EP1810149A1

EP1810149A1 - Method and device for mode switching and signal comparison in a computer system comprising at least two processing units

Info

Publication number: EP1810149A1
Application number: EP05801271A
Authority: EP
Inventors: Reinhard Weiberle; Bernd Mueller; Yorck Collani; Rainer Gmehlich; Eberhard Boehl
Original assignee: Robert Bosch GmbH
Current assignee: Robert Bosch GmbH
Priority date: 2004-10-25
Filing date: 2005-10-25
Publication date: 2007-07-25
Also published as: WO2006045785A1; JP2008518304A; KR20070083732A; CN101048752A; CN101048755A; US20080320340A1

Abstract

The invention relates to a method and a device for mode switching and data comparison in a computer system comprising at least two processing units, switching means for switching between at least two operating modes, and comparison means, a first operating mode corresponding to a comparison mode and a second operating mode corresponding to a performance mode. The invention is characterised in that a vote is carried out in the comparison mode, at least as a 2 out of 3 evaluation, control means being provided for the execution of said vote.

Description

METHOD AND DEVICE FOR MODE CHANGING AND SIGNAL COMPARISON IN A COMPUTER SYSTEM WITH AT LEAST TWO PROCESSING UNITS

State of the art

A method for detecting errors in a comparison mode is described in Wo 01/46806 A1. The data is processed and compared in parallel in a processing unit with two processing units ALUs. In the event of a fault (soft error, transient error), both ALUs work there independently of each other until the faulty data have been removed and a repeated (partially repeated) redundant processing can be carried out. This presupposes that both ALUs work synchronously with each other and that the results can be compared in exact time.

In the prior art methods are known, such as between a comparison mode for error detection, in which tasks are processed redundantly, and a performance mode to achieve a higher performance can be switched.

The prerequisite here is that the processing units for the comparison mode are mutually synchronized. For this purpose, it is necessary that both processing units can be stopped and work in sync with clock timing in order to be able to compare the result data when writing to the memory. This requires intervention in the hardware, individual solutions are proposed.

In the patent EP 0969373 A2, on the other hand, a comparison of the results of redundantly operating processing units or processing units is also then guaranteed, even if they work asynchronously to each other, ie not equal in time or with an unknown clock offset.

Voting systems are known from the aircraft industry, which can use inputs from standard computers and process them safely by a majority decision and thus trigger safety-relevant actions. A system that combines inter-processing unit and inter-control-unit communication is the FME system, which maintains the system still operational by a high degree of redundancy even in the case of single or even multiple faults and by the DASA for Space has been developed (Urban, et al: A survivable avionics System for Space applications, Int. Symposium on Fault-tolerant

Computing, FTCS-28 (1998), pp.372-381). This system can even tolerate Byzantine errors (ie particularly malicious errors in which not all components receive the same information but are even "intentionally" distributing differently wrong information to different components by an intriguer.) Such a system is commercial because of the high cost It is not known to produce a cost-effective solution which can be produced in large numbers and which also has the possibility to switch over, and the task is therefore to create a switching and comparison unit that can do it It is possible to switch the operating mode of two or more processing units, without any intervention in the structure of these processing units, nor does it require any additional signals for this purpose, it being possible to use different digital or analogue signals from ver different processing units in a comparison mode to compare each other. Under certain circumstances, this comparison should also be possible if the processing units are operated with different clock signals and do not work synchronously with one another. In particular, it is the object to provide means and methods that enable general voting using more than two execution units.

advantages

Advantageously, a method for switching and data comparison is used in a computer system having at least three processing units, wherein switching means are provided and switched between at least two operating modes, wherein comparison means are provided and a first operating mode a comparison mode and a second mode of operation corresponds to a performance mode, characterized in that in the comparison mode a voting, at least as a 2 out of 3 rating, is provided, wherein control means are provided by which the voting is adjustable.

Advantageously, a method is used in which the control means at least

Memory means contain or are designed as such and in the memory means an identifier for setting the Votings, in particular a bit, is stored.

Advantageously, a method is used in which an error detection and an error localization is carried out by the voting and corresponding error states are stored.

Advantageously, a method is used in which the error status is stored next to the identifier in the at least one memory means.

Advantageously, a method is used in which the identifier is written by at least one processing unit and the error status by the comparison unit and / or the switching unit in the storage means.

Advantageously, a method is used in which the data is input to the

Comparative means are buffered.

Advantageously, a method is used in which a computer system internal source, in particular a processing unit specifies the identifier for setting the voting.

Advantageously, a method is used in which an external source to the computer system specifies the identifier for setting the voting.

Advantageously, a method is used in which a plurality of sources for specifying the

Identification are provided and all sources are connected such that the setting of the voting is configured by an OR operation. - A -

Advantageously, a device for switching and for data comparison is used in a computer system having at least three processing units, wherein switching means are provided and switched between at least two operating modes, wherein comparing means are provided and a first operating mode corresponds to a comparison mode and a second operating mode to a performance mode, characterized in that in the comparison mode a voting, at least as a 2 out of 3 rating, is provided, wherein control means are contained by which the voting is settable.

Advantageously, a device is used in which the control means at least

Advantageously, a device is used in which the storage means is at least one control register.

Advantageously, use is made of a device in which at least one input buffer memory is provided, which is designed such that the data is buffered before entry into the comparison means.

Advantageously, use is made of a device in which the control means are provided as a source for setting the voting external to the computer system.

Advantageously, a device is used in which the control means are provided as a source for setting the voting internally to the computer system.

Advantageously, use is made of a device in which all the sources for setting the voting are linked in such a way that the setting of the voting is configured by an OR operation.

