Reset Manager HWIP Technical Specification

Overview

This document describes the functionality of the reset controller and its interaction with the rest of the OpenTitan system.

Features

• Stretch incoming POR.
• Cascaded system resets.
• Peripheral system reset requests.
• RISC-V non-debug-module reset support.
• Limited and selective software controlled module reset.
• Always-on reset information register.
• Always-on alert crash dump register.
• Always-on CPU crash dump register.
• Reset consistency checks.

Theory of Operation

The OpenTitan reset topology and reset controller block diagram are shown in the diagram below. The reset controller is closely related to the power controller, please refer to that spec for details on how reset controller inputs are controlled.

Reset Topology

The topology can be summarized as follows:

• There are two reset domains
  • Test Domain - Driven by TRSTn
  • Core Domain - Driven by internal POR circuitry.
• Test domain is comprised of the following components
  • SOC TAP and related DFT circuits
  • RISC-V TAP (part of the rv_dm module)

The test domain does not have sub reset trees. TRSTn is used directly by all components in the domain.

The Core domain consists of all remaining logic and contains 4 sub reset trees, see table below.

The reset trees are cascaded upon one another in this order:

• rst_por_n -> rst_lc_n -> rst_module_n
• rst_por_n -> rst_sys_n -> rst_module_n This means when a particular reset asserts, all downstream resets also assert.

The primary difference between rst_lc_n and rst_sys_n is that the former controls the reset state of most logic in the system, while the latter controls the reset state only of the debug domain. This separation is required because the debug domain may request the system to reset while retaining debug info and control. This is particularly useful if one wanted to debug something early during the boot flow, and thus needed to set a break point after requesting a debug reset.

The reset topology also contains additional properties:

• Selective processor HART resets, such as hartreset in dmcontrol, are not implemented, as it causes a security policy inconsistency with the remaining system.
  • Specifically, these selective resets can cause the cascaded property shown above to not be obeyed.
• Modules do not implement local resets that wipe configuration registers, especially if there are configuration locks.
  • Modules are allowed to implement local soft resets that clear datapaths; but these are examined on a case by case basis for possible security side channels.
• In a production system, the Test Reset Input (TRSTn) should be explicitly asserted through system integration.
  • In a production system, TRSTn only needs to be released for RMA transitions and nothing else. .

Reset Manager

The reset manager handles the reset of the core domain, and also holds relevant reset information in CSR registers, such as:

• RESET_INFO indicates why the system was reset.
• ALERT_INFO contains the recorded alert status prior to system reset.
  • This is useful in case the reset was triggered by an alert escalation.
• CPU_INFO contains recorded CPU state prior to system reset.
  • This is useful in case the reset was triggered by a watchdog where the host hung on a particular bus transaction.

Additionally, the reset manager, along with the power manager, accepts requests from the system and asserts resets for the appropriate clock trees. These requests primarily come from the following sources:

• Peripherals capable of reset requests: such as sysrst_ctrl and always on timers .
• Debug modules such as rv_dm.
• Power manager request for low power entry and exit.
• Escalation reset requests such as those from alert_handler or pwrmgr itself.
• Direct software request for reset.

Shadow Resets

OpenTitan supports the shadow configuration registers. These are registers stored in two constantly checking copies to ensure the values are not maliciously or accidentally disturbed. For these components, the reset manager outputs a shadow reset dedicated to resetting only the shadow storage. This reset separation ensures that a targetted attack on the reset line cannot easily defeat shadow registers.

Reset Consistency Checks

The reset manager implements reset consistency checks to ensure that triggered resets are supposed to happen and not due to some fault in the system. Every leaf reset in the system has an associated consistency checker.

The consistency check ensures that when a leaf reset asserts, either its parent reset must have asserted, or the software request, if available, has asserted. While this sounds simple in principle, the check itself crosses up to 3 clock domains and must be carefully managed.

First, the parent and leaf resets are used to asynchronously assert a flag indication. This flag indication is then synchronized into the reset manager’s local clock domain.

The reset manager then checks as follows:

• If a leaf reset has asserted, check to see either its parent or software request (synchronous to the local domain) has asserted.

• If the condition is not true, it is possible the parent reset indication is still being synchronized, thus we wait for the parent indication.

• It is also possible the parent indication was seen first, but the leaf condition was not, in this case, we wait for the leaf indication.

• A timeout period corresponding to the maximum synchronization delay is used to cover both waits.

  • If the appropriate pairing is not seen in the given amount of time, signal an error, as the leaf reset asserted without cause.

• If all reset conditions are satisfied, wait for the reset release to gracefully complete the cycle.

