qemu_esp32 Add tensilica esp32 cpu and a board to qemu and dump the rom to learn more about esp-idf
This documents how to add an esp32 cpu and a simple esp32 board to qemu in order to run an app compiled with the SDK (esp-idf) in QEMU. Esp32 is a 240 MHz dual core Tensilica LX6 microcontroller. It is a good way to learn about qemu , esp32 and the esp32 rom.
1. Build qemu
git clone git://github.com/Ebiroll/qemu-xtensa-esp32
mkdir qemu_esp32
cd qemu_esp32
../qemu-xtensa-esp32/configure --disable-werror --prefix=`pwd`/root --target-list=xtensa-softmmu,xtensaeb-softmmu
make
2. Dump rom1 and rom.bin
~/esp/esp-idf/components/esptool_py/esptool/esptool.py --chip esp32 -b 921600 -p /dev/ttyUSB0 dump_mem 0x40000000 0x000C2000 rom.bin
~/esp/esp-idf/components/esptool_py/esptool/esptool.py --chip esp32 -b 921600 -p /dev/ttyUSB0 dump_mem 0x3FF90000 0x00010000 rom1.bin
3. Build an esp32 application
Open a new window to build example app, and set env variables
export PATH=$PATH:$HOME/esp/xtensa-esp32-elf/bin
export IDF_PATH=~/esp/esp-idf
cd esp
git clone git://github.com/Ebiroll/qemu_esp32
cd qemu_esp32
make menuconfig
> Set [*] Run FreeRTOS only on first core
> [ ] Make exception and panic handlers JTAG/OCD aware
make
4. Build the qemu_flash app to create the qemu flash image
Dont forget to check the code in toflash.c as it uses hardcoded paths.
gcc toflash.c -o qemu_flash
5. Flash the esp32flash.bin file
> ./qemu_flash build/app-template.bin
Note that this will create an 4MB empty file filled with 0xff and write
bootloader, partition table and application to this file called esp32flash.bin
This is then copied to /home/user/qemu_esp32, this file will be read by qemu at upstart.
6. In the first window, start qemu
> xtensa-softmmu/qemu-system-xtensa -d guest_errors,unimp -cpu esp32 -M esp32 -m 4M -s > io.txt
Add -S if you want halt execution until debugger is attached.
7. Optional step (start debugger)
cp qemu_esp32/bin/xtensa-esp32-elf-gdb $HOME/esp/xtensa-esp32-elf/bin/xtensa-esp32-elf-gdb.qemu
> xtensa-esp32-elf-gdb.qemu build/app-template.elf -ex 'target remote:1234'
(gdb) b bootloader_main
(gdb) b app_main
(gdb) c
When breakpoint is hit try Ctrl-X then O
Or to debug the bootloader,
xtensa-esp32-elf-gdb.qemu build/bootloader/bootloader.elf -ex 'target remote:1234'
(gdb) b bootloader_main
Remember to start qemu with -S (capital S) this will halt qemu execusion until debugger is attached.
A fun example with an emulated 1306 display
cd examples/06_1306_interactive
ln -s ../../components/ components
make
gcc ../toflash.c -o qemu_flash
./qemu_flash build/app-template.bin
Start qemu
Connect to emulated UART1
nc 127.0.0.1 8881
Press return
However the latest changes in esp-idf makes i2c emulation very slow.
A much faster but non-interactive example is the 06_duino example.
It runs fine with the latest version of esp-idf
The latest version of esp-idf has some asserts regarding the RTC. To run in qemu withe latest esp-idf you can remove these asserts. components/soc/esp32/rtc_clk_init.c
120: bool res = rtc_clk_cpu_freq_mhz_to_config(cfg.cpu_freq_mhz, &new_config);
121: //assert(res && "invalid CPU frequency value");
122: rtc_clk_cpu_freq_set_config(&new_config);
components/soc/esp32/rtc_time.c
74: // The required amount of slowclk_cycles can produce in a counter TIMG a overflow error. Decrease the slowclk_cycles for fix it.
75: //assert(us_time_estimate < us_timer_max);
76: /* Start calibration */
components/esp32/clk.c
126: bool res = rtc_clk_cpu_freq_mhz_to_config(new_freq_mhz, &new_config);
127: //assert(res);
Only if you are using head. Version 3.1 of esp-idf seems to work ok. qemu will be improved to return better values in order to avoid these asserts.
The latest version of qemu rtc emulation is a little better, then this is no longer a problem. With the 3.1 version of esp-idf you might get stuck here..
W (2061) clk: still waiting for source selection RTC
or
(40) clk: RTC: Not found External 32 kHz XTAL. Switching to Internal 150 kHz RC chain
This is configure in menuconfig → Component config → ESP32-specific
b rtc_clk_cal
In end of rtc_clk_cal
s period=100
Now execusion should continue
As the bootloader uses HW accelerated sha256 calculations this is now emulated by qemu,checksum must be correct and should not be overwritten with 000
Booting from emulated flash! This is very cool. Now IT IS MANDATORY to have a proper flash file , ets_unpack_flash_code is not longer patched , qemu now relies on flash and MMU emulation.
Not that esp-idf is constantly being updated and you must keep the qemu source updated to keep it working.
Serial flasher config ---> Flash size (4 MB)
However it seems that latest chages in the bootloader (esp-idf) breaks this, flash hash caclulations dont match. This is because bootloader always use hardware acceleration to calculate the hash and qemu does not emulate hardware SHA5 calculation.
Component config --->
mbedTLS --->
[ ] Enable hardware SHA acceleration
So now the toflash.c program did overwrite the checksum with 00000
this must not happen anymore as
the bootloader relies on proper hash of the image
if (patch_hash==1) {
for (j=0;j<33;j++)
{
tmp_data[file_size-j]=0;
}
}
To create a esp32flash.bin file build qemu_flash tool found in this repository.
gcc toflash.c -o qemu_flash
Then run the ./qemu_flash program.
./qemu_flash build/app-template.bin
Note that you have to use the .bin file as argument. This will generate a flash image with bootloader, partition information and flash file that the bootloder can use boot the proper application from.
You no longer need to give application as kernel parameter.
-kernel /home/olas/esp/qemu_esp32/examples/07_flash_mmap/build/mmap_test.elf
STARTING QEMU, with forward of localhost port 10080 to qemu port 80
xtensa-softmmu/qemu-system-xtensa -d guest_errors,unimp -cpu esp32 -M esp32 -m 4M -net nic,model=vlan0 -net user,id=simnet,ipver4=on,net=192.168.4.0/24,host=192.168.4.40,hostfwd=tcp::10080-192.168.4.3:80 -net dump,file=/tmp/vm0.pcap -s > io.txt
This will reload the application and prevent the crash as the code is reloaded
on the correct locations.
> (gdb) b app_main
> (gdb) c
Press Ctrl-X o to open the source n Press return to reuse latest command
It is sometimes possible to run only from the generated flash,
xtensa-softmmu/qemu-system-xtensa -cpu esp32 -M esp32 -s > io.txt
TRYING to MAP esp32flash.binMAPPED esp32flash.binI (14) boot: ESP-IDF v3.0-dev-20-g9b955f4c 2nd stage bootloader
I (14) boot: compile time 12:24:53
I (14) boot: Enabling RNG early entropy source...
I (14) boot: SPI Speed : 40MHz
I (14) boot: SPI Mode : DIO
I (14) boot: SPI Flash Size : 2MB
I (14) boot: Partition Table:
I (14) boot: ## Label Usage Type ST Offset Length
I (15) boot: 0 nvs WiFi data 01 02 00009000 00006000
I (15) boot: 1 phy_init RF data 01 01 0000f000 00001000
I (15) boot: 2 factory factory app 00 00 00010000 00100000
I (15) boot: End of partition table
I (15) boot: Disabling RNG early entropy source...