Further advantages and advantageous embodiments will become apparent from the features of the claims and the description. characters

FIG. 1 shows the basic function of a switching and comparison unit for two processing units

FIG. 1a shows a generalized representation of a comparator

FIG. 1c shows an expanded representation of a comparator

Figure Ib shows a generalized representation of a switching and comparison unit

Figure 2 shows a more detailed representation of the switching and comparison unit for two processing units

FIG. 3 shows a possible realization of a switching and comparison unit for two

processing units

FIG. 4 shows a more detailed representation of a switching and comparison unit for more than two processing units

FIG. 5 shows a possible implementation of a switching and comparison unit for more than two processing units

FIG. 6 shows a possible realization of a control register

FIG. 7 shows a voting unit for central voting

FIG. 8 shows a voting unit for decentralized voting

FIG. 9 shows a synchronization element

Figure 10 shows a handshake interface

FIG. 11 shows a differential amplifier FIG. 12 shows a comparator for positive voltage difference

FIG. 13 shows a comparator for negative voltage difference

Figure 14 shows a circuit for storing an error

Figure 15 shows an analog-to-digital converter with output registers

FIG. 16 shows the representation of a digitally converted analog value with kung and

Analogbit

FIG. 17 shows the representation of a digital value as a digital word with digital bit

Description of the embodiments

An execution unit or processing unit may refer in the following to a processor / core / CPU as well as an FPU (floating point unit), DSP (digital signal processor), coprocessor or ALU (Arithmetic Logical Unit).

A system of two or more processing units is considered. Basically, in safety-relevant systems, it is possible to use such resources either to increase performance, by providing the various processing units as possible with different tasks. Alternatively, you can also use some of the resources redundantly to each other, by using the same

Task supplies and detects unequal results on an error. Depending on how many processing units there are, several modes are conceivable. In a two-tier system, the two modes "comparison" and "performance" exist as described above. In a threefold system, in addition to the pure performance mode, in which all three processing units work in parallel, and the pure comparison mode, in which all three processing units are redundantly calculated and compared, one can also implement a 2out3 voting mode, in which all three Processing units redundant computing and a majority selection is made. Next you can also realize a mixed mode in which, for example, two of the processing units redundant to each other and calculate the Results are compared while the third processing unit is processing a different, parallel task. In a four or more processing unit system obviously further combinations are conceivable.

The object to be solved is that the available processing units in a system can be used variably in operation without necessitating an intervention in the existing structure of these processing units (for example for synchronization purposes). In a particular embodiment, each processing unit should be able to operate with its own clock, i. The execution of identical tasks for the purpose of comparison can also work asynchronously to each other.

This object is achieved in that a universal, widely deployable IP is created, which allows switching of the operating modes (eg comparison, performance or voting mode) at arbitrary times without previously switching off the processing units and possibly the comparison or the voting of each other manages asynchronous data streams. This IP may be implemented as a chip, or it may be integrated with one or more processing units on a chip. Further, it is not a prerequisite that this chip consists of only one piece of silicon, it is also quite possible that this is realized from separate components.

To ensure the synchronicity between different processing units are

Signals are necessary to prevent the continuous processing of individual processing units. For this purpose, a WAIT signal is usually provided. If an execution unit does not have a wait signal, it can also be synchronized via an interrupt. For this purpose, the synchronization signal (for example, M 140 in FIG. 2) is not routed to a wait input, but is set to an interrupt. This interrupt must have a sufficiently high priority over the processing program and also against other interrupts to interrupt normal operation. The associated interrupt routine only executes a certain number of NOPs (empty commands with no effect on data) before jumping back into the interrupted program, thereby delaying further processing of the processing program. If necessary, the usual memory operations at the beginning and at the end must be carried out in the interrupt routine in order not to impair the normal program execution by the interrupt. This process continues until the synchronicity has been established (eg, other processing units will provide the expected comparison data). However, precise clock synchronization and in particular phase equality with other processing units can only be guaranteed to a limited extent with this method. It is therefore recommended that when using the interrupt signal for synchronization, the data to be compared is temporarily stored in the UVE before being compared.

The advantage of the invention is that any commercially available standard structures can be used, because no additional signals are needed (no interference with the hardware structure) and any output signals of these components can be monitored, which are for example used directly to drive actuators. This includes the verification of transducer structures such as DACs and PWMs, which have not been directly testable by comparison so far in the prior art. If the check is not required for individual tasks or software tasks, it is also possible to switch to a performance mode in which different tasks are distributed to different processing units.

Another advantage is that not all data has to be compared in a comparison or voting mode. Only the data to be compared or voted are synchronized with each other in the switching and comparison unit. The selection of these data is variable (programmable) by the targeted response of the switching and comparison unit and can be adapted to the respective processing unit architecture as well as to the application. Thus, the use of diverse μC or software parts is easily possible, since only results that can reasonably be compared, actually compared. Further, any access to a (e.g., external) memory can be monitored, or even just driving external I / O modules. Internal signals can be checked via the software-controlled additional output to the switching module on the external data and / or address bus.

All control signals for the comparison operations are generated in the preferably programmable switching and voting unit and the comparison also takes place there. The processing units (eg, processors) whose outputs are to be compared with each other may use the same program, a duplicate program (which additionally allows recognition of memory access errors), or a diversified program for detecting software errors. Not all of It is also possible to provide certain signals for comparison by means of an identifier (address or control signals) or not. This identifier is evaluated in the switching and comparison device and thus controlled the comparison. Separate timers monitor deviations in the time response beyond a specifiable limit.

Some or even all modules of the switching and comparison unit can be integrated on a chip, be housed on a common board or spatially separated. In the latter case, the data and control signals are exchanged with each other via suitable bus systems. On-site registers are then described via the bus system and control the operations by means of the data and / or addresses / control signals stored therein.

FIG. 1 shows the basic function of the switching unit BO1 according to the invention for the

Application shown in conjunction with two processing units BIO and BI l.

Various output signals, such as data, control and address signals B20 and B21 of the processing units BIO and BI l are connected to the switching unit BO 1.

In addition, there is at least one synchronization signal, in an embodiment of the arrangement according to the invention, the two output signals B40 and B41, with one of the

Comparative units are connected.

The switching unit includes at least one control register B 15 having at least one binary-character (bit) memory element B16 that switches the mode of the comparison unit. B16 can take at least the two values 0 and 1 and can be detected both by the signals B20 or B21 of the processing units or by internal processes of the

Switching unit set or reset.