Hardware Interfaces

Parameters

The following table lists the instantiation parameters of rstmgr.

Signals

Referring to the Comportable guideline for peripheral device functionality, the module RSTMGR has the following hardware interfaces defined.

Primary Clock: clk_i

Other Clocks: clk_aon_i, clk_io_div4_i, clk_main_i, clk_io_i, clk_io_div2_i, clk_usb_i, clk_por_i

Bus Device Interfaces (TL-UL): tl

Bus Host Interfaces (TL-UL): none

Peripheral Pins for Chip IO: none

Inter-Module Signals: Reference

Interrupts: none

Security Alerts:

Security Countermeasures:

Design Details

The reset manager generates the resets required by the system by synchronizing reset tree components to appropriate output clocks. As a result, a particular reset tree (for example rst_lc_n) may have multiple outputs depending on the clock domains of its consumers.

Each reset tree is discussed in detail below.

POR Reset Tree

The POR reset tree, rst_por_n, is the root reset of the entire device. If this reset ever asserts, everything in the design is reset.

The ast input aon_pok is used as the root reset indication. It is filtered and stretched to cover any slow voltage ramp scenarios. The stretch parameters are design time configurations.

• The filter acts as a synchronizer and is by default 3 stages.
• The count by default is 32.
  • The counter increments only when all stages of the filter are 1.
  • If any stage at any point becomes ‘0’, the reset counter returns to 0 and downstream logic is driven to reset again.
• Both functions are expected to operate on slow, always available KHz clocks.

Life Cycle Reset Tree

Life cycle reset, rst_lc_n asserts under the following conditions:

• Whenever rst_por_n asserts.
• Whenever a peripheral reset request (always on timer watchdog, rbox reset request, alert handler escalation, direct software request) is received.

The rst_lc_n tree contains both always-on and non-always-on versions. How many non-always-on versions is dependent on how many power domains are supported by the system.

System Reset Tree

System reset, rst_sys_n , assertion depends on life cycle state.

When in PROD and PROD_END states, rst_sys_n is identical to rst_lc_n.

When in TEST, RMA and DEV states, rst_sys_n is identical to rst_lc_n unless the reset request is ndmreset_req. ndmreset_req is issued by the debug module of the system, it requests for all logic, except those needed to maintain debug state to reset.

Since ndmreset_req is valid only during TEST, RMA and DEV states, it is the only place where the reset is differentiated. During these states, when ndmreset_req is issued, all logic except the debug module and associated glue logic are reset.

The rst_sys_n tree contains both always-on and non-always-on versions. How many non-always-on versions is dependent on how many power domains are supported by the system.

Output Leaf Resets

The reset trees discussed above are not directly output to the system for consumption. Instead, the output leaf resets are synchronized versions of the various root resets. How many leaf resets there are and to which clock is decided by the system and templated through the reset manager module.

Assuming a leaf output has N power domains and M clock domains, it potentially means one reset tree may output NxM outputs to satisfy all the reset scenario combinations.

Power Domains and Reset Trees

It is alluded above that reset trees may contain both always-on and non-always-on versions. This distinction is required to support power manager’s various low power states. When a power domain goes offline, all of its components must reset, regardless of the reset tree to which it belongs.

For example, assume a system with two power domains - Domain A is always-on, and Domain B is non-always-on. When Domain B is powered off, all of Domain B’s resets, from rst_lc_n, rst_sys_n to rst_module_n are asserted. However, the corresponding resets for Domain A are left untouched because it has not been powered off.

Software Controlled Resets

Certain leaf resets can be directly controlled by software. Due to security considerations, most leaf resets cannot be controlled, only a few blocks are given exceptions. The only blocks currently allowed to software reset are spi_device, usbdev, spi_host and i2c.

The criteria for selecting which block is software reset controllable is meant to be overly restrictive. Unless there is a clear need, the default option is to not provide reset control.

In general, the following rules apply:

• If a module has configuration register lockdown, it cannot be software resettable.
• If a module operates on secret data (keys), it cannot be software resettable.
  • Or a software reset should render the secret data unusable until some initialization routine is run to reduce the Hamming leakage of secret data.
• If a module can alter the software’s perception of time or general control flow (timer or interrupt aggregator), it cannot be software resettable.
• If a module contains sensor functions for security, it cannot be software resettable.
• If a module controls life cycle or related function, it cannot be software resettable.

Summary

The following table summarizes the different reset requests and which part of each reset tree, along with what power domain is affected.

Reset Information

The reset information register is a reflection of the reset state from the perspective of the system. In OpenTitan, since there is only 1 host, it is thus from the perspective of the processor. This also suggests that if the design had multiple processors, there would need to be multiple such registers.

