[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <426BE71F-FAF6-4C68-944D-9F559FF42737@zytor.com>
Date: Fri, 07 Mar 2025 03:13:55 -0800
From: "H. Peter Anvin" <hpa@...or.com>
To: Uros Bizjak <ubizjak@...il.com>
CC: linux-kernel@...r.kernel.org,
tip-bot2 for Uros Bizjak <tip-bot2@...utronix.de>,
linux-tip-commits@...r.kernel.org, Ingo Molnar <mingo@...nel.org>,
David Woodhouse <dwmw@...zon.co.uk>, Baoquan He <bhe@...hat.com>,
Vivek Goyal <vgoyal@...hat.com>, Dave Young <dyoung@...hat.com>,
Ard Biesheuvel <ardb@...nel.org>, x86@...nel.org
Subject: Re: [tip: x86/asm] x86/kexec: Merge x86_32 and x86_64 code using macros from <asm/asm.h>
On March 7, 2025 1:20:17 AM PST, Uros Bizjak <ubizjak@...il.com> wrote:
>On Fri, Mar 7, 2025 at 4:00 AM H. Peter Anvin <hpa@...or.com> wrote:
>>
>> On March 6, 2025 1:33:43 PM PST, tip-bot2 for Uros Bizjak <tip-bot2@...utronix.de> wrote:
>> >The following commit has been merged into the x86/asm branch of tip:
>> >
>> >Commit-ID: aa3942d4d12ef57f031faa2772fe410c24191e36
>> >Gitweb: https://git.kernel.org/tip/aa3942d4d12ef57f031faa2772fe410c24191e36
>> >Author: Uros Bizjak <ubizjak@...il.com>
>> >AuthorDate: Thu, 06 Mar 2025 15:52:11 +01:00
>> >Committer: Ingo Molnar <mingo@...nel.org>
>> >CommitterDate: Thu, 06 Mar 2025 22:04:48 +01:00
>> >
>> >x86/kexec: Merge x86_32 and x86_64 code using macros from <asm/asm.h>
>> >
>> >Merge common x86_32 and x86_64 code in crash_setup_regs()
>> >using macros from <asm/asm.h>.
>> >
>> >The compiled object files before and after the patch are unchanged.
>> >
>> >Signed-off-by: Uros Bizjak <ubizjak@...il.com>
>> >Signed-off-by: Ingo Molnar <mingo@...nel.org>
>> >Cc: David Woodhouse <dwmw@...zon.co.uk>
>> >Cc: Baoquan He <bhe@...hat.com>
>> >Cc: Vivek Goyal <vgoyal@...hat.com>
>> >Cc: Dave Young <dyoung@...hat.com>
>> >Cc: Ard Biesheuvel <ardb@...nel.org>
>> >Cc: "H. Peter Anvin" <hpa@...or.com>
>> >Link: https://lore.kernel.org/r/20250306145227.55819-1-ubizjak@gmail.com
>> >---
>> > arch/x86/include/asm/kexec.h | 58 +++++++++++++++--------------------
>> > 1 file changed, 25 insertions(+), 33 deletions(-)
>> >
>> >diff --git a/arch/x86/include/asm/kexec.h b/arch/x86/include/asm/kexec.h
>> >index 8ad1874..e3589d6 100644
>> >--- a/arch/x86/include/asm/kexec.h
>> >+++ b/arch/x86/include/asm/kexec.h
>> >@@ -18,6 +18,7 @@
>> > #include <linux/string.h>
>> > #include <linux/kernel.h>
>> >
>> >+#include <asm/asm.h>
>> > #include <asm/page.h>
>> > #include <asm/ptrace.h>
>> >
>> >@@ -71,41 +72,32 @@ static inline void crash_setup_regs(struct pt_regs *newregs,
>> > if (oldregs) {
>> > memcpy(newregs, oldregs, sizeof(*newregs));
>> > } else {
>> >+ asm volatile("mov %%" _ASM_BX ",%0" : "=m"(newregs->bx));
>> >+ asm volatile("mov %%" _ASM_CX ",%0" : "=m"(newregs->cx));
>> >+ asm volatile("mov %%" _ASM_DX ",%0" : "=m"(newregs->dx));
>> >+ asm volatile("mov %%" _ASM_SI ",%0" : "=m"(newregs->si));
>> >+ asm volatile("mov %%" _ASM_DI ",%0" : "=m"(newregs->di));
>> >+ asm volatile("mov %%" _ASM_BP ",%0" : "=m"(newregs->bp));
>> >+ asm volatile("mov %%" _ASM_AX ",%0" : "=m"(newregs->ax));
>> >+ asm volatile("mov %%" _ASM_SP ",%0" : "=m"(newregs->sp));
>> >+#ifdef CONFIG_X86_64
>> >+ asm volatile("mov %%r8,%0" : "=m"(newregs->r8));
>> >+ asm volatile("mov %%r9,%0" : "=m"(newregs->r9));
>> >+ asm volatile("mov %%r10,%0" : "=m"(newregs->r10));