I (16) boot: Loading app partition at offset 00010000
I (328) boot: segment 0: paddr=0x00010018 vaddr=0x00000000 size=0x0ffe8 ( 65512)
I (328) boot: segment 1: paddr=0x00020008 vaddr=0x3f400010 size=0x1077c ( 67452) map
I (328) boot: segment 2: paddr=0x0003078c vaddr=0x3ffb0000 size=0x023d8 ( 9176) load
I (329) boot: segment 3: paddr=0x00032b6c vaddr=0x40080000 size=0x00400 ( 1024) load
I (329) boot: segment 4: paddr=0x00032f74 vaddr=0x40080400 size=0x12fb0 ( 77744) load
I (332) boot: segment 5: paddr=0x00045f2c vaddr=0x400c0000 size=0x00000 ( 0) load
I (332) boot: segment 6: paddr=0x00045f34 vaddr=0x00000000 size=0x0a0d4 ( 41172)
I (333) boot: segment 7: paddr=0x00050010 vaddr=0x400d0018 size=0x4c9a4 (313764) map
I (333) cpu_start: Pro cpu up.
I (333) cpu_start: Single core mode
I (334) heap_alloc_caps: Initializing. RAM available for dynamic allocation:
I (334) heap_alloc_caps: At 3FFAE2A0 len 00001D60 (7 KiB): DRAM
I (334) heap_alloc_caps: At 3FFB7C40 len 000283C0 (160 KiB): DRAM
I (334) heap_alloc_caps: At 3FFE0440 len 00003BC0 (14 KiB): D/IRAM
I (335) heap_alloc_caps: At 3FFE4350 len 0001BCB0 (111 KiB): D/IRAM
I (335) heap_alloc_caps: At 400933B0 len 0000CC50 (51 KiB): IRAM
I (335) cpu_start: Pro cpu start user code
I (478) cpu_start: Starting scheduler on PRO CPU.
Running in qemu
Running in qemu
ethoc: num_tx: 8 num_rx: 8
TCP/IP initializing...
TCP/IP initialized.
E (487) wifi: esp_wifi_internal_set_sta_ip 1281 wifi is not init
E (487) event: system_event_sta_got_ip_default 162 esp_wifi_internal_set_sta_ip ret=12289
Applications started.
I (1507) 1306: Setting UART configuration number 1...
I (1507) 1306: Setting UART configuration number 1...
I (1517) uart: queue free spaces: 10
1. clear display.
2. Set page and col to 0.
3. Set page to 4.
4. Set column to 4.
5. Send 8*0xff.
6. Send 8*0x0f.
7. Incremet nuUart Menu
1. clear display.
2. Set page and col to 0.
3. Set page to 4.
4. Set column to 4.
5. Send 8*0xff.
6. Send 8*0x0f.
Works pretty well in linux now
By following the instructions here, I added esp32 to qemu. http://wiki.linux-xtensa.org/index.php/Xtensa_on_QEMU
prerequisites Debian/Ubuntu:
sudo apt-get install libpixman-1-0 libpixman-1-dev
prerequisites Redhat/Centos yum install glib2-devel pixman-devel
git clone git://github.com/Ebiroll/qemu-xtensa-esp32
Requires that you run a patched gdb to run in qemu. If not you will get Remote 'g' packet reply is too long: The key to using the original gdb, is to set num_regs = 104, in core-esp32.c However debugging c-functions in the elf file requires a patched qemu. In the bin directory there are patched versions of xtensa-esp32-elf-gdb. You can also build your own. Just apply this patch, https://github.com/jcmvbkbc/xtensa-toolchain-build/blob/master/fixup-gdb.sh Another option is to build from this prepatched gdb crosstool source, https://github.com/Ebiroll/gdb One has to create a link or copy ar to x86_64-build_pc-linux-gnu-ar Also dont forget to prepare this version for use with with qemu. cd gdb cp xtensa-config.c.qemu xtensa-config.c ./configure --without-guile --target=xtensa-esp32-elf
mkdir ../qemu_esp32
cd ../qemu_esp32
Then run configure as,
../qemu-xtensa-esp32/configure --disable-werror --prefix=`pwd`/root --target-list=xtensa-softmmu,xtensaeb-softmmu
../qemu-xtensa-esp32/configure --disable-werror --prefix=`pwd`/root --target-list=xtensa-softmmu,xtensaeb-softmmu --python=/usr/bin/python2
The goal was to run the app-template.elf and connect the debugger to qemu to allow debugging of the application. It works quite well and as the rom is patched slightly it its possible to run the elf file that is generated by the esp-idf qemu. However you must use requir theese configutation items.
Dont use autodetect fom Main XTAL frequency
Main XTAL frequency (40 MHz) --->
[ ] Make exception and panic handlers JTAG/OCD aware
This is no longer required but recomended.
[*] Run FreeRTOS only on first core,
[ ] Initialize PHY in startup code
[ ] Make exception and panic handlers JTAG/OCD aware
Optimization level (Debug)
Not sure if this is necessary,
Xtensa timer to use as the FreeRTOS tick source (Timer 0 (int 6, level 1))
Now I have also added the possibility to initialize PHY in startup code.
This however reqires that you match the EFUSE mac adresses with the ones dumped from the flash dump. (esp32.c)
case 0x5a004:
printf("EFUSE MAC\n");
return 0xc400c870;
break;
case 0x5a008:
printf("EFUSE MAC\n");
return 0xffda240a;
break;
Starting wifi works sometimes but its better to do like this, if you want to be able to flash and run
the same app in qemu.
#include "emul_ip.h"
if (is_running_qemu()) {
printf("Running in qemu\n");
ESP_ERROR_CHECK( esp_event_loop_init(esp32_wifi_eventHandler, NULL) );
//task_lwip_init(NULL);
xTaskCreatePinnedToCore(task_lwip_init, "lwip_init", 2*4096, NULL, 14, NULL, 0);
}
else {
wifi_task(NULL);
}
UART emulation is not so good as output is primaraly on stderr. Also interrupts are not emulated so well. esp32.c creates a 3 threads with socket on port 8880-8882 tto allow connect to the serial port over a socket. In the starting terminal you can see output from all uarts mixed. However if you want to see output from only one uart, do, nc 127.0.0.1 8880 or nc 127.0.0.1 8881 for uart 0.
When running I advise that you build a patched gdb as described in this document or use the gdb provided in the /bin directory (linux 64 bit). This improves the debugging exerience quite a bit. The xtensa-esp32-elf-gdb.arch is built for arch linux.
When debugging in the rom you can use,
(gdb) add-symbol-file rom.elf 0x40000000
This also works for the original gdb and gives you all names of the functions in rom0.
You can get and compile your own elf file from rom.S found here, https://github.com/cesanta/mongoose-os/tree/master/common/platforms/esp32/rom Or build your own from esp-idf/components/esp32/ld/esp32.rom.ld: // sort -k 5 esp32.rom.ld | perl -npe 's/=/,/; s/;//'
To setup esp-idf do,
git clone --recursive https://github.com/espressif/esp-idf.git
git pull & git submodule update --recursive git submodule update --init
export PATH=$PATH:$HOME/esp/xtensa-esp32-elf/bin
export IDF_PATH=~/esp/esp-idf
You can most likely use the latest version of esp-idf, just dont start wifi and run on one core only. Dual core should work because esp_crosscore_int_send_yield is implemented. Also rtc_clk_xtal_freq_get works. However one thread sometimes seems to get stuck waiting for s_flash_op_complete, Workaround until the problem is fixed,
(gdb) thread 2
(gdb) set s_flash_op_complete=1
(gdb) c
You can probably use head of esp-idf
This version works anyways,
git checkout b540322dc10ee2a0ce773086da42627468897325
For a short time there was a divide by zero.
