x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

When using eBPF to probe CS CALL instructions within a function,
a crash can be triggered.

The eBPF tool probes offset 257 of the __hrtimer_run_queues()
function:

<__hrtimer_run_queues+249>:  nopl   0x0(%rax,%rax,1)
<__hrtimer_run_queues+254>:  mov    %r14,%rdi
<__hrtimer_run_queues+257>:  cs call <__x86_indirect_thunk_r12>
<__hrtimer_run_queues+263>:  mov    %eax,%r12d
<__hrtimer_run_queues+266>:  xchg   %ax,%ax
<__hrtimer_run_queues+268>:  mov    %r13,%rdi

Which triggers this crash:

  BUG: unable to handle page fault for address: 00000000000f41c9
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] SMP NOPTI
  CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P
  RIP: 0010:__hrtimer_run_queues+0x106/0x230

Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is
at the 6th byte of the above CS CALL instruction. Since the CS CALL
instruction occupies 6 bytes, the exception occurred in the middle of that
call instruction.

The root cause is that when using eBPF tools to probe in the middle of a
function, a kprobe with INT3 is used as the underlying implementation.

During single-step emulation of the original CALL instruction,
int3_emulate_call() assumes that the probed CALL instruction is 5 bytes
long. However, the actual CS-prefixed CALL instruction occupies 6 bytes,
so it constructs an incorrect exception return address. When the CPU
returns from the kprobe handler, the next instruction to be executed is at
the address of the last byte of that CS CALL instruction. Coincidentally,
starting from that address, the CPU fetches and decodes a completely
different instruction, which ultimately triggers a kernel crash.

Fix the issue by using the actual instruction length obtained from
the instruction decoder when constructing the exception return
address, rather than relying on the hardcoded CALL_INSN_SIZE macro.

[ mingo: Refined the changelog ]

Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

1 files changed, 213 insertions, 0 deletions

diff --git a/arch/powerpc/kvm/timing.c b/arch/powerpc/kvm/timing.c
new file mode 100644
index 000000000..25071331f
--- /dev/null
+++ b/arch/powerpc/kvm/timing.c
@@ -0,0 +1,213 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ *
+ * Copyright IBM Corp. 2008
+ *
+ * Authors: Hollis Blanchard <hollisb@us.ibm.com>
+ *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
+ */
+
+#include <linux/kvm_host.h>
+#include <linux/fs.h>
+#include <linux/seq_file.h>
+#include <linux/debugfs.h>
+#include <linux/uaccess.h>
+#include <linux/module.h>
+
+#include <asm/time.h>
+#include <asm-generic/div64.h>
+
+#include "timing.h"
+
+void kvmppc_init_timing_stats(struct kvm_vcpu *vcpu)
+{
+	int i;
+
+	/* Take a lock to avoid concurrent updates */
+	mutex_lock(&vcpu->arch.exit_timing_lock);
+
+	vcpu->arch.last_exit_type = 0xDEAD;
+	for (i = 0; i < __NUMBER_OF_KVM_EXIT_TYPES; i++) {
+		vcpu->arch.timing_count_type[i] = 0;
+		vcpu->arch.timing_max_duration[i] = 0;
+		vcpu->arch.timing_min_duration[i] = 0xFFFFFFFF;
+		vcpu->arch.timing_sum_duration[i] = 0;
+		vcpu->arch.timing_sum_quad_duration[i] = 0;
+	}
+	vcpu->arch.timing_last_exit = 0;
+	vcpu->arch.timing_exit.tv64 = 0;
+	vcpu->arch.timing_last_enter.tv64 = 0;
+
+	mutex_unlock(&vcpu->arch.exit_timing_lock);
+}
+
+static void add_exit_timing(struct kvm_vcpu *vcpu, u64 duration, int type)
+{
+	u64 old;
+
+	mutex_lock(&vcpu->arch.exit_timing_lock);
+
+	vcpu->arch.timing_count_type[type]++;
+
+	/* sum */
+	old = vcpu->arch.timing_sum_duration[type];
+	vcpu->arch.timing_sum_duration[type] += duration;
+	if (unlikely(old > vcpu->arch.timing_sum_duration[type])) {
+		printk(KERN_ERR"%s - wrap adding sum of durations"
+			" old %lld new %lld type %d exit # of type %d\n",
+			__func__, old, vcpu->arch.timing_sum_duration[type],
+			type, vcpu->arch.timing_count_type[type]);
+	}
+
+	/* square sum */
+	old = vcpu->arch.timing_sum_quad_duration[type];
+	vcpu->arch.timing_sum_quad_duration[type] += (duration*duration);
+	if (unlikely(old > vcpu->arch.timing_sum_quad_duration[type])) {
+		printk(KERN_ERR"%s - wrap adding sum of squared durations"
+			" old %lld new %lld type %d exit # of type %d\n",
+			__func__, old,
+			vcpu->arch.timing_sum_quad_duration[type],
+			type, vcpu->arch.timing_count_type[type]);
+	}
+
+	/* set min/max */
+	if (unlikely(duration < vcpu->arch.timing_min_duration[type]))
+		vcpu->arch.timing_min_duration[type] = duration;
+	if (unlikely(duration > vcpu->arch.timing_max_duration[type]))
+		vcpu->arch.timing_max_duration[type] = duration;
+
+	mutex_unlock(&vcpu->arch.exit_timing_lock);
+}
+
+void kvmppc_update_timing_stats(struct kvm_vcpu *vcpu)
+{
+	u64 exit = vcpu->arch.timing_last_exit;
+	u64 enter = vcpu->arch.timing_last_enter.tv64;
+
+	/* save exit time, used next exit when the reenter time is known */
+	vcpu->arch.timing_last_exit = vcpu->arch.timing_exit.tv64;
+
+	if (unlikely(vcpu->arch.last_exit_type == 0xDEAD || exit == 0))
+		return; /* skip incomplete cycle (e.g. after reset) */
+
+	/* update statistics for average and standard deviation */
+	add_exit_timing(vcpu, (enter - exit), vcpu->arch.last_exit_type);
+	/* enter -> timing_last_exit is time spent in guest - log this too */
+	add_exit_timing(vcpu, (vcpu->arch.timing_last_exit - enter),
+			TIMEINGUEST);
+}
+
+static const char *kvm_exit_names[__NUMBER_OF_KVM_EXIT_TYPES] = {
+	[MMIO_EXITS] =              "MMIO",
+	[SIGNAL_EXITS] =            "SIGNAL",
+	[ITLB_REAL_MISS_EXITS] =    "ITLBREAL",
+	[ITLB_VIRT_MISS_EXITS] =    "ITLBVIRT",
+	[DTLB_REAL_MISS_EXITS] =    "DTLBREAL",
+	[DTLB_VIRT_MISS_EXITS] =    "DTLBVIRT",
+	[SYSCALL_EXITS] =           "SYSCALL",
+	[ISI_EXITS] =               "ISI",
+	[DSI_EXITS] =               "DSI",
+	[EMULATED_INST_EXITS] =     "EMULINST",
+	[EMULATED_MTMSRWE_EXITS] =  "EMUL_WAIT",
+	[EMULATED_WRTEE_EXITS] =    "EMUL_WRTEE",
+	[EMULATED_MTSPR_EXITS] =    "EMUL_MTSPR",
+	[EMULATED_MFSPR_EXITS] =    "EMUL_MFSPR",
+	[EMULATED_MTMSR_EXITS] =    "EMUL_MTMSR",
+	[EMULATED_MFMSR_EXITS] =    "EMUL_MFMSR",
+	[EMULATED_TLBSX_EXITS] =    "EMUL_TLBSX",
+	[EMULATED_TLBWE_EXITS] =    "EMUL_TLBWE",
+	[EMULATED_RFI_EXITS] =      "EMUL_RFI",
+	[DEC_EXITS] =               "DEC",
+	[EXT_INTR_EXITS] =          "EXTINT",
+	[HALT_WAKEUP] =             "HALT",
+	[USR_PR_INST] =             "USR_PR_INST",
+	[FP_UNAVAIL] =              "FP_UNAVAIL",
+	[DEBUG_EXITS] =             "DEBUG",
+	[TIMEINGUEST] =             "TIMEINGUEST"
+};
+
+static int kvmppc_exit_timing_show(struct seq_file *m, void *private)
+{
+	struct kvm_vcpu *vcpu = m->private;
+	int i;
+	u64 min, max, sum, sum_quad;
+
+	seq_puts(m, "type	count	min	max	sum	sum_squared\n");
+
+	for (i = 0; i < __NUMBER_OF_KVM_EXIT_TYPES; i++) {
+
+		min = vcpu->arch.timing_min_duration[i];
+		do_div(min, tb_ticks_per_usec);
+		max = vcpu->arch.timing_max_duration[i];
+		do_div(max, tb_ticks_per_usec);
+		sum = vcpu->arch.timing_sum_duration[i];
+		do_div(sum, tb_ticks_per_usec);
+		sum_quad = vcpu->arch.timing_sum_quad_duration[i];
+		do_div(sum_quad, tb_ticks_per_usec);
+
+		seq_printf(m, "%12s	%10d	%10lld	%10lld	%20lld	%20lld\n",
+			kvm_exit_names[i],
+			vcpu->arch.timing_count_type[i],
+			min,
+			max,
+			sum,
+			sum_quad);
+
+	}
+	return 0;
+}
+
+/* Write 'c' to clear the timing statistics. */
+static ssize_t kvmppc_exit_timing_write(struct file *file,
+				       const char __user *user_buf,
+				       size_t count, loff_t *ppos)
+{
+	int err = -EINVAL;
+	char c;
+
+	if (count > 1) {
+		goto done;
+	}
+
+	if (get_user(c, user_buf)) {
+		err = -EFAULT;
+		goto done;
+	}
+
+	if (c == 'c') {
+		struct seq_file *seqf = file->private_data;
+		struct kvm_vcpu *vcpu = seqf->private;
+		/* Write does not affect our buffers previously generated with
+		 * show. seq_file is locked here to prevent races of init with
+		 * a show call */
+		mutex_lock(&seqf->lock);
+		kvmppc_init_timing_stats(vcpu);
+		mutex_unlock(&seqf->lock);
+		err = count;
+	}
+
+done:
+	return err;
+}
+
+static int kvmppc_exit_timing_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, kvmppc_exit_timing_show, inode->i_private);
+}
+
+static const struct file_operations kvmppc_exit_timing_fops = {
+	.owner   = THIS_MODULE,
+	.open    = kvmppc_exit_timing_open,
+	.read    = seq_read,
+	.write   = kvmppc_exit_timing_write,
+	.llseek  = seq_lseek,
+	.release = single_release,
+};
+
+int kvmppc_create_vcpu_debugfs_e500(struct kvm_vcpu *vcpu,
+				    struct dentry *debugfs_dentry)
+{
+	debugfs_create_file("timing", 0666, debugfs_dentry,
+			    vcpu, &kvmppc_exit_timing_fops);
+	return 0;
+}