README.txt
==========
This directory holds a port of NuttX to the NXP/Freescale Sabre board
featuring the iMX 6Quad CPU.
Contents
========
- Status
- Platform Features
- Serial Console
- LEDs and Buttons
- Using U-Boot to Run NuttX
- Debugging with the Segger J-Link
- SMP
- Configurations
Status
======
2016-02-28: The i.MX6Q port is just beginning. A few files have been
populated with the port is a long way from being complete or even ready to
begin any kind of testing.
2016-03-12: The i.MX6Q port is code complete including initial
implementation of logic needed for CONFIG_SMP=y . There is no clock
configuration logic. This is probably not an issue if we are loaded into
SDRAM by a bootloader (because we cannot change the clocking anyway in
that case).
There is a lot of testing that could be done but, unfortunately, I still
have no i.MX6 hardware to test on.
In additional to the unexpected issues, I do expect to run into some
cache coherency issues when I get to testing an SMP configuration.
2016-03-28: I now have a used MCIMX6Q-SDB which is similar to the target
configuration described below except that it does not have the 10.1" LVDS
display. Next step: Figure out how to run a copy of NuttX using U-Boot.
2016-03-31: Most all of the boot of the NSH configuration seems to be
working. It gets to NSH and NSH appears to run normally. Non-interrupt
driver serial output to the VCOM console is working (llsyslog). However,
there does not appear to be any interrupt activity: No timer interrupts,
no interrupt driver serial console output (syslog, printf).
2016-05-16: I now get serial interrupts (but not timer interrupts). This
involves a few changes to GIC bit settings that I do not fully understand.
With this change, the NSH serial console works:
MX6Q SABRESD U-Boot > ABEFGHILMN
NuttShell (NSH)
nsh>
But there are still no timer interrupts. LEDs do not appear to be working.
2016-05-17: Timer interrupts now work. This turned out to be just a minor
bit setting error in the timer configuration. LEDs were not working simply
because board_autoled_initialize() was not being called in the board startup
logic.
At this point, I would say that the basic NSH port is complete.
2016-05-18: Started looking at the SMP configuration. Initially, I verfied
that the NSH configuration works with CONFIG_SMP_NCPUS=1. Not a very
interesting case, but this does exercise a lot of the basic SMP logic.
When more than one CPU is configured, then there are certain failures that
appear to be stack corruption problem. See the open issues below under
SMP.
2016-05-22: In a simple NSH case, SMP does not seem to be working. But there
are known SMP open issues so I assume if the tasking were stressed more there
would be additional failures. See the open issues below under SMP.
An smp configuration was added. This is not quite the same as the
configuration that I used for testing. I enabled DEBUG output, ran with
only 2 CPUS, and disabled the RAMLOG:
+CONFIG_DEBUG_FEATURES=y
+CONFIG_DEBUG_INFO=y
+CONFIG_DEBUG_SCHED=y
+CONFIG_DEBUG_SYMBOLS=y
-CONFIG_DEBUG_FULLOPT=y
+CONFIG_DEBUG_NOOPT=y
-CONFIG_SMP_NCPUS=4
+CONFIG_SMP_NCPUS=2
-CONFIG_RAMLOG=y
-CONFIG_RAMLOG_SYSLOG=y
-CONFIG_RAMLOG_BUFSIZE=16384
-CONFIG_RAMLOG_NONBLOCKING=y
-CONFIG_RAMLOG_NPOLLWAITERS=4
I would also disable debug output from CPU0 so that I could better see the
debug output from CPU1. In drivers/syslog/vsyslog.c:
+if (up_cpu_index() == 0) return 17; // REMOVE ME
Platform Features
=================
Processor:
- i.MX 6Quad or 6DualLite 1 GHz ARM Cortex-A9 processor
Memory/storage:
- 1 GB DDR3 SDRAM up to 533 MHz (1066 MTPS) memory
- 8 GB eMMC flash
- 4 MB SPI NOR flash
Display:
- 10.1" 1024 x 768 LVDS display with integrated P-cap sensing
- HDMI connector
- LVDS connector (for optional second display)
- LCD expansion connector (parallel, 24-bit)
- EPDC expansion connector (for 6DualLite only)
- MIPI DSI connector (two data lanes, 1 GHz each)
User Interface:
- 10.1" capacitive multitouch display
- Buttons: power, reset, volume
Power Management:
- Proprietary PF0100 PMIC
Audio:
- Audio codec
- 2x digital microphones
