Problem with SBRK not returning from call on bare metal risk
Damien Towning
connolly.damien@gmail.com
Wed Mar 8 01:23:04 GMT 2023
Thank you for the helpful response and the link. Having understood this
helped a lot. I went back to using my own _sbrk and the newlib malloc seems
to work okay with it. I think this is mostly about the variable names for
heap and stack matching what the newlib expects _and_ avoiding any of the
re-entrant stuff. Am able to load up and free complex memory blocks out of
the heap.
Now what prompted my question to start with is that I am concerned about
implementing functions I do not ( or should not ) implement myself. So for
example I have had to implement the srand and rand functions myself because
those appear to be re-entrant in the newlib. I've now moved on and got
another question to ask now about the math library which I think I will put
in a separate post :)
On Tue, Mar 7, 2023 at 12:24 AM Dave Nadler <drn@nadler.com> wrote:
> Hi Damien - You need to verify the sbrk you're using actually provides the
> correct addresses.
> In cases I'm familiar with, sbrk is provided by the chip-specific
> toolchain and not newlib.
> Providing your own sbrk is not all bad...
> Also, you need to be sure how newlib was built in the toolchain you are
> using;
> you didn't show us the newlib configuration?
> This might provide some help:
> https://nadler.com/embedded/newlibAndFreeRTOS.html
> Hope this helps!
> Best Regards, Dave
>
> On 3/6/2023 2:36 AM, Damien Towning wrote:
>
> So a bit of a background. I have the bonfire Risc-V 32 core running on a
> T-35 FPGA board. The metal part of this is working great. The processor
> runs. I've got it to turn around the segment display on the board. I've
> also implemented my own _sbrk and got that to work I think consistently
> with malloc. Now in my Risc-v 32 toolchain I have the newlib and I have
> brought that in and with my own wrapped _sbrk I am able to use the system
> defined malloc. But I am not entirely happy with this because I don't want
> to be wrapping every function I encounter. So for example today I found
> myself wrapping srand and rand. So I am pretty sure the reason my
> nano_newlib isn't working is something to do with either not having the
> correct variable names in my firmware.ld or something to do with disabling
> the re-entrant behaviour of newlib. I've included in this email my
> firmware.ld,platform.h,c code example. What happens is I call the _sbrk(0)
> and ( in the simulator I am also running this in ) I never see a bad opcode
> generated. It just seems like the processor goes away and never comes back.
> Does anybody have any suggestions as to what might cause this sort of
> problem ? I'd like to use the sbrk and other funcitons within the newlib
> itself if possible without rolling my own.
>
> ------------- firmware.ld
> MEMORY
> {
>
> RAM (rwx) : ORIGIN = 0x0C000000, LENGTH = 512K
> }
>
> ENTRY(_start)
>
> SECTIONS
> {
> _rombase = 0x0C000000;
>
> . = 0x0;
>
> .text : {
> *(.init)
> *(.text.*)
>
> } > RAM
>
> .data ALIGN(4) : {
> *(.rodata.*)
> *(.rodata)
> *(.eh_frame)
> *(.sdata)
> *(.data)
> *(.sbss)
> *(.sbss.*)
> *(.gnu.linkonce.sb.*)
> *(.bss)
> *(.bss.*)
> *(.gnu.linkonce.b.*)
> *(COMMON)
>
> end = .; _end = .; __end = .;
>
> } > RAM
>
> _heap_end = _end;
> _heap_prev_end = _end;
>
> }
>
> PROVIDE(__heap_end = _end);
> PROVIDE(__prev_heap_end = _end);
> PROVIDE(__heap_start = _end+128K);
>
> PROVIDE(__stack_top = ORIGIN(RAM) + LENGTH(RAM));
> PROVIDE(__stack_start = ORIGIN(RAM) + LENGTH(RAM));
>
> PROVIDE(__heap_size = 128K );
> PROVIDE(__stack_size = 128K );
>
>
>
> /*
> 0x0C000000 +-------------------+
> | |
> | ROM / Flash |
> 0x0C080000 +-------------------+
> | |
> | .data |
> |-------------------|
> | .bss / .sbss |
> |-------------------|
> | _end |
> |-------------------|
> | Heap |
> |-------------------|
