POSIX-UEFI

We hate that horrible and ugly UEFI API, we want the usual POSIX!

---

NOTE: a smartypants on reddit is worried that I'm supposedly "hiding something" because of the use of -Wno-builtin-declaration-mismatch (gcc) and -Wno-incompatible-library-redeclaration (Clang) flags. Here's my answer to you: I hide nothing, that flag is only needed because you can disable transparent UTF-8 conversion. You see, under the hood UEFI uses 16 bit characters, and for example strlen(wchar_t *str) or main(int argc, wchar_t *argv) isn't exactly POSIX-standard, that's why there's a need for that flag. You should know that had you have spent more time learning or just reading this README instead of falsely accusing others on reddit.

---

This is a very small build environment that helps you to develop for UEFI under Linux (and other POSIX systems). It was greatly inspired by gnu-efi (big big kudos to those guys), but it is a lot smaller, easier to integrate (works with LLVM Clang and GNU gcc both) and easier to use because it provides a POSIX like API for your UEFI application.

An UEFI environment consist of two parts: a firmware with GUID protocol interfaces and a user library. We cannot change the former, but we can make the second friendlier. That's what POSIX-UEFI does for your application. It is a small API wrapper library around the GUID protocols, not a fully blown POSIX compatible libc implementation.

You have two options on how to integrate it into your project:

Distributing as Static Library

Same method as with gnu-efi, not really recommended. In the uefi directory, run

$ USE_GCC=1 make

This will create build/uefi with all the necessary files in it. These are:

• crt0.o, the run-time that bootstraps POSIX-UEFI
• link.ld, the linker script you must use with POSIX-UEFI (same as with gnu-efi)
• libuefi.a, the library itself
• uefi.h, the all-in-one C / C++ header

You can use this and link your application with it, but you won't be able to recompile it, plus you're on your own with the linking and converting.

Strictly speaking you'll only need crt0.o and link.ld, that will get you started and will call your application's "main()", but to get libc functions like memcmp, strcpy, malloc or fopen, you'll have to link with libuefi.a too.

Clang is configured in a way to directly create PE files, so the resulting library's name is libuefi.dll.a.

Distributing as Source

This is the preferred way, as it also provides a Makefile to set up your toolchain properly.

1. simply copy the uefi directory into your source tree (or set up a git submodule and a symlink). A dozen files, about 132K in total.
2. create an extremely simple Makefile like the one below
3. compile your code for UEFI by running make

TARGET = helloworld.efi
include uefi/Makefile

An example helloworld.c goes like this:

#include <uefi.h>

int main(int argc, char **argv)
{
    printf("Hello World!\n");
    return 0;
}

By default it uses Clang + lld, and PE is generated directly without conversion. If USE_GCC is set, then the host native's GNU gcc + ld is used to create a shared object and get converted into an .efi file with objcopy, just like how gnu-efi does.

NOTE: if you don't want to clone this entire repo, just the uefi directory,

git clone --no-checkout https://gitlab.com/bztsrc/posix-uefi.git .
git sparse-checkout set --no-cone '/uefi/*'
git checkout

Available Makefile Options

Variable	Description
TARGET	the target application (required)
SRCS	list of source files you want to compile (defaults to *.c *.S)
CFLAGS	compiler flags you want to use (empty by default, like "-Wall -pedantic -std=c99")
LDFLAGS	linker flags you want to use (I don't think you'll ever need this, just in case)
LIBS	additional libraries you want to link with (like "-lm", only static .a libraries allowed)
EXTRA	any additional object files you might want to link with, these are also called as makefile rules before compiling
ALSO	additional makefile rules to be called after compiling
OUTDIR	if given, then your project's object files are generated into this directory (by default not set)
USE_GCC	set this if you want native GNU gcc + ld + objcopy instead of LLVM Clang + Lld
ARCH	the target architecture

Here's a more advanced Makefile example:

