Sandbox

Native Execution of U-Boot

The ‘sandbox’ architecture is designed to allow U-Boot to run under Linux on almost any hardware. To achieve this it builds U-Boot (so far as possible) as a normal C application with a main() and normal C libraries.

All of U-Boot’s architecture-specific code therefore cannot be built as part of the sandbox U-Boot. The purpose of running U-Boot under Linux is to test all the generic code, not specific to any one architecture. The idea is to create unit tests which we can run to test this upper level code.

Sandbox allows development of many types of new features in a traditional way, rather than needing to test each iteration on real hardware. Many U-Boot features were developed on sandbox, including the core driver model, most uclasses, verified boot, bloblist, logging and dozens of others. Sandbox has enabled many large-scale code refactors as well.

CONFIG_SANDBOX is defined when building a native board.

The board name is ‘sandbox’ but the vendor name is unset, so there is a single board in board/sandbox.

CONFIG_SANDBOX_BIG_ENDIAN should be defined when running on big-endian machines.

There are two versions of the sandbox: One using 32-bit-wide integers, and one using 64-bit-wide integers. The 32-bit version can be build and run on either 32 or 64-bit hosts by either selecting or deselecting CONFIG_SANDBOX_32BIT; by default, the sandbox it built for a 32-bit host. The sandbox using 64-bit-wide integers can only be built on 64-bit hosts.

Note that standalone/API support is not available at present.

Prerequisites

Install the dependencies noted in Building with GCC.

Basic Operation

To run sandbox U-Boot use something like:

make sandbox_defconfig all
./u-boot

Note: If you get errors about ‘sdl-config: Command not found’ you may need to install libsdl2.0-dev or similar to get SDL support. Alternatively you can build sandbox without SDL (i.e. no display/keyboard support) by removing the CONFIG_SANDBOX_SDL line in include/configs/sandbox.h or using:

make sandbox_defconfig all NO_SDL=1
./u-boot

U-Boot will start on your computer, showing a sandbox emulation of the serial console:

U-Boot 2014.04 (Mar 20 2014 - 19:06:00)

DRAM:  128 MiB
Using default environment

In:    serial
Out:   lcd
Err:   lcd
=>

You can issue commands as your would normally. If the command you want is not supported you can add it to include/configs/sandbox.h.

To exit, type ‘poweroff’ or press Ctrl-C.

Console / LCD support

Assuming that CONFIG_SANDBOX_SDL is defined when building, you can run the sandbox with LCD and keyboard emulation, using something like:

./u-boot -d u-boot.dtb -l

This will start U-Boot with a window showing the contents of the LCD. If that window has the focus then you will be able to type commands as you would on the console. You can adjust the display settings in the device tree file - see arch/sandbox/dts/sandbox.dts.

Command-line Options

Various options are available, mostly for test purposes. Use -h to see available options. Some of these are described below:

-t, --terminal <arg>

The terminal is normally in what is called ‘raw-with-sigs’ mode. This means that you can use arrow keys for command editing and history, but if you press Ctrl-C, U-Boot will exit instead of handling this as a keypress. Other options are ‘raw’ (so Ctrl-C is handled within U-Boot) and ‘cooked’ (where the terminal is in cooked mode and cursor keys will not work, Ctrl-C will exit).

-l

Show the LCD emulation window.

-d <device_tree>

A device tree binary file can be provided with -d. If you edit the source (it is stored at arch/sandbox/dts/sandbox.dts) you must rebuild U-Boot to recreate the binary file.

-D

To use the default device tree, use -D.

-T

To use the test device tree, use -T.

-c [<cmd>;]<cmd>

To execute commands directly, use the -c option. You can specify a single command, or multiple commands separated by a semicolon, as is normal in U-Boot. Be careful with quoting as the shell will normally process and swallow quotes. When -c is used, U-Boot exits after the command is complete, but you can force it to go to interactive mode instead with -i.

-i

Go to interactive mode after executing the commands specified by -c.

Environment Variables

UBOOT_SB_TIME_OFFSET

This environment variable stores the offset of the emulated real time clock to the host’s real time clock in seconds. The offset defaults to zero.

Memory Emulation

Memory emulation is supported, with the size set by CONFIG_SANDBOX_RAM_SIZE_MB. The -m option can be used to read memory from a file on start-up and write it when shutting down. This allows preserving of memory contents across test runs. You can tell U-Boot to remove the memory file after it is read (on start-up) with the –rm_memory option.

To access U-Boot’s emulated memory within the code, use map_sysmem(). This function is used throughout U-Boot to ensure that emulated memory is used rather than the U-Boot application memory. This provides memory starting at 0 and extending to the size of the emulation.

