 and the dynamically linked dynamic library 33.

The execution information collecting unit 22 is called by a signal from a timer at regular time intervals. The execution information collecting unit 22 knows which address of the main load module 30 or the dynamic library 33 is to be executed next from the value of the program counter of the restoration destination, and the count area of the information collecting area 100 corresponding to the address. Count up C by 1. This process is repeated each time it is called.

The collected information output unit 24 has information of the information collecting area 100, address information of the main load module (start address and end address of text portion) 101, address information of each dynamic library (start address of text portion and End addresses) 102, 103, ... Are stored in the collection information file 70.

When displaying the collected information, the name of the execution module (name of the main load module) from which the information was collected is specified by a command or the like, and the information display processing unit 80 is started. The information display processing unit 80 acquires a list of the dynamic libraries 33 required for execution in the designated execution module. Similarly, for each dynamic library 33 obtained there, a list of dynamic libraries required by the dynamic library 33 at the time of execution is acquired. In the same way, obtain all the dynamic library names required by the process to be measured at the time of execution. Next, the symbol table 71 of the main load module and the symbol tables 72a, 7a of all the dynamic libraries thus obtained.
2b, ... Is acquired.

Next, the information display processing unit 80 reads the information output by the collected information output unit 24 from the collected information file 70. By using the address information (the start address and the end address of each text part) of the function written in the information read from the collection information file 70, from the information collected by the execution information collecting part 22 to where is the main load module 30. It is checked whether it is information or from where to where information of each dynamic library 33. Then, for each (main load module, each dynamic library), the start address of the function of the symbol table is used to check which function information is from where to where,
The execution time is calculated by calculating the information for each function. Finally, the obtained result is output to the output device 52 or the storage device 53.
Output to.

As a result, as shown in FIG. 3, information such as the execution ratio and time for each function can be obtained. That is, the execution time of each function is known by multiplying the count information for each address and the fixed time for collecting the information, and the execution ratio of each function is obtained by dividing the execution time of each function by their total. Although not shown in FIG. 3, by using the execution count information for each function obtained by the function execution trace unit 23, it is possible to obtain information about the execution time per function of each function. it can.

FIGS. 4 and 5 are diagrams for explaining an operation example of the dynamic linker according to the embodiment of the present invention. FIG. 4 shows an operation example at a normal time (when no measurement is performed), and FIG. An operation example of (when measuring) is shown.

First, the call of the dynamic library function at the normal time will be described with reference to FIGS. The operation in the normal case is exactly the same as that of the conventional dynamic linker.

When the dynamic library function f is called from the instruction text portion 31 of the main load module for the first time, first, the corresponding entry of the PLT 32 is jumped to. The entry of the PLT 32 is set so as to call the function of the dynamic linker 10 first.
Control is transferred to.

The dynamic linker 10 rewrites the corresponding entry of the PLT 32 so as to jump to the address of the dynamic library function f, and then calls the function f. ~ After that (from the second time onward), the instruction text part 31
When the dynamic library function f is called from PLT3
The corresponding entry of 2'is called and the control is directly passed from the corresponding entry of PLT 32 'to the function f of the dynamic library 33.

On the other hand, the operation during the function execution trace is as shown in FIGS. 5 of FIG.
The operation when the function is called for the second time, and to indicate the operation when the function is called for the second time and thereafter.

When the dynamic library function f is called from the instruction text portion 31 of the main load module for the first time, first, the corresponding entry of the PLT 32 is jumped to. The entry of the PLT 32 is set so as to call the function of the dynamic linker 10 first.
Control is transferred to.

The dynamic linker 10 rewrites the corresponding entry of the PLT 32 so as to call the function execution trace unit (trace function) 23 instead of the dynamic library function f. After that, the function execution trace unit (trace function) 23 is called.

The tracing function collects and records information about the dynamic library function f, and then jumps to the address of the dynamic library function f. After that (after the second time), when the dynamic library function f is called from the instruction text part 31, the corresponding entry of PLT32 ′ is called, and from the corresponding entry of PLT32 ′,
The function execution trace unit (trace function) 23 is called. After information is collected and recorded by the function execution trace unit (trace function) 23, control is passed to the target function f of the dynamic library 33.

Although not shown, in the dynamic linker 10, in order to transfer control from the trace function to the target dynamic library function, the purpose of each entry is rewritten when the PLT entry is rewritten. A process of saving the address of the dynamic library function to be performed in a predetermined work area is performed.

FIG. 6 is a processing flowchart of the measurement monitoring processing section. In this embodiment, commands for instructing start / preparation / end of the trace, profile, and both are prepared for measurement. When these commands are input, the measurement monitoring processing unit 40 is activated.
First, the measurement monitoring processing unit 40 acquires the measurement target setting shared memory 61 in the processing of step S11. Next, in step S12, the measurement instruction flag of the measurement target setting shared memory 61 is set according to the instruction information. The measurement instruction flag includes, for example, a trace / profile / trace and a flag indicating whether or not to perform a profile. In step S13, measurement / non-measurement is set for the in-kernel flag 46.

