x86/callthunks: Add call patching for call depth tracking

Mitigating the Intel SKL RSB underflow issue in software requires to
track the call depth. That is every CALL and every RET need to be
intercepted and additional code injected.

The existing retbleed mitigations already include means of redirecting
RET to __x86_return_thunk; this can be re-purposed and RET can be
redirected to another function doing RET accounting.

CALL accounting will use the function padding introduced in prior
patches. For each CALL instruction, the destination symbol's padding
is rewritten to do the accounting and the CALL instruction is adjusted
to call into the padding.

This ensures only affected CPUs pay the overhead of this accounting.
Unaffected CPUs will leave the padding unused and have their 'JMP
__x86_return_thunk' replaced with an actual 'RET' instruction.

Objtool has been modified to supply a .call_sites section that lists
all the 'CALL' instructions. Additionally the paravirt instruction
sites are iterated since they will have been patched from an indirect
call to direct calls (or direct instructions in which case it'll be
ignored).

Module handling and the actual thunk code for SKL will be added in
subsequent steps.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20220915111147.470877038@infradead.org

1 files changed, 251 insertions, 0 deletions

diff --git a/arch/x86/kernel/callthunks.c b/arch/x86/kernel/callthunks.c
new file mode 100644
index 000000000000..e5275d6e674d
--- /dev/null
+++ b/arch/x86/kernel/callthunks.c
@@ -0,0 +1,251 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+#define pr_fmt(fmt) "callthunks: " fmt
+
+#include <linux/kallsyms.h>
+#include <linux/memory.h>
+#include <linux/moduleloader.h>
+
+#include <asm/alternative.h>
+#include <asm/cpu.h>
+#include <asm/ftrace.h>
+#include <asm/insn.h>
+#include <asm/kexec.h>
+#include <asm/nospec-branch.h>
+#include <asm/paravirt.h>
+#include <asm/sections.h>
+#include <asm/switch_to.h>
+#include <asm/sync_core.h>
+#include <asm/text-patching.h>
+#include <asm/xen/hypercall.h>
+
+static int __initdata_or_module debug_callthunks;
+
+#define prdbg(fmt, args...)					\
+do {								\
+	if (debug_callthunks)					\
+		printk(KERN_DEBUG pr_fmt(fmt), ##args);		\
+} while(0)
+
+static int __init debug_thunks(char *str)
+{
+	debug_callthunks = 1;
+	return 1;
+}
+__setup("debug-callthunks", debug_thunks);
+
+extern s32 __call_sites[], __call_sites_end[];
+
+struct thunk_desc {
+	void		*template;
+	unsigned int	template_size;
+};
+
+struct core_text {
+	unsigned long	base;
+	unsigned long	end;
+	const char	*name;
+};
+
+static bool thunks_initialized __ro_after_init;
+
+static const struct core_text builtin_coretext = {
+	.base = (unsigned long)_text,
+	.end  = (unsigned long)_etext,
+	.name = "builtin",
+};
+
+static struct thunk_desc callthunk_desc __ro_after_init;
+
+extern void error_entry(void);
+extern void xen_error_entry(void);
+extern void paranoid_entry(void);
+
+static inline bool within_coretext(const struct core_text *ct, void *addr)
+{
+	unsigned long p = (unsigned long)addr;
+
+	return ct->base <= p && p < ct->end;
+}
+
+static inline bool within_module_coretext(void *addr)
+{
+	bool ret = false;
+
+#ifdef CONFIG_MODULES
+	struct module *mod;
+
+	preempt_disable();
+	mod = __module_address((unsigned long)addr);
+	if (mod && within_module_core((unsigned long)addr, mod))
+		ret = true;
+	preempt_enable();
+#endif
+	return ret;
+}
+
+static bool is_coretext(const struct core_text *ct, void *addr)
+{
+	if (ct && within_coretext(ct, addr))
+		return true;
+	if (within_coretext(&builtin_coretext, addr))
+		return true;
+	return within_module_coretext(addr);
+}
+
+static __init_or_module bool skip_addr(void *dest)
+{