- 2x 3.5 mm audio ports
- Dual 1 watt speakers
Expansion Connector:
- Camera MIPI CSI port
- I2C, SPI signals
Connectivity:
- 2x full-size SD/MMC card slots
- 7-pin SATA data connector
- 10/100/1000 Ethernet port
- 1x USB 2.0 OTG port (micro USB)
Debug:
- JTAG connector (20-pin)
- 1x Serial-to-USB connector (for JTAG)
OS Support:
- Linux<75> and Android<69> from NXP/Freescale
- Others supported via third party (QNX, Windows Embedded)
Tools Support:
- Manufacturing tool from NXP/Freescale
- IOMUX tool from NXP/Freescale
- Lauterbach, ARM (DS-5), IAR and Macraigor
Additional Features:
- Proprietary 3-axis accelerometer
- Proprietary 3D magnetometer
- Ambient light sensor
- GPS receiver module
- 2x 5MP cameras
- Battery charger
- Battery connectors (battery not included)
Serial Console
==============
A DEBUG VCOM is available MICRO USB AB 5 J509. This corresponds to UART1
from the i.MX6. UART1 connects to J509 via the CSIO_DAT10 and CSIO_DAT11
pins
LEDs and Buttons
================
LEDs
----
A single LED is available driven GPIO1_IO02. On the schematic this is
USR_DEF_RED_LED signal to pin T1 (GPIO_2). This signal is shared with
KEY_ROW6 (ALT2). A high value illuminates the LED.
This LED is not used by the board port unless CONFIG_ARCH_LEDS is
defined. In that case, the usage by the board port is defined in
include/board.h and src/sam_autoleds.c. The LED is used to encode
OS-related events as follows:
------------------- ----------------------- ------
SYMBOL Meaning LED
------------------- ----------------------- ------
LED_STARTED NuttX has been started OFF
LED_HEAPALLOCATE Heap has been allocated OFF
LED_IRQSENABLED Interrupts enabled OFF
LED_STACKCREATED Idle stack created ON
LED_INIRQ In an interrupt N/C
LED_SIGNAL In a signal handler N/C
LED_ASSERTION An assertion failed N/C
LED_PANIC The system has crashed FLASH
Thus if the LED is statically on, NuttX has successfully booted and is,
apparently, running normally. If the LED is flashing at approximately
2Hz, then a fatal error has been detected and the system has halted.
Buttons
-------
Using U-Boot to Run NuttX
=========================
The MCIMX6Q-SDB comes with a 8GB SD card containing the U-Boot and Android.
You simply put the SD card in the SD card slot SD3 (on the bottom of the
board next to the HDMI connect) and Android will boot.
But we need some other way to boot NuttX. Here are some things that I have
experimented with.
Building U-Boot (Failed Attempt #1)
-----------------------------------
I have been unsuccessful getting building a working version of u-boot from
scratch. It builds, but it does not run. Here are the things I did:
1. Get a copy of the u-boot i.MX6 code via:
https://github.com/boundarydevices/u-boot-imx6/tree/production
or
$ git clone git://git.denx.de/u-boot.git
2. Build U-Boot for the i.MX6Q Sabre using the following steps. This
assumes that you have the path to your arm-none-eabi- toolchain at the
beginning of your PATH variable:
$ cd u-boot
$ export ARCH=arm
$ export CROSS_COMPILE=arm-none-eabi-
$ make mx6qsabresd_config
$ make
This should create a number of files, including u-boot.imx
3. Format an SD card
Create a FAT16 partition at an offset of about 1MB into the SD card.
This is where we will put nuttx.bin.
4. Put U-Boot on SD. U-boot should reside at offset 1024B of your SD
card. To put it there, do:
$ dd if=u-boot.imx of=/dev/<your-sd-card> bs=1k seek=1
$ sync
Your SD card device is typically something in /dev/sd<X> or
/dev/mmcblk<X>. Note that you need write permissions on the SD card
for the command to succeed, so you might need to su - as root, or use
sudo, or do a chmod a+w as root on the SD card device node to grant
permissions to users.
Using the Other SD Card Slot (Failed Attempt #2)
------------------------------------------------
Another option is to use the version u-boot that came on the 8GB but put
NuttX on another SD card inserted in the other SD card slot at the opposite
corner of the board.
To make a long story short: This doesn't work. As far as I can tell,
U-Boot does not support any other other SC card except for mmc 2 with is the
boot SD card slot.