#5 __udivdi3 (n=128, d=4611190139757918081) at /home/ivan/e/crosstool-NG/.build/src/gcc-5.2.0/libgcc/libgcc2.c:1288
#6 0x400df781 in __udivmoddi4 (rp=0x0, d=4611190964391638913, n=0) at /home/ivan/e/crosstool-NG/.build/src/gcc-5.2.0/libgcc/libgcc2.c:1088
#7 __udivdi3 (n=0, d=4611190964391638913) at /home/ivan/e/crosstool-NG/.build/src/gcc-5.2.0/libgcc/libgcc2.c:1288
#8 0x400d0708 in select_rtc_slow_clk (slow_clk=RTC_SLOW_FREQ_RTC) at /home/olof/esp/esp-idf/components/esp32/./clk.c:120
#9 0x400d0721 in esp_clk_init () at /home/olof/esp/esp-idf/components/esp32/./clk.c:50
#10 0x40080bd5 in start_cpu0_default () at /home/olof/esp/esp-idf/components/esp32/./cpu_start.c:207
#11 0x40080e6a in call_start_cpu0 () at /home/olof/esp/esp-idf/components/esp32/./cpu_start.c:168
The emulation has also been some problems with do_global_ctors(). should be available at this location... 0x3f409fb8 __init_array_start , however nvs reuses 0x3f40000 as I have not gotten it to recognize that this is already mmapped by the bootloader. However when running full bootloader
Or you can try version 3.1
git checkout release/v3.1
git submodule update --init
Or you can try version 2.0
git checkout v2.0
git submodule update --init
Anyways, things will probably work much better now.
cd ..
git clone --recursive https://github.com/espressif/esp-idf.git
# dump_mem ROM0(776KB) to rom.bin
~/esp/esp-idf/components/esptool_py/esptool/esptool.py --chip esp32 -b 921600 -p /dev/ttyUSB0 dump_mem 0x40000000 0x000C2000 rom.bin
# dump_mem ROM1(64KB) to rom1.bin
~/esp/esp-idf/components/esptool_py/esptool/esptool.py --chip esp32 -b 921600 -p /dev/ttyUSB0 dump_mem 0x3FF90000 0x00010000 rom1.bin
Note that rom0 is smaller than the actual dump.
Those two files will be loaded by qemu and must be in same directory as you start qemu.
If you see this,
I (440) system_api: Base MAC address is not set, read default base MAC address from BLK0 of EFUSE
I (1540) system_api: Base MAC address is not set, read default base MAC address from BLK0 of EFUSE
W (2650) phy_init: failed to load RF calibration data (0x1102), falling back to full calibration
I (3740) phy: error: pll_cal exceeds 2ms!!!
I (4300) phy: error: pll_cal exceeds 2ms!!!
I (4860) phy: error: pll_cal exceeds 2ms!!!
I (5420) phy: error: pll_cal exceeds 2ms!!!
I (5980) phy: error: pll_cal exceeds 2ms!!!
I (6540) phy: error: pll_cal exceeds 2ms!!!
I (7100) phy: error: pll_cal exceeds 2ms!!!
I (7660) phy: error: pll_cal exceeds 2ms!!!
It is because you try to start wifi
Now there is simple flash emulation in qemu. You need the file esp32flash.bin to be in the same directory as rom.bin & rom1.bin. If no flashfile exists, an empty file will be created.
~/esp/esp-idf/components/esptool_py/esptool/esptool.py --baud 920600 read_flash 0 0x400000 esp32flash.bin
Another possibility in order to create a proper flash file is by running the following.
Double check the toflash.c program as it copies the generated esp32flash.bin file to the directory where we start qemu
It also assumes partitioning according to partitions_singleapp.bin
gcc toflash.c -o qemu_flash
./qemu_flash build/app-template.bin
To get more correct mmap behaviour now the bootloader runs first.
This emulation reqires that you generate a proper flash file first. Serial flasher config ---> Flash size (4 MB)
Then run the ./qemu_flash program. (described earlier) ./qemu_flash build/app-template.bin
Note that you have to use the .bin file as argument. This will generate a flash image with bootloader, partition information and flash file that the bootloder can use boot the proper application from.
To debug you can use, xtensa-esp32-elf-gdb.qemu build/app-template.elf -ex 'target remote:1234' (gdb) b app_main
Or to debug the bootloader, xtensa-esp32-elf-gdb.qemu build/bootloader/bootloader.elf -ex 'target remote:1234' (gdb) b bootloader_main (gdb) b bootloader_mmap
#define DPORT_PRO_FLASH_MMU_TABLE ((volatile uint32_t) 0x3FF10000)
PAGE 0 is the content of 0x3FF10000 the virtual adress ranged mapped would be adress 0x3f400000 - 0x3f410000
Page 1 would be adress 0x3f410000 - 0x3f420000
I guess paddr=2 would means that this virtual range is mapped to flash content at 0x20000?
This is output from spi_flash_mmap_dump(); after running the second stage bootloader but before nv_init().
Note that the enable phy at startup would call nv_init before app_main.
spi_flash_mmap_dump()
Display information about mapped regions.
This function lists handles obtained using spi_flash_mmap, along with range of pages allocated to each handle. It also lists all non-zero entries of MMU table and corresponding reference counts.
page 0: refcnt=1 paddr=2
page 77: refcnt=1 paddr=4
page 78: refcnt=1 paddr=5
Here is the bootloader output from loading the app partition.
I (80) boot: Loading app partition at offset 00010000
I (435) boot: segment 0: paddr=0x00010018 vaddr=0x00000000 size=0x0ffe8 ( 65512)
I (435) boot: segment 1: paddr=0x00020008 vaddr=0x3f400010 size=0x03824 ( 14372) map
I (440) boot: segment 2: paddr=0x00023834 vaddr=0x3ffb2ac0 size=0x012a0 ( 4768) load
I (451) boot: segment 3: paddr=0x00024adc vaddr=0x40080000 size=0x00400 ( 1024) load
I (458) boot: segment 4: paddr=0x00024ee4 vaddr=0x40080400 size=0x0fdd4 ( 64980) load
I (496) boot: segment 5: paddr=0x00034cc0 vaddr=0x400c0000 size=0x00000 ( 0) load
I (497) boot: segment 6: paddr=0x00034cc8 vaddr=0x00000000 size=0x0b340 ( 45888)
I (502) boot: segment 7: paddr=0x00040010 vaddr=0x400d0018 size=0x11dcc ( 73164) map
I (510) boot: segment 8: paddr=0x00051de4 vaddr=0x50000000 size=0x00008 ( 8) load
Now you can also debug the bootloder with qemu. There used to be some problem with the elf file, but its fixed with the later versions of esp-idf,
Set Bootloader config
(X) Verbose
Before running bootloader, locate all // TO TEST BOOTLOADER in (esp32.c)
and add or remove comments.
xtensa-softmmu/qemu-system-xtensa -cpu esp32 -M esp32 -m 4M -kernel ~/esp/qemu_esp32/build/bootloader/bootloader.elf -s > io.txt
Jun 8 2016 00:22:57
rst:0x10 (RTCWDT_RTC_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT)
I (0) boot: Espressif ESP32 2nd stage bootloader v. V0.1
I (0) boot: compile time 23:25:16
D (1) esp_image: reading image header @ 0x1000
D (1) bootloader_flash: mmu set block paddr=0x00000000 (was 0xffffffff)
(qemu) Flash map 00000000 to memory, 3F720000
(qemu) Flash partition data is loaded.
D (1) boot: magic e9
D (1) boot: segments 04
D (1) boot: spi_mode 02
D (1) boot: spi_speed 00
D (1) boot: spi_size 01
I (1) boot: SPI Speed : 40MHz
I (1) boot: SPI Mode : DIO
I (1) boot: SPI Flash Size : 2MB
I (1) boot: Partition Table:
I (2) boot: ## Label Usage Type ST Offset Length
D (2) bootloader_flash: mmu set paddr=00000000 count=1
D (2) boot: mapped partition table 0x8000 at 0x3f408000
D (2) boot: load partition table entry 0x3f408000
D (2) boot: type=0 subtype=0
I (2) boot: End of partition table
E (2) boot: nothing to load
user code done
Running the bootloader without dumpimg the flash content will give an output like this,
TRYING to MAP esp32flash.bin MAPPED esp32flash.bin
ets Jun 8 2016 00:22:57
rst:0x10 (RTCWDT_RTC_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT)
I (0) boot: Espressif ESP32 2nd stage bootloader v. V0.1
I (0) boot: compile time 23:25:16
D (1) esp_image: reading image header @ 0x1000
D (1) bootloader_flash: mmu set block paddr=0x00000000 (was 0xffffffff)
(qemu) Flash map 00000000 to memory, 3F720000
Flash partition data is loaded.