+	if (dest == error_entry)
+		return true;
+	if (dest == paranoid_entry)
+		return true;
+	if (dest == xen_error_entry)
+		return true;
+	/* Does FILL_RSB... */
+	if (dest == __switch_to_asm)
+		return true;
+	/* Accounts directly */
+	if (dest == ret_from_fork)
+		return true;
+#ifdef CONFIG_HOTPLUG_CPU
+	if (dest == start_cpu0)
+		return true;
+#endif
+#ifdef CONFIG_FUNCTION_TRACER
+	if (dest == __fentry__)
+		return true;
+#endif
+#ifdef CONFIG_KEXEC_CORE
+	if (dest >= (void *)relocate_kernel &&
+	    dest < (void*)relocate_kernel + KEXEC_CONTROL_CODE_MAX_SIZE)
+		return true;
+#endif
+#ifdef CONFIG_XEN
+	if (dest >= (void *)hypercall_page &&
+	    dest < (void*)hypercall_page + PAGE_SIZE)
+		return true;
+#endif
+	return false;
+}
+
+static __init_or_module void *call_get_dest(void *addr)
+{
+	struct insn insn;
+	void *dest;
+	int ret;
+
+	ret = insn_decode_kernel(&insn, addr);
+	if (ret)
+		return ERR_PTR(ret);
+
+	/* Patched out call? */
+	if (insn.opcode.bytes[0] != CALL_INSN_OPCODE)
+		return NULL;
+
+	dest = addr + insn.length + insn.immediate.value;
+	if (skip_addr(dest))
+		return NULL;
+	return dest;
+}
+
+static const u8 nops[] = {
+	0x90, 0x90, 0x90, 0x90, 0x90, 0x90, 0x90, 0x90,
+	0x90, 0x90, 0x90, 0x90, 0x90, 0x90, 0x90, 0x90,
+	0x90, 0x90, 0x90, 0x90, 0x90, 0x90, 0x90, 0x90,
+	0x90, 0x90, 0x90, 0x90, 0x90, 0x90, 0x90, 0x90,
+};
+
+static __init_or_module void *patch_dest(void *dest, bool direct)
+{
+	unsigned int tsize = callthunk_desc.template_size;
+	u8 *pad = dest - tsize;
+
+	/* Already patched? */
+	if (!bcmp(pad, callthunk_desc.template, tsize))
+		return pad;
+
+	/* Ensure there are nops */
+	if (bcmp(pad, nops, tsize)) {
+		pr_warn_once("Invalid padding area for %pS\n", dest);
+		return NULL;
+	}
+
+	if (direct)
+		memcpy(pad, callthunk_desc.template, tsize);
+	else
+		text_poke_copy_locked(pad, callthunk_desc.template, tsize, true);
+	return pad;
+}
+
+static __init_or_module void patch_call(void *addr, const struct core_text *ct)
+{
+	void *pad, *dest;
+	u8 bytes[8];
+
+	if (!within_coretext(ct, addr))
+		return;
+
+	dest = call_get_dest(addr);
+	if (!dest || WARN_ON_ONCE(IS_ERR(dest)))
+		return;
+
+	if (!is_coretext(ct, dest))
+		return;
+
+	pad = patch_dest(dest, within_coretext(ct, dest));
+	if (!pad)
+		return;
+
+	prdbg("Patch call at: %pS %px to %pS %px -> %px \n", addr, addr,
+		dest, dest, pad);
+	__text_gen_insn(bytes, CALL_INSN_OPCODE, addr, pad, CALL_INSN_SIZE);
+	text_poke_early(addr, bytes, CALL_INSN_SIZE);
+}
+
+static __init_or_module void
+patch_call_sites(s32 *start, s32 *end, const struct core_text *ct)
+{
+	s32 *s;
+
+	for (s = start; s < end; s++)
+		patch_call((void *)s + *s, ct);
+}
+
+static __init_or_module void
+patch_paravirt_call_sites(struct paravirt_patch_site *start,
+			  struct paravirt_patch_site *end,
+			  const struct core_text *ct)
+{
+	struct paravirt_patch_site *p;
+
+	for (p = start; p < end; p++)
+		patch_call(p->instr, ct);
+}
+
+static __init_or_module void
+callthunks_setup(struct callthunk_sites *cs, const struct core_text *ct)
+{
+	prdbg("Patching call sites %s\n", ct->name);
+	patch_call_sites(cs->call_start, cs->call_end, ct);
+	patch_paravirt_call_sites(cs->pv_start, cs->pv_end, ct);
+	prdbg("Patching call sites done%s\n", ct->name);
+}
+
+void __init callthunks_patch_builtin_calls(void)
+{
+	struct callthunk_sites cs = {
+		.call_start	= __call_sites,
+		.call_end	= __call_sites_end,
+		.pv_start	= __parainstructions,
+		.pv_end		= __parainstructions_end
+	};
+
+	if (!cpu_feature_enabled(X86_FEATURE_CALL_DEPTH))
+		return;
+
+	pr_info("Setting up call depth tracking\n");
+	mutex_lock(&text_mutex);
+	callthunks_setup(&cs, &builtin_coretext);
+	thunks_initialized = true;
+	mutex_unlock(&text_mutex);
+}