-
Notifications
You must be signed in to change notification settings - Fork 51
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge branch 'work/AM64x-cortexr5-runtime' into 'master'
Add TI AM64x/AM263x Arm Cortex-R5 Light-Tasking Runtime See merge request eng/toolchain/bb-runtimes!50
- Loading branch information
Showing
24 changed files
with
3,795 additions
and
98 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,274 @@ | ||
TI AM64x/AM263x Arm Cortex-R5 Runtime | ||
===================================== | ||
|
||
Runtimes Supported | ||
------------------ | ||
|
||
* Light-Tasking | ||
|
||
Targets Supported | ||
----------------- | ||
|
||
* TI AM64x/AM263x Arm Cortex-R5F Subsystem (R5FSS) | ||
- The runtime can run on any R5FSS CPU without modification. | ||
- Symmetric multiprocessing (SMP) is not supported. | ||
- Separate applications can be run on different CPUs. | ||
|
||
Using the runtimes | ||
------------------ | ||
|
||
As a prerequisite, GNAT Pro for ARM ELF needs to be installed. To then use the | ||
runtime copy it to the `<arm-elf gnat install>/arm-eabi/lib/gnat/` folder. | ||
You can then select the runtime in GNAT Studio from the Toolchain pane of the | ||
Project Properties dialog box or set the following in your GPR file: | ||
|
||
for Runtime ("ada") use "light-tasking-am64xr5"; | ||
|
||
Alternatively, you can locate the runtime in another directory and then in the | ||
project file provide the full or relative path from the project file: | ||
|
||
for Runtime ("ada") use "/path/to/runtime/light-tasking-am64xr5"; | ||
|
||
Rebuilding the Runtime | ||
---------------------- | ||
|
||
If you need to make changes to the runtime's BSP you can rebuild it as follows: | ||
|
||
For the Light-Tasking runtime: gprbuild -P ravenscar_build.gpr | ||
|
||
Note: if you were provided the runtime in a certification context, please reach | ||
out to AdaCore before modifying the runtime. | ||
|
||
Resources used by the runtime | ||
----------------------------- | ||
|
||
- The Global Timebase Counter as the source for Ada.Real_Time.Clock; | ||
- Timers 0 and 1 as the alarm source for R5FSS0 CPU 0 and 1 respectively; | ||
- Timers 2 and 3 as the alarm source for R5FSS1 CPU 0 and 1 respectively; | ||
- The local Vectored Interrupt Manager (VIM) controller for interrupt support; | ||
- UART0 for Ada.Text_IO. | ||
|
||
Interrupt sources TIMER0_INTR_PEND_0, TIMER1_INTR_PEND_0, TIMER2_INTR_PEND_0, | ||
TIMER3_INTR_PEND_0 are reserved for runtime use and should not be used by user | ||
applications. | ||
|
||
See the respective sections below for further details on each resource used by | ||
the GNAT runtime. | ||
|
||
System Clocks | ||
------------- | ||
|
||
The AM64x/AM263x R5FSS runtime uses two clock sources: | ||
|
||
* Global Timebase Counter (GTC): high resolution counter used for | ||
Ada.Real_Time.Clock. | ||
* Timers: used for delays and timing events. | ||
|
||
The GTC is a read-only monotonic clock shared among all CPUs. The runtime will | ||
enable the GTC if it's not enabled. User software should not disable the GTC nor | ||
modify the GTC after the runtime starts, instead configuration of the GTC | ||
should be performed by the SBL. The GTC clock frequency is defined in the | ||
runtime package System.BB.Board_Parameters (`s-bbbopa.ads`) in the gnat | ||
directory by the constant GTC_Frequency. By default, the GTC_Frequency is | ||
defined as 225 MHz and can be updated by users by modifying this value and | ||
rebuilding the runtime. | ||
|
||
Timers 0 and 1 are used as the alarm source when the runtime is run on R5FSS0 | ||
CPU 0 and CPU 1 respectively, while Timers 2 and 3 are used as the alarm source | ||
for R5FSS1 CPU 0 and CPU 1. The runtime will determine the appropriate Timer | ||
to use at runtime, no configuration is necessary. The runtime configures each | ||
Timer to use HFOSC0_CLKOUT as its clock input source. The frequency of | ||
HFOSC0_CLKOUT is defined in System.BB.Board_Parameters as 25 MHz and can be | ||
changed by users to suit their board if required. | ||
|
||
Startup Code | ||
------------ | ||
|
||
The TI AM64x/AM263x R5FSS runtime provides low-level startup code in `crt0.S` | ||
that contains the entry point of the application to be called from your | ||
bootloader and performs the low-level runtime initialization required before | ||
high-level Ada code is called. The startup code performs the following | ||
operations: | ||
|
||
* initialization of core registers; | ||
* Enabling of the FPU; | ||
