Thread (10 messages) read the whole thread 10 messages, 4 authors, 2024-01-08

Re: [PATCH v1 1/1] arm64: remove unnecessary ifdefs around is_compat_task()

From: Leonardo Bras <hidden>
Date: 2024-01-08 17:09:31
Also in: lkml

On Mon, Jan 08, 2024 at 05:26:26PM +0100, Arnd Bergmann wrote:
quoted hunk ↗ jump to hunk
On Mon, Jan 8, 2024, at 17:04, Leonardo Bras wrote:
quoted
On Mon, Jan 08, 2024 at 12:07:48PM -0300, Leonardo Bras wrote:
quoted
On Fri, Jan 05, 2024 at 03:38:05PM +0100, Arnd Bergmann wrote:
quoted
I suspect it's enough to remove all of the other
"#ifdef CONFIG_COMPAT" checks in this file and rely on
dead code elimination to remove the rest, but there might
be additional problems if some extern declarations are
hidden in an #ifdef as well.
I could remove all CONFIG_COMPAT ifdefs from this file, and for compiling 
it required a few extra defines (in other files) to be moved outside of
their #ifdef CONFIG_COMPAT. Those being:

 #define VFP_STATE_SIZE         ((32 * 8) + 4)
 #define VFP_FPSCR_STAT_MASK    0xf800009f
 #define VFP_FPSCR_CTRL_MASK    0x07f79f00

 #define COMPAT_ELF_NGREG               18
 typedef unsigned int                   compat_elf_greg_t;
 typedef compat_elf_greg_t              compat_elf_gregset_t[COMPAT_ELF_NGREG];


OTOH, the size of the final arch/arm64/kernel/ptrace.o went from 44768 to 
56328 bytes, which I understand to be undesired.
Right, unfortunately it seems that compat_arch_ptrace() is
globally visible and consequently not dropped by the compiler
in dead code elimination.
quoted
A different (and simpler) solution is to have an empty struct in case of 
!CONFIG_COMPAT, that will be optimized out in compile-time:
diff --git a/arch/arm64/kernel/ptrace.c b/arch/arm64/kernel/ptrace.c
index 9f8781f1fdfda..d2f275d8a3e6e 100644
--- a/arch/arm64/kernel/ptrace.c
+++ b/arch/arm64/kernel/ptrace.c
@@ -2107,6 +2107,9 @@ long compat_arch_ptrace(struct task_struct 
*child, compat_long_t request,

        return ret;
 }
+#else
+static const struct user_regset_view user_aarch32_view = {};
+static const struct user_regset_view user_aarch32_ptrace_view = {};
 #endif /* CONFIG_COMPAT */

 const struct user_regset_view *task_user_regset_view(struct task_struct *task)

With this the patch will build successfully and arch/arm64/kernel/ptrace.o 
will be able to keep it's original size.

Arnd, is that ok?
I don't see it being worth it if you add extra unused lines
in order to remove one more #ifdef. I would either leave the
task_user_regset_view() function unchanged here, or (if this
works) move the #ifdef down a few lines so the existing
user_regset_view structures can be shared:
@@ -1595,7 +1595,6 @@ static const struct user_regset_view user_aarch64_view = {
        .regsets = aarch64_regsets, .n = ARRAY_SIZE(aarch64_regsets)
 };
 
-#ifdef CONFIG_COMPAT
 enum compat_regset {
        REGSET_COMPAT_GPR,
        REGSET_COMPAT_VFP,
@@ -1852,6 +1851,7 @@ static const struct user_regset_view user_aarch32_ptrace_view = {
        .regsets = aarch32_ptrace_regsets, .n = ARRAY_SIZE(aarch32_ptrace_regsets)
 };
 
+#ifdef CONFIG_COMPAT
 static int compat_ptrace_read_user(struct task_struct *tsk, compat_ulong_t off,
                                   compat_ulong_t __user *ret)
 {
Ok, with the previous defines moved outside !CONFIG_COMPAT, this works as a 
charm and keeps arch/arm64/kernel/ptrace.o with the same size. 

I will send a v2 soon.

Thanks!
Leo
     Arnd

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help