E (1) esp_image: image at 0x1000 has invalid magic byte
E (1) boot: failed to load bootloader header!
user code done
This because qemu will create an empty file esp32flash.bin and there is no partition table in this flash.
> xtensa-softmmu/qemu-system-xtensa -d guest_errors,unimp -cpu esp32 -M esp32 -m 4M -net nic,model=vlan0 -net user,id=simnet,ipver4=on,net=192.168.1.0/24,host=192.168.1.40,hostfwd=tcp::10077-192.168.1.3:7 -net dump,file=/tmp/vm0.pcap -kernel ~/esp/qemu_esp32/build/app-template.elf -s -S > io.txt
> xtensa-esp32-elf-gdb build/app-template.elf -ex 'target remote:1234'
(gdb) x/10i $pc
(gdb) x/10i 0x40000400
(gdb) p/x $a0
Call0 will set a0 with return adress, this will list the instructions at the call
(gdb) x/10i $a0-3
(gdb) info symbol 0x400d0eb1
(gdb) layout next
(gdb) b app_main
(gdb) continue
(gdb) list *$pc
Ctrl-X and the O opens the source
To disassemble a specific function you can set pc from gdb,
(gdb) set $pc=0x40080804
(gdb) set $pc=<tab> to let gdb list available functions
(gdb) layout asm
(gdb) si
(gdb) ni (skips calls)
(gdb) finish (run until ret)
Setting programcounter (pc) to a function,
(gdb) set $pc=call_start_cpu0
(gdb) x/20i $pc
(gdb) p/x $a3 (register as hex)
If you dump the flash from an actual hardware device,
~/esp/esp-idf/components/esptool_py/esptool/esptool.py --baud 920600 read_flash 0 0x400000 esp32flash.bin
You might get this problem
E (37) esp_image: Image hash failed - image is corrupt
E (37) boot: Factory app partition is not bootable
E (37) boot: No bootable app partitions in the partition table
This is due to incorrect dump or some flash emulation error.If you still want to run, you can replace the bootloader ony and patch the bootloader to skip the check. These are some hints to get you started.
./qemu_flash -bl
xtensa-esp32-elf-gdb.qemu build/bootloader/bootloader.elf -ex 'target remote:1234'
(gdb) b bootloader_main
Dont forget to configure flashsize to < 4MB Also remember that the bootloader has changed slightly during the different versions of the esp-id.
Dump mixed source/disassemply listing,
xtensa-esp32-elf-objdump -d -S build/bootloader/bootloader.elf
xtensa-esp32-elf-objdump -d build/main/main.o
Try adding simple assembly or functions and looḱ at the generated code,
void retint() {
asm volatile (\
"RFDE\n");
}
int test(int in) {
return (in+1);
}
void jump() {
asm volatile (\
"jx a0\n");
retint();
}
00000000 <retint>:
0: 004136 entry a1, 32
3: 003200 rfde
6: f01d retw.n
Disassembly of section .text.test:
00000000 <test>:
0: 004136 entry a1, 32
3: 221b addi.n a2, a2, 1
5: f01d retw.n
Disassembly of section .text.jump:
00000000 <jump>:
0: 004136 entry a1, 32
3: 0000a0 jx a0
6: 000081 l32r a8, fffc0008 <jump+0xfffc0008>
9: 0008e0 callx8 a8
c: f01d retw.n
Emulation of i2c0 is started but not yet finished. The files involved in the emulation are,
hw/i2c/i2c_esp32.c
hw/xtensa/tmpbme280.c
hw/display/ssd1306.c
The idea was to emulate a 1306 display over i2c. Almost there but need improved emulation to get 06_duino example running. This example uses more ssd1306 commands than the other examples. [[https://github.com/Ebiroll/qemu_esp32/blob/master/img/1306.jpg] ]
If you get something like this,
Illegal entry instruction(pc = 40080a4c), PS = 0000001f
Illegal entry instruction(pc = 40080a4c), PS = 0000001f
Then you have probably not downloaded rom files or put them in another directory than where you start qemu.
I got my ESP32-dev board from Adafruit. https://dl.espressif.com/dl/schematics/ESP32-Core-Board-V2_sch.pdf I have made two dumps and mapped the dumps into the files rom.bin & rom1.bin The code in esp32.c also patches the function ets_unpack_flash_code. Instead of loading the code it sets the user_code_start variable (0x3ffe0400) This will later be used by the rom to start the loaded elf file. The functions SPIEraseSector & SPIWrite are also patched to return 0. The rom function -- ets_unpack_flash_code is located at 0x40007018.
The serial device should also be emulated differently. Now serial output goes to stderr. All io calls are printed on stdout thats why you must do > io.txt Opening anoter terminal and doing tail -f io.txt allows you to see what io ports are accessed.
xtensa-softmmu/qemu-system-xtensa -d guest_errors,page,unimp -cpu esp32 -M esp32 -m 4M -kernel ~/esp/qemu_esp32/build/app-template.elf -S -s > io.txt
(gdb) b start_cpu0_default
(gdb) b app_main
(gdb) c
SR 97 is not implemented
SR 97 is not implemented
ets Jun 8 2016 00:22:57
rst:0x10 (RTCWDT_RTC_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT)
I (1) heap_alloc_caps: Initializing heap allocator:
I (1) heap_alloc_caps: Region 19: 3FFB3D10 len 0002C2F0 tag 0
I (1) heap_alloc_caps: Region 25: 3FFE8000 len 00018000 tag 1
check b=0x3ffb3d1c size=180948 ok
check b=0x3ffdfff0 size=0 ok
check b=0x3ffe800c size=98276 ok
I (2) cpu_start: Pro cpu up.
I (2) cpu_start: Single core mode
I (2) cpu_start: Pro cpu start user code
rtc v112 Sep 26 2016 22:32:10
XTAL 0M
TBD(pc = 40083be6): /home/olas/qemu/target-xtensa/translate.c:1354
case 6: /*RER*/ Not implemented in qemu simulation
I (3) cpu_start: Starting scheduler on PRO CPU.
WUR 234 not implemented, TBD(pc = 400912c3): /home/olas/qemu/target-xtensa/translate.c:1887
WUR 235 not implemented, TBD(pc = 400912c8): /home/olas/qemu/target-xtensa/translate.c:1887
WUR 236 not implemented, TBD(pc = 400912cd): /home/olas/qemu/target-xtensa/translate.c:1887
RUR 234 not implemented, TBD(pc = 40091301): /home/olas/qemu/target-xtensa/translate.c:1876
RUR 235 not implemented, TBD(pc = 40091306): /home/olas/qemu/target-xtensa/translate.c:1876
RUR 236 not implemented, TBD(pc = 4009130b): /home/olas/qemu/target-xtensa/translate.c:1876
The RER, WUR and RUR instructions are not implemented in qemu yet,
My version has some hadling of these instructions.
When configuring, choose
Component config --->
FreeRTOS --->
[*] Run FreeRTOS only on first core
ESP32-specific config --->
[ ] Initialize PHY in startup code
However, it might not be necessary to run o only first core anymore. This patch seems to do the trick. https://github.com/espressif/esp-idf/commit/47b8f78cb0e15fa43647788a808dac353167a485
Before the rom patches we got this
xtensa-softmmu/qemu-system-xtensa -d guest_errors,int,page,unimp -cpu esp32 -M esp32 -m 4M -kernel ~/esp/qemu_esp32/build/app-template.elf > io.txt
ets Jun 8 2016 00:22:57
rst:0x10 (RTCWDT_RTC_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT)
flash read err, 1000
Falling back to built-in command interpreter.
