US 20060117167 A1
Within a data processing systems supporting conditional write processing operations, a trash register is provided such that when non-write conditions are encountered a register write is made to the trash register rather than the data register specified by the conditional write operation. Thus the power signature associated with whether or not a register write does or does not occur is masked. The trash register activity may be programmable enabled and disabled by a configuration parameter stored within a system configuration register.
1. Apparatus for processing data, said-apparatus comprising:
a process core (4) operable to execute data processing instructions to generate result data values; and
data processing registers (12) holding data values defining state of said processor core to which said result data values are written; wherein
at least one data processing instruction executed by said processor core is a conditional write data processing instruction encoding condition codes (26) specifying conditions under which said conditional write data processing instruction will or will not be permitted to write data to effect a change in state of said processor core: and further comprising
a trash register (51) to which a result data value may be written instead of a data processing register upon execution of said conditional write data processing instruction when said condition codes within said conditional write data processing instruction do not permit a write to effect a change in state of said processor core.
2. Apparatus as claim in
3. Apparatus as claimed in
4. Apparatus as claimed in
5. Apparatus as claimed in
6. A method of processing data, said method comprising the steps of:
generating result data values upon execution by a processor core (4) of data processing instructions, at least one data processing instruction executed being a conditional write data processing instruction encoding condition codes (26) specifying conditions under which said conditional write data processing instruction will or will not be permitted to write data to effect a change in state of said processor core and wherein
a result data value is not written to a data processing register holding a data value defining state of said processor core when condition codes within said condition write data processing instruction do not permit a write to effect a change in state of said processor core but is instead written to a trash register (51).
7. A method as claimed in
8. A method as claimed in
9. A method as claimed in
10. A method as claimed in
This invention relates to the field of data processing systems. More particularly, this invention relates to the masking of processing activity within data processing systems, for example, in order to increase security.
It is known to provide data processing systems which manipulate secure data and for which it is desirable to ensure a high degree of security. As an example, it is known to provide smart cards which include a data processing system which manipulates secure data, such as secret cryptographic keys, and this data must be kept secret in order to prevent fraud.
Known ways of attacking the security of such systems include timing analysis and power analysis. By observing the tiling behaviour and/or the power consumption behaviour of such a system in response to inputs, information concerning the processing being performed and the data being manipulated can be determined in a way that can compromise security. It is strongly advantageous to provide resistance against such security attacks.
Viewed from one aspect the present invention provides apparatus for processing data, said apparatus comprising:
a result data value generating circuit operable to generate a result data value upon execution of a data processing operation; and
a data processing register to which said result data value is written; wherein
at least one data processing operation executed by said result data generating circuit is a conditional write data processing operation for which a result data value is not written to said data processing register when non-write conditions are met; and further comprising
a trash register to which a result data value may be written upon execution of said conditional write data processing operation when said non-write conditions are met.
This invention recognises that there is a characteristic power consumption signature associated with a write to a data processing register and accordingly information concerning the data processing being performed in association with conditional write data processing operations can be externally observed, i.e. information upon whether or not the conditional write did or did not occur. The invention address this problem by providing a trash register to which a result value (which is preferably the true value) is written when the conditional write data processing operation meets its non-write conditions and a write would not otherwise occur. Accordingly, a write to a register whether the true register or the trash register, always occurs irrespective of whether or not the write conditions or non-write conditions are met and thus the security of this system is enhanced.
The data register to which the write is normally made when the write conditions are met is preferably part of a register bank containing a plurality of such registers. In this circumstance, a common trash register(s) may be used for dummy writes irrespective of how many real data registers are provided within the register bank.
Preferably, the trash register is physically located as part of the register bank so as to avoid leakage of information by observing which part of a circuit is active at any given time.
Whilst the present technique is applicable to a wide variety of systems, such a microprocessor based systems, digital signal processing system and the like, the invention is particularly well suited to systems including a processor core responsive to data processing instructions as these often incorporate conditional write instructions which give rise to the potential vulnerability addressed by this technique.
It will be appreciated that a conditional write operation may be arranged to either occur or not occur when particular conditions are met.
It will be appreciated that the normal technical prejudice in this field is to reduce power consumption as much as possible. Accordingly, it would conventionally be considered that not performing a register write when a conditional write operation did not require one would be an advantageous feature since it would reduce the amount of power consumed. The present technique moves against this technical prejudice in the field by deliberately performing a trash register write and consuming power even though this is not required for the real processing activities of the system.
In preferred embodiments of the invention the trash register activity can be selectively enabled and disabled depending upon a control signal stored in a system configuration register. This allows programmable activity of the trash register activity such that power can be saved by disabling this feature when non secure processing is taking place and yet security improved when required, such as when handling cryptographic keys, decoding passwords etc.
As mentioned above, whilst the trash register may be physically located within the register bank with the normal data registers, in preferred embodiments the trash register is unmapped to a register number such that it cannot be specified by any program instruction and accordingly is invisible as a register from the programmer's model point of view. The trash register is however visible in the sense that its activity can be enable or disabled in preferred embodiments by a configuration parameter.