Storing State

With sandbox you can write drivers which emulate the operation of drivers on real devices. Some of these drivers may want to record state which is preserved across U-Boot runs. This is particularly useful for testing. For example, the contents of a SPI flash chip should not disappear just because U-Boot exits.

State is stored in a device tree file in a simple format which is driver- specific. You then use the -s option to specify the state file. Use -r to make U-Boot read the state on start-up (otherwise it starts empty) and -w to write it on exit (otherwise the stored state is left unchanged and any changes U-Boot made will be lost). You can also use -n to tell U-Boot to ignore any problems with missing state. This is useful when first running since the state file will be empty.

The device tree file has one node for each driver - the driver can store whatever properties it likes in there. See ‘Writing Sandbox Drivers’ below for more details on how to get drivers to read and write their state.

Running and Booting

Since there is no machine architecture, sandbox U-Boot cannot actually boot a kernel, but it does support the bootm command. Filesystems, memory commands, hashing, FIT images, verified boot and many other features are supported.

When ‘bootm’ runs a kernel, sandbox will exit, as U-Boot does on a real machine. Of course in this case, no kernel is run.

It is also possible to tell U-Boot that it has jumped from a temporary previous U-Boot binary, with the -j option. That binary is automatically removed by the U-Boot that gets the -j option. This allows you to write tests which emulate the action of chain-loading U-Boot, typically used in a situation where a second ‘updatable’ U-Boot is stored on your board. It is very risky to overwrite or upgrade the only U-Boot on a board, since a power or other failure will brick the board and require return to the manufacturer in the case of a consumer device.

Supported Drivers

U-Boot sandbox supports these emulations:

  • Block devices

  • Chrome OS EC

  • GPIO

  • Host filesystem (access files on the host from within U-Boot)

  • I2C

  • Keyboard (Chrome OS)

  • LCD

  • Network

  • Serial (for console only)

  • Sound (incomplete - see sandbox_sdl_sound_init() for details)

  • SPI

  • SPI flash

  • TPM (Trusted Platform Module)

A wide range of commands are implemented. Filesystems which use a block device are supported.

Also sandbox supports driver model (CONFIG_DM) and associated commands.

Sandbox Variants

There are unfortunately quite a few variants at present:

sandbox:

should be used for most tests

sandbox64:

special build that forces a 64-bit host

sandbox_flattree:

builds with dev_read_…() functions defined as inline. We need this build so that we can test those inline functions, and we cannot build with both the inline functions and the non-inline functions since they are named the same.

sandbox_spl:

builds sandbox with SPL support, so you can run spl/u-boot-spl and it will start up and then load ./u-boot. It is also possible to run ./u-boot directly.

Of these sandbox_spl can probably be removed since it is a superset of sandbox.

Most of the config options should be identical between these variants.

Linux RAW Networking Bridge

The sandbox_eth_raw driver bridges traffic between the bottom of the network stack and the RAW sockets API in Linux. This allows much of the U-Boot network functionality to be tested in sandbox against real network traffic.

For Ethernet network adapters, the bridge utilizes the RAW AF_PACKET API. This is needed to get access to the lowest level of the network stack in Linux. This means that all of the Ethernet frame is included. This allows the U-Boot network stack to be fully used. In other words, nothing about the Linux network stack is involved in forming the packets that end up on the wire. To receive the responses to packets sent from U-Boot the network interface has to be set to promiscuous mode so that the network card won’t filter out packets not destined for its configured (on Linux) MAC address.

The RAW sockets Ethernet API requires elevated privileges in Linux. You can either run as root, or you can add the capability needed like so:

sudo /sbin/setcap "CAP_NET_RAW+ep" /path/to/u-boot

The default device tree for sandbox includes an entry for eth0 on the sandbox host machine whose alias is “eth1”. The following are a few examples of network operations being tested on the eth0 interface.

sudo /path/to/u-boot -D

DHCP
....

setenv autoload no
setenv ethrotate no
setenv ethact eth1
dhcp

PING
....

setenv autoload no
setenv ethrotate no
setenv ethact eth1
dhcp
ping $gatewayip

TFTP
....

setenv autoload no
setenv ethrotate no
setenv ethact eth1
dhcp
setenv serverip WWW.XXX.YYY.ZZZ
tftpboot u-boot.bin

The bridge also supports (to a lesser extent) the localhost interface, ‘lo’.

The ‘lo’ interface cannot use the RAW AF_PACKET API because the lo interface doesn’t support Ethernet-level traffic. It is a higher-level interface that is expected only to be used at the AF_INET level of the API. As such, the most raw we can get on that interface is the RAW AF_INET API on UDP. This allows us to set the IP_HDRINCL option to include everything except the Ethernet header in the packets we send and receive.

Because only UDP is supported, ICMP traffic will not work, so expect that ping commands will time out.