ARCH = x86_64
TARGET = helloworld.efi
SRCS = $(wildcard *.c)
CFLAGS = -pedantic -Wall -Wextra -Werror --std=c11 -O2
LDFLAGS =
LIBS = -lm
OUTDIR = build/loader

USE_GCC = 1
include uefi/Makefile

The build environment configurator was created in a way that it can handle any number of architectures, however only x86_64 crt0 has been throughfully tested for now. There's an aarch64 and a riscv64 crt0 too, but since I don't have neither an ARM UEFI, nor a RISC-V UEFI board, these haven't been tested at all. Should work though. If you want to port it to another architecture, all you need is a setjmp struct in uefi.h, and an appropriate crt_X.c file. That's it. Everything else was coded in an architecture independent way.

Available Configuration Options

These can be set at the beginning of uefi.h.

Define	Description
UEFI_NO_UTF8	Do not use transparent UTF-8 conversion between the application and the UEFI interface
UEFI_NO_TRACK_ALLOC	Do not keep track of allocated buffers (faster, but causes out of bound reads on realloc)

Notable Differences to POSIX libc

This library is nowhere near as complete as glibc or musl for example. It only provides the very basic libc functions for you, because simplicity was one of its main goals. It is the best to say this is just wrapper around the UEFI API, rather than a POSIX compatible libc.

All strings in the UEFI environment are stored with 16 bits wide characters. The library provides wchar_t type for that, and the UEFI_NO_UTF8 define to convert between char and wchar_t transparently. If you have UEFI_NO_UTF8, then for example your main() will NOT be like main(int argc, char **argv), but main(int argc, wchar_t **argv) instead. All the other string related libc functions (like strlen() for example) will use this wide character type too. For this reason, you must specify your string literals with L"" and characters with L''. To handle both configurations, char_t type is defined, which is either char or wchar_t, and the CL() macro which might add the L prefix to constant literals. Functions that are supposed to handle characters in int type (like getchar, putchar), do not truncate to unsigned char, rather to wchar_t.

Sadly UEFI has no concept of reallocation. AllocatePool does not accept input, and there's no way to query the size of an already allocated buffer. So we are left with two bad options with realloc:

1. we keep track of sizes ourselves, which means more complexcity and a considerable overhead, so performance loss.
2. make peace with the fact that copying data to the new buffer unavoidably reads out of bounds from the old buffer. You can choose this latter with the UEFI_NO_TRACK_ALLOC define.

File types in dirent are limited to directories and files only (DT_DIR, DT_REG), but for stat in addition to S_IFDIR and S_IFREG, S_IFIFO (for console streams: stdin, stdout, stderr), S_IFBLK (for Block IO) and S_IFCHR (for Serial IO) also returned.

Note that getenv and setenv aren't POSIX standard, because UEFI environment variables are binary blobs.

That's about it, everything else is the same.

List of Provided POSIX Functions

dirent.h

Function	Description
opendir	as usual, but might accept wide char strings
readdir	as usual
rewinddir	as usual
closedir	as usual

Because UEFI has no concept of device files nor of symlinks, dirent fields are limited and only DT_DIR and DT_REG supported.

stdlib.h

Function	Description
atoi	as usual, but might accept wide char strings and understands "0x" prefix
atol	as usual, but might accept wide char strings and understands "0x" prefix
strtol	as usual, but might accept wide char strings
malloc	as usual
calloc	as usual
realloc	as usual
free	as usual
abort	as usual
exit	as usual
exit_bs	leave this entire UEFI bullshit behind (exit Boot Services)
mbtowc	as usual (UTF-8 char to wchar_t)
wctomb	as usual (wchar_t to UTF-8 char)
mbstowcs	as usual (UTF-8 string to wchar_t string)
wcstombs	as usual (wchar_t string to UTF-8 string)
srand	as usual
rand	as usual, but uses EFI_RNG_PROTOCOL if possible
getenv	pretty UEFI specific
setenv	pretty UEFI specific

int exit_bs();