The command interpreter is Basic, Here you can read about it http://hackaday.com/2016/10/27/basic-interpreter-hidden-in-esp32-silicon/
If you get an exception like this
Guru Meditation Error of type StoreProhibited occurred on core 0. Exception was unhandled.
Register dump:
PC : 4008189e PS : 00060030 A0 : 800d05f6 A1 : 3ffe3e20
A2 : 3ffb1134 A3 : 0002e49c A4 : 3ffb008c A5 : fffffffc
A6 : 3ffb008c A7 : ffffffff A8 : 3ffe8000 A9 : bffcffdc
A10 : 3ffb12c4 A11 : 3ffe8000 A12 : 00000019 A13 : 00000001
A14 : 7ffe7fe8 A15 : 00000000 SAR : 00000004 EXCCAUSE: 0000001d
EXCVADDR: bffcffe0 LBEG : 4000c46c LEND : 4000c477 LCOUNT : ffffffff
Look at PC for the error then set a breakpoint there
(gdb) b *0x4008189e
And reset qemu.
We break here , components/freertos/./heap_regions.c
(gdb) Pressing Ctrl-X and the o will open the source code if it exists.
(gdb) where
(gdb) up
Some i/o register name mapping in esp32.c is probably wrong. The values returned are also many times wrong. I did this mapping very quickly with grep to get a better understanding of what the rom was doing. I have made the emulation based on what the code expects, should have done more by reading the document.
Terminal 1
>xtensa-softmmu/qemu-system-xtensa -d guest_errors,int,mmu,page,unimp -cpu esp32 -M esp32 -m 4M -kernel ~/esp/qemu_esp32/build/app-template.elf -S -s
Terminal 2
>xtensa-esp32-elf-gdb build/app-template.elf
(gdb) target remote:1234
//Uart_Init, 0x40009120
(gdb) b *0x40009120
(gdb) layout asm
xtensa-softmmu/qemu-system-xtensa -d guest_errors,int,mmu,page,unimp -cpu esp32 -M esp32 -m 4M -kernel ~/esp/qemu_esp32/build/app-template.elf -S -s
Elf entry 400807BC
serial: event 2
tlb_fill(40000400, 2, 0) -> 40000400, ret = 0
tlb_fill(400004b3, 2, 0) -> 400004b3, ret = 0
tlb_fill(400004e3, 2, 0) -> 400004e3, ret = 0
SR 97 is not implemented
SR 97 is not implemented
tlb_fill(4000d4f8, 0, 0) -> 4000d4f8, ret = 0
tlb_fill(3ffae010, 1, 0) -> 3ffae010, ret = 0
tlb_fill(4000f0d0, 0, 0) -> 4000f0d0, ret = 0
tlb_fill(3ffe0010, 1, 0) -> 3ffe0010, ret = 0
tlb_fill(3ffe1000, 1, 0) -> 3ffe1000, ret = 0
tlb_fill(400076c4, 2, 0) -> 400076c4, ret = 0
tlb_fill(4000c2c8, 2, 0) -> 4000c2c8, ret = 0
tlb_fill(3ff9918c, 0, 0) -> 3ff9918c, ret = 0
tlb_fill(3ffe3eb0, 1, 0) -> 3ffe3eb0, ret = 0
tlb_fill(3ff00044, 0, 0) -> 3ff00044, ret = 0
io read 00000044 DPORT 3ff00044=8E6
tlb_fill(400081d4, 2, 0) -> 400081d4, ret = 0
tlb_fill(3ff48034, 0, 0) -> 3ff48034, ret = 0
io write 00000044 000008E6 io read 00048034 RTC_CNTL_RESET_STATE_REG 3ff48034=3390
io read 00048088 RTC_CNTL_DIG_ISO_REG 3ff48088=00000000
io write 00048088 00000000 RTC_CNTL_DIG_ISO_REG 3ff48088
io read 00048088 RTC_CNTL_DIG_ISO_REG 3ff48088=00000000
io write 00048088 00000400 RTC_CNTL_DIG_ISO_REG 3ff48088
tlb_fill(40009000, 2, 0) -> 40009000, ret = 0
tlb_fill(3ff40010, 1, 0) -> 3ff40010, ret = 0
serial: write addr=0x4 val=0x1
tlb_fill(3ff50010, 1, 0) -> 3ff50010, ret = 0
tlb_fill(400067fc, 2, 0) -> 400067fc, ret = 0
tlb_fill(4000bfac, 2, 0) -> 4000bfac, ret = 0
tlb_fill(3ff9c2b9, 0, 0) -> 3ff9c2b9, ret = 0
tlb_fill(3ff5f06c, 0, 0) -> 3ff5f06c, ret = 0
io write 00050010 00000001
io read 0005f06c TIMG_RTCCALICFG1_REG 3ff5f06c=25
tlb_fill(3ff5a010, 0, 0) -> 3ff5a010, ret = 0
io read 0005a010 TIMG_T0ALARMLO_REG 3ff5a010=01
tlb_fill(40005cdc, 0, 0) -> 40005cdc, ret = 0
io read 000480b4 RTC_CNTL_STORE5_REG 3ff480b4=00000000
io read 000480b4 RTC_CNTL_STORE5_REG 3ff480b4=00000000
io write 000480b4 18CB18CB RTC_CNTL_STORE5_REG 3ff480b4
----------- break Uart_Init, 0x40009120
(gdb) finish
Please use other method (esptool.py), its easier and faster. This is saved for historical reasons and for the screen instructions. It can also be used to dump the content as seen by cpu1. Set the environment properly. Build the romdump app and flash it. Use i.e screen as serial terminal.
screen /dev/ttyUSB0 115200
Ctrl-A then H (save output)
Press and hold the boot button, then press reset. This puts the ESP to download mode.
Ctrl-A then \ (exit, detach)
make flash
screen /dev/ttyUSB0 115200
Ctrl-A then H (save output)
press reset on ESP to get start.
When finished trim the capturefile (remove all before and after the dump data) and call it, test.log Notice that there are two dumps search for ROM and ROM1 Compile the dump to rom program.
gcc torom.c -o torom
torom
If successfull you will have a binary rom dump, rom.bin
Then also create the file rom1.bin
If you start with second part and then do
mv rom.bin rom1.bin
Then you can do the first part
Those two files will be loaded by qemu and must be in same directory as you start qemu.
Now it also works for esp32 debugging. (2.10.0)
git clone git://git.qemu.org/qemu.git
cd qemu
This is probably no longer necessary. git submodule update --init dtc
However uart emulation may cause this behaviour.
ERROR:/home/olas/qemu-2.10.1/accel/tcg/tcg-all.c:42:tcg_handle_interrupt: assertion failed: (qemu_mutex_iothread_locked())
Another version for qemu exists here https://github.com/OSLL/qemu-xtensa,
However I think the important stuff has been mereged into HEA of qemu.
git clone https://github.com/OSLL/qemu-xtensa -b xtensa-esp32
but requires that you run a patched gdb to run gdb with qemu. If not you will get Remote 'g' packet reply is too long:
The key to using the original gdb, is to set num_regs = 104, in core-esp32.c
git clone https://github.com/Ebiroll/qemu_esp32.git qemu_esp32_patch
cd qemu_esp32_patch/qemu-patch
./maketar.sh
cp qemu-esp32.tar ../../qemu
cd ../../qemu
tar xvf qemu-esp32.tar
in qemu source tree, manually add to makefiles:
hw/xtensa/Makefile.objs
obj-y += esp32.o
obj-y += MemoryMapped.o
target/xtensa/Makefile.objs
obj-y += core-esp32.o
If running qemu head (2.9.0) make sure to disable this. from the esp32 options
[ ] Make exception and panic handlers JTAG/OCD aware
Interesting info. To be investigated.
static HeapRegionTagged_t regions[]={
{ (uint8_t *)0x3F800000, 0x20000, 15, 0}, //SPI SRAM, if available
{ (uint8_t *)0x3FFAE000, 0x2000, 0, 0}, //pool 16 <- used for rom code
{ (uint8_t *)0x3FFB0000, 0x8000, 0, 0}, //pool 15 <- can be used for BT
{ (uint8_t *)0x3FFB8000, 0x8000, 0, 0}, //pool 14 <- can be used for BT
/* Use first 50 blocks in MMU for bootloader_mmap, 50th block for bootloader_flash_read */
#define MMU_BLOCK0_VADDR 0x3f400000
#define MMU_BLOCK50_VADDR 0x3f720000
#define MMU_FLASH_MASK 0xffff0000
#define MMU_BLOCK_SIZE 0x00010000
Here are some functions with the associated io instructions, to help me improve flash emulation.
