Skip to content
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?
Go to file
Cannot retrieve contributors at this time

Linking Picolibc applications

Linking embedded applications requires significant target-specific information, including the location of various memory sections along with a range of application-specific memory settings.

You can create your own custom linker script, or you can use the linker script provided by Picolibc. This document describes how to do either.

Creating a Custom Linker Script

Aside from the application and hardware specific aspects of creating a linker script, if your application is using the Picolib startup code, then you need to define the addresses used in that code, and set up the data as required. Checkout the Initializers in Picolibc document for details on what names to declare.

To use a custom linker script when linking with gcc using -specs=picolibc.specs, you'll need to use the gcc -T option instead of using the -Wl,-T linker pass through option. This causes picolibc.specs to not add the picolibc linker script along with your custom one:

gcc -specs=picolibc.specs -Tcustom.ld

Using picolibc.ld

Picolibc provides a default linker script which can often be used to link applications, providing that your linking requirements are fairly straightforward. To use picolibc.ld, you'll create a custom linker script that sets up some variables and then INCLUDE's picolibc.ld. Here's a sample custom linker script sample.ld:

__flash = 0x08000000;
__flash_size = 128K;
__ram = 0x20000000;
__ram_size = 16k;
__stack_size = 512;

INCLUDE picolibc.ld

This is for an STM32L151 SoC with 128kB of flash and 16kB of RAM. We want to make sure there's space for at least 512 bytes of stack. To use this with gcc, the command line would look like this:

gcc -specs=picolibc.specs -Tsample.ld

Alternatively, you can specify those values using --defsym and use picolibc.ld as the linker script:

cc -Wl,--defsym=__flash=__0x08000000 -Wl,--defsym=__flash_size=128K ... -Tpicolibc.ld

Defining Memory Regions

Picolibc.ld defines only two memory regions: flash and ram. Flash is an addressable region of read-only memory which holds program text, constant data and initializers for read-write data. Ram is read-write memory which needs to be initialized before your application starts.

As shown above, you declare the base and size of both memory regions in your linker script:

  • __flash specifies the lowest address in read-only memory used by your application. This needs to be in flash, but need not be the start of actual flash in the device.

  • __flash_size specifies the amount of read-only memory you want to allow the application to fill. This need not be all of the available memory.

  • __ram specifies the lowest address you want the linker to allocate to read-write data for the application.

  • __ram_size specifies the maximum amount of read-write memory you want to permit the application to use.

  • __stack_size reserves this much space at the top of ram for the initial stack.

Arranging Code and Data in Memory

Where bits of code and data land in memory can be controlled to some degree by placing variables and functions in various sections by decorating them with __attribute__ ((section(name))). You'll find '*' used in the following defintions; that can be replaced with any string. For instance, when you use -ffunction-sections or -fdata-sections with gcc, that creates a section named .text.function-name for each function and .data.variable-name for each variable. Here are all of the section names used in picolibc.ld:

Flash contents

These are stored in flash and used directly from flash.

  1. Contents located first in flash. These can be used for interrupt vectors or startup code.

    • .text.init.enter
    • .data.init.enter
    • .init, .init.*
  2. The bulk of the application code

    • .text.unlikely, .text.unlikely.*
    • .text.startup, .text.startup.*
    • .text, .text.*
    • .gnu.linkonce.t.*
  3. Cleanup routines

    • .fini, .fini.*
  4. Read-only data

    • .rdata
    • .rodata, .rodata.*
    • .gnu.linkonce.r.*
    • .srodata.cst16
    • .srodata.cst8
    • .srodata.cst4
    • .srodata.cst2
    • .srodata, .srodata.*
    • .got, .got.*
  5. Addresses of pre-initialization functions. Each of the addresses in the list is called during program initialization, before _init().

    • .preinit_array
  6. Addresses of initializer/constructor functions. Each of the addresses in the list is called during program initialization, before main().

    • .init_array, .ctors
  7. Addresses of de-initializer/destructor functions. Each of the addresses in the list is called after the program finishes, after main().

    • .fini_array, .dtors

Uninitialized ram contents

You can place items in RAM that is not initialized by picolibc. These can be handy if you need values in memory to survive reset, perhaps as a way to communicate from the application to a boot loader or similar. These are placed first in RAM and are sorted by name so that the order is consistent across linking operations:

  1. .preserve.*

  2. .preserve

Initialized ram contents

picolibc.ld places values for variables with explicit initializers in flash and marks the location in flash and in RAM. At application startup, picocrt uses those recorded addresses to copy data from flash to RAM. As a result, any initialized data takes twice as much memory; the initialization values stored in flash and the runtime values stored in ram. Making values read-only where possible saves the RAM.

  1. .data, .data.*

  2. .gnu.linkonce.d.*

  3. .sdata, .sdata.*, .sdata2.*

  4. .gnu.linkonce.s.*

Picolibc uses native toolchain TLS support for values which should be per-thread. This means that variables like errno will be referenced using TLS mechanisms. To make these work even when the application doesn't support threading, Picolibc allocates a static copy of the TLS data in RAM. Picocrt initializes the architecture TLS mechanism to reference this static copy.

By arranging the static copy of initialized and zero'd TLS data right at the data/bss boundary, picolibc can initialize the TLS block as a part of initializing RAM with no additional code. This requires a bit of a trick as the linker doesn't allocate any memory for TLS bss segments; picolibc.ld makes space by simply advancing the memory location by the size of the TLS bss segment.

  1. .tdata, .tdata.*,*

Cleared ram contents

Variables without any explicit initializers are set to zero by picocrt at startup time. The first chunk of these is part of the TLS block:

  1. .tbss, .tbss.*, .gnu.linkonce.tb.*
  2. .tcommon

After the TLS bss section comes the regular BSS variables:

  1. .sbss*
  3. .bss, .bss.*
  4. .gnu.linkonce.b.*