Replace Boot SD Card (Successful Attempt #3)
--------------------------------------------
What if you remove the SD card after U-boot has booted, then then insert
another SD card containing the nuttx.bin image?
1. Build nuttx.bin and copy it only a FAT formated SD card. Insert the SD
card containing NuttX into the "other" SD card slot. Insert the 8GB SD
card with U-boot already on it in the normal, boot SD card slot.
2. Connect the VCOM port using the USB port next to the boot SD card slot.
3. Start a console at 11500 8N1 on the VCOM port
4. Power up the board with the 8GB SD card in place. U-Boot will start and
countdown before starting Linux. Press enter to break into U-Boot before
Linux is started.
5. Remove the 8GB U-Boot SD card; insert in its place.
6. Rescan the SD card:
MX6Q SABRESD U-Boot > mmc dev 2
mmc2 is current device
MX6Q SABRESD U-Boot > mmc rescan
MX6Q SABRESD U-Boot > fatls mmc 2
system volume information/
87260 nuttx.bin
1 file(s), 1 dir(s)
7. Then we can boot NuttX off the rescanned SD card:
MX6Q SABRESD U-Boot > fatload mmc 2 0x10800000 nuttx.bin
reading nuttx.bin
87260 bytes read
MX6Q SABRESD U-Boot > go 0x10800040
## Starting application at 0x10800040 ...
That seems to work okay.
Use the FAT Partition on the 8GB SD Card (Untested Idea #4)
-----------------------------------------------------------
Partition 4 on the SD card is an Android FAT file system. So one thing you
could do would be put the nuttx.bin file on that partition, then boot like:
MX6Q SABRESD U-Boot > fatload mmc 2:4 0x10800000 nuttx.bin
SD Card Image Copy (Successful Attempt #5)
-------------------------------------
You can use the 'dd' command to copy the first couple of megabytes from the
8GB SD card and copy that to another SD card. You then have to use 'fdisk'
to fix the partition table and to add a single FAT16 partition at an offset
of 1MB or so.
1. Insert the 8GB boot SD card into your PC: Copy the first 2Mb from the SD
card to a file:
$ dd if=/dev/sdh of=sdh.img bs=512 count=4096
2. Remove the 8GB boot SD card and replace it with a fresh SD card. Copy the
saved file to the first the new SD card:
$ dd of=/dev/sdh if=sdh.img bs=512 count=4096
3. Then use 'fdisk' to:
- Remove all of the non-existent partitions created by the 'dd' copy.
- Make a single FAT16 partition at the end of the SD card.
You will also need to format the partion for FAT.
4. You can put nuttx.bin here and then boot very simply with:
MX6Q SABRESD U-Boot > fatload mmc 2:1 0x10800000 nuttx.bin
MX6Q SABRESD U-Boot > go 0x10800040
A little hokey, but not such a bad solution.
Debugging with the Segger J-Link
================================
These procedures work for debugging the boot-up sequence when there is a
single CPU running and not much else going on. If you want to do higher
level debugger, you will need something more capable. NXP/Freescale suggest
some other debuggers that you might want to consider.
These instructions all assume that you have built NuttX with debug symbols
enabled. When debugging the nuttx.bin file on the SD card, it is also
assumed the the nuttx ELF file with the debug symbol addresses is from the
same build so that the symbols match up.
Debugging the NuttX image on the SD card
----------------------------------------
1. Connect the J-Link to the 20-pin JTAG connector.
2. Connect the "USB TO UART" USB VCOM port to the host PC. Start a
terminal emulation program like TeraTerm on Minicom. Select the USB
VCOM serial port at 115200 8N1.
When you apply power to the board, you should see the U-Boot messages in
the terminal window. Stop the U-Boot countdown to get to the U-Boot
prompt.
2. Start the Segger GDB server:
Target: MCIMX6Q6
Target Interface: JTAG
If the GDB server starts correctly you should see the following in the
Log output:
Waiting for GDB Connection
3. In another Xterm terminal window, start arm-none-eabi-gdb and connect to
the GDB server.
From the Xterm Window:
$ arm-none-eabi-gdb
You will need to have the path to the arm-none-eabi-gdb program in your
PATH variable.
Then from GDB:
gdb> target connect localhost:2331
gdb> mon halt
4. Start U-boot under GDB control:
From GDB:
gdb> mon reset
gdb> mon go
Again stop the U-Boot countdown to get to the U-Boot prompt.