From boatloader
Cache_Read_Disable(0);
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000020
io write 40,20
DPORT_PRO_CACHE_CTRL_REG 3ff00040
io read 58 DPORT_APP_CACHE_CTRL_REG 3ff00058=00000020
1 esp32_spi_read: +0xf8: 0x00000000
1 esp32_spi_read: +0x50: 0x00000004
1 esp32_spi_write: +0x50 = 0x00000004
written
---
Cache_Flush(0);
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000020
io write 40,20
DPORT_PRO_CACHE_CTRL_REG 3ff00040
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000020
io write 40,30
DPORT_PRO_CACHE_CTRL_REG 3ff00040
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000030
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000030
io write 40,20 DPORT_PRO_CACHE_CTRL_REG 3ff00040
---
/**
* @brief Set Flash-Cache mmu mapping.
* Please do not call this function in your SDK application.
*
* @param int cpu_no : CPU number, 0 for PRO cpu, 1 for APP cpu.
*
* @param int pod : process identifier. Range 0~7.
*
* @param unsigned int vaddr : virtual address in CPU address space.
* Can be IRam0, IRam1, IRom0 and DRom0 memory address.
* Should be aligned by psize.
*
* @param unsigned int paddr : physical address in Flash.
* Should be aligned by psize.
*
* @param int psize : page size of flash, in kilobytes. Should be 64 here.
*
* @param int num : pages to be set.
*
* @return unsigned int: error status
* 0 : mmu set success
* 1 : vaddr or paddr is not aligned
* 2 : pid error
* 3 : psize error
* 4 : mmu table to be written is out of range
* 5 : vaddr is out of range
*/
unsigned int cache_flash_mmu_set(int cpu_no, int pid, unsigned int vaddr, unsigned int paddr, int psize, int num);
#define MMU_BLOCK50_VADDR 0x3f720000
int e = cache_flash_mmu_set(0, 0, MMU_BLOCK50_VADDR, map_at (0), 64, 1);
io read 44 DPORT_PRO_CACHE_CTRL1_REG 3ff00044=8E6
io read 5c DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008E6
io read 5c DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008E6
io write 5c,8ff DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008FF
io read 44 DPORT_PRO_CACHE_CTRL1_REG 3ff00044=8E6
io write 44,8ff DPORT_PRO_CACHE_CTRL1_REG 3ff00044 8ff
io write 100c8,0
io write 44,8e6 DPORT_PRO_CACHE_CTRL1_REG 3ff00044 8e6
io write 5c,8e6 DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008E6
io read 44 DPORT_PRO_CACHE_CTRL1_REG 3ff00044=8E6
io write 44,8e6 DPORT_PRO_CACHE_CTRL1_REG 3ff00044 8e6
---
Cache_Read_Enable(0)
1 esp32_spi_read: +0x50: 0x00000004
1 esp32_spi_write: +0x50 = 0x00000005
written
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000020
io write 40,28
DPORT_PRO_CACHE_CTRL_REG 3ff00040
---
partitions = bootloader_mmap(ESP_PARTITION_TABLE_ADDR, ESP_PARTITION_TABLE_DATA_LEN);
cache_flash_mmu_set( 0, 0, MMU_BLOCK0_VADDR, src_addr_aligned(0), 64, count );
D (2) bootloader_flash: mmu set paddr=00000000 count=1
io read 44 DPORT_PRO_CACHE_CTRL1_REG 3ff00044=8E6
io read 5c DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008E6
io read 5c DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008E6
io write 5c,8ff
DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008FF
io read 44 DPORT_PRO_CACHE_CTRL1_REG 3ff00044=8E6
io write 44,8ff
DPORT_PRO_CACHE_CTRL1_REG 3ff00044 8ff
io write 10000,0
MMU CACHE 3ff10000 0
io write 44,8e6
DPORT_PRO_CACHE_CTRL1_REG 3ff00044 8e6
io write 5c,8e6
DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008E6
io read 44 DPORT_PRO_CACHE_CTRL1_REG 3ff00044=8E6
io write 44,8e6
DPORT_PRO_CACHE_CTRL1_REG 3ff00044 8e6
---
Cache_Read_Disable,
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000028
io write 40,20
DPORT_PRO_CACHE_CTRL_REG 3ff00040
io read 58 DPORT_APP_CACHE_CTRL_REG 3ff00058=00000028
---
Cache_Flush,
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000020
io write 40,28
DPORT_PRO_CACHE_CTRL_REG 3ff00040
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000028
io write 40,38
DPORT_PRO_CACHE_CTRL_REG 3ff00040
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000038
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000038
io write 40,28
DPORT_PRO_CACHE_CTRL_REG 3ff00040
---
cache_flash_mmu_set,
io read 44 DPORT_PRO_CACHE_CTRL1_REG 3ff00044=8E6
io read 5c DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008E6
io read 5c DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008E6
io write 5c,8ff
DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008FF
io read 44 DPORT_PRO_CACHE_CTRL1_REG 3ff00044=8E6
io write 44,8ff
DPORT_PRO_CACHE_CTRL1_REG 3ff00044 8ff
io write 10000,0
MMU CACHE 3ff10000 0
io write 44,8e6
DPORT_PRO_CACHE_CTRL1_REG 3ff00044 8e6
io write 5c,8e6
DPORT_APP_CACHE_CTRL1_REG 3ff0005C=000008E6
io read 44 DPORT_PRO_CACHE_CTRL1_REG 3ff00044=8E6
io write 44,8e6
DPORT_PRO_CACHE_CTRL1_REG 3ff00044 8e6
---
Cache_Read_Enable
1 esp32_spi_read: +0x50: 0x00000005
1 esp32_spi_write: +0x50 = 0x00000005
written
io read 40 DPORT_PRO_CACHE_CTRL_REG 3ff00040=00000028
io write 40,28
DPORT_PRO_CACHE_CTRL_REG 3ff00040
---
We still get
flash read err, 1000
If unpatching the ets_unpack_flash_code rom function.
This does not work.
head -c 4M /dev/zero > esp32.flash
xtensa-softmmu/qemu-system-xtensa -d guest_errors,page,unimp -cpu esp32 -M esp32 -m 4M -pflash esp32.flash -kernel ~/esp/qemu_esp32/build/app-template.elf -s -S > io.txt
This does not work at all. I dont understand qemu enough to get this working.
To add flash checkout line 502 in esp32.c
dinfo = drive_get(IF_PFLASH, 0, 0);
...