> | Stack |
> 0xXXXXXXXX +-------------------+
> */
> --------------------------------------------------------------------------------------
>
> ---------platform.h
> ----------------------------------------------------------------
> #ifndef BASIC_PLATFORM_H
> #define BASIC_PLATFORM_H
>
> // New Defintions for new bonfire-soc-io core
> #define IO_BASE 0x04000000
> #define SOC_IO_OFFSET 0x10000 // Offset from one I/O Device to the next
> (64K range)
>
> #define UART0_BASE IO_BASE
> #define SPIFLASH_BASE (IO_BASE+SOC_IO_OFFSET)
> #define GPIO_BASE (IO_BASE+3*SOC_IO_OFFSET)
> #define UART1_BASE (IO_BASE+2*SOC_IO_OFFSET)
>
> #define UART_BASE UART0_BASE // Backwards compatiblity
>
> #define MTIME_BASE 0x0FFFF0000
>
> #define DRAM_BASE 0x0
> #define DRAM_SIZE 0
> #define DRAM_TOP (DRAM_BASE+DRAM_SIZE-1)
> #define SRAM_BASE 0x0C000000
> #define SRAM_SIZE 524288
> #define SRAM_TOP (SRAM_BASE+SRAM_SIZE-1)
>
> //#define SYSCLK 12000000
> //#define SYSCLK 33333333
> //#define SYSCLK 24997500
> //#define SYSCLK 88786000
> #define EXT_SYSCLK 89015000
>
> #define SYSCLK 24997500
>
> //#define EXT_SYSCLK 92370000 /* For reasons not clear to me when any
> change is made to the interface */
> /* configuration in Efinity and a new synthesis is done a clock time will
> */
> /* be calculated so you must build your efinity project. Take that new
> */
> /* time value. Update it here. Rebuild your hex file. Then rebuild your
> */
> /* efinity project AGAIN
> */
>
>
> #define TIME_UNIT 1e+9; // A billion
>
> #ifndef SIMULATOR
> #define SCALAR 1 // Relative Scalar division for efinity build
> #define BAUD_SCALAR 1 // Relative Scalar division for efinity build
> #else
> #define SCALAR 10000 // Relative Scalar division for simulator. Use for
> any of your own timing since
> #define BAUD_SCALAR 20 // simulator runs slower than real hardware
> #endif
>
> #define CLK_PERIOD (TIME_UNIT / EXT_SYSCLK) // in ns...
>
> //#define DCACHE_SIZE 0 // (2048*4) // DCache Size in Bytes
>
> // Parameters for SPI Flash
>
> #define FLASHSIZE (8192*1024)
> #define MAX_FLASH_IMAGESIZE (2024*1024) // Max 2MB of flash used for boot
> image
> #define FLASH_IMAGEBASE (1024*3072) // Boot Image starts at 3MB in Flash
>
> #define PLATFORM_BAUDRATE (115200*BAUD_SCALAR)
>
> #define NO_RECURSION
> #define NO_SYSCALL
> #define NO_FLASH
> #define NO_XMODEM
> #define NO_DRAMTEST
>
> //#define GPIO_TEST
>
> #endif
> ---------------------------------------------------------------------------------------------
>
> ---- sim_hello.c
> ------------------------------------------------------------------------
> // ==============================================
> // T35 S100 Module Test Code
> // Damien Towning - 2022
> // ==============================================
> #include <stdioSo a bit of a background. I have the bonfire Risc-V 32 core
> running on a T-35 FPGA board. The metal part of this is working great. The
> processor runs. I've got it to turn around the segment display on the
> board. I've also implemented my own _sbrk and got that to work I think
> consistently with malloc. Now in my Risc-v 32 toolchain I have the newlib
> and I have brought that in and with my own wrapped _sbrk I am able to use
> the system defined malloc. But I am not entirely happy with this because I
> don't want to be wrapping every function I encounter. So for example today
> I found myself wrapping srand and rand. So I am pretty sure the reason my
> nano_newlib isn't working is something to do with either not having the
> correct variable names in my firmware.ld or something to do with disabling
> the re-entrant behaviour of newlib. I've included in this email my
> firmware.ld,platform.h,c code example. What happens is I call the _sbrk(0)
> and ( in the simulator I am also running this in ) I never see a bad opcode
> generated. It just seems like the processor goes away and never comes back.