If B 16 is set to the first value, the changeover unit operates in comparison mode. In this mode, all incoming data signals from B20 are compared with the data signals from B21, provided certain predeterminable comparison conditions of the control and / or

Address signals from the signals B20 and B21 are met, which signal the validity of the data and the intended comparison for this data.

If these comparison conditions are satisfied at both signals B20 and B21 at the same time, the data from these signals are compared directly and in case of inequality becomes

Error signal B17 is set. If only the comparison condition from either the signals B20 or B21 is fulfilled, then the corresponding synchronization signal B40 or B41 is set.

This signal causes in the corresponding processing unit BIO or BI l stopping the processing, and thus preventing the forwarding of the corresponding Signals that could not previously be compared. The signal B40 or B41 remains set until the corresponding comparison condition of the respective other processing unit B21 or B20 is fulfilled. In this case, the comparison is carried out and the corresponding synchronization signal is reset. To compare the comparison in the described non-simultaneous provision of the

To ensure data by the two processing units, it is either necessary that the data and comparison conditions of the corresponding processing unit are kept at the corresponding values until the corresponding synchronization signal B40 or B41 has been reset, or the data provided first in the Switching unit to be stored for comparison.

Depending on which processing unit first provides data, it must wait until the other processing unit provides the corresponding comparison data with the further processing of its program or its processes. In a special embodiment of the switching unit according to FIG. 1, it is possible to dispense with one of the signals B40 or B41, if it is always ensured that the associated one

Processing unit does not provide comparative data rather than the other processing unit.

If B16 is set to the second value, then the synchronization signals B20 and B21 and the error signal B 17 are always inactive, for example set to the value 0. There is no comparison and both processing units work independently.

An essential component in the system according to the invention is the comparator. In the simplest form it is shown in FIG. 1a. The comparison component M500 can receive two input signals M510 and M511. It then compares these to equality, in the context presented here, preferably in the sense of a bit-wise equality. If it detects inequality, the error signal M530 is activated and the signal M520 is deactivated. In the same case, the value of the input signals M510, M511 is given to the output signal M520 and the error signal M530 is not active, ie it signals the 'Guf' state. From this basic system, a variety of advanced embodiments are conceivable. First, the component M500 can be executed as a so-called TSC component (totally seif checking). In this case, the error signal M530 is routed to at least two lines ("dual rail") to the outside, and it is ensured by internal design and error detection measures that in any possible error case of the comparison component this signal is correct or recognizable incorrect preferred embodiment in the use of the system according to the invention is to use such a TSC comparator.

A second class of embodiments can be distinguished as to what degree of synchronicity the two inputs M510, M511 (or M610, M611) must have. One possible variant is characterized by intermittent synchronicity, i. the comparison of the data can be done in one cycle. A slight change arises from the fact that a synchronous delay element is used with a fixed phase offset between the inputs, which delays the corresponding signals, for example, by integer or half clock periods. Such a phase offset is useful to avoid common cause errors, i. these are errors that can affect multiple processing units simultaneously. In FIG. 1c, component M640, which delays the previous input by the phase offset, is therefore inserted beyond the components in FIG. Preferably, this delay element is accommodated in the comparator to use this element only in the comparison mode. Alternatively or additionally, in order to be able to tolerate asynchronisms as well, intermediate buffers can be placed in the input chain.

Preferably, these are designed as FIFO memory. If such a buffer exists, one can also tolerate asynchronisms up to the maximum depth of the buffer. In this case, an error signal must be output even if the buffer overflows. Further, in the comparator embodiments, it can be distinguished according to how the signal M520 (or M620) is generated. A preferred embodiment is the input signals

M510, M511 (or M610, M611) to the output and make the connection interruptible by switch. The particular advantage of this variant is that the same switches can be used to switch between the performance mode and possible different comparison modes. Alternatively, the signals can also be generated from internal comparator buffers.

A final class of embodiments may be distinguished as to how many inputs are present on the comparator and how the comparator should react. With three inputs, a majority voting, a comparison of all three or a comparison of only two signals can be made. With four or more inputs, correspondingly more variants are conceivable. These variants are preferably to be coupled with the various operating modes of the overall system.

In order to explain the general case, a generalized representation of a switching and comparison unit is shown in FIG. 1b, as it is to be preferably used. From For the execution units to be considered, n signals N140,..., N14n go to the switching and comparison component N100. This can generate up to n output signals N160, ..., N16n from these input signals. In the simplest case, the "pure performance mode", all signals N14i are directed to the corresponding output signals N16i. In the opposite limit case, the "pure comparison mode", all signals are transmitted

N 140, ..., N14n only to exactly one of the output signals N16i passed.

This figure shows how the various conceivable modes can arise. For this purpose, the logical component of a switching logic Nl 10 is included in this figure. The component does not have to exist as such, it is crucial that its function is present. It first determines how many output signals there are. Furthermore, the switching logic NI lO determines which of the input signals contribute to which of the output signals. An input signal can contribute to exactly one output signal. Formulated in a different mathematical form, a function is defined by the switching logic which assigns an element of the set {N 160, ..., N 16n} to each element of the set {N 140, ..., N 14n}.

The function of the processing logic N120 then determines to which of the outputs N16i the form in which the inputs contribute to this output signal. Also, this component does not have to exist as a separate component. It is again crucial that the functions described are implemented in the system. To exemplify the different

To describe variation possibilities, it is assumed without restriction of generality that the output N 160 is generated by the signals N141, ..., N 14m. If m = 1, this simply means that the signal is switched through, if m = 2 then the signals N141, N 142 are compared. This comparison can be performed synchronously or asynchronously, it can be performed bitwise or only on significant bits or even with a tolerance band.

If m> = 3, there are several possibilities.

A first possibility is to compare all signals and to detect an error in the presence of at least two different values, which can be optionally signaled.

A second possibility is to make a k out of m selection (k> m / 2). This can be realized by using comparators. Optionally, an error signal can be generated if one of the signals is detected as deviating. One possibly different error signal can be generated if all three signals are different.