5. Load NuttX from the SD card into RAM
From U-Boot:
MX6Q SABRESD U-Boot > fatload mmc 2:1 0x10800000 nuttx.bin
6. Load symbols and set a breakpoint
From GDB:
gdb> mon halt
gdb> file nuttx
gdb> b __start
gdb> c
__start is the entry point into the NuttX binary at 0x10800040. You can,
of course, use a different symbol if you want to start debugging later
in the boot sequence.
7. Start NuttX
From U-Boot:
MX6Q SABRESD U-Boot > go 0x10800040
8. You should hit the breakpoint that you set above and be off and
debugging.
Debugging a Different NuttX Image
---------------------------------
Q: What if I want do run a different version of nuttx than the nuttx.bin
file on the SD card. I just want to build and debug without futzing with
the SD card. Can I do that?
A: Yes with the following modifications to the prodecure above.
- Skip step 5, don't bother to load NuttX into RAM
- In step 6, load NuttX into RAM like this:
gdb> mon halt
gdb> load nuttx <-- Loads NuttX into RAM at 0x010800000
gdb> file nuttx
gdb> b __start
gdb> c
- Then after step 7, you should hit the breakpoint at the instruction you
just loaded at address 0x10800040.
- Or, in step 6, instead of continuing ('c') which will resume U-Boot,
even just:
gdb> mon halt
gdb> load nuttx <-- Loads NuttX into RAM at 0x010800000
gdb> file nuttx
gdb> mon set pc 0x10800040
gdb> s
The final single will then step into the freshly loaded program.
You can then forget about steps 7 and 8.
This is, in fact, my preferred way to debug.
You can restart the debug session at any time at the gdb> prompt by:
gdb> mon reset
gdb> mon go
That will restart U-Boot and you have to press ENTER in the terminal
window to stop U-Boot. Restarting U-Boot is a necesary part of the
restart process because you need to put the hardware back in its initial
state before running NuttX
Then this will restart the debug session just as before:
gdb> mon halt
gdb> load nuttx <-- Loads NuttX into RAM at 0x010800000
gdb> file nuttx
gdb> mon set pc 0x10800040
gdb> s
SMP
===
The i.MX6 6Quad has 4 CPUs. Support is included for testing an SMP
configuration. That configuration is still not yet ready for usage but can
be enabled with the following configuration settings:
RTOS Features -> Tasks and Scheduling
CONFIG_SPINLOCK=y
CONFIG_SMP=y
CONFIG_SMP_NCPUS=4
CONFIG_SMP_IDLETHREAD_STACKSIZE=2048
Open Issues:
1. Currently all device interrupts are handled on CPU0 only. Critical sections will
attempt to disable interrupts but will now disable interrupts only on the current
CPU (which may not be CPU0). There is a spinlock to prohibit entrance into these critical sections in interrupt handlers of other CPUs.
When the critical section is used to lock a resource that is also used by
interupt handling, the interrupt handling logic must also take the spinlock.
This will cause the interrupt handlers on other CPUs to spin until
leave_critical_section() is called. More verification is needed, however.
2. Cache Concurency. This is a complex problem. There is logic in place now to
clean CPU0 D-cache before starting a new CPU and for invalidating the D-Cache
when the new CPU is started. REVISIT: Seems that this should not be necessary.
If the Shareable bit set in the MMU mappings and my understanding is that this
should keep cache coherency at least within a cluster. I need to study more
how the inner and outer shareable attribute works to control cacheing
But there may are many, many more such cache coherency issues if I cannot find
a systematic way to manage cache coherency.
http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dht0008a/CJABEHDA.html
http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.den0024a/CEGDBEJE.html
Try:
--- mmu.h.orig 2016-05-20 13:09:34.773462000 -0600
+++ mmu.h 2016-05-20 13:03:13.261978100 -0600
@@ -572,8 +572,14 @@
#define MMU_ROMFLAGS (PMD_TYPE_SECT | PMD_SECT_AP_R1 | PMD_CACHEABLE | \
PMD_SECT_DOM(0))
-#define MMU_MEMFLAGS (PMD_TYPE_SECT | PMD_SECT_AP_RW1 | PMD_CACHEABLE | \
+#ifdef CONFIG_SMP
+
+# define MMU_MEMFLAGS (PMD_TYPE_SECT | PMD_SECT_AP_RW1 | PMD_CACHEABLE | \
+ PMD_SECT_S | PMD_SECT_DOM(0))
+#else
+# define MMU_MEMFLAGS (PMD_TYPE_SECT | PMD_SECT_AP_RW1 | PMD_CACHEABLE | \
PMD_SECT_DOM(0))
+#endif
#define MMU_IOFLAGS (PMD_TYPE_SECT | PMD_SECT_AP_RW1 | PMD_DEVICE | \
PMD_SECT_DOM(0) | PMD_SECT_XN)
#define MMU_STRONGLY_ORDERED (PMD_TYPE_SECT | PMD_SECT_AP_RW1 | \
Another alternative would be to place all spinlocks in a non-cachable memory
region. That is problem what will have to be done.