The default device tree for sandbox includes an entry for lo on the sandbox host machine whose alias is “eth5”. The following is an example of a network operation being tested on the lo interface.

TFTP
....

setenv ethrotate no
setenv ethact eth5
tftpboot u-boot.bin

SPI Emulation

Sandbox supports SPI and SPI flash emulation.

The device can be enabled via a device tree, for example:

spi@0 {
        #address-cells = <1>;
        #size-cells = <0>;
        reg = <0 1>;
        compatible = "sandbox,spi";
        cs-gpios = <0>, <&gpio_a 0>;
        spi.bin@0 {
                reg = <0>;
                compatible = "spansion,m25p16", "jedec,spi-nor";
                spi-max-frequency = <40000000>;
                sandbox,filename = "spi.bin";
        };
};

The file must be created in advance:

$ dd if=/dev/zero of=spi.bin bs=1M count=2
$ u-boot -T

Here, you can use “-T” or “-D” option to specify test.dtb or u-boot.dtb, respectively, or “-d <file>” for your own dtb.

With this setup you can issue SPI flash commands as normal:

=>sf probe
SF: Detected M25P16 with page size 64 KiB, total 2 MiB
=>sf read 0 0 10000
SF: 65536 bytes @ 0x0 Read: OK

Since this is a full SPI emulation (rather than just flash), you can also use low-level SPI commands:

=>sspi 0:0 32 9f
FF202015

This is issuing a READ_ID command and getting back 20 (ST Micro) part 0x2015 (the M25P16).

Block Device Emulation

U-Boot can use raw disk images for block device emulation. To e.g. list the contents of the root directory on the second partion of the image “disk.raw”, you can use the following commands:

=>host bind 0 ./disk.raw
=>ls host 0:2

The device can be marked removeable with ‘host bind -r’.

A disk image can be created using the following commands:

$> truncate -s 1200M ./disk.raw
$> echo -e "label: gpt\n,64M,U\n,,L" | /usr/sbin/sgdisk  ./disk.raw
$> lodev=`sudo losetup -P -f --show ./disk.raw`
$> sudo mkfs.vfat -n EFI -v ${lodev}p1
$> sudo mkfs.ext4 -L ROOT -v ${lodev}p2

or utilize the device described in test/py/make_test_disk.py:

#!/usr/bin/python
import make_test_disk
make_test_disk.makeDisk()

Writing Sandbox Drivers

Generally you should put your driver in a file containing the word ‘sandbox’ and put it in the same directory as other drivers of its type. You can then implement the same hooks as the other drivers.

To access U-Boot’s emulated memory, use map_sysmem() as mentioned above.

If your driver needs to store configuration or state (such as SPI flash contents or emulated chip registers), you can use the device tree as described above. Define handlers for this with the SANDBOX_STATE_IO macro. See arch/sandbox/include/asm/state.h for documentation. In short you provide a node name, compatible string and functions to read and write the state. Since writing the state can expand the device tree, you may need to use state_setprop() which does this automatically and avoids running out of space. See existing code for examples.

VPL (Verifying Program Loader)

Sandbox provides an example build of vpl called sandbox_vpl. This can be run using:

/path/to/sandbox_vpl/tpl/u-boot-tpl -D

It starts up TPL (first-stage init), then VPL, then runs SPL and finally U-Boot proper, following the normal flow for a verified boot. At present, no verification is actually implemented.

Debugging the init sequence

If you get a failure in the initcall sequence, like this:

initcall sequence 0000560775957c80 failed at call 0000000000048134 (err=-96)

Then you use can use grep to see which init call failed, e.g.:

$ grep 0000000000048134 u-boot.map
stdio_add_devices

Of course another option is to run it with a debugger such as gdb:

$ gdb u-boot
...
(gdb) br initcall.h:41
Breakpoint 1 at 0x4db9d: initcall.h:41. (2 locations)

Note that two locations are reported, since this function is used in both board_init_f() and board_init_r().

(gdb) r
Starting program: /tmp/b/sandbox/u-boot
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".

U-Boot 2018.09-00264-ge0c2ba9814-dirty (Sep 22 2018 - 12:21:46 -0600)

DRAM:  128 MiB
MMC:

Breakpoint 1, initcall_run_list (init_sequence=0x5555559619e0 <init_sequence_f>)
    at /scratch/sglass/cosarm/src/third_party/u-boot/files/include/initcall.h:41
41                              printf("initcall sequence %p failed at call %p (err=%d)\n",
(gdb) print *init_fnc_ptr
$1 = (const init_fnc_t) 0x55555559c114 <stdio_add_devices>
(gdb)

This approach can be used on normal boards as well as sandbox.

For debugging with GDB or LLDB, it is preferable to reduce the compiler optimization level (CONFIG_CC_OPTIMIZE_FOR_DEBUG=y) and to disable Link Time Optimization (CONFIG_LTO=n).