If a reset does not cause the processor to reset, there is no reason for the reset information to change (this is also why there is a strong security link between the reset of the processor and the rest of the system). The following are the currently defined reset reasons and their meaning:

The reset info register is write 1 clear. It is software responsibility to clear old reset reasons; the reset manager simply records based on the rules below.

Excluding power on reset, which is always recorded when the device POR circuitry is triggered, the other resets are recorded when authorized by the reset manager. Reset manager authorization is based on reset categories as indicated by the power manager. The power manager has three reset categories that are mutually exclusive:

• No reset has been triggered by pwrmgr.
• Low power entry reset has been triggered by pwrmgr.
• Software or peripheral reset request has been triggered by pwrmgr.

The reset categories are sent to the reset manager so that it can decide which reason to record when the processor reset is observed. Non-debug-module resets are allowed only when no resets have been triggered by pwrmgr.

Since a reset could be motivated by multiple reasons (a security escalation during low power transition for example), the reset information registers constantly record all reset causes in which it is allowed. The only case where this is not done is POR, where active recording is silenced until the first processor reset release.

Even though four reset causes are labeled as warm boot, their effects on the system are not identical.

• When the reset cause is LOW_POWER_EXIT, it means only the non-always-on domains have been reset.
  • Always-on domains retain their pre-low power values.
• When the reset cause is NDM_RESET, it means only the rst_sys_n tree has asserted for all power domains.
• When the reset cause is HW_REQ or SW_REQ, it means everything other than power / clock / reset managers have reset.

This behavioral difference may be important to software, as it implies the configuration of the system may need to be different.

Crash Dump Information

The reset manager manages crash dump information for software debugging across unexpected resets and watchdogs. When enabled, the latest alert information and latest CPU information are captured in always-on registers.

When the software resumes after the reset, it is then able to examine the last CPU state or the last set of alert information to understand why the system has reset.

The enable for such debug capture can be locked such that it never captures.

Alert Information

The alert information register contains the value of the alert crash dump prior to a triggered reset. Since this information differs in length between system implementation, the alert information register only displays 32-bits at a time. The ALERT_INFO_ATTR register indicates how many 32-bit data segments must be read.

To enable alert crash dump capture, set ALERT_INFO_CTRL.EN to 1. Once the system has reset, check ALERT_INFO_ATTR.CNT_AVAIL for how many reads need to be done. Set ALERT_INFO_CTRL.INDEX to the desired segment, and then read the output from ALERT_INFO.

CPU Information

The CPU information register contains the value of the CPU state prior to a triggered reset. Since this information differs in length between system implementation, the information register only displays 32-bits at a time.

For more details on the CPU dump details, please see crash dump.

The CPU_INFO_ATTR register indicates how many 32-bit data segments must be read. Software then simply needs to write in CPU_INFO_CTRL.INDEX which segment it wishes and then read out the CPU_INFO register.

Programmers Guide

Device Interface Functions (DIFs)

To use this DIF, include the following C header:

#include "sw/device/lib/dif/dif_rstmgr.h"

This header provides the following device interface functions:

• dif_rstmgr_alert_info_dump_read Reads the entire alert info crash dump.
• dif_rstmgr_alert_info_get_enabled Retrieves the alert info crash dump capture state.
• dif_rstmgr_alert_info_get_size Get alert info dump size.
• dif_rstmgr_alert_info_set_enabled Enables or disables the alert crash dump capture.
• dif_rstmgr_cpu_info_dump_read Reads the entire CPU info crash dump.
• dif_rstmgr_cpu_info_get_enabled Retrieves the CPU info crash dump capture state.
• dif_rstmgr_cpu_info_get_size Get cpu info dump size.
• dif_rstmgr_cpu_info_set_enabled Enables or disables the CPU crash dump capture.
• dif_rstmgr_fatal_err_code_get_codes Read the fatal error codes.
• dif_rstmgr_reset Resets the Reset Manager registers to sane defaults.
• dif_rstmgr_reset_info_clear Clears the reset information in Reset Manager.
• dif_rstmgr_reset_info_get Obtains the complete Reset Manager reset information.
• dif_rstmgr_reset_is_locked Checks whether requested peripheral reset functionality is locked.
• dif_rstmgr_reset_lock Locks out requested peripheral reset functionality.
• dif_rstmgr_software_device_reset Software request for system reset.
• dif_rstmgr_software_reset Asserts or de-asserts software reset for the requested peripheral.
• dif_rstmgr_software_reset_is_held Queries whether the requested peripheral is held in reset.

Register Table