mirror of
https://github.com/lowRISC/ibex.git
synced 2025-04-24 22:07:43 -04:00
124 lines
7.3 KiB
ReStructuredText
124 lines
7.3 KiB
ReStructuredText
.. _security:
|
|
|
|
Security Features
|
|
=================
|
|
|
|
Ibex implements a set of extra features (when the SecureIbex parameter is set) to support security-critical applications.
|
|
All features are runtime configurable via bits in the **cpuctrl** custom CSR.
|
|
|
|
Outputs
|
|
-------
|
|
|
|
Ibex has three alert outputs for signalling security issues.
|
|
The internal major alert (**alert_major_internal_o**) indicates a critical security issue from which the core cannot recover which was detected internally in `ibex_top`.
|
|
The bus major alert (**alert_major_internal_o**) indicates a critical security issue from which the core cannot recover which was detected on incoming bus data.
|
|
The minor alert (**alert_minor_o**) indicates potential security issues which can be monitored over time by a system.
|
|
|
|
Data Independent Timing
|
|
-----------------------
|
|
|
|
When enabled (via the **data_ind_timing** control bit in the **cpuctrl** register), execution times and power consumption of all instructions shall be independent of input data.
|
|
This makes it more difficult for an external observer to infer secret data by observing power lines or exploiting timing side-channels.
|
|
|
|
In Ibex, most instructions already execute independent of their input operands.
|
|
When data-independent timing is enabled:
|
|
|
|
* Branches execute identically regardless of their taken/not-taken status
|
|
* Early completion of multiplication by zero/one is removed
|
|
* Early completion of divide by zero is removed
|
|
|
|
Note that data memory operations to unaligned addresses might result in multiple bus accesses being made.
|
|
This in turn could expose information about the address as a timing side-channel.
|
|
It is therefore recommended to stick to aligned memory accesses when using this feature for critical code regions.
|
|
|
|
When Ibex is configured to use an instruction cache, stalls on instruction fetch can see variable latency (depending on whether or not they hit in the cache).
|
|
Software that has need of data independent timing may wish to disable the instruction cache to avoid this or to carefully analyse execution to determine if variable latency introduced by the cache causes unacceptable leakage.
|
|
The instruction cache is controlled by the **icache_enable** bit in the **cpuctrl** register.
|
|
Precise details of fetch timing will depend upon the memory system Ibex is connected to.
|
|
|
|
Dummy Instruction Insertion
|
|
---------------------------
|
|
|
|
When enabled (via the **dummy_instr_en** control bit in the **cpuctrl** register), dummy instructions will be inserted at random intervals into the execution pipeline.
|
|
The dummy instructions have no functional impact on processor state, but add some difficult-to-predict timing and power disruption to executed code.
|
|
This disruption makes it more difficult for an attacker to infer what is being executed, and also makes it more difficult to execute precisely timed fault injection attacks.
|
|
|
|
The frequency of injected instructions can be tuned via the **dummy_instr_mask** bits in the **cpuctrl** register.
|
|
|
|
+----------------------+----------------------------------------------------------+
|
|
| **dummy_instr_mask** | Interval |
|
|
+----------------------+----------------------------------------------------------+
|
|
| 000 | Dummy instruction every 0 - 4 real instructions |
|
|
+----------------------+----------------------------------------------------------+
|
|
| 001 | Dummy instruction every 0 - 8 real instructions |
|
|
+----------------------+----------------------------------------------------------+
|
|
| 011 | Dummy instruction every 0 - 16 real instructions |
|
|
+----------------------+----------------------------------------------------------+
|
|
| 111 | Dummy instruction every 0 - 32 real instructions |
|
|
+----------------------+----------------------------------------------------------+
|
|
|
|
Other values of **dummy_instr_mask** are legal, but will have a less predictable impact.
|
|
|
|
The interval between instruction insertion is randomized in the core using an LFSR.
|
|
The initial seed and output permutation for this LFSR can be set using parameters from the top-level of Ibex.
|
|
Sofware can periodically re-seed this LFSR with true random numbers (if available) via the **secureseed** CSR.
|
|
This will make the insertion interval of dummy instructions much harder for an attacker to predict.
|
|
|
|
Note that the dummy instruction feature inserts multiply and divide instructions.
|
|
The core must be configured with a multiplier (`RV32M != ibex_pkg::RV32MNone`) or errors will occur using this feature.
|
|
|
|
Bus integrity checking
|
|
----------------------
|
|
|
|
Extra signals are available alongside the instruction and data side memory channels to support bus integrity checking.
|
|
When the SecureIbex parameter is set, incoming data will be checked against the supplied checkbits.
|
|
An :ref:`internal interrupt<internal-interrupts>` will be generated and a bus major alert signalled if there is a mismatch.
|
|
Where load data has bad checkbits the write to the load's destination register will be suppressed.
|
|
Write data can be checked against the supplied checkbits at its destination to confirm integrity.
|
|
|
|
Register file ECC
|
|
-----------------
|
|
|
|
When Ibex is configured with the SecureIbex parameter, ECC checking is added to all reads of the register file.
|
|
This can be useful to detect fault injection attacks since the register file covers a reasonably large area.
|
|
No attempt is made to correct detected errors, but an internal major alert is signaled for the system to take action.
|
|
|
|
Register file write enable glitch detection
|
|
-------------------------------------------
|
|
|
|
When Ibex is configured with the SecureIbex parameter, the write enable signal into the register file is checked to be one-hot.
|
|
This can be useful to detect fault injection attacks.
|
|
No attempt is made to correct detected errors, but an internal major alert is signaled for the system to take action.
|
|
|
|
ICache ECC
|
|
----------
|
|
|
|
The ICache can be configured with ECC protection.
|
|
When an ECC error is detected a minor alert is signaled.
|
|
See :ref:`icache-ecc` for more information.
|
|
|
|
Hardened PC
|
|
-----------
|
|
|
|
This adds a check that the PC driven from the IF stage has not been modified.
|
|
A check is asserted that the current IF stage PC equals the previous PC plus the correct increment.
|
|
The check is disabled after branches and after reset.
|
|
If a mismatch is detected, an internal major alert is signaled.
|
|
|
|
Shadow CSRs
|
|
-----------
|
|
|
|
Certain critical CSRs (`mstatus`, `mtvec`, `cpuctrl`, `pmpcfg` and `pmpaddr`) have extra glitch detection enabled.
|
|
This creates a second copy of the register which stores a complemented version of the main CSR data.
|
|
A constant check is made that the two copies are consistent, and an internal major alert is signalled if not.
|
|
Note that this feature is not currently used when the SecureIbex parameter is set due to overlap with dual core lockstep.
|
|
|
|
Dual core lockstep
|
|
------------------
|
|
|
|
This configuration option instantiates a second copy of the core logic, referred to as the shadow core.
|
|
The shadow core executes using a delayed version of all inputs supplied to the main core.
|
|
All outputs of the shadow core are compared against a delayed version of the outputs of the main core.
|
|
Any mismatch between the two sets of outputs will trigger an internal major alert.
|
|
|
|
Note that the register file and icache RAMs are not duplicated since these units are covered by ECC protection.
|