SDL_CONFIG

If sdl-config is on a different path from the default, set the SDL_CONFIG environment variable to the correct pathname before building U-Boot.

Using valgrind / memcheck

It is possible to run U-Boot under valgrind to check memory allocations:

valgrind ./u-boot

However, this does not give very useful results. The sandbox allocates a memory pool via mmap(). U-Boot’s internal malloc() and free() work on this memory pool. Custom allocators and deallocators are invisible to valgrind by default. To expose U-Boot’s malloc() and free() to valgrind, enable CONFIG_VALGRIND. Enabling this option will inject placeholder assembler code which valgrind interprets. This is used to annotate sections of memory as safe or unsafe, and to inform valgrind about malloc()s and free()s. There are currently no standard placeholder assembly sequences for RISC-V, so this option cannot be enabled on that architecture.

Malloc’s bookkeeping information is marked as unsafe by default. However, this will generate many false positives when malloc itself accesses this information. These warnings can be suppressed with:

valgrind --suppressions=scripts/u-boot.supp ./u-boot

Additionally, you may experience false positives if U-Boot is using a smaller pointer size than your host architecture. This is because the pointers used by U-Boot will only contain 32 bits of addressing information. When interpreted as 64-bit pointers, valgrind will think that they are not initialized properly. To fix this, enable CONFIG_SANDBOX64 (such as via sandbox64_defconfig) when running on a 64-bit host.

Additional options

The following valgrind options are useful in addition to the above examples:

--trace-childen=yes

tells valgrind to keep tracking subprocesses, such as when U-Boot jumps from TPL to SPL, or from SPL to U-Boot proper.

--track-origins=yes

will (for a small overhead) tell valgrind to keep track of who allocated some troublesome memory.

--error-limit

will enable printing more than 1000 errors in a single session.

--vgdb=yes --vgdb-error=0

will let you use GDB to attach like:

gdb -ex "target remote | vgdb" u-boot

This is very helpful for inspecting the program state when there is an error.

The following U-Boot option are also helpful:

-Tc 'ut all'

lets U-Boot run unit tests automatically. Note that not all unit tests will succeed in the default configuration.

-t cooked

will keep the console in a sane state if you terminate it early (instead of having to run tset).

Future work

The biggest limitation to the current approach is that supressions don’t “un-taint” uninitialized memory accesses. Currently, dlmalloc’s bookkeeping information is marked as a “red zone.” This means that all reads to that zone are marked as illegal by valgrind. This is fine for regular code, but dlmalloc really does need to access this area, so we suppress its violations. However, if dlmalloc then passes a result calculated from a “tainted” access, that result is still tainted. So the first accessor will raise a warning. This means that every construct like

foo = malloc(sizeof(*foo));
if (!foo)
    return -ENOMEM;

will raise a warning when we check the result of malloc. Whoops.

There are at least four possible ways to address this:

  • Don’t mark dlmalloc bookkeeping information as a red zone. This is the simplest solution, but reduces the power of valgrind immensely, since we can no longer determine that (e.g.) access past the end of an array is undefined.

  • Implement red zones properly. This would involve growing every allocation by a fixed amount (16 bytes or so) and then using that extra space for a real red zone that neither regular code nor dlmalloc needs to access. Unfortunately, this would probably some fairly intensive surgery to dlmalloc to add/remove the offset appropriately.

  • Mark bookkeeping information as valid before we use it in dlmalloc, and then mark it invalid before returning. This would be the most correct, but it would be very tricky to implement since there are so many code paths to mark. I think it would be the most effort out of the three options here.

  • Use the host malloc and free instead of U-Boot’s custom allocator. This will eliminate the need to annotate dlmalloc. However, using a different allocator for sandbox will mean that bugs in dlmalloc will only be tested when running on read (or emulated) hardware.

Until one of the above options are implemented, it will remain difficult to sift through the massive amount of spurious warnings.

Testing

U-Boot sandbox can be used to run various tests, mostly in the test/ directory.

See Sandbox tests for more information and Introduction to testing for information about testing generally.

Memory Map

Sandbox has its own emulated memory starting at 0. Here are some of the things that are mapped into that memory:

Addr

Config

Usage

0

CONFIG_SYS_FDT_LOAD_ADDR

Device tree

c000

CONFIG_BLOBLIST_ADDR

Blob list

10000

CONFIG_MALLOC_F_ADDR

Early memory allocation

f0000

CONFIG_PRE_CON_BUF_ADDR

Pre-console buffer

100000

CONFIG_TRACE_EARLY_ADDR

Early trace buffer (if enabled). Also used as the SPL load buffer in spl_test_load().

200000

CONFIG_SYS_TEXT_BASE

Load buffer for U-Boot (sandbox_spl only)