Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Ammar Faizi <ammar.faizi@students.amikom.ac.id>
To:     Willy Tarreau <w@1wt.eu>
Cc:     Ammar Faizi <ammar.faizi@students.amikom.ac.id>,
        Paul Walmsley <paul.walmsley@sifive.com>,
        Palmer Dabbelt <palmer@dabbelt.com>,
        Albert Ou <aou@eecs.berkeley.edu>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Andy Lutomirski <luto@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, x86@kernel.org,
        "H. Peter Anvin" <hpa@zytor.com>,
        David Laight <David.Laight@ACULAB.COM>,
        Peter Cordes <peter@cordes.ca>,
        Bedirhan KURT <windowz414@gnuweeb.org>,
        Louvian Lyndal <louvianlyndal@gmail.com>
Subject: Re: [PATCH 2/2] tools/nolibc: x86-64: Fix startup code bug
Date:   Fri, 15 Oct 2021 15:57:01 +0700
Message-Id: <6sZ9qpcJvtqCksJQVaiZyA-ammarfaizi2@gnuweeb.org>
In-Reply-To: <CtypPaXXhVINRV0090UVzA-ammarfaizi2@gnuweeb.org>
References: <CtypPaXXhVINRV0090UVzA-ammarfaizi2@gnuweeb.org>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Hi,

This is a code to test.

Compile with:
  gcc -O3 -ggdb3 -nostdlib -o test test.c

Technical explanation:
The System V ABI mandates the %rsp must be 16-byte aligned before
performing a function call, but the current nolibc.h violates it.

This %rsp alignment violation makes the callee can't align its stack
properly. Note that the callee may have a situation where it requires
vector aligned move. For example, `movaps` with memory operand w.r.t.
xmm registers, it requires the src/dst address be 16-byte aligned.

Since the callee can't align its stack properly, it will segfault when
executing `movaps`. The following C code is the reproducer and test
to ensure the bug really exists and this patch fixes it.

```
/* SPDX-License-Identifier: LGPL-2.1 OR MIT */

/*
 * Bug reproducer and test for Willy's nolibc (x86-64) by Ammar.
 *
 * Compile with:
 *  gcc -O3 -ggdb3 -nostdlib -o test test.c
 */

#include "tools/include/nolibc/nolibc.h"

static void dump_argv(int argc, char *argv[])
{
	int i;
	const char str[] = "\nDumping argv...\n";

	write(1, str, strlen(str));
	for (i = 0; i < argc; i++) {
		write(1, argv[i], strlen(argv[i]));
		write(1, "\n", 1);
	}
}

static void dump_envp(char *envp[])
{
	int i = 0;
	const char str[] = "\nDumping envp...\n";

	write(1, str, strlen(str));
	while (envp[i] != NULL) {
		write(1, envp[i], strlen(envp[i]));
		write(1, "\n", 1);
		i++;
	}
}

#define read_barrier(PTR) __asm__ volatile(""::"r"(PTR):"memory")

__attribute__((__target__("sse2")))
static void test_sse_move_aligned(void)
{
	int i;
	int arr[32] __attribute__((__aligned__(16)));

	/*
	 * The assignment inside this loop is very likely
	 * performed with aligned move, thus if we don't
	 * align the %rsp properly, it will fault!
	 *
	 * If we fault due to misaligned here, then there
	 * must be a caller below us that violates SysV
	 * ABI w.r.t. to %rsp alignment before func call.
	 */
	for (i = 0; i < 32; i++)
		arr[i] = 1;

	read_barrier(arr);
}

int main(int argc, char *argv[], char *envp[])
{
	dump_argv(argc, argv);
	dump_envp(envp);
	test_sse_move_aligned();
	return 0;
}
```

-- 
Ammar Faizi