> Does anybody have any suggestions as to what might cause this sort of
> problem ? I'd like to use the sbrk and other funcitons within the newlib
> itself if possible without rolling my own.
>
> ------------- firmware.ld
> MEMORY
> {
>
> RAM (rwx) : ORIGIN = 0x0C000000, LENGTH = 512K
> }
>
> ENTRY(_start)
>
> SECTIONS
> {
> _rombase = 0x0C000000;
>
> . = 0x0;
>
> .text : {
> *(.init)
> *(.text.*)
>
> } > RAM
>
> .data ALIGN(4) : {
> *(.rodata.*)
> *(.rodata)
> *(.eh_frame)
> *(.sdata)
> *(.data)
> *(.sbss)
> *(.sbss.*)
> *(.gnu.linkonce.sb.*)
> *(.bss)
> *(.bss.*)
> *(.gnu.linkonce.b.*)
> *(COMMON)
>
> end = .; _end = .; __end = .;
>
> } > RAM
>
> _heap_end = _end;
> _heap_prev_end = _end;
>
> }
>
> PROVIDE(__heap_end = _end);
> PROVIDE(__prev_heap_end = _end);
> PROVIDE(__heap_start = _end+128K);
>
> PROVIDE(__stack_top = ORIGIN(RAM) + LENGTH(RAM));
> PROVIDE(__stack_start = ORIGIN(RAM) + LENGTH(RAM));
>
> PROVIDE(__heap_size = 128K );
> PROVIDE(__stack_size = 128K );
>
>
>
> /*
> 0x0C000000 +-------------------+
> | |
> | ROM / Flash |
> 0x0C080000 +-------------------+
> | |
> | .data |
> |-------------------|
> | .bss / .sbss |
> |-------------------|
> | _end |
> |-------------------|
> | Heap |
> |-------------------|
> | Stack |
> 0xXXXXXXXX +-------------------+
> */
> --------------------------------------------------------------------------------------
>
> ---------platform.h
> ----------------------------------------------------------------
> #ifndef BASIC_PLATFORM_H
> #define BASIC_PLATFORM_H
>
> // New Defintions for new bonfire-soc-io core
> #define IO_BASE 0x04000000
> #define SOC_IO_OFFSET 0x10000 // Offset from one I/O Device to the next
> (64K range)
>
> #define UART0_BASE IO_BASE
> #define SPIFLASH_BASE (IO_BASE+SOC_IO_OFFSET)
> #define GPIO_BASE (IO_BASE+3*SOC_IO_OFFSET)
> #define UART1_BASE (IO_BASE+2*SOC_IO_OFFSET)
>
> #define UART_BASE UART0_BASE // Backwards compatiblity
>
> #define MTIME_BASE 0x0FFFF0000
>
> #define DRAM_BASE 0x0
> #define DRAM_SIZE 0
> #define DRAM_TOP (DRAM_BASE+DRAM_SIZE-1)
> #define SRAM_BASE 0x0C000000
> #define SRAM_SIZE 524288
> #define SRAM_TOP (SRAM_BASE+SRAM_SIZE-1)
>
> //#define SYSCLK 12000000
> //#define SYSCLK 33333333
> //#define SYSCLK 24997500
> //#define SYSCLK 88786000
> #define EXT_SYSCLK 89015000
>
> #define SYSCLK 24997500
>
> //#define EXT_SYSCLK 92370000 /* For reasons not clear to me when any
> change is made to the interface */
> /* configuration in Efinity and a new synthesis is done a clock time will
> */
> /* be calculated so you must build your efinity project. Take that new
> */
> /* time value. Update it here. Rebuild your hex file. Then rebuild your
> */
> /* efinity project AGAIN
> */
>
>
> #define TIME_UNIT 1e+9; // A billion
>
> #ifndef SIMULATOR
> #define SCALAR 1 // Relative Scalar division for efinity build
> #define BAUD_SCALAR 1 // Relative Scalar division for efinity build
> #else
> #define SCALAR 10000 // Relative Scalar division for simulator. Use for
> any of your own timing since
> #define BAUD_SCALAR 20 // simulator runs slower than real hardware
> #endif
>
> #define CLK_PERIOD (TIME_UNIT / EXT_SYSCLK) // in ns...