A third option is to apply these values to an algorithm. This may be, for example, the formation of an average, a median, or the use of a Fault Tolerant Algorithm (FTA). Such an FTA is based on extreme values of the

Delete input values and make a kind of averaging over the remaining values. This averaging can be done over the entire set of residual values, or preferably over a subset that is easy to form in HW. In this case, it is not always necessary to actually compare the values. For averaging, for example, you just have to add and divide FTM, FTA or

Median require a partial sorting. If necessary, an error signal can optionally also be output at sufficiently large extreme values

These various possibilities of processing a plurality of signals into one signal are referred to as comparison operations for the sake of brevity. The task of the processing logic is thus to determine the exact shape of the comparison operation for each output signal - and thus also for the associated input signals. The combination of the information of the switching logic NI 10 (ie the above mentioned function) and the processing logic (ie the determination of the comparison operation per output signal, ie per function value) is the mode information and sets the mode. Of course, in the general case this information is multivalued, ie not representable only via a logical bit. Not all the theoretically conceivable modes are useful in a given implementation, it is preferable to restrict the number of modes allowed. It should be emphasized that in the case of only two execution units, where there is only one compare mode, all the information can be condensed to only one logical bit. Switching from a performance mode to a comparison mode is characterized in the general case by the fact that execution units that are displayed in the performance mode on different outputs are mapped in the compare mode to the same output. This is preferably realized in that there is a subsystem of execution units in which in the performance mode all input signals N14i to be considered in the subsystem are switched directly to corresponding output signals N16i, while in the comparison mode they are all mapped to one output. Alternatively, such switching can also be realized by changing pairings. It is represented by the fact that in the general case one can not speak of the performance mode and the comparison mode, although in a given Embodiment of the invention can limit the amount of allowed modes so that this is the case. However, one can always speak of switching from the performance to the comparison mode (and vice versa).

Between these modes can be controlled by software, dynamically switched during operation. The switching is triggered, for example, by the execution of special switching instructions, special instruction sequences, explicitly marked instructions or by the access to specific addresses by at least one of the execution units of the multiprocessor system.

FIG. 2 shows a detailed two-processor or two μC system with a switching and comparison unit M100 according to the invention, in which optionally also different signals can be dispensed with. It consists of two processing units (Ml 10, Ml I l) and a switching and comparison unit M100. From each processing unit, data signals (M120, M121) and address / control signals (M130, M131) go to the switching unit, and each processing unit optionally also gets from the

Switching unit data (M150, M151) and control signals (M140, M141) back. The unit M100 outputs data (M160, M161) and status information M169 and receives signals such as data (M170, M171) and control signals M179, which can also be forwarded to the processing units. Via M 170, M 171 and M 179, optionally, the operating mode of the unit M 100 can also be set independently of the processing units; Likewise, the processors can set the operating mode via the outputs M120, M121 (eg data bus) and the control and address signals M130, M131 (eg Write) in the unit M100 - eg performance mode (without comparison) or comparison mode (with comparison of the signals M 120 , M 121 and / or the signals M 170, Ml 71, coming from eg peripheral units). In the performance mode, the outputs M120, M121 are forwarded to the outputs M 160, M161 if necessary in connection with control signals, and conversely the inputs M170, M171 to M150, M151. In comparison mode, the outputs are compared and advantageously forwarded to M 160, M161 only in the error-free case, where either both outputs are used, or only one of them. Similarly, a check of input data M 170, M171 is possible, which are forwarded to the processing units. In the case of a faulty comparison of the signals in comparison mode, an error signal is generated and signaled to the outside (eg by means of double-rail signals: fail-safe) (part of status information M169). The status M169 can also indicate the operating mode or information about the time offset of the signals Execution units include. In the case of non-provision of comparison data of a processing unit in a predetermined (programmable) time interval, the error signal is also activated. In case of an error, the outputs M 160, M161 can be disabled (fail silent behavior). This can affect both digital and analog signals. These output driver stages can also be the non-cached (not cached)

Output signals M120, M121 of a processing unit, with the possibility of subsequent error detection. This is tolerated by a safety relevant system as long as the fault tolerance time is not exceeded, i. the time that a (sluggish) system does not yet catastrophically react to errors and therefore there is still the possibility of correction.

Also, output signals M 180, Ml 81 which are not led to the UVE and internal signals of a processing unit can be compared, at least with respect to their calculated value, by outputting this value on the outputs M120, M121 for the purpose of comparison. The same can be done with input signals M 190, M 191, which do not come via M100.

In order to monitor the unit M100, it may be possible for selected or also all signals M 160, M 161 to read them back via M 170, M 171 or also M 190, M191. In this way, it is also possible to ensure in the comparison mode that erroneous signals from the unit M100 are detected. By means of a suitable switch-off path to which (in an OR link) M100, MI10, M1 I1 have access, this can result in a fail-silence behavior of the entire

Systems are manufactured.

FIG. 3 shows a possible implementation of the switching and comparison unit M100 from FIG. 2 in detail. The unit M100 contains a control register M200 with at least one bit representing the mode (performance / comparison) and a status register

M220 with at least one bit representing the error state in comparison mode. The wait and interrupt signals are controlled by further bits in the control register for both processing units. It may also be necessary to differentiate between different interrupts, for example for synchronization purposes, for preparing for the operating mode switches and for error handling.

Optionally, there are further control registers, such as M240, which contains the maximum permitted time difference (in number of clock periods) between the processing units for controlling an internal or external watchdog, and M241 with the time difference value (clock cycle number), from which the fastest processor is assigned by means of WAIT or interrupt signals should be temporarily stopped or delayed, for example, to prevent overflow of data registers.

In status register M220, e.g. In addition to the error bit also stored, how large the clock offset between the processing units is currently. For this purpose, e.g. at least one timer M230 always started by a processing unit when a (via address and

Control signals, e.g. particular address value) is first provided and the value of the timer is always transferred to the status register when the corresponding data value is provided by the second processing unit. In addition, the timer is preferably set so that even with different program sequences according to the WCET (worst case execution time) guaranteed all

Processing units must supply a date. If the preset value is exceeded by the timer, an error signal is output.