* Clearing the BSS section; | ||
* Loading of the DATA section to RAM if required; | ||
* Configuration of the MPU. | ||
|
||
Outside of the startup code, the runtime will also configure the CPU's | ||
respective Timer and local VIM. All other SoC initialization and configuration | ||
needs to be carried out by the Secondary Bootloader (SBL) prior to passing | ||
control over to the application. The SBL is also required to load the | ||
application into memory if required. Any peripherals the application may use | ||
should be configured by the SBL or the application directly prior to their use. | ||
|
||
The configuration of the MPU is located in System.Board_Parameters | ||
(`s-boapar`). This package can be modified directly by non-certification users | ||
to suit your needs. For certification users who cannot modify the runtime you | ||
should configure the MPU directly yourself by implementing your own version of | ||
the parameterless function `__gnat_mpu_init` in your application. At link time | ||
your application will use your version of `__gnat_mpu_init` instead of the | ||
runtime provided version. | ||
|
||
In limited circumstances, `crt0.S` may be replaced with your own code that | ||
performs the above requirement and directly calls `main`, but please contact | ||
AdaCore if you need to modify the FPU settings as doing so may affect the | ||
conformance of the runtime to the Ada Standard. | ||
|
||
Memory | ||
------ | ||
|
||
The application memory is broken down into three primary memory regions: | ||
|
||
* Boot and Vector Code ("Boot Region") | ||
* Code, read-only data and the initial read-write data ("Code Region") | ||
* Data and bss (zeroed data) sections ("Data Region") | ||
|
||
A set of default linker scripts are provided with the runtime that locate | ||
the Boot Region at the start of TCM A and the Code and Data regions into MSRAM. | ||
These linker scripts can be found in the `ld` directory and consist of the | ||
following scripts: | ||
|
||
* `memmap.ld`: defines the layout of the AM64x/AM263x memory. | ||
* `ram.ld`: maps AM64x/AM263x memory to the memory regions defined above. This | ||
is the main linker script that includes the other two scripts. | ||
* `common.ld`: defines the application entry point and sections. It uses the | ||
memory regions defined in `ram.ld` to allow this script to be used in | ||
different scenarios without modification. | ||
|
||
While you can modify these scripts directly, it's recommended to copy the | ||
scripts you need to modify to your project folder and modify them there. This | ||
allows you to preserve your scripts between runtime updates and is required if | ||
you will be using the runtime to run applications on different cores (as each | ||
application will need to be mapped to different memory locations). | ||
|
||
To use your own linker scripts, specify the main linker script via the `-T` | ||
linker switch and use the `-XLOADER=USER` switch when building your application | ||
with GPRbuild. | ||
|
||
When relocating the primary memory regions, please keep in mind the following | ||
guidance for each region. | ||
|
||
Boot Region | ||
~~~~~~~~~~~~ | ||
|
||
The Boot Region contains the exception vector table, start code and hardware | ||
exception handlers. The exception vector table appears at the start of this | ||
region and consequently the region needs to be mapped to 0x00000000 if the | ||
R5FSS core is booted from TCM. The AM64x/AM263x also supports remapping the | ||
boot vector elsewhere in the SoC memory map by programming the | ||
CTRLMMR_SEC_CLSTR0_CORE0_BOOTVECT_LO and CTRLMMR_SEC_CLSTR0_CORE0_BOOTVECT_HI | ||
registers. In this scenario, the Boot Region should be placed at the start of | ||
this boot vector address. | ||
|
||
This region needs to be loaded into memory by the SBL. | ||
|
||
Code Region | ||
~~~~~~~~~~~ | ||
|
||
The Code Region contains the application code, read-only data and the initial | ||
state of the data section. This region can be located in RAM or read-only | ||
memory like FLASH. If located in RAM it is the bootloader's responsibility to | ||
load the Code Region to memory. | ||
|
||
Data Region | ||
~~~~~~~~~~~ | ||
|
||
The Data Region contains the data and bss sections. The region needs to be | ||
located in RAM, which can either be the SoC's internal TCM or SRAM, or external | ||
DDR memory. During runtime startup the runtime will zero the bss section and | ||
copy the initial state of the data section from the Code Region if the two | ||