Also checkout m25p80.c driver in qemu
Dump with od,
od -t x4 partitions_singleapp.bin
xtensa-softmmu/qemu-system-xtensa -d guest_errors,page,unimp -cpu esp32 -M esp32 -m 4M -kernel ~/esp/qemu_esp32/build/app-template.elf > io.txt
Running two cores, is default behaviour now, try (gdb) info threads The app cpu starts stalled and will be unstalled by the pro cpu.
xtensa-softmmu/qemu-system-xtensa -d guest_errors,page,unimp -cpu esp32 -M esp32 -m 4M -smp 2 -pflash esp32.flash -kernel ~/esp/qemu_esp32/build/app-template.elf -s -S > io.txt
ets Jun 8 2016 00:22:57
rst:0x10 (RTCWDT_RTC_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT)
I (97134) heap_alloc_caps: Initializing heap allocator:
I (97134) heap_alloc_caps: Region 19: 3FFB5B8C len 0002A474 tag 0
I (97134) heap_alloc_caps: Region 25: 3FFE8000 len 00018000 tag 1
check b=0x3ffb5b98 size=173144 ok
check b=0x3ffdfff0 size=0 ok
check b=0x3ffe800c size=98276 ok
I (97135) cpu_start: Pro cpu up.
I (97135) cpu_start: Starting app cpu, user_code_start is 0x00000000
I (97135) cpu_start: Starting app cpu, entry point is 0x40080b50
I (68034) cpu_start: App cpu up.
I (97140) cpu_start: App cpu started, user_code_start is 0x40080b50
I (97140) cpu_start: Pro cpu start user code
I (97140) rtc: rtc v160 Nov 22 2016 19:00:05
I (97141) rtc: XTAL 40M
I (97142) cpu_start: Starting scheduler on PRO CPU.
I (68253) cpu_start: Starting scheduler on APP CPU.
There used to be some problem with the scheduling and probably some other error as well.
When calling nvs_flash_init()qemu always hangs here,
(gdb) info threads
Id Target Id Frame
2 Thread 2 (CPU#1 [running]) 0x40081a75 in spi_flash_op_block_func (arg=0x1)
at /home/olas/esp/esp-idf/components/spi_flash/./cache_utils.c:67
* 1 Thread 1 (CPU#0 [halted ]) 0x400d2444 in esp_vApplicationIdleHook ()
at /home/olas/esp/esp-idf/components/esp32/./freertos_hooks.c:52
This was added to call_start_cpu0(),cpu_start.c for the extra user code start info
int test= (void (*)(void))REG_READ(DPORT_APPCPU_CTRL_D_REG);
void *user_code_start =(void *) test;
ESP_EARLY_LOGI(TAG, "Starting app cpu, user_code_start is %p", user_code_start);
This seems fixed now, https://github.com/espressif/esp-idf/commit/47b8f78cb0e15fa43647788a808dac353167a485 You should be able to run on two cores.
This is now emulated better,
void esp_crosscore_int_send_yield(int coreId) {
assert(coreId<portNUM_PROCESSORS);
//Mark the reason we interrupt the other CPU
portENTER_CRITICAL(&reasonSpinlock);
reason[coreId]|=REASON_YIELD;
portEXIT_CRITICAL(&reasonSpinlock);
//Poke the other CPU.
if (coreId==0) {
WRITE_PERI_REG(DPORT_CPU_INTR_FROM_CPU_0_REG, DPORT_CPU_INTR_FROM_CPU_0);
} else {
WRITE_PERI_REG(DPORT_CPU_INTR_FROM_CPU_1_REG, DPORT_CPU_INTR_FROM_CPU_1);
}
This results in
io write dc,1 if coreId==0
io write e0,1 if coreId==1
Good information here, https://github.com/espressif/esp-idf/blob/master/components/freertos/readme_xtensa.txt
The timer tick is handled here,
_frxt_timer_int ()
This calls the
xTaskIncrementTick function.
More about the boot of the romdumps
In the bin directory there is an improved gdb version. The following patch was applied, https://github.com/jcmvbkbc/xtensa-toolchain-build/blob/master/fixup-gdb.sh To use it properly you must increase the num_regs = 104 to i.e. 172 in core-esp32.c
The base of all this is done by Max Filippov https://github.com/OSLL/qemu-xtensa
#Another version of qemu with esp32 exists here, However the work is not yet finished.
git clone https://github.com/OSLL/qemu-xtensa
cd qemu-xtensa
git checkout xtensa-esp32
git submodule update --init dtc
cd ..
mkdir build-qemu-xtensa
cd build-qemu-xtensa
../qemu-xtensa/configure --disable-werror --prefix=`pwd`/root --target-list=xtensa-softmmu
There exists networking support of an emulated opencore network device,
components/emul_ip/lwip_ethoc.c is the driver.
All files in the net/ direcory is just for reference, they are currently not used.
To get emulated network to work with 3.1 version of esp-idf you must patch esp-idf components/lwip/port/freertos/sys_arch.c
sys_init(void) ...
//esp_vfs_lwip_sockets_register();
Dont try running emulated_net(); on actual hardware.
Probably not harmful but I put the emulated hardware here, i uesd the DR_REG_EMAC_BASE
#define OC_BASE 0x3ff69000
#define OC_DESC_START 0x3ff69400
#define OC_BUF_START 0x3FFF8000 //pool 6 blk 1 <- can be used as trace memory
TODO! Check if it is safe to use this memory??
This could possibly screw up your data.
Note this one components/esp32/include/soc/soc.h
DR_REG_EMAC_BASE 0x3ff69000
Here are the pins needed to get ethernet support.
http://www.smartarduino.com/esp32/esp32_chip_pin_list_en.pdf
Note that this could be outdated.
The examples/14_basic_webserver worked when I tried it,
Start qemu like this,
xtensa-softmmu/qemu-system-xtensa -d guest_errors,unimp -cpu esp32 -M esp32 -m 4M -net nic,model=vlan0 -net user,id=simnet,ipver4=on,net=192.168.4.0/24,host=192.168.4.40,hostfwd=tcp::10080-192.168.4.3:80 -net dump,file=/tmp/vm0.pcap -s > io.txt
Then start browser on http://127.0.0.1:10080
It is a good idea to save the original xtensa-esp32-elf-gdb as the one in the bin directory works best with qemu
Component config --->
ESP32-specific config --->
Panic handler behaviour (Invoke GDBStub) --->
esp32.c creates a 3 threads with socket on port 8880-8882 tto allow connect to the serial port over a socket.
This allows connecting to panic handler gdbstub. You can also use this to debug the debugstub code.
xtensa-esp32-elf-gdb.sav build/app-template.elf -ex 'target remote:8880'
Another solution if you are running qemu-gdb is to set a breakpoint in the stub.
(gdb)b esp_gdbstub_panic_handler
or if you use default settings
(gdb)b commonErrorHandler
(gdb) where (or up) will show you what the problem is
(gdb)b _xt_panic
This gdbstub panic handler is also nice to have when running on target.
xtensa-esp32-elf-gdb build/app-template.elf -b 115200 -ex 'target remote /dev/ttyUSB0'
I noticed that memory got overwritten as priv_ethoc suddenly contained faulty iobase. Memory access breakpoints comes very handy for these tyoes of errors.
To find a variable based on location in memory
(gdb) info symbol 0x40154e6c
To track memory writes
(gdb) watch priv_ethoc.io_base
To track memory reads
(gdb) rwatch priv_ethoc.io_base
To track memory reads and writes
(gdb) awatch *0x3ffe0400
Together with add-symbol-file rom.elf 0x40000000 this allows you to easy find where a register is accessed.
Here is an important register. DPORT_PRO_CACHE_CTRL_REG
(gdb) awatch *0x3ff00040
(gdb) info watchpoints
Dont forget that you can reset qemu hardware and start again.
