x86/debug: Prevent data breakpoints on __per_cpu_offset
authorLai Jiangshan <laijs@linux.alibaba.com>
Thu, 4 Feb 2021 15:27:06 +0000 (23:27 +0800)
committerThomas Gleixner <tglx@linutronix.de>
Fri, 5 Feb 2021 19:13:11 +0000 (20:13 +0100)
When FSGSBASE is enabled, paranoid_entry() fetches the per-CPU GSBASE value
via __per_cpu_offset or pcpu_unit_offsets.

When a data breakpoint is set on __per_cpu_offset[cpu] (read-write
operation), the specific CPU will be stuck in an infinite #DB loop.

RCU will try to send an NMI to the specific CPU, but it is not working
either since NMI also relies on paranoid_entry(). Which means it's
undebuggable.

Fixes: eaad981291ee3("x86/entry/64: Introduce the FIND_PERCPU_BASE macro")
Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20210204152708.21308-1-jiangshanlai@gmail.com
arch/x86/kernel/hw_breakpoint.c

index 6694c0f8e6c16895b001208b757b25834bc6cd37..012ed82e5bd324d491ba83f666bfc0088dd63df3 100644 (file)
@@ -269,6 +269,20 @@ static inline bool within_cpu_entry(unsigned long addr, unsigned long end)
                        CPU_ENTRY_AREA_TOTAL_SIZE))
                return true;
 
+       /*
+        * When FSGSBASE is enabled, paranoid_entry() fetches the per-CPU
+        * GSBASE value via __per_cpu_offset or pcpu_unit_offsets.
+        */
+#ifdef CONFIG_SMP
+       if (within_area(addr, end, (unsigned long)__per_cpu_offset,
+                       sizeof(unsigned long) * nr_cpu_ids))
+               return true;
+#else
+       if (within_area(addr, end, (unsigned long)&pcpu_unit_offsets,
+                       sizeof(pcpu_unit_offsets)))
+               return true;
+#endif
+
        for_each_possible_cpu(cpu) {
                /* The original rw GDT is being used after load_direct_gdt() */
                if (within_area(addr, end, (unsigned long)get_cpu_gdt_rw(cpu),