The outputs M120, M121 of the processing units are to be stored in M100, in particular for the comparison mode in a buffer memory M250, M251, as far as they are digital data and can not be provided in a clock-accurate manner. Preferably, this can

Memory be executed as a FIFO. If this memory has only a depth of 1 (register), then e.g. Wait signals are used to delay the output of further values until the comparison has been made in order to avoid data loss. There is also a comparison unit M210, which compares the digital data from the input memories M250, M251, the direct inputs M120, M121 or M170, M171. This comparison unit may also compare serial digital data (e.g., PWM signals), e.g. in the memory unit M250, M251 can receive the serial data and convert it into parallel data, which are then compared in M210. Similarly, asynchronous digital input signals M 170, M 171 can be synchronized via additional memory units M270, M271. As well as for the

Input signals 120, 121 are preferably buffered in a FIFO. Switching between performance and compare modes is accomplished by setting or resetting the mode bit in the control register, causing e.g. corresponding interrupts are caused in the two processing units. The comparison itself is provided by the provided data M 120, M121 and the associated addresses and control signals

M130, M131 everanlasst. In this case, certain signals from M120 and M130 or M121 and M131 can act as an identifier indicating whether a comparison of the assigned data is to take place. This is a further embodiment compared to the simple switching in FIG. 1. In this case, during the transition into a comparison mode by means of the interrupt routines, different preparations are advantageously to be made so that identical initial conditions are created for both processing units. If the processing unit is finished with it, the processor-specific ready bit is set in the control register and the processing unit remains in the wait state until the other processing unit indicates its readiness by its ready bit (see also description of the control register in FIG. 6).

In this comparison unit analogue data can also be compared with one another in a suitable analog comparison unit M211 (analogue compare unit). However, this presupposes that the output of the analog signals is sufficiently synchronous with one another or that the analogue compare unit provides for storage of the data digitized by an ADC implemented there (see further comments on FIGS. 12 to 14). Synchronicity can be achieved by comparing the digital outputs of the processing units (data, address and control signals) as described above and maintaining the processing unit too fast. For this purpose, the digital signals which are processed as the source of the analog signals in the processing unit can also be fed via the outputs M 120, M121 to the unit M100, although these signals are otherwise not needed externally. This redundant comparison in addition to the comparison of the analog signals ensures that an error in the calculation can be detected earlier and also facilitates the synchronization of the processing units. The comparison of the analog signals causes additional error detection for the DAC (digital to analog converter) of the processing unit. In other structures of the DCSL architectures such a possibility does not exist. For analog input signals from the peripheral units, a comparison is also possible.

In particular, when it comes to redundant sensor signals of the same system parameter, then you need no additional synchronization measures, but only optionally a control signal indicating the validity of the sensor signals. The realization of a comparison of analog signals will be shown in detail.

FIG. 4 shows a multiprocessor system with at least n + 1 processing units, wherein each of these components may in turn also consist of several sub-processing units (CPUs, ALUs, DSPs with corresponding additional components). The signals of these processing units are also connected to a switching and comparison unit, as described in the two-person system of Figure 2. All components and signals in this figure therefore have the same meaning in content as the corresponding components and signals in FIG. 2. In the multiprocessor system, the switching and comparison unit M300 can distinguish between the performance mode (all processing units execute different tasks), different comparison modes (FIG. the

Data of two or more processing units should be compared and in the case of deviations an error should be signaled) and different voting modes (majority decision in case of deviation according to different predefinable algorithms). For each processing unit can be decided separately, in which mode it works and with what other processing units together they may work in these modes. Exactly how the switchover takes place is subsequently explained further in the description of the control register according to FIG.

FIG. 5 shows a possible implementation of a switching unit for a multiprocessor system with n + 1 processing units. For each processing unit, at least one control register M44i is provided in the control unit of the switching and comparison module. A preferred set of control registers is shown and described in detail in FIG. M44i corresponds to the control register Ci. Various embodiments in the control register are conceivable. It may be described by appropriate bit combinations whether to use an error detection or fault tolerance pattern. Depending on the effort involved in the M300 unit, it is also possible to specify which type of fault tolerance pattern (2 out of 3, median, 2 out of 4, 3 out of 4, FTA, FTM ...) you want to use. Next you can make it configurable, which output one goes through. One can then also form embodiments, which components for which date can influence this configuration.

The output signals of the processing units involved are then compared with one another in the switching unit. Since the signals are not necessarily processed clock-accurate, a caching of the data is required. In this case, data can also be compared in the switching unit, which are given with a larger time difference from the various processing units to the switching unit. By using a temporary memory (eg designed as FIFO memory: first in - out-of-memory or else in a different buffer form), a plurality of data can initially also be received by one processing unit, while other processing units do not yet receive any data provide. In this case, a measure of the synchronicity of the two processing units is the fill level of the FIFO memory. If a certain predefinable fill level is exceeded, the processing unit furthest advanced in the processing is temporarily stopped either by an existing WAIT signal or by suitable interrupt routines, in order to wait for the slower processing units progressing in the processing. The monitoring should be extended to all externally available signals of a processing unit; this also includes analog signals or PWM signals. In the switching unit structures are provided to allow a comparison of such signals. In addition, it is proposed to specify a maximum time deviation between the data to be compared and to monitor it by means of at least one timer.

In the general case, if more than two processing units are interconnected by a common switching unit, a control register is required for each of these processing units or processing units. A specific embodiment of these control registers is explained in FIG.

The (n + 1) lower bits B500x to B50nx of the respective control register Cx are uniquely assigned to the n + 1 processors / processing units. The bit B514x of the control register Cx switches between compare / vote on the one hand and parallel work on the other hand and corresponds to the value of B16 of Figure 1. The bit B513x indicates whether the processing unit concerned is ready for comparison, bit B512x controls this

Synchronization signal (WAIT or INTERRUPT) and bit B511x may be used to prepare the corresponding processing unit x for the comparison by an interrupt. Similarly, bit B5110x controls an interrupt that returns the processing unit to parallel mode.

If B50ik and B50kk of the control register Ck are set to one (0 <i, k <n)., This means in this embodiment that the outputs of the processing unit i are to be compared with those of the processing unit k. If, in addition, B50jk equals one, then vv should be voted between i, j and k and the voting result is output at the output k of the EIS (0 <i, j, k <n). For this purpose, for each group of processing units, a special type of vote or even a majority comparison can be determined, as already enumerated in the explanation for picture M4. In general, all bits B50ik must be set for the processing units i to be compared / voted (in the control register Ck) be, if at the output k of the UVE the voting result is to be output. Parallel output on other outputs is possible.