This is a VERIFIED PROBLEM: I have seen cases where CPU0 sets a spinlock=1 then
tries to lock the spinlock. CPU0 will wait in this case until CPU1 unlocks the
spinlock. Most of this happens correctly; I can see that CPU1 does set the
spinlock=0, but CPU0 never sees the change and spins forever. That is surely
a consequence of cache issues.
This was observed between up_cpu_pause() and arm_pause_handler() with the
spinlock "g_cpu_paused[cpu]". CPU1 correctly sets g_cpu_paused[cpu] to zero
but CPU0 never sees the change.
3. Caching probabaly interferes with spinlocks as they are currently implemented.
Waiting on a cached copy of the spinlock may result in a hang or a failure to
wait.
Should all spinlocks go into a special "strongly ordered" memory region?
Update: Cache inconsistencies seem to be the root cause of all current SMP
issues.
Configurations
==============
Information Common to All Configurations
----------------------------------------
Each Sabre-6Quad configuration is maintained in a sub-directory and
can be selected as follow:
cd tools
./configure.sh sabre-6quad/<subdir>
cd -
. ./setenv.sh
Before sourcing the setenv.sh file above, you should examine it and perform
edits as necessary so that TOOLCHAIN_BIN is the correct path to the directory
than holds your toolchain binaries.
And then build NuttX by simply typing the following. At the conclusion of
the make, the nuttx binary will reside in an ELF file called, simply, nuttx.
make oldconfig
make
The <subdir> that is provided above as an argument to the tools/configure.sh
must be is one of the following.
NOTES:
1. These configurations use the mconf-based configuration tool. To
change any of these configurations using that tool, you should:
a. Build and install the kconfig-mconf tool. See nuttx/README.txt
see additional README.txt files in the NuttX tools repository.
b. Execute 'make menuconfig' in nuttx/ in order to start the
reconfiguration process.
2. Unless stated otherwise, all configurations generate console
output on UART1 which is a available to the host PC from the USB
micro AB as a VCOM part.
3. All of these configurations are set up to build under Windows using the
"GNU Tools for ARM Embedded Processors" that is maintained by ARM
(unless stated otherwise in the description of the configuration).
https://launchpad.net/gcc-arm-embedded
That toolchain selection can easily be reconfigured using
'make menuconfig'. Here are the relevant current settings:
Build Setup:
CONFIG_HOST_WINDOWS=y : Window environment
CONFIG_WINDOWS_CYGWIN=y : Cywin under Windows
System Type -> Toolchain:
CONFIG_ARMV7M_TOOLCHAIN_GNU_EABIW=y : GNU ARM EABI toolchain
Configuration sub-directories
-----------------------------
nsh
---
This is a NuttShell (NSH) configuration that uses the NSH library
at apps/nshlib with the start logic at apps/examples/nsh.
NOTES:
1. This configuration assumes that we are loaded into SDRAM and
started via U-Boot.
2. The serial console is configured by default for use UART1, the
USB VCOM port (UART1), same as the serial port used by U-Boot.
You will need to reconfigure if you want to use a different UART.
3. NSH built-in applications are supported, but no built-in
applications are enabled.
Binary Formats:
CONFIG_BUILTIN=y : Enable support for built-in programs
Application Configuration:
CONFIG_NSH_BUILTIN_APPS=y : Enable starting apps from NSH command line
4. The RAMLOG is enabled. All SYSLOG (DEBUG) output will go to the
RAMLOG and will not be visible unless you use the nsh 'dmesg'
command. To disable this RAMLOG feature, disable the following:
Device Drivers: CONFIG_RAMLOG
smp
---
This is a configuration of testing the SMP configuration. It is
essentially equivalent to the SMP configuration except has SMP enabled.
NOTES:
1. See the notest for the nsh configuration. Since this configuration
is essentially the same all of those comments apply.
2. SMP is not fully functional. See the STATUS and SMP sections above
for detailed SMP-related issues.