FIG. 7 is a processing flowchart of the kernel. When the process command is started, the kernel 45
First, in step S21, the in-kernel flag 46
Is ON (measured) or OFF (non-measured). OF
In the case of F, normal process activation is performed in step S23. If the in-kernel flag 46 is ON, in step S22, information indicating that the in-kernel flag 46 is ON is set in a predetermined register 34, and then step S23 is executed.

FIG. 8 is a processing flowchart in the process including the dynamic linker. When the process is started, control is first passed to the dynamic linker 10. In the dynamic linker 10,
First, in step S31, ON (measurement) / OFF (non-measurement) is determined by referring to a predetermined register 34. If the register is ON, the process proceeds to step S33
If it is FF, the process proceeds to step S32, and the normal dynamic linker process similar to the conventional process is performed.

In step S33, first, the information collecting library 20 for collecting the function execution information is mapped,
Make it possible to call those processing units. Next, in step S34, initialization processing of the dynamic linker such as securing a work area required for measurement is performed. Step S
At 35, the measurement instruction flag in the measurement target setting shared memory 61 is referred to, and it is determined whether or not profiling is performed. If profiling is performed, the process of step S36 is performed. If profiling is not performed, step S36 is performed.
Proceed to step 37.

In step S36, the initialization processing unit 21 is called, and as the initialization processing for periodically activating the execution information collecting unit 22, the activation of a timer is set. In the process of step S37, control is transferred to the instruction text part 31 of the main load module 30. As a result, the user program is executed (step S3).
8). When the first call of the dynamic library function is made in the instruction text portion 31 (step S39), the dynamic linker 10 is passed via the PLT 32 in step S40.
Is called, and control is transferred to the dynamic linker 10 in step S41.

The dynamic linker 10 refers to the measurement instruction flag in the process of step S42 and determines whether the trace preparation / trace start / not trace is performed. If the trace preparation or the trace start is performed, the process of step S43 is performed. If the trace is not performed, the normal PLT 32 entry is rewritten by the process of step S45.

In step S43, the corresponding entry of PLT 32 is rewritten for log collection (address of function execution trace section 23). Further, when the trace is started, the function execution trace unit 23 is called and the function execution information is collected in step S42. After that, step S
Returning to 37, control is transferred to the instruction text part of the called dynamic library function.

FIG. 9 is a processing flowchart of the execution information collecting unit. The execution information collection unit 22 is called when a timer interrupt is generated by the setting of the initial processing unit 21. The execution information collecting unit 22 firstly executes step S5.
By referring to 1, the measurement instruction flag is referred to and it is determined whether profile preparation / profile start / no profile. If the profile is ready, the process returns to the caller at the time of the timer interrupt without doing anything in step S54.
If the profile is started, the execution information regarding the program counter is sampled by the process of step S52, and the process returns to the calling source by step S54. If the profile is not set (profile end), the activation of the execution information collecting unit 22 by the timer is stopped in step S53, and the process is restored in step S54.

FIG. 10 is a processing flow chart of the function execution tracing section. When the function execution trace unit 23 is called once when the function whose PLT 32 has been rewritten for tracing has already been called, the measurement instruction flag is referred to in the processing of step S61 to determine whether or not to start tracing. When starting the trace, the process of step S62 is performed, and the function execution information is collected (log collection).
I do. Then, in step S63, control is transferred to the instruction text portion of the dynamic library function. When the measurement instruction flag indicates that the trace is not being started or the trace is not started at the end of the trace, the control is transferred to the instruction text portion of the dynamic library function without performing the process of step S62.

[Second Embodiment] In the first embodiment, the collection information output unit 24 is called only when the process ends and when an error occurs in the execution information collection unit 22. When the sampling unit 22 calls the measurement target determination unit 11 and determines whether or not to perform measurement at that time, when the end instruction of the measurement that has started once is detected, the sampling information output unit 24 is called and until then. It is also possible to process such as outputting the information of.

[Third Embodiment] For writing to the collection information file 70 shown in FIG. 2, the shared memory information output unit 42 of the measurement monitoring processing unit 40 monitors the log amount, not the dynamic linker 10. Alternatively, or by notification from the process to be measured.

[0077]

As described above, according to the present invention,
Execution information can be collected in the normal load module state without re-translation / recombination of the measurement target, and execution information of the dynamic library can also be obtained. Also, it is possible to collect only the information about the requested process and the requested part. Furthermore, the information of a plurality of processes can be collected at the same time, and by collecting the information, it is possible to obtain the information of the entire process to be measured without merging work. In addition, the dynamic linker
When the measurement is not performed, the same process as the conventional process is executed with the overhead of only determining the flag notified from the kernel, so that the performance deterioration can be made extremely small.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration example of the present invention.

FIG. 2 is a configuration diagram of an embodiment of the present invention.

FIG. 3 is a diagram illustrating a method of collecting execution information by an execution information collection unit and its output.

FIG. 4 is a diagram illustrating an operation example of a dynamic linker at a normal time according to the embodiment of this invention.

FIG. 5 is a diagram illustrating an operation example of a dynamic linker at the time of tracing according to the exemplary embodiment of the present invention.