A one in B50ii of the control register i (0 <i, <n) indicates that the output i of the comparison unit should be active. Carry all control registers Ci only in the appropriate

Memory locations B50ii a one (i = 0, 1, ... n), all processing units work in the

Performance mode with arbitrarily different programs and own output signals.

If all the n + 1 lower bits B50ik are equal to one (i = 0, 1, ... n) and, moreover, B514k is set, then the output signals of all the processing units are selected by majority vote and output to the output k of the UVE; with n = 1 only one finds

Comparison instead.

In the following explanations, it is described by way of example what a procedure for the transition to a comparison / voting in a system with several processing units can look like.

The bit B514i in the control register Ci is set to activate the comparison or the voting. This bit may be set by the processing unit itself as well as by the switching and comparing unit depending on certain system conditions, timing conditions or other conditions (such as accessing particular memory areas, errors or implausibilities). Become the bits with B514i

B50ii and B5 (M), the UVE automatically sets bits B51 Ii and B51 Ik, thereby triggering interrupts in the processing units i and k These interrupts cause the processing units to jump to a particular program location, to initiate certain initialization steps to the comparison mode and then output a feedback (Ready) to the switching and comparison unit

Ready signal causes an automatic reset of the interrupt bit B511i in the respective control register Ci of the processing unit and at the same time the setting of the wait bit B512i. If all wait bits of the processing units involved are set, they are reset by the switchover and compare unit simultaneously. The processing units then start to process the program parts to be monitored. In an advantageous

Embodiment is prevented writing to a control register Ci with set bit B514i by locking (HW or SW). This makes it sensible that the configuration of the comparison can not be changed during execution. A change in the control register Ci is possible only after resetting the bit B514i. This reset causes interrupts in the respective processing units by setting the bits B510x in the control registers of all involved processing units to transition to normal mode (parallel operation).

The consistency of all control registers relative to one another is monitored in accordance with user specifications, and in the event of an error, an error signal is generated which forms part of the

Status information is. For example, it may not happen that one processing unit is used concurrently for multiple independent comparison or voting processes, because then the synchronization is not guaranteed. It is conceivable, however, to compare several processing units without an output of the data signals, but only for the purpose of generating an error signal in the event of inequality.

In a further embodiment, the entry is to be made similar in several or all control registers of the processing units involved in a comparison or voting, i. the corresponding bits of these processing units are to be set there identically, with the possible exception of their own bit i, which controls the output.

FIG. 7 shows the voting unit Q100 for central voting. Voting can be carried out both by means of suitable hardware and by software. The voting algorithm (for example, bit-precise voting) is to be specified. The voting unit Q100 receives several signals Q1, Q1, Q1, and Q112, and from these forms an output signal Q 120, which is produced by voting (for example, an m out of n selection).

If an error occurs during the comparison, the error bit is set in the respective control register. In a voting, the date of the processing unit concerned is ignored; in a simple comparison the output is locked. Any data not available in time before the programmed time expires will be treated as error. The resetting of the error bits is system-dependent and, if necessary, enables reintegration of the respective processing unit

In the event that the processing units and / or the voter are not spatially concentrated angeordnert, a decentralized voting in conjunction with a suitable bus system according to Figure 8 is possible. In FIG. 8, a decentralized voting unit Q200 is controlled by a control unit Q210. It is connected via bus systems Q221, Q222

Data over these bus systems and gives them out there again.

The reset of the comparison and voting bits in a control register with active output bit causes an interrupt in the participating processing units, which are then returned to a parallel operation. In this case, each processing unit have a different entry address, which is managed separately. The program execution can also take place from the same program memory. The accesses are separate and usually to different addresses. If the security-relevant part is small in comparison to the parallel modes, it must be weighed whether a separate program memory with duplicated security part may be less expensive.

The data memory can also be shared in performance mode. The accesses are then successively, for example by means of the AHB / ABP bus.

A special feature is that the error bits have to be evaluated by the system. In order to ensure a safe shutdown in the event of a fault, the safety-relevant signals must be implemented redundantly in a suitable form (for example, in the 1-out-2 code).

In the previous UVEs according to FIGS. 1, 2, 3, 4 and 5, it was initially assumed that the processing units operate with identical or derived clocks which are in a constant phase relation to one another. If clocks from different oscillators and generators are used for the processing devices, and the phase relationships change, then the signals generated with them must be synchronized when they change the clock domain. A synchronization element M800 is shown in FIG. 9 for this purpose. In order to securely store and compare the digital data in particular, synchronization devices M800 are required, which can be attached anywhere in the signal flow. These once ensure the storage of data M820 with the clock M830 of the processing unit that provides this data. For reading, the clock is then used with which the date M840 is further processed. Such a synchronization stage M800 can be developed as a FIFO in order to store a plurality of data (see FIG. 9). In the general case, the synchronization of the data alone is not sufficient, but it is also the sync signal of the data to synchronize with the receive clock.

For this purpose, moreover, a handshake interface is required (FIG. 10) that ensures acceptance by request signals M850 and acknowledgment signals M880. Such an interface is necessary whenever the clock domain changes to ensure secure transmission of data from one clock domain to another. When writing, the data M820 from the area Q305 are provided synchronized with the clock M830 in the register cells M800 and a write request signal M850 indicates the provision of the data. This write request signal is taken from the area Q306 with the clock M860 in a memory element M801 and as a synchronized signal M870 indicates the provision of the data. With the next active clock edge of clock M860, the synchronized data M840 is then accepted and in the process an acknowledgment signal M880 is sent back. This confirmation signal is synchronized by the clock M830 in another memory element M801 to the signal M890 and thus the provision of the data is terminated. New data can then be written to the relevant register. Such interfaces are state of the art and known and can work in special embodiments by an additional coding very fast, without having to wait for an acknowledgment signal.