Exit Boot Services. Returns 0 on success. You won't be able to return from main() after calling this successfully, you must transfer control directly.

uint8_t *getenv(char_t *name, uintn_t *len);

Query the value of environment variable name. On success, len is set, and a malloc'd buffer returned. It is the caller's responsibility to free the buffer later. On error returns NULL.

int setenv(char_t *name, uintn_t len, uint8_t *data);

Sets an environment variable by name with data of length len. On success returns 1, otherwise 0 on error.

stdio.h

Function	Description
remove	as usual, but might accept wide char strings
fopen	as usual, but might accept wide char strings, also for mode
fclose	as usual
fflush	as usual
fread	as usual, only real files and blk io accepted (no stdin)
fwrite	as usual, only real files and blk io accepted (no stdout nor stderr)
fseek	as usual, only real files and blk io accepted (no stdin, stdout, stderr)
ftell	as usual, only real files and blk io accepted (no stdin, stdout, stderr)
feof	as usual, only real files and blk io accepted (no stdin, stdout, stderr)
fprintf	as usual, might be wide char strings, BUFSIZ, files, ser, stdout, stderr
printf	as usual, might be wide char strings, max BUFSIZ, stdout only
sprintf	as usual, might be wide char strings, max BUFSIZ
vfprintf	as usual, might be wide char strings, BUFSIZ, files, ser, stdout, stderr
vprintf	as usual, might be wide char strings, max BUFSIZ, stdout only
vsprintf	as usual, might be wide char strings, max BUFSIZ
snprintf	as usual, might be wide char strings
vsnprintf	as usual, might be wide char strings
getchar	as usual, blocking, stdin only (no stream redirects), returns UNICODE
getchar_ifany	non-blocking, returns 0 if there was no key press, UNICODE otherwise
putchar	as usual, stdout only (no stream redirects)

String formating is limited; only supports padding via positive number prefixes, %d, %i, %x, %X, %c, %s, %q and %p (no %e, %f, %g, no asterisk and dollar). When UEFI_NO_UTF8 is defined, then formating operates on wchar_t, so it also supports the non-standard %S (printing an UTF-8 string) and %Q (printing an escaped UTF-8 string). These functions don't allocate memory, but in return the total length of the output string cannot be longer than BUFSIZ (8k if you haven't defined otherwise), except for the variants which have a maxlen argument. For convenience, %D requires efi_physical_address_t as argument, and it dumps memory, 16 bytes or one line at once. With the padding modifier you can dump more lines, for example %5D gives you 5 lines (80 dumped bytes).

File open modes: "r" read, "w" write, "a" append. Because of UEFI peculiarities, "wd" creates directory.

Special "device files" you can open:

Name	Description
/dev/stdin	returns ST->ConIn
/dev/stdout	returns ST->ConOut, fprintf
/dev/stderr	returns ST->StdErr, fprintf
/dev/serial(baud)	returns Serial IO protocol, fread, fwrite, fprintf
/dev/disk(n)	returns Block IO protocol, fseek, ftell, fread, fwrite, feof

With Block IO, fseek and buffer size for fread and fwrite is always truncated to the media's block size. So fseek(513) for example will seek to 512 with standard block sizes, and 0 with large 4096 block sizes. To detect the media's block size, use fstat.

if(!fstat(f, &st))
    block_size = st.st_size / st.st_blocks;

To interpret a GPT, there are typedefs like efi_partition_table_header_t and efi_partition_entry_t which you can point to the read data.