>
> //#define DCACHE_SIZE 0 // (2048*4) // DCache Size in Bytes
>
> // Parameters for SPI Flash
>
> #define FLASHSIZE (8192*1024)
> #define MAX_FLASH_IMAGESIZE (2024*1024) // Max 2MB of flash used for boot
> image
> #define FLASH_IMAGEBASE (1024*3072) // Boot Image starts at 3MB in Flash
>
> #define PLATFORM_BAUDRATE (115200*BAUD_SCALAR)
>
> #define NO_RECURSION
> #define NO_SYSCALL
> #define NO_FLASH
> #define NO_XMODEM
> #define NO_DRAMTEST
>
> //#define GPIO_TEST
>
> #endif
> ---------------------------------------------------------------------------------------------
>
> ---- sim_hello.c
> ------------------------------------------------------------------------
> // ==============================================
> // T35 S100 Module Test Code
> // Damien Towning - 2022
> // ==============================================
> #include <stdio.h>
> #include <stdlib.h>
>
> //#include <newlib.h>
> //#include <nano_malloc.h>
> #include <errno.h>
> #include <sys/types.h>
> #include <string.h>
> #include "bonfire.h"
> #include "uart.h"
> #include "bonfire_gpio.h"
> #include "mem_rw.h"
> #include "console.h"
>
> extern void* sbrk(incr);
>
>
> // ========================================
> // Main function
> // ========================================
> int main() {
>
> setBaudRate(PLATFORM_BAUDRATE);
>
> caddr_t *ptr = sbrk(0);
> // code never returns from sbrk
> if (ptr == (void *)-1) {
> printk("Failed to allocate memory\r\n");
> }
> else {
> printk("Allocated memory at : %p\r\n", (void *) &ptr);
> }
> return 0;
> }
> ------------------------------------------------------------------------------------
> --- Makefile
> --------------------------------------------------------------------
>
> .PHONY: all clean
>
> ARCH ?= rv32ia
> ABI=ilp32
> PLATFORM ?= S100_T35
>
> #TARGET_PREFIX ?= riscv32-unknown-linux-gnu
> TARGET_PREFIX ?= riscv32-unknown-elf
> TARGET_CC := $(TARGET_PREFIX)-gcc
> TARGET_LD := $(TARGET_PREFIX)-gcc
> TARGET_SIZE := $(TARGET_PREFIX)-size
> TARGET_OBJCOPY := $(TARGET_PREFIX)-objcopy
> HEXDUMP ?= hexdump
> DATA2MEM = /opt/Xilinx/14.7/ISE_DS/ISE/bin/lin64/data2mem
>
> #PROJROOT = ../../..
> #TOPLEVEL = ~/riscv/ise/bonfire/papilio_pro_dram_toplevel.bit
> PLATFORMDIR=../platform
>
> LINKDEF?=$(PLATFORMDIR)/$(PLATFORM)/firmware.ld
>
> TARGET_CFLAGS += -march=$(ARCH) -mabi=$(ABI) -Wall -Os -g
> -fomit-frame-pointer \
> -ffreestanding -fno-builtin \
> -Wall -Werror=implicit-function-declaration \
> -D$(PLATFORM)
>
> ifeq ($(SIMULATOR), 1)
> TARGET_CFLAGS+=-DSIMULATOR
> endif
>
> TARGET_CFLAGS+=-I$(PLATFORMDIR) -I$(PLATFORMDIR)/$(PLATFORM) -I../riscv
>
> # If we wish to build with no newlib
> #TARGET_LDFLAGS += -march=$(ARCH) -mabi=$(ABI) -nostartfiles \
> -Wl,-m,elf32lriscv,--wrap=malloc,--wrap=sbrk --specs=nosys.specs
> -Wl,-T$(LINKDEF) \
> -Wl,--gc-sections
>
> TARGET_LDFLAGS += -march=$(ARCH) -mabi=$(ABI) -nostartfiles \
> -Wl,-m,elf32lriscv,--wrap=_srand,--wrap=_rand --specs=nano.specs