FIG. 6 is a processing flowchart of a measurement monitoring processing unit.

FIG. 7 is a processing flowchart of a kernel.

FIG. 8 is a process flowchart in a process including a dynamic linker.

FIG. 9 is a processing flowchart of an execution information collection unit.

FIG. 10 is a processing flowchart of a function execution trace unit.

FIG. 11 is an explanatory diagram of a conventional technique.

[Explanation of symbols]

 1 Process 10 Dynamic Linker 11 Measurement Target Judgment Part 12 PLT Entry Rewriting Part 20 Information Collection Library 21 Initial Processing Part 22 Execution Information Collection Part 23 Function Execution Trace Part 24 Collection Information Output Part 25 Error Processing Part 30 Main Load Module 31 Instruction Text part 32 Link table (PLT) 33a, 33b Dynamic library 40 Measurement monitoring processing part 41 Flag setting part 42 Shared memory information output part 45 Kernel 46 Kernel flag 51 Input device 52 Output device 53 Storage device 60 Shared memory 63 process Internal memory

Claims (10)

[Claims] 1. A process execution information collecting device in an information processing system comprising a main load module, a dynamic library, and a dynamic linker that performs a process of combining the dynamic libraries at the time of process execution, and collects a value of a program counter. Then, the execution information collecting unit that counts the number of times the address indicated by the program counter is collected according to the value of the program counter, and the setting that is invoked by the dynamic linker and starts the execution information collecting unit every predetermined time An initial processing unit for performing the above, and the collected information output for outputting the information collected by the execution information collecting unit and the address information of the instruction text part of the main load module and the dynamic library related to the measurement target to a predetermined memory, storage device or output device. And a process execution information collecting device. 2. The process execution information collecting device according to claim 1, further comprising a measurement monitoring processing unit that sets, in a kernel, instruction information as to whether or not to collect the program counter information by an instruction input from the outside, The dynamic linker has a means for deciding whether to perform only a normal dynamic binding process or to call the initial processing unit according to the instruction information obtained from the kernel, and the process execution information collecting device. 3. The process execution information collecting device according to claim 2, wherein the initial processing unit, the execution information collecting unit, or both are configured as modules for collecting information, which are different from the dynamic library. The process execution information collecting device, wherein the dynamic linker has means for mapping the information collecting library so that control transfer can be performed according to the instruction information set by the measurement monitoring processing unit. 4. The process execution information collecting device according to claim 2 or 3, wherein the measurement monitoring processing unit shares information on a time when the program counter information is collected by an instruction input from the outside or a process to be measured. A measuring object having means for setting in the memory, and determining whether or not to perform the processing by the initial processing section or the processing by the execution information collecting section based on the measurement time or the information of the measuring object set in the shared memory An apparatus for collecting process execution information, comprising a determination unit. 5. A process execution information collecting device in an information processing system comprising a main load module, a dynamic library, and a dynamic linker that performs a process of linking the dynamic libraries during process execution. A function execution trace unit that is called at the time of control transfer and collects the execution information of the dynamic library function, and the dynamic library function at the time of the first control transfer to the dynamic library function to be measured in the dynamic linker. The PLT entry rewriting unit that rewrites the corresponding entry in the link table in which the address information is set to the function execution trace unit, and the information collected by the function execution trace unit that sets the address information so that control is transferred to the function execution trace unit. It is provided with a collection information output unit for outputting to a predetermined memory, storage device or output device. Characterized process execution information collection device. 6. The process execution information collecting device according to claim 5, further comprising a measurement monitoring processing unit for setting, in the kernel, instruction information as to whether or not to collect dynamic library function execution information by an instruction input from the outside. The dynamic linker has means for determining whether to perform a normal dynamic binding process or a dynamic binding process via the function execution trace unit based on the instruction information obtained from the kernel. A device for collecting process execution information. 7. The process execution information collecting device according to claim 6, wherein the function execution trace unit is configured as a module different from the dynamic library as an information collecting library, and the dynamic linker is configured to perform the measurement. A process execution information collecting device having means for mapping the information collecting library so that control transfer is possible according to the instruction information set by the monitoring processing unit. 8. The process execution information collecting device according to claim 6 or 7, wherein the measurement monitoring processing unit is a process to be measured by a command input from the outside, a dynamic library, or a function in the dynamic library. And a measurement target determining unit that selects an entry rewriting target of the link table by the PLT entry rewriting unit based on the measurement target information set in the shared memory. A process execution information collection device characterized by the above. 9. The process execution information collection device according to claim 1, further comprising an error processing unit that is called when an error occurs during information collection by the execution information collection unit or the function execution trace unit. Then
The process execution information collecting device, wherein the error processing unit calls the collected information output unit so that the collected information output unit outputs information collected until an error occurs. 10. The process execution information collection device according to claim 1, wherein the execution information collection unit or the function execution trace unit is configured to record the collected measurement information in a shared memory, and perform measurement. A shared memory information output unit that outputs the measurement information to a predetermined storage device or output device according to the accumulation state of the measurement information in the shared memory is provided in a process different from the target process. Process execution information collection device.