Re: [PATCH v2 0/5] kexec: use IS_ENABLED(CONFIG_KEXEC_CORE) instead of #ifdef

[Jisheng Zhang <jszhang@kernel.org>]

On Wed, Jan 19, 2022 at 05:33:22PM +0800, Baoquan He wrote:
> On 01/19/22 at 09:52am, Alexandre Ghiti wrote:
> > Hi Baoquan,
> > 
> > On Wed, Jan 19, 2022 at 9:11 AM Baoquan He <bhe@redhat.com> wrote:
> > >
> > > On 01/18/22 at 10:13pm, Jisheng Zhang wrote:
> > > > On Sun, Jan 16, 2022 at 09:38:47PM +0800, Baoquan He wrote:
> > > > > Hi Jisheng,
> > > >
> > > > Hi Baoquan,
> > > >
> > > > >
> > > > > On 12/07/21 at 12:05am, Jisheng Zhang wrote:
> > > > > > Replace the conditional compilation using "#ifdef CONFIG_KEXEC_CORE"
> > > > > > by a check for "IS_ENABLED(CONFIG_KEXEC_CORE)", to simplify the code
> > > > > > and increase compile coverage.
> > > > >
> > > > > I go through this patchset, You mention the benefits it brings are
> > > > > 1) simplity the code;
> > > > > 2) increase compile coverage;
> > > > >
> > > > > For benefit 1), it mainly removes the dummy function in x86, arm and
> > > > > arm64, right?
> > > >
> > > > Another benefit: remove those #ifdef #else #endif usage. Recently, I
> > > > fixed a bug due to lots of "#ifdefs":
> > > > http://lists.infradead.org/pipermail/linux-riscv/2021-December/010607.html
> > >
> > > Glad to know the fix. While, sometime the ifdeffery is necessary. I am
> > > sorry about the one in riscv and you have fixed, it's truly a bug . But,
> > > the increasing compile coverage at below you tried to make, it may cause
> > > issue. Please see below my comment.
> > >
> > > >
> > > > >
> > > > > For benefit 2), increasing compile coverage, could you tell more how it
> > > > > achieves and why it matters? What if people disables CONFIG_KEXEC_CORE in
> > > > > purpose? Please forgive my poor compiling knowledge.
> > > >
> > > > Just my humble opinion, let's compare the code::
> > > >
> > > > #ifdef CONFIG_KEXEC_CORE
> > > >
> > > > code block A;
> > > >
> > > > #endif
> > > >
> > > > If KEXEC_CORE is disabled, code block A won't be compiled at all, the
> > > > preprocessor will remove code block A;
> > > >
> > > > If we convert the code to:
> > > >
> > > > if (IS_ENABLED(CONFIG_KEXEC_CORE)) {
> > > >       code block A;
> > > > }
> > > >
> > > > Even if KEXEC_CORE is disabled, code block A is still compiled.
> > >
> > > This is what I am worried about. Before, if CONFIG_KEXEC_CORE is
> > > unset, those relevant codes are not compiled in. I can't see what
> > > benefit is brought in if compiled in the unneeded code block. Do I miss
> > > anything?
> > >
> > 
> > This is explained in Documentation/process/coding-style.rst "21)
> > Conditional Compilation".
> 
> Thanks for the pointer, Alex.
> 
> I read that part, while my confusion isn't gone. With the current code,
> CONFIG_KEXEC_CORE is set,
>   - reserve_crashkernel_low() and reserve_crashkernel() compiled in.

Although the code block will be compiled, but the code block will be
optimized out.

> CONFIG_KEXEC_CORE is unset,
>   - reserve_crashkernel_low() and reserve_crashkernel() compiled out. 
> 
> After this patch applied, does it have the same effect as the old code?

I compared the .o, and can confirm they acchieve the same effect.

> 
> arch/x86/kernel/setup.c:
> 
> before
> ======
> #ifdef CONFIG_KEXEC_CORE
> static int __init reserve_crashkernel_low(void)
> {
> 	......
> }
> static void __init reserve_crashkernel(void)
> {
> 	......
> }
> #else
> static void __init reserve_crashkernel(void)
> {
> }
> #endif
> 
> after
> =======
> static int __init reserve_crashkernel_low(void)
> {
> 	......
> }
> static void __init reserve_crashkernel(void)
> {
> 	......
> 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
>                 return;
> 	......
> }
>