>> >+ asm volatile("mov %%r11,%0" : "=m"(newregs->r11));
>> >+ asm volatile("mov %%r12,%0" : "=m"(newregs->r12));
>> >+ asm volatile("mov %%r13,%0" : "=m"(newregs->r13));
>> >+ asm volatile("mov %%r14,%0" : "=m"(newregs->r14));
>> >+ asm volatile("mov %%r15,%0" : "=m"(newregs->r15));
>> >+#endif
>> >+ asm volatile("mov %%ss,%k0" : "=a"(newregs->ss));
>> >+ asm volatile("mov %%cs,%k0" : "=a"(newregs->cs));
>> > #ifdef CONFIG_X86_32
>> >- asm volatile("movl %%ebx,%0" : "=m"(newregs->bx));
>> >- asm volatile("movl %%ecx,%0" : "=m"(newregs->cx));
>> >- asm volatile("movl %%edx,%0" : "=m"(newregs->dx));
>> >- asm volatile("movl %%esi,%0" : "=m"(newregs->si));
>> >- asm volatile("movl %%edi,%0" : "=m"(newregs->di));
>> >- asm volatile("movl %%ebp,%0" : "=m"(newregs->bp));
>> >- asm volatile("movl %%eax,%0" : "=m"(newregs->ax));
>> >- asm volatile("movl %%esp,%0" : "=m"(newregs->sp));
>> >- asm volatile("movl %%ss, %%eax;" :"=a"(newregs->ss));
>> >- asm volatile("movl %%cs, %%eax;" :"=a"(newregs->cs));
>> >- asm volatile("movl %%ds, %%eax;" :"=a"(newregs->ds));
>> >- asm volatile("movl %%es, %%eax;" :"=a"(newregs->es));
>> >- asm volatile("pushfl; popl %0" :"=m"(newregs->flags));
>> >-#else
>> >- asm volatile("movq %%rbx,%0" : "=m"(newregs->bx));
>> >- asm volatile("movq %%rcx,%0" : "=m"(newregs->cx));
>> >- asm volatile("movq %%rdx,%0" : "=m"(newregs->dx));
>> >- asm volatile("movq %%rsi,%0" : "=m"(newregs->si));
>> >- asm volatile("movq %%rdi,%0" : "=m"(newregs->di));
>> >- asm volatile("movq %%rbp,%0" : "=m"(newregs->bp));
>> >- asm volatile("movq %%rax,%0" : "=m"(newregs->ax));
>> >- asm volatile("movq %%rsp,%0" : "=m"(newregs->sp));
>> >- asm volatile("movq %%r8,%0" : "=m"(newregs->r8));
>> >- asm volatile("movq %%r9,%0" : "=m"(newregs->r9));
>> >- asm volatile("movq %%r10,%0" : "=m"(newregs->r10));
>> >- asm volatile("movq %%r11,%0" : "=m"(newregs->r11));
>> >- asm volatile("movq %%r12,%0" : "=m"(newregs->r12));
>> >- asm volatile("movq %%r13,%0" : "=m"(newregs->r13));
>> >- asm volatile("movq %%r14,%0" : "=m"(newregs->r14));
>> >- asm volatile("movq %%r15,%0" : "=m"(newregs->r15));
>> >- asm volatile("movl %%ss, %%eax;" :"=a"(newregs->ss));
>> >- asm volatile("movl %%cs, %%eax;" :"=a"(newregs->cs));
>> >- asm volatile("pushfq; popq %0" :"=m"(newregs->flags));
>> >+ asm volatile("mov %%ds,%k0" : "=a"(newregs->ds));
>> >+ asm volatile("mov %%es,%k0" : "=a"(newregs->es));
>> > #endif
>> >+ asm volatile("pushf\n\t"
>> >+ "pop %0" : "=m"(newregs->flags));
>> > newregs->ip = _THIS_IP_;
>> > }
>> > }
>>
>> Incidentally, doing this in C code is obviously completely broken, especially doing it in multiple statements. You have no idea what the compiler has messed with before you get there.
>
>These are "asm volatile" statemets, so at least they won't be
>scheduled in a different way. OTOH, please note that the patch is very
>carefully written to not change code flow, usage of hardregs in the
>inline asm is usually the sign of fragile code.
>
>Uros.
>
That doesn't matter, though; that only means they can't be moved relatively to each other, but the compiler is perfectly capable of inserting code before, or in between.
Your patch might be a functional null, but the code is broken on a deep and fundamental basis.
Powered by blists - more mailing lists