> -Wl,-Map=output.map,-T$(LINKDEF) \
> -Wl,--gc-sections
>
> all: sim_hello.hex memtest.hex nanotest.hex
>
> %.o : %.S
> $(TARGET_CC) $(TARGET_CFLAGS) -c $<
>
> %.o : %.c
> $(TARGET_CC) $(TARGET_CFLAGS) -c $<
>
> %.hex : %.elf
> $(TARGET_OBJCOPY) -S -O binary $< $(basename $@).bin
> $(HEXDUMP) -v -e '1/4 "%08x\n"' $(basename $@).bin >$@
> $(TARGET_PREFIX)-objdump -S -d $< >$(basename $@).lst
> #$(TARGET_PREFIX)-objdump -D $< >$(basename $@).S
> $(TARGET_PREFIX)-objdump -s $< >$(basename $@).dmp
> $(TARGET_PREFIX)-size $<
>
> #sbrk.o: libwrap/sbrk.c
> # $(TARGET_CC) $(TARGET_CFLAGS) -DSIM -c libwrap/sbrk.c
>
> rand.o: libwrap/rand.c
> $(TARGET_CC) $(TARGET_CFLAGS) -DSIM -c libwrap/rand.c
>
> uart.o: uart.c
> $(TARGET_CC) $(TARGET_CFLAGS) -DSIM -c uart.c
>
> #sim_hello.o : sim_hello.c
> # $(TARGET_CC) $(TARGET_CFLAGS) -DSIM -c sim_hello.c
>
> sim_hello.elf: start.o uart.o sim_hello.o snprintf.o console.o rand.o
> $(TARGET_LD) -o $@ $(TARGET_LDFLAGS) start.o uart.o sim_hello.o snprintf.o
> console.o rand.o
>
> #memtest.elf: start.o uart.o memtest.o snprintf.o console.o mempattern.o
> sbrk.o
> # $(TARGET_LD) -o $@ $(TARGET_LDFLAGS) start.o uart.o memtest.o snprintf.o
> console.o mempattern.o sbrk.o
>
> #nanotest.elf: start.o uart.o nanotest.o snprintf.o console.o mempattern.o
> sbrk.o
> # $(TARGET_LD) -o $@ $(TARGET_LDFLAGS) start.o uart.o nanotest.o snprintf.o
> console.o mempattern.o sbrk.o
>
> #jump0.elf : jump0.S
> # $(TARGET_LD) -o $@ -march=RV32IM -nostartfiles -Wl,-T$(BOOTLINK) $<
>
> clean:
> rm -f *.o
> ---------------------------------------------------------------------------------------------.h>
> #include <stdlib.h>
>
> //#include <newlib.h>
> //#include <nano_malloc.h>
> #include <errno.h>
> #include <sys/types.h>
> #include <string.h>
> #include "bonfire.h"
> #include "uart.h"
> #include "bonfire_gpio.h"
> #include "mem_rw.h"
> #include "console.h"
>
> extern void* sbrk(incr);
>
>
> // ========================================
> // Main function
> // ========================================
> int main() {
>
> setBaudRate(PLATFORM_BAUDRATE);
>
> caddr_t *ptr = sbrk(0);
> // code never returns from sbrk
> if (ptr == (void *)-1) {
> printk("Failed to allocate memory\r\n");
> }
> else {
> printk("Allocated memory at : %p\r\n", (void *) &ptr);
> }
> return 0;
> }
> ------------------------------------------------------------------------------------
> --- Makefile
> --------------------------------------------------------------------
>
> .PHONY: all clean
>
> ARCH ?= rv32ia
> ABI=ilp32
> PLATFORM ?= S100_T35
>
> #TARGET_PREFIX ?= riscv32-unknown-linux-gnu
> TARGET_PREFIX ?= riscv32-unknown-elf
> TARGET_CC := $(TARGET_PREFIX)-gcc
> TARGET_LD := $(TARGET_PREFIX)-gcc
> TARGET_SIZE := $(TARGET_PREFIX)-size
> TARGET_OBJCOPY := $(TARGET_PREFIX)-objcopy
> HEXDUMP ?= hexdump
> DATA2MEM = /opt/Xilinx/14.7/ISE_DS/ISE/bin/lin64/data2mem
>
> #PROJROOT = ../../..