In a particular embodiment, the memory elements M800 are designed as FIFO memories (first-in, first-out).

The circuits for comparing analog signals of Figure 11 to Figure 14 assume that the processing units that provide the analog signals to be compared, are synchronized with each other so that the comparison makes sense. The synchronization can be achieved by the corresponding signals B40 and B41 of FIG.

FIG. 11 shows a differential amplifier. With the help of this element two voltages can be compared.

BlOO is an operational amplifier, to the negative input BlOl a signal B 141 is connected, which is connected via a resistor BI lO with the value R _1n to the input signal BlI l, at which the voltage value Vi is present. The positive input B 102 is connected to the signal B 142, which is connected via the resistor B 120 with the value R _1n to the input B 121, to which the voltage value V ₂ is applied. The output B 103 of this operational amplifier is connected to the output signal B 190 _having the voltage _value V _0Ut . The signal B190 is via the resistor B140 with the value R _f with the signal

B141 connected and the signal B142 is connected via the resistor B130 with the value R _f with the signal B131 carrying the voltage value of the analog reference point V _{AGN (j.} The output voltage can be calculated using the above-mentioned voltage and resistance values according to the following formula : V _0nJ = R _f Z R j _n (V ₂ - V ₁ ). (1)

If the differential _{amplifier is} only operated with a positive operating voltage, as is usually the case with CMOS, the analog ground V _{agn (} j is a voltage between the operating voltage and the digital ground, usually the mean potential.) If the two analog input voltages Vi and V _{2 are} only slightly different, so will the Output voltage V _{out have} only a small difference V _< ji _ff to the analog ground (positive or negative).

With the aid of 2 comparators it is now checked whether the output voltage above V _agnd + V _d i _ff (FIG. 12) or below V _agnd - V _< ji _{ff is} also the analogue reference point (FIG. 13). In this case, in FIG. 12, the input signal B221 is connected via the resistor B150 with the value Ri to the signal B242, which is connected to the positive input B202 of the operational amplifier B200. Furthermore, the signal B242 is connected via the resistor B 160 with the value R ₂ to the signal B231, which is used as the digital reference potential V _dgng . The negative input B201 of the operational amplifier is connected to the input signal 211, which carries the voltage value of a reference voltage V _ref . The output B203 of the

B200 operational amplifier is connected to the output signal B290 carrying ben the voltage value V _O.

In Fig. 13, accordingly, the input signal B321 is connected through the resistor B 170 of the value R ₃ to the signal B342 which is connected to the negative input B301 of the operational amplifier B300. This signal B342 is still above the

Resistor B 180 connected to the value R4 to the signal B331, which also carries the digital reference potential V _dgnd . The positive input B302 of the operational amplifier B300 is connected to the input signal B311 which carries voltage value of a reference voltage V _ref . The output B303 of the operational amplifier B300 is connected to the output signal B390, which is the voltage value

This is achieved by dimensioning the resistors B150, B160, B170 and B180 with their values Ri, R ₂ , R ₃ and R ₄ in relation to the fixed reference voltage V _ref applied to the signals B211 and B311 as follows : V _ref = (V _agnd + V _1U a) * R ₂ / (R ₁ + R ₂ ) (2)

V _ref = (V _agnd - V ^ * R ₄ / (R ₃ + R ₄ ) (3) V _d i _ff = ((V _2max - V _lmin ) * R _f / R _ta ) - V _agnd (4)

V _{2max denotes} the maximum tolerated voltage _value of V ₂ at signal B 121 and Vi _m i _n the minimum tolerated voltage value of Vi at signal Bl I l. The reference voltage source can be provided externally, or realized by an internally realized bandgap (temperature-compensated and operating voltage-independent reference voltage). In equation (4), the maximum tolerated difference V _d i _{ff is determined} from the maximum positive deviation V _2max and the associated maximum negative deviation Vi _n J _n , ie (V _2max - Vi _n J _n ) is the maximum tolerated Voltage deviation of redundant analog signals to each other to be compared.

If one of the voltage values at the two signals B290 or B390 (V _o ben or Vun _te n) becomes positive, there is a greater deviation of the analog signals than it should be tolerated. As long as the processors that supply these analog signals are synchronized, then there is one

Error that must be stored and possibly leads to switching off the output signals. The synchronicity is given, for example, when the ready signal in the control register of the corresponding processing units is active, or certain digital signals are sent to the UVE, which signal a particular state of the relevant analog signal and thus also the value to be compared in the sense of an identifier. A circuit that the

Error stores is shown in Figure 14. In this circuit, the two input signals B390 and B290 are linked to the output signal B411 via a NOR circuit (logical OR circuit with subsequent inversion) B410. This signal B411 is combined with the input signal B421 in another NOR element B420 to the output signal B421. This signal B421 is input in an OR circuit B430 with the signal B401 to

Connected signal B431, which serves as input to the memory element (D flip-flop) B400. The output signal B401 of this element B400 indicates an error with the value 1. The D flip-flop B400 stores with the clock B403 is a 1 if one of the two voltage values V, thus contributes positively _bottom or V _{at the top} on the signals B390 or B290 as a digital signal the value is high, the signal B421 is not active and no reset signal B402 is present.

The error remains stored until the signal Reset was active at least once. It should be noted in the dimensioning of the circuits Figure 11 to Figure 13 that match the resistors to each other, ie that the resistance ratios of R _f and R _1n , Ri and R ₂ and R ₃ and R _{4 are} as independent of manufacturing tolerances constant. With the signal B421 you can control, whether the circuit should be active, or just one

Synchronization of the processing units takes place, in which no comparison is to be made. The signal B402 resets a previous error and therefore allows a new comparison.