In order to invoke the correct interrupts this is some useful info
esp_intr_alloc(ETS_TIMER1_INTR_SOURCE, 0, &frc_timer_isr, NULL, NULL);
io write 1e4,2
esp_intr_alloc(ETS_TG0_WDT_LEVEL_INTR_SOURCE, 0, task_wdt_isr, NULL, NULL);
io write 144,3
ret = esp_intr_alloc(ETS_I2C_EXT0_INTR_SOURCE, intr_alloc_flags, fn, arg, handle);
int intr=get_free_int(flags, cpu, force);
intr=12
xt_set_interrupt_handler(intr, handler, arg);
intr_matrix_set(cpu, source, intr);
io write 1c8,c
To make debugging the rom functions there is a file rom.elf that contains debug information for the rom file. It was created with the help from this project, https://github.com/jcmvbkbc/esp-elf-rom This rom.elf also works with the original gdb with panic handler gdbstub.
xtensa-softmmu/qemu-system-xtensa -d guest_errors,unimp -cpu esp32 -M esp32 -m 4M -kernel ~/esp/qemu_esp32/build/app-template.elf -s -S > io.txt
xtensa-esp32-elf-gdb build/app-template.elf -ex 'target remote:1234'
(gdb) add-symbol-file rom.elf 0x40000000
(gdb) b start_cpu0_default
(gdb) c
(gdb) b app_main
The .gdbinit file contains code to list all tasks, freertos_show_threads It will list all the tasks, hande,name,core,stack,stack_usage It will also list current location of the $pc Take care that because of how the Xtensa core works, the leftmost digit of the hex number may be off and you have to manually correct it to '4'. For example, for $pc = 0x800820a2, you would want to look up 0x400820a2. list *0x400820a2, then you can see what the ipc0 thread is doing. Note that --terminating-- , in this example it is empty, no threads are terminating.
(gdb) freertos_show_threads
0x3ffb80a0 IDLE 0x0 0x3ffb7b44 0x3ffb7f00 956
0x400d0d9e <esp_task_wdt_feed+6>: or a2, a10, a10
0x400d0da1 <esp_task_wdt_feed+9>: l32r a3, 0x400d00c0 <_stext+168>
0x400d0da4 <esp_task_wdt_feed+12>: l32i a3, a3, 0
0x3ffb78b4 main 0x0 0x3ffb6898 0x3ffb7750 3768
$1 = 0x80083070
0x3ffb9798 tiT 0x7fffffff 0x3ffb8f7c 0x3ffb9620 1700
$2 = 0x800820a2
0x3ffba140 poll_task 0x7fffffff 0x3ffb9924 0x3ffba060 1852
$3 = 0x80083070
0x3ffb8890 Tmr Svc 0x0 0x3ffb8334 0x3ffb8790 1116
$4 = 0x80083b9e
--suspended--
0x3ffbaae8 echo_thread 0x7fffffff 0x3ffba2cc 0x3ffba8b0 1508
$5 = 0x800820a2
0x3ffb670c ipc0 0x0 0x3ffb61b0 0x3ffb6600 1104
$6 = 0x800820a2
--terminating--
The MMU will map flash data/instruction to the processors depending on how these are set.
/* Flash MMU table for PRO CPU */
//#define DPORT_PRO_FLASH_MMU_TABLE ((volatile uint32_t*) 0x3FF10000)
4 regions with 64
/* Flash MMU table for APP CPU */
//#define DPORT_APP_FLASH_MMU_TABLE ((volatile uint32_t*) 0x3FF12000)
MMU_BLOCK0_VADDR 0x3f400000 // Dont use this one. Must be mapped to flash.rodata
MMU_BLOCK1_VADDR 0x3f410000
..
MMU_BLOCK50_VADDR 0x3f720000
..
MMU_BLOCK63_VADDR 0x3fA30000
qemu emulates flash mapping by looking at the X_FLASH_MMU_TABLE[]
A value of 2 in DPORT_PRO_FLASH_MMU_TABLE[1] means that
contents of the flash file at location 0x20000-0x30000 will be mapped to virtual address 0x3f410000-0x3f420000
I think that from BLOCK50 and up memory is mapped as instructions. BLOCK0-49 is data.
Offset, length, name , data
0x1000 , 2000 , Bootloader
0x8000 , >100 , Partition table
0x9000 , 6000 nvs , wifi auth data
0xf000 , 1000 phy_init rf data, calibration data
0x10000 , >50000 factory, application
0x3E8000 is the end for a 4MB flash
Old locations? Used in olfer version of the idf.
0x6000 , 6000 nvs , wifi auth data
Note that the esp32 can save rf calibration data for subsequent startups
qemu uses a simplified memory map, (subject to change)
0x2000_0000 - 0x4000_0000 ram
0x4000_0000 - 0x40BF_FFFF ram1
0x3ff0_0000 - 0x3ff0_0000 system_io
The rom dumps are just copied into ram, but is probably protected.
Also some of the ram was also dumped and is restored, this is probably a good thing.
0x3FF9_0000 - 0x3FF9_FFFF rom1.bin
0x4000_0000 - 0x40c1_FFFF rom.bin
0x5000_0000 - 0x5008_0000 ulp ram
0x6000_0000 - wifi io, undocumented i/o
flash mmu emulation is done by copying from the file esp32flash.bin when the registers DPORT_PRO_FLASH_MMU_TABLE are written to.
ROM0 384 KB 0x4000_0000 0x4005_FFFF Static MPU
ROM1 64 KB 0x3FF9_0000 0x3FF9_FFFF Static MPU
SRAM0 64 KB 0x4007_0000 0x4007_FFFF Static MPU
128 KB 0x4008_0000 0x4009_FFFF SRAM0 MMU
SRAM1 128 KB 0x3FFE_0000 0x3FFF_FFFF Static MPU
128 KB 0x400A_0000 0x400B_FFFF Static MPU
SRAM2 72 KB 0x3FFA_E000 0x3FFB_FFFF Static MPU
128 KB 0x3FFC_0000 0x3FFD_FFFF SRAM2 MMU
RTC FAST 8 KB 0x3FF8_0000 0x3FF8_1FFF RTC FAST MPU
8 KB 0x400C_0000 0x400C_1FFF RTC FAST MPU
RTC SLOW 8 KB 0x5000_0000 0x5000_1FFF RTC SLOW MPU
The on-chip memory is governed by fixed-function MPUs, configurable MPUs, and MMUs:
Both the upper 128 KB of SRAM0 and the upper 128 KB of SRAM2 are governed by an MMU. 128 KB 0x4008_0000 0x4009_FFFF SRAM0 MMU 128 KB 0x3FFC_0000 0x3FFD_FFFF SRAM2 MMU
The internal RAM MMUs divide the memory range they govern into 16 pages. The page size is configurable as 8 KB, 4 KB and 2 KB. When the page size is 8 KB, the 16 pages span the entire 128 KB memory region; when the page size is 4 KB or 2 KB, a non-MMU-covered region of 64 or 96 KB, respectively, will exist at the end of the memory space.
The address range of the first 32 KB of the ROM 0 (0x4000_0000 ~ 0x4000_7FFF) can be remapped in order to access a part of Internal SRAM 1 that normally resides in a memory range of 0x400B_0000 ~ 0x400B_7FFF. While remapping, the 32 KB SRAM cannot be accessed by an address range of 0x400B_0000 ~ 0x400B_7FFF any more, but it can still be accessible through the data bus (0x3FFE_8000 ~ 0x3FFE_FFFF
#Layout
Bus Type | Boundary Address | Size | Target
Low Address | High Add
0x0000_0000 0x3F3F_FFFF Reserved
Data 0x3F40_0000 0x3F7F_FFFF 4 MB External Memory
Data 0x3F80_0000 0x3FBF_FFFF 4 MB External Memory
0x3FC0_0000 0x3FEF_FFFF 3 MB Reserved
Data 0x3FF0_0000 0x3FF7_FFFF 512 KB Peripheral (io)
Data 0x3FF8_0000 0x3FFF_FFFF 512 KB Embedded Memory
Instruction 0x4000_0000 0x400C_1FFF 776 KB Embedded Memory rom
Instruction 0x400C_2000 0x40BF_FFFF 11512 KB External Memory
0x40C0_0000 0x4FFF_FFFF 244 MB Reserved
0x5000_0000 0x5000_1FFF 8 KB Embedded Memory
0x5000_2000 0xFFFF_FFFF Reserved
The linker scripts contains information of how to layout the code and variables. componenets/esp32/ld
/* IRAM for PRO cpu. Not sure if happy with this, this is MMU area... */
iram0_0_seg (RX) : org = 0x40080000, len = 0x20000