Viewed from another aspect the present invention provides a method of processing data, said method comprising the steps of:
generating a result data value upon execution of a data processing operation, at least one data processing operation executed being a conditional write data processing operation, wherein
a result data value is not written to a data processing register when non-write conditions are met but is instead written to a trash register.
Embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings in which:
In operation, the processor core 4 fetches instructions and data from the memory 8. The instructions are fed to an instruction pipeline 10 where they occupy successive pipeline stages such as, for example, fetch, decode, execute, memory and write back on successive processing cycles. Pipelined processors are in themselves well known as a way of effectively executing a number of program instructions in a partially overlapping fashion in order to improve processor performance.
The data values read from the memory 8 by the processor core 4 are supplied to a register bank 12 from where they may be manipulated under program instruction control using one or more of a multiplier 14, a shifter 16 and an adder 18. Other data manipulating circuits may be provided, such as circuits performing logical operations, such as ANDs, Ors, count leading zeros etc.
The coprocessor 6 is a system configuration coprocessor containing a number of configuration registers 22 which may be written under program control to set up configuration controlling parameters. These configuration controlling parameters can specify many aspects of the configuration of the processing system 2, such as for example the endianess and the like. Included within one of these configuration controlling registers 22 is a bit which specifies whether or not the processor core should operate in a fixed timing mode or a variable timing mode. This bit is illustrated as being supplied as an input to the instruction decoder 20, but it will be appreciated that this bit may also be supplied to various other points within the processor core 4 as required to control their behaviour. In dependence upon this fixed/variable bit, the processor core 4 operates in either a fixed timing mode or a variable timing mode. When in the fixed timing mode at least one program instruction which has a variable timing (i.e. takes a variable number of processing cycles to complete) in the variable timing mode, is instead forced to have a fixed timing (e.g. take the maximum possible number of processing cycles to complete irrespective of whether or not it could have been suppressed in its entirety or completed in less than the maximum number of processing cycles. As the instruction decoder 20 is primarily responsible for decoding the program instructions and instructing the activity of the other elements of the processor core 4, the instruction decoder 20 can take the major role in controlling the processor core 4 to either operate in the fixed timing mode or the variable timing mode. Not all variable timing instruction need be provided with a fixed timing mode type of operation.
It will be appreciated that in the above description a single bit in the configuration controlling register 22 is shown as switching between fixed and variable timing modes. Alternatively, multiple bits within the configuration controlling register 22 may be provided to separately enable and disable the fixed or variable timing behaviour of different types of instruction, such as conditional instruction behaviour, uniform branch behaviour, disabling early terminate, etc.
At step 28, the instruction decoder 20 waits for a new instruction to execute. When a new instruction is received processing proceeds to step 30 at which the condition codes associated with the new instruction are read. At step 32 these condition codes are compared with the currently existing condition codes in the system. These condition codes currently existing in the system are the result of previous processing activity, either in the immediately preceding instruction or in the last instruction which would have updated those condition codes.
At step 34, a check is made for a match between the condition codes 26 of the current instruction being executed and the existing condition codes. If a match does not occur, then processing proceeds to step 36 where execution of the current instruction is started. It will be appreciated that
Following step 36, processing proceeds to step 38 where a check is made as to whether or not early termination of the instruction is possible. This early termination may, for example, be because one of the operands has a particular value, such as zero or unity, or on subsequent processing cycles that a particular partial result has been produced. If early termination is possible, then processing proceeds to step 40 where a check is made as to whether or not the processor core 4 is currently operating in the fixed or variable timing mode. If the processor is in the variable timing mode, then processing proceeds to step 42 and the instruction concerned is early terminated with the result being returned as appropriate and processing returns to step 28.
If the determination at step 40 is that the system is in the fixed timing mode, then processing proceeds to step 44 irrespective of the fact that early termination is possible. Step 44, which may also be reached by a determination at step 38 that early termination is not possible, executes the instruction concerned for one processing cycle. In the case of a multicycle processing instruction, such as a multiplication, a divide, an add or a subtraction, these typically take several cycles to execute and so after step 44 processing proceeds to step 46 at which a determination is made as to whether or not the maximum number of cycles associated with that instruction has yet been performed. If the maximum number of cycles has been performed, then the result will have been generated. If early termination was possible and the system was being forced to continue to execute for further processing cycles, then step 46 will still indicate that this forced execution should cease when the maximum possible number of processing cycles for that type of instruction has been reached. If the maximum number of processing cycles has not yet been performed, then processing is returned to step 38.
