On Wed, Mar 03, 2021 at 06:57:09PM +0100, Christophe Leroy wrote:

Le 03/03/2021 à 18:46, Will Deacon a écrit :

quoted

On Wed, Mar 03, 2021 at 06:38:16PM +0100, Christophe Leroy wrote:

quoted

Le 03/03/2021 à 18:28, Will Deacon a écrit :

quoted

On Tue, Mar 02, 2021 at 05:25:17PM +0000, Christophe Leroy wrote:

quoted

This code provides architectures with a way to build command line
based on what is built in the kernel and what is handed over by the
bootloader, based on selected compile-time options.

Signed-off-by: Christophe Leroy <redacted>
---
   include/linux/cmdline.h | 62 +++++++++++++++++++++++++++++++++++++++++
   1 file changed, 62 insertions(+)
   create mode 100644 include/linux/cmdline.h

diff --git a/include/linux/cmdline.h b/include/linux/cmdline.h
new file mode 100644
index 000000000000..ae3610bb0ee2
--- /dev/null
+++ b/include/linux/cmdline.h

@@ -0,0 +1,62 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _LINUX_CMDLINE_H
+#define _LINUX_CMDLINE_H
+
+static __always_inline size_t cmdline_strlen(const char *s)
+{
+	const char *sc;
+
+	for (sc = s; *sc != '\0'; ++sc)
+		; /* nothing */
+	return sc - s;
+}
+
+static __always_inline size_t cmdline_strlcat(char *dest, const char *src, size_t count)
+{
+	size_t dsize = cmdline_strlen(dest);
+	size_t len = cmdline_strlen(src);
+	size_t res = dsize + len;
+
+	/* This would be a bug */
+	if (dsize >= count)
+		return count;
+
+	dest += dsize;
+	count -= dsize;
+	if (len >= count)
+		len = count - 1;
+	memcpy(dest, src, len);
+	dest[len] = 0;
+	return res;
+}

Why are these needed instead of using strlen and strlcat directly?

Because on powerpc (at least), it will be used in prom_init, it is very
early in the boot and KASAN shadow memory is not set up yet so calling
generic string functions would crash the board.

Hmm. We deliberately setup a _really_ early shadow on arm64 for this, can
you not do something similar? Failing that, I think it would be better to
offer the option for an arch to implement cmdline_*, but have then point to
the normal library routines by default.

I don't think it is possible to setup an earlier early shadow.

At the point we are in prom_init, the code is not yet relocated at the
address it was linked for, and it is running with the MMU set by the
bootloader, I can't imagine being able to setup MMU entries for the early
shadow KASAN yet without breaking everything.

That's very similar to us; we're not relocated, although we are at least
in control of the MMU (which is using a temporary set of page-tables).

Is it really worth trying to point to the normal library routines by default
? It is really only a few lines of code hence only not many bytes, and
anyway they are in __init section so they get discarded at the end.

I would prefer to use the normal routines by default and allow architectures
to override them based on their needs, otherwise we'll end up trying to
maintain a "lowest-common-dominator" set of string routines that can be
safely run in whatever different constraints different architectures have.

Will