On 9/7/21 2:00 AM, Jan Beulich wrote:
On 06.09.2021 18:22, Andrew Cooper wrote:
On 06/09/2021 16:52, Jan Beulich wrote:
On 03.09.2021 21:06, Daniel P. Smith wrote:
--- /dev/null
+++ b/xen/include/xen/alternative-call.h
@@ -0,0 +1,63 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef XEN_ALTERNATIVE_CALL
+#define XEN_ALTERNATIVE_CALL
+
+/*
+ * Some subsystems in Xen may have multiple implementions, which can be
+ * resolved to a single implementation at boot time. By default, this will
+ * result in the use of function pointers.
+ *
+ * Some architectures may have mechanisms for dynamically modifying .text.
+ * Using this mechnaism, function pointers can be converted to direct calls
+ * which are typically more efficient at runtime.
+ *
+ * For architectures to support:
+ *
+ * - Implement alternative_{,v}call() in asm/alternative.h. Code generation
+ * requirements are to emit a function pointer call at build time, and stash
+ * enough metadata to simplify the call at boot once the implementation has
+ * been resolved.
+ * - Select ALTERNATIVE_CALL in Kconfig.
+ *
+ * To use:
+ *
+ * Consider the following simplified example.
+ *
+ * 1) struct foo_ops __alt_call_maybe_initdata ops;
+ *
+ * 2) const struct foo_ops __initconst foo_a_ops = { ... };
+ * const struct foo_ops __initconst foo_b_ops = { ... };
+ *
+ * void foo_init(void)
+ * {
+ * ...
+ * if ( use_impl_a )
+ * ops = *foo_a_ops;
+ * else if ( use_impl_b )
+ * ops = *foo_b_ops;
+ * ...
+ * }
+ *
+ * 3) alternative_call(ops.bar, ...);
+ *
+ * There needs to a single ops object (1) which will eventually contain the
+ * function pointers. This should be populated in foo's init() function (2)
+ * by one of the available implementations. To call functions, use
+ * alternative_{,v}call() referencing the main ops object (3).
+ */
+
+#ifdef CONFIG_ALTERNATIVE_CALL
+
+#include <asm/alternative.h>
+
+#define __alt_call_maybe_initdata __initdata
My v3 comment here was:
"I think it wants (needs) clarifying that this may only be used if
the ops object is used exclusively in alternative_{,v}call()
instances (besides the original assignments to it, of course)."
I realize this was slightly too strict, as other uses from .init.*
are of course also okay, but I continue to think that - in
particular with the example using it - there should be a warning
about this possible pitfall. Or am I merely unable to spot the
wording change somewhere in the comment?
Such a comment is utterly pointless. initdata has a well known meaning,
and a comment warning about the effects of it is just teaching
developers to suck eggs[1]
Well, okay then - at least the definition of __alt_call_maybe_initdata
isn't far away from the comment. (What I'm not convinced of is that
people knowing __initdata's meaning necessarily need to correctly
infer __alt_call_maybe_initdata's.)
Two other observations about the comment though, which I'd like to be
taken care of (perhaps while committing):
- __initconst wants to become __initconstrel.
- foo_init(), seeing that there are section annotations elsewhere,
wants to be marked __init.
Then
Acked-by: Jan Beulich <jbeul...@suse.com>
Daniel, you having made changes (even if just minor ones) imo requires
you S-o-b on the patch alongside Andrew's.
Ack, I realized after sending I didn't SoB it, my apologies on that.
v/r,
dps