string.h

Function	Description
memcpy	as usual, works on bytes
memmove	as usual, works on bytes
memset	as usual, works on bytes
memcmp	as usual, works on bytes
memchr	as usual, works on bytes
memrchr	as usual, works on bytes
memmem	as usual, works on bytes
memrmem	as usual, works on bytes
strcpy	might work on wide char strings
strncpy	might work on wide char strings
strcat	might work on wide char strings
strncat	might work on wide char strings
strcmp	might work on wide char strings
strncmp	might work on wide char strings
strdup	might work on wide char strings
strchr	might work on wide char strings
strrchr	might work on wide char strings
strstr	might work on wide char strings
strtok	might work on wide char strings
strtok_r	might work on wide char strings
strlen	might work on wide char strings

sys/stat.h

Function	Description
stat	as usual, but might accept wide char strings
fstat	UEFI doesn't have fd, so it uses FILE*
mkdir	as usual, but might accept wide char strings, and mode unused

Because UEFI has no concept of device major and minor number nor of inodes, struct stat's fields are limited. The actual implementation of fstat is in stdio.c, because it needs to access static variables defined there.

time.h

Function	Description
localtime	argument unused, always returns current time in struct tm
mktime	as usual
time	as usual

unistd.h

Function	Description
usleep	as usual (uses BS->Stall)
sleep	as usual
unlink	as usual, but might accept wide char strings
rmdir	as usual, but might accept wide char strings

Accessing UEFI Services

It is very likely that you want to call UEFI specific functions directly. For that, POSIX-UEFI specifies some globals in uefi.h:

Global Variable	Description
*BS, gBS	efi_boot_services_t, pointer to the Boot Time Services
*RT, gRT	efi_runtime_t, pointer to the Runtime Services
*ST, gST	efi_system_table_t, pointer to the UEFI System Table
IM	efi_handle_t of your Loaded Image

The EFI structures, enums, typedefs and defines are all converted to ANSI C standard POSIX style, for example BOOLEAN -> boolean_t, UINTN -> uintn_t, EFI_MEMORY_DESCRIPTOR -> efi_memory_descriptor_t, and of course EFI_BOOT_SERVICES -> efi_boot_services_t.

Calling UEFI functions is as simple as with EDK II, just do the call, no need for "uefi_call_wrapper":

    ST->ConOut->OutputString(ST->ConOut, L"Hello World!\r\n");

(Note: unlike with printf, with OutputString you must use L"" and also print carrige return L"\r" before L"\n". These are the small things that POSIX-UEFI does for you to make your life comfortable.)

There are two additional, non-POSIX calls in the library. One is exit_bs() to exit Boot Services, and the other is a non-blocking version getchar_ifany().

Unlike gnu-efi, POSIX-UEFI does not pollute your application's namespace with unused GUID variables. It only provides header definitions, so you must create each GUID instance if and when you need them.

Example:

efi_guid_t gopGuid = EFI_GRAPHICS_OUTPUT_PROTOCOL_GUID;
efi_gop_t *gop = NULL;

status = BS->LocateProtocol(&gopGuid, NULL, (void**)&gop);

Also unlike gnu-efi, POSIX-UEFI does not provide standard EFI headers. It expects that you have installed those under /usr/include/efi from EDK II or gnu-efi, and POSIX-UEFI makes it possible to use those system wide headers without naming conflicts. POSIX-UEFI itself ships the very minimum set of typedefs and structs (with POSIX-ized names).

#include <efi.h>
#include <uefi.h> /* this will work as expected! Both POSIX-UEFI and EDK II / gnu-efi typedefs available */

The advantage of this is that you can use the simplicity of the POSIX-UEFI library and build environment, while getting access to the most up-to-date protocol and interface definitions at the same time.

IMPORTANT NOTE

In some cases the combination of GNU-EFI headers and Clang might incorrectly define uint64_t as 32 bits. If this happens, then

#undef __STDC_VERSION__
#include <efi.h>
#include <uefi.h>

should workaround the problem by avoiding the inclusion of stdint.h and defining uint64_t by efibind.h as unsigned long long.

License

POSIX_UEFI is licensed under the terms of the MIT license.

Contributors

I'd like to say thanks to @vladimir132 for a through testing and very accurate and detailed feedbacks.

Cheers,

bzt