Figure 15 shows an ADC. Depending on the existing requirements, such as conversion speed, accuracy, resolution, noise immunity, linearity and frequency spectrum, this ADC can be implemented using the various known conversion methods. For example, you can choose the principle of successive approximation, where you get the analog signal with a generated signal from a Digital-to-Analog Converter (DAC). A comparator compares the digital input bits of the DAC systematically from MSB (most significant bit) to LSB (least significant bit) to high and then resets exactly when the DAC's analog output signal goes high higher value than the analog input signal (the signal to be converted). The DAC controls with its digital bits from the LSB to the MSB either

Resistors or capacitances with the weights 1, 2, 4, 8, 16, ... in such a way that the setting of the next higher bit always has twice the effect on the analog value as the previous one. After all bits have been set and, if necessary, reset again, the value of the digital word corresponds to the digital representation of the analog input signal. For higher speed requirements, continuous data streams can also be used

Converter can be used, which continuously processes the analog signal and outputs a serial digital signal, which approximates this analog data stream through the serial bit sequence. The digital word is here represented by the bit sequence stored in a shift register. However, such transducers presuppose that changes in the analog signal are constantly made during the conversion period because they can not process constant values.

For lower speed requirements also converters can be used according to the counting principle, for example, by means of the input voltage or the input current cause a corresponding constant charging or discharging a capacitor connected to an integrator. The time required for this is measured and put into relation to the time in the opposite direction to the charge or charge of the same capacitor

(Integrators) by means of a reference voltage source or a corresponding reference current is necessary. The time unit is measured in cycles and the number of clocks required is a measure of the analog input value. Such a method is, for example, the dual-slope method in which the one slope is determined by the discharge corresponding to the analog value and the second edge is determined by the recharge corresponding to the reference value (see also http: //www.exstrom .com / joumaVadc / dsadc.html). The ADC B600 of FIG. 15 is controlled by a trigger signal B602, which is typically an output signal of the processor providing the analog signal and optionally an identifier B603 which provides information about the type of analog signal being provided to distinguish it from allow multiple analog signals. With the

Trigger signal B602, the converted analog word is taken into memory area B640 as a digital value in a register B610 and optionally together with the identifier B603, which is stored in B620 and possibly an additional signal B604 (which is 1 for the identification of an analogernwert), in the Memory B630 is stored. Of the Memory area B640 can advantageously also be realized as FIFO (first-in, furth-out) if a plurality of values are to be stored and the value stored first is to be re-evaluated first as well. If memory area B640 is used for both digital and digitized analog values, then advantageously all digital values are supplemented by one bit A = O at the MSB location, as appropriate

B630 to distinguish them from digitized analog values with A = I (B630) (see Figures 16 and 17). Both B602 and B603 are part of the digital output data O _{1 of} a processor i. In Figure 16, the parts of the stored digitized analog value are shown separately as they are stored in the Speichj erbereich. B710 is the digitized analogue value itself, B720 is the associated identifier and B730 is the analogue value

Analog bit, which in this case is to be stored as 1. FIG. 17 shows a variant of a digital value stored in the same memory area. In B810, the digital value itself is stored, in B820 an optiona option is provided which, for example, indicates whether the digital value is to be compared at all or whether it may also contain further conditions for the comparison. In B830, the value 0 is then stored to indicate that it is a digital value.

To compare the buffered digital and analog signals, the sequence of the storage and possibly the A bit (B730 or B830) as well as the identifier B720 or B820 in conjunction with the converted digital value B710 or the digital value B810 are checked. There is also the possibility, e.g. because of different bit width, to accommodate the analog and digital signals in separate memories (two FIFOs). The comparison is then event-controlled: whenever a value of a processor is transmitted to the UVE, it is checked whether the other participating processors have already provided such a value. If this is not the case, the value in the corresponding FIFO or

Stored memory, in the other case, the comparison is carried out directly, in which case the FIFO can serve as a memory. For example, a comparison is always completed if the FIFOs involved are not empty. If there are more than two processors or comparison signals involved, it can be determined by a voting whether all signals are allowed to be distributed (fail silent behavior) or, if necessary, only by an error signal of the

Error state is signaled.

Claims

claims

1. A method for switching and for data comparison in a computer system having at least three processing units, wherein switching means are provided and switched between at least two operating modes, wherein comparing means are provided and a first operating mode corresponds to a comparison mode and a second operating mode a performance mode, characterized in that in the comparison mode a voting, at least as a 2 out of 3 rating, is provided, wherein control means are provided by which the voting is adjustable.

2. The method according Anspruchl, characterized in that the control means comprise at least storage means or are designed as such and in the

Memory means an identifier for setting the voting, in particular a bit, is stored.

3. The method according to claim 1, characterized in that an error detection and a fault location is performed by the voting and corresponding error statuses are stored.

4. The method according to claim 2 and 3, characterized in that the Fehlerstati next to the identifier in the at least one memory means is stored.

5. The method according to claim 4, characterized in that the identifier is written by at least one processing unit and the error status by the comparison unit and / or the switching unit in the storage means.

6. The method according to claim 1, characterized in that the data is buffered before the input to the comparison means.

7. The method according to claim 2, characterized in that an internal to the computer system source, in particular a processing unit, the identifier for setting the

Voting pretends.

8. The method according to claim 2, characterized in that an external to the computer system source specifies the identifier for setting the voting.

9. The method according to claim 7 and / or 8, characterized in that a plurality of sources are provided for specifying the identifier and all sources are connected such that the setting of the voting is configured by an OR operation.

10. A device for switching and for data comparison in a computer system having at least three processing units, wherein switching means are provided and switched between at least two operating modes, wherein comparison means are provided and a first operating mode a comparison mode and a second operating mode corresponds to a performance mode, characterized in that in comparison mode a voting, at least as a 2 out of 3 rating, is provided, wherein

Control means are contained by which the voting is adjustable.

11. The device according to claim 10, characterized in that the control means comprise at least memory means or are formed as such and in the memory means an identifier for setting the Votings, in particular a bit, is stored.

12. The device according to claim 11, characterized in that it is the storage means to at least one control register.

13. The device according to claim 10, characterized in that at least one input buffer memory is provided, which is designed such that the data is buffered before entering the comparison means.

14. The apparatus according to claim 10, characterized in that the control means are provided as a source for setting the voting external to the computer system.

15. The apparatus according to claim 10, characterized in that the control means are provided as a source for setting the Voting internally to the computer system.

16. The apparatus of claim 14 and / or 15, characterized in that all the sources are set to adjust the Voting such that the setting of voting by an OR operation is configured.