regions are not located together. There is no need for the SBL to load the | ||
Data Region into RAM. | ||
|
||
Caches | ||
~~~~~~ | ||
|
||
The runtime enables the data and instruction cache during initialization. | ||
|
||
Interrupts | ||
---------- | ||
|
||
External Interrupts | ||
~~~~~~~~~~~~~~~~~~~ | ||
|
||
The AM64x/AM263x R5FSS runtime supports connecting IRQ interrupts generated by | ||
the CPU's Vectored Interrupt Manager (VIM) to protected procedure handlers as | ||
documented in the Ada Reference Manual C.3.1. The list of interrupt names can | ||
be found in the package Ada.Interrupts.Names (`gnarl/a-intnam.ads`). | ||
|
||
Due to the design of the VIM, all interrupts are mapped to a single Ada | ||
Interrupt Priority and nested priorities are not supported. Interrupts handled | ||
by the runtime use a dedicated stack and the FPU is enabled, allowing the | ||
use of floating-point instructions within an interrupt handler. The runtime | ||
will initialize the CPU's VIM during runtime initialization and the VIM is | ||
configured to use the vector interface (VIC). | ||
|
||
Additional interrupt functionality is provided through the GNAT.Interrupts | ||
(`gnarl/g-interr.ads`) package: | ||
|
||
* Enabling and disabling interrupts at the VIM; | ||
* Raising a software interrupt (for testing); | ||
* Setting the type of interrupt event (level (default) or pulse); | ||
* Setting the priority of the interrupt at the VIM. | ||
|
||
Note that since the runtime does not support nested priorities, the VIM | ||
interrupt priorities are only used to determine which interrupt has precedence | ||
to be serviced when there are multiple pending interrupts. | ||
|
||
Additionally, GNAT.Interrupts supports direct attachment of IRQ and FIQ | ||
interrupt handlers to the VIM that bypasses the GNAT runtime. This allows users | ||
to provide low latency handlers. Direct attachment of handlers comes with the | ||
following limitations: | ||
|
||
* Nesting of interrupts is not supported; | ||
* Interrupts must not be reenabled while in a handler; | ||
* Handler *must* not make any runtime calls, including calling protected | ||
subprograms; | ||
* Handler should not use the FPU; | ||
* Users are responsible for saving and restoring register state, and | ||
returning from the interrupt. This can be done by using the "interrupt" | ||
machine attribute on the handler procedure; | ||
* Users must follow TI's documentation on "Servicing IRQ Through Vector | ||
Interface" for IRQ handlers and "Servicing FIQ" for FIQ handlers. | ||
Use the provided procedures Clear_VIM_Status and Clear_Priority_Mask to | ||
clear the interrupt status and VIM priority. | ||
|
||
Cortex R5 Core Interrupts | ||
~~~~~~~~~~~~~~~~~~~~~~~~~ | ||
|
||
Undefined instruction exceptions, prefetch abort and data abort exceptions | ||
will call the GNAT Last Chance Handler with a message indicating which | ||
exception was raised. By default, the Last Chance Handler will print out this | ||
exception message and loop. It is expected that the default Last Chance Handler | ||
is replaced with your application specific Last Chance Handler. | ||
|
||
Power Management | ||
---------------- | ||
|
||
The runtime will place the CPU into standby mode if there are no tasks eligible | ||
to run. | ||
|
||
Text_IO | ||
------- | ||
|
||
The runtime provides a minimal version of the Ada.Text_IO package supporting | ||
character- and string-based input and output routines for basic I/O needs. It | ||
is recommended to implement your own I/O packages based around your I/O | ||
channel of choice. | ||
|
||
The bodies of the Ada.Text_IO routines call through to a device-specific I/O | ||
package named System.Text_IO. See the package body in the file `s-textio.adb` | ||
in the gnat directory for more details. | ||
|
||
System.Text_IO on the AM64x/AM263x R5FSS runtime is configured by default to | ||
use UART0. UART0 is configured for 115200 baud rate, one stop bit, no parity | ||
using the 48 MHz clock source. | ||
|
||
Note: System.Text_IO does not provide any synchronised access to UART0 across | ||
R5FSS and A53SS CPUs. If you deploy multiple applications using the runtime, | ||
ensure only one application uses Ada.Text_IO. If your other applications | ||
require Text_IO services, have these applications interface directly with | ||
different I/O peripherals, or use the AM64x/AM243x IPC facilities to relay | ||
messages through the application designated to communicate with UART0. |
Oops, something went wrong.