linux-kernel - Re: [PATCH v4 00/45] x86: Kernel IBT

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 10 Mar 2022 10:05:04 +0100
From:   Peter Zijlstra <peterz@...radead.org>
To:     Nick Desaulniers <ndesaulniers@...gle.com>
Cc:     Alexei Starovoitov <alexei.starovoitov@...il.com>, x86@...nel.org,
        joao@...rdrivepizza.com, hjl.tools@...il.com, jpoimboe@...hat.com,
        andrew.cooper3@...rix.com, linux-kernel@...r.kernel.org,
        keescook@...omium.org, samitolvanen@...gle.com,
        mark.rutland@....com, alyssa.milburn@...el.com, mbenes@...e.cz,
        rostedt@...dmis.org, mhiramat@...nel.org, daniel@...earbox.net,
        andrii@...nel.org, bpf@...r.kernel.org, llvm@...ts.linux.dev
Subject: Re: [PATCH v4 00/45] x86: Kernel IBT

On Wed, Mar 09, 2022 at 04:30:28PM -0800, Nick Desaulniers wrote:

> I observed the following error when building with
> CONFIG_LTO_CLANG_FULL=y enabled:
> 
> ld.lld: error: ld-temp.o <inline asm>:7:2: symbol 'ibt_selftest_ip' is
> already defined
>         ibt_selftest_ip:
>         ^
> 
> Seems to come from
> commit a802350ba65a ("x86/ibt: Add IBT feature, MSR and #CP handling")
> 
> Commenting out the label in the inline asm, I then observed:
> vmlinux.o: warning: objtool: identify_cpu()+0x6d0: sibling call from
> callable instruction with modified stack frame
> vmlinux.o: warning: objtool: identify_cpu()+0x6e0: stack state
> mismatch: cfa1=4+64 cfa2=4+8
> These seemed to disappear when I kept CONFIG_LTO_CLANG_FULL=y but then
> disabled CONFIG_X86_KERNEL_IBT. (perhaps due to the way I hacked out
> the ibt_selftest_ip label).

Urgh.. I'm thikning this is a clang bug :/

The code in question is:


void ibt_selftest_ip(void); /* code label defined in asm below */

DEFINE_IDTENTRY_ERRORCODE(exc_control_protection)
{
	/* ... */

	if (unlikely(regs->ip == (unsigned long)ibt_selftest_ip)) {
		regs->ax = 0;
		return;
	}

	/* ... */
}

bool ibt_selftest(void)
{
	unsigned long ret;

	asm ("	lea ibt_selftest_ip(%%rip), %%rax\n\t"
	     ANNOTATE_RETPOLINE_SAFE
	     "	jmp *%%rax\n\t"
	     "ibt_selftest_ip:\n\t"
	     UNWIND_HINT_FUNC
	     ANNOTATE_NOENDBR
	     "	nop\n\t"

	     : "=a" (ret) : : "memory");

	return !ret;
}

There is only a single definition of that symbol, the one in the asm.
The other is a declaration, which is used in the exception handler to
compare against regs->ip.

So what this code does is trigger an explicit #CP and special case that
in the handler. For that the handler needs to know the special IP that
will trigger the failure, this is cummunicated with that symbol.

> Otherwise defconfig and CONFIG_LTO_CLANG_THIN=y both built and booted
> in a vm WITHOUT IBT support.
> 
> Any idea what's the status of IBT emulation in QEMU, and if it exists,
> what's the necessary `-cpu` flag to enable it?

I have a very ugly kvm patch that goes with a very ugly qemu patch to
make it work. I would very much not recommend those getting merged.

Someone with some actual kvm/qemu foo should do one. The complicating
factor is that IA32_S_CET also contains SHSTK enable bits, so a straight
passthrough like I use relies on the guest never setting those bits or
keeping the pieces. It either needs to filter the MSR or implement the
full CET mess.