[PATCH 3/3] rcutorture/nolibc: add a bit of documentation to explain how to use nolibc

From: Willy Tarreau
Date: Sat Dec 29 2018 - 10:58:55 EST


Ingo rightfully asked for a bit more documentation in the nolibc header,
so this patch adds some explanation about its purpose, how it's made, and
how to use it.

Cc: Ingo Molnar <mingo@xxxxxxxxxx>
Cc: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
Signed-off-by: Willy Tarreau <w@xxxxxx>
---
tools/testing/selftests/rcutorture/bin/nolibc.h | 90 +++++++++++++++++++++----
1 file changed, 78 insertions(+), 12 deletions(-)

diff --git a/tools/testing/selftests/rcutorture/bin/nolibc.h b/tools/testing/selftests/rcutorture/bin/nolibc.h
index 985364c..20d15a6 100644
--- a/tools/testing/selftests/rcutorture/bin/nolibc.h
+++ b/tools/testing/selftests/rcutorture/bin/nolibc.h
@@ -3,6 +3,84 @@
* Copyright (C) 2017-2018 Willy Tarreau <w@xxxxxx>
*/

+/*
+ * This file is designed to be used as a libc alternative for minimal programs
+ * with very limited requirements. It consists of a small number of syscall and
+ * types definitions, and the minimal startup code needed to call main().
+ * All syscalls are declared as static functions so that they can be optimized
+ * away by the compiler when not used.
+ *
+ * Syscalls are split between 3 levels :
+ * - the lower level is the arch-specific syscall() definition, consisting in
+ * assembly code in compound expressions. These ones are called
+ * my_syscall0() to my_syscall6() depending on the number of arguments. The
+ * MIPS implementation is limited to 5 arguments. All input arguments are
+ * cast to a long stored in a register. These expressions always return the
+ * syscall's return value as a signed long value which is often either a
+ * pointer or the negated errno value.
+ *
+ * - the second level is mostly architecture-independent. It is made of
+ * static functions called sys_<name>() which rely on my_syscallN()
+ * depending on the syscall definition. These functions are responsible
+ * for exposing the appropriate types for the syscall arguments (int,
+ * pointers, etc) and for setting the appropriate return type (often int).
+ * A few of them are architecture-specific because the syscalls are not all
+ * mapped exactly the same between architecture. For example, some archs do
+ * not implement select() and need pselect6() instead, so the sys_select()
+ * function will have to abstract this.
+ *
+ * - the third level is the libc call definition. It exposes the lower raw
+ * sys_<name>() calls in a way that looks like what a libc usually does,
+ * takes care of specific input values, and of setting errno upon error.
+ * There can be minor variations compared to standard libc calls. For
+ * example the open() call always takes 3 args here.
+ *
+ * The errno variable is declared static and unused. This way it can be
+ * optimized away if not used. However this means that a program made of
+ * multiple C files may observe different errno values (one per C file). For
+ * the type of programs this project targets it usually is not a problem. The
+ * resulting program may even be reduced by defining the NOLIBC_IGNORE_ERRNO
+ * macro, in which case the errno value will never be assigned.
+ *
+ * Some stdint-like integer types are defined. These ones are valid on all
+ * currently supported architecture, because signs are enforced, ints are
+ * assumed to be 32 bits, longs the size of a pointer and long long 64 bits.
+ * If more architectures have to be supported, this may need to be adapted.
+ *
+ * Some macro definitions like the O_* values passed to open(), and some
+ * structures like the sys_stat struct depend on the architecture.
+ *
+ * The definitions start with the architecture-specific parts, which are picked
+ * based on what the compiler knows about the target architecture, and are
+ * completed with the generic code. Since it is the compiler which sets the
+ * target architecture, cross-compiling normally works out of the box without
+ * having to specify anything.
+ *
+ * Finally some very common libc-level functions are provided. It is the case
+ * for a few functions usually found in string.h, ctype.h, or stdlib.h. Nothing
+ * is currently provided regarding stdio emulation.
+ *
+ * The macro NOLIBC is always defined, so that it is possible for a program to
+ * check this macro to know if it is being built against and decide to disable
+ * some features or simply not to include some standard libc files.
+ *
+ * Ideally this file should be split in multiple files for easier long term
+ * maintenance, but provided as a single file as it is now, it's quite
+ * convenient to use. Maybe some variations involving a set of includes at the
+ * top could work.
+ *
+ * A simple static executable may be built this way :
+ * $ gcc -fno-asynchronous-unwind-tables -fno-ident -s -Os -nostdlib \
+ * -static -include nolibc.h -lgcc -o hello hello.c
+ *
+ * A very useful calling convention table may be found here :
+ * http://man7.org/linux/man-pages/man2/syscall.2.html
+ *
+ * This doc is quite convenient though not necessarily up to date :
+ * https://w3challs.com/syscalls/
+ *
+ */
+
/* some archs (at least aarch64) don't expose the regular syscalls anymore by
* default, either because they have an "_at" replacement, or because there are
* more modern alternatives. For now we'd rather still use them.
@@ -19,18 +97,6 @@

#define NOLIBC

-/* Build a static executable this way :
- * $ gcc -fno-asynchronous-unwind-tables -fno-ident -s -Os -nostdlib \
- * -static -include nolibc.h -lgcc -o hello hello.c
- *
- * Useful calling convention table found here :
- * http://man7.org/linux/man-pages/man2/syscall.2.html
- *
- * This doc is even better :
- * https://w3challs.com/syscalls/
- */
-
-
/* this way it will be removed if unused */
static int errno;

--
2.9.0