> #TOPLEVEL = ~/riscv/ise/bonfire/papilio_pro_dram_toplevel.bit
> PLATFORMDIR=../platform
>
> LINKDEF?=$(PLATFORMDIR)/$(PLATFORM)/firmware.ld
>
> TARGET_CFLAGS += -march=$(ARCH) -mabi=$(ABI) -Wall -Os -g
> -fomit-frame-pointer \
> -ffreestanding -fno-builtin \
> -Wall -Werror=implicit-function-declaration \
> -D$(PLATFORM)
>
> ifeq ($(SIMULATOR), 1)
> TARGET_CFLAGS+=-DSIMULATOR
> endif
>
> TARGET_CFLAGS+=-I$(PLATFORMDIR) -I$(PLATFORMDIR)/$(PLATFORM) -I../riscv
>
> # If we wish to build with no newlib
> #TARGET_LDFLAGS += -march=$(ARCH) -mabi=$(ABI) -nostartfiles \
> -Wl,-m,elf32lriscv,--wrap=malloc,--wrap=sbrk --specs=nosys.specs
> -Wl,-T$(LINKDEF) \
> -Wl,--gc-sections
>
> TARGET_LDFLAGS += -march=$(ARCH) -mabi=$(ABI) -nostartfiles \
> -Wl,-m,elf32lriscv,--wrap=_srand,--wrap=_rand --specs=nano.specs
> -Wl,-Map=output.map,-T$(LINKDEF) \
> -Wl,--gc-sections
>
> all: sim_hello.hex memtest.hex nanotest.hex
>
> %.o : %.S
> $(TARGET_CC) $(TARGET_CFLAGS) -c $<
>
> %.o : %.c
> $(TARGET_CC) $(TARGET_CFLAGS) -c $<
>
> %.hex : %.elf
> $(TARGET_OBJCOPY) -S -O binary $< $(basename $@).bin
> $(HEXDUMP) -v -e '1/4 "%08x\n"' $(basename $@).bin >$@
> $(TARGET_PREFIX)-objdump -S -d $< >$(basename $@).lst
> #$(TARGET_PREFIX)-objdump -D $< >$(basename $@).S
> $(TARGET_PREFIX)-objdump -s $< >$(basename $@).dmp
> $(TARGET_PREFIX)-size $<
>
> #sbrk.o: libwrap/sbrk.c
> # $(TARGET_CC) $(TARGET_CFLAGS) -DSIM -c libwrap/sbrk.c
>
> rand.o: libwrap/rand.c
> $(TARGET_CC) $(TARGET_CFLAGS) -DSIM -c libwrap/rand.c
>
> uart.o: uart.c
> $(TARGET_CC) $(TARGET_CFLAGS) -DSIM -c uart.c
>
> #sim_hello.o : sim_hello.c
> # $(TARGET_CC) $(TARGET_CFLAGS) -DSIM -c sim_hello.c
>
> sim_hello.elf: start.o uart.o sim_hello.o snprintf.o console.o rand.o
> $(TARGET_LD) -o $@ $(TARGET_LDFLAGS) start.o uart.o sim_hello.o snprintf.o
> console.o rand.o
>
> #memtest.elf: start.o uart.o memtest.o snprintf.o console.o mempattern.o
> sbrk.o
> # $(TARGET_LD) -o $@ $(TARGET_LDFLAGS) start.o uart.o memtest.o snprintf.o
> console.o mempattern.o sbrk.o
>
> #nanotest.elf: start.o uart.o nanotest.o snprintf.o console.o mempattern.o
> sbrk.o
> # $(TARGET_LD) -o $@ $(TARGET_LDFLAGS) start.o uart.o nanotest.o snprintf.o
> console.o mempattern.o sbrk.o
>
> #jump0.elf : jump0.S
> # $(TARGET_LD) -o $@ -march=RV32IM -nostartfiles -Wl,-T$(BOOTLINK) $<
>
> clean:
> rm -f *.o
> ---------------------------------------------------------------------------------------------
>
>
> --
> Dave Nadler, USA East Coast voice (978) 263-0097, drn@nadler.com, Skype
> Dave.Nadler1
>
>
--
Damien Towning
More information about the Newlib
mailing list