If the match tested for at step 34 was positive, then processing proceeds to step 48. In this example, the positive detection of a match at step 34 indicates that execution of the particular instruction should be suppressed. Step 48 determines whether or not the system is currently in the forced execution mode. If in the forced execution mode, then processing proceeds to step 50 where a forced dummy execution of the instruction will occur. When dummy execution is performed the result is written to a trash register (see trash register 51 in
It will be appreciated that
Associated with the multiplier 56 is a dummy activity enable circuit 76 which serves to enable dummy activity in the multiplier 56 as appropriate or alternatively pass the required activity enable signals to activate the multiplier 56 when the program instruction being executed so requires. Similar dummy activity enables circuits 78, 80, 82, 84 are associated with the respective other circuit portions 58, 60, 62, 64 previously mentioned.
In operation, instructions to be executed are to be passed to the instruction pipeline 66 and on to the instruction decoder 68 to generate instruction driven enable signals which are applied to respective circuit portions. These enable signal serve to select the data path through the data processing system 52 and to activate the circuit portions concerned to read their inputs, to perform the specified processing and to generate their associated output signals. As an example, a multiplier-accumulate operation might read data values from the register bank 54, apply these to the multiplier 56 and adder 60 and then write back the result to the register bank 54. Thus, the register bank 54, the multiplier 56 and the adder 60 would all be subject to required activity enable signals which both enabled their operation and selected them to form a complete data path. The different circuit portions have different power consumption characteristics and timing characteristics such that external observation could reveal which instruction were being executed by observing such parameters. Accordingly, pseudo random dummy activity of the other circuit portions not required for the instruction being executed is also enabled. Thus, even though the shifter 58 may not being used by the particular multiplier accumulate instruction being executed, it may nevertheless be pseudo randomly enabled such that it will consume power by shifting whatever value is applied to its input. Its output latches will not be enabled to avoid this dummy activity altering the circuit state an undesired way which may interfere with required operation e.g some circuit portions might assume persistence of output values. The dummy activity is enabled for time periods matching the normal operation timings for the circuit portions concerned.
The random clock gating circuit 70 serves to receive a plurality of pseudo random enable signals for different respective circuit portions and gate these for applying to respective circuit portions under the control of configuration parameters read from the system configuration register 74 within the system configuration coprocessor 72. These configuration flags may indicate that dummy activity should be enabled for the shifter 58, ALU 62 and multiplier 56, but not for the adder 60 or the load store unit 64. The different pseudo random enable signals allow different pseudo random characteristics to be applied in a manner that can match these respective circuit portion concerned. As an example, there may a different minimum enable time associated with the normal timing of the different circuit portions.
At an overall level, it will be seen that the instruction decoder 68 will serve as a required activity enabling circuit which will enable the circuit portions required to perform the data processing operation specified by the instruction currently being executed. superimposed upon this required activity, various dummy activities within the other circuit portions will be enabled/stimulated by the dummy activity control circuitry provided in various places in the rest of the data processing system 52. The dummy activity serves to mask the power consumption and timing characteristics associated with the required activity.
It will be appreciated that the process illustrated in
As well as the trash data register RT, further registers 98, 100 are also provided for the purpose of balancing the high to low and low to high transitions. Dedicated dummy registers 98 are provided in respect of the data registers R0 to R12 as well as the trash data register RT. Shared dummy registers 100 are provided for storing an exclusive OR value as well as the inverse of the exclusive OR value in response to each write to a data register subject to the transition balancing technique. A register write control circuit 102 serves to generate the appropriate data values to be written to the further registers 98, 100 in response to a data value write to a data register. This symmetric write control is selectively enabled and disabled by an appropriate system configuring controlling flag signal from the system configuring coprocessor 72.
If the register to which the write is being made is potentially subject to symmetric register writing then step 110 serves to determine whether or not this feature is currently enabled. If this feature is not currently enabled, then processing proceeds to step 108. If this feature is enabled, then processing proceeds to step 112.
At step 112, the register controls circuit calculates for each bit position within the data value a value being the inverse exclusive OR of the current bit being written at that position and the previously stored bit at that position which is then exclusive Ored with the previously stored dummy register value for that bit position (see
At step 114 the data value is written to the data register in a similar manner to step 108. At step 116 the three further values determined for each bit position within the registers concerned are written to the three further registers. Steps 114 and 116 take place simultaneously. As will be discussed in relation to
The shared dummy registers 100 are indicated in
The Java bytecode decoder 130 may be selectively enabled and disabled by an input signal. When the Java bytecode decoder 130 is disabled a received Java bytecode will trigger an exception which starts execution of software emulation code for handling Java bytecodes using the native ARM Thumb instruction sets. This support code is stored within memory in area 132 as illustrated. The Java bytecode program 134 is also stored within memory. When it is desired to obscure the nature of the Java program execution, the Java bytecode decoder 130 may be subject to a pseudo random signal which selectively enables and disables this element so as to effectively switch the instruction execution mechanism for the Java bytecodes between a mixed hardware and emulation execution mechanism and a purely emulation mechanism. Configuration controlling values within a system configuration register 136 specify whether or not the Java decoder 130 is present and whether or not random enabling and disabling of this Java decode 130 is permitted.