x86/mm, tracing: Fix CR2 corruption

Despite the current efforts to read CR2 before tracing happens there still
exist a number of possible holes:

  idtentry page_fault             do_page_fault           has_error_code=1
    call error_entry
      TRACE_IRQS_OFF
        call trace_hardirqs_off*
          #PF // modifies CR2

      CALL_enter_from_user_mode
        __context_tracking_exit()
          trace_user_exit(0)
            #PF // modifies CR2

    call do_page_fault
      address = read_cr2(); /* whoopsie */

And similar for i386.

Fix it by pulling the CR2 read into the entry code, before any of that
stuff gets a chance to run and ruin things.

Reported-by: He Zhe <zhe.he@windriver.com>
Reported-by: Eiichi Tsukata <devel@etsukata.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Andy Lutomirski <luto@kernel.org>
Cc: bp@alien8.de
Cc: rostedt@goodmis.org
Cc: torvalds@linux-foundation.org
Cc: hpa@zytor.com
Cc: dave.hansen@linux.intel.com
Cc: jgross@suse.com
Cc: joel@joelfernandes.org
Link: https://lkml.kernel.org/r/20190711114336.116812491@infradead.org

Debugged-by: Steven Rostedt <rostedt@goodmis.org>

1 files changed, 18 insertions, 17 deletions

diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
index 95ae05f0edf2..7cb2e1f1ec09 100644
--- a/arch/x86/entry/entry_64.S
+++ b/arch/x86/entry/entry_64.S
@@ -864,7 +864,7 @@ apicinterrupt IRQ_WORK_VECTOR			irq_work_interrupt		smp_irq_work_interrupt
  */
 #define CPU_TSS_IST(x) PER_CPU_VAR(cpu_tss_rw) + (TSS_ist + (x) * 8)
 
-.macro idtentry_part do_sym, has_error_code:req, paranoid:req, shift_ist=-1, ist_offset=0
+.macro idtentry_part do_sym, has_error_code:req, read_cr2:req, paranoid:req, shift_ist=-1, ist_offset=0
 
 	.if \paranoid
 	call	paranoid_entry
@@ -874,12 +874,21 @@ apicinterrupt IRQ_WORK_VECTOR			irq_work_interrupt		smp_irq_work_interrupt
 	.endif
 	UNWIND_HINT_REGS
 
-	.if \paranoid
+	.if \read_cr2
+	GET_CR2_INTO(%rdx);			/* can clobber %rax */
+	.endif
+
 	.if \shift_ist != -1
 	TRACE_IRQS_OFF_DEBUG			/* reload IDT in case of recursion */
 	.else
 	TRACE_IRQS_OFF
 	.endif
+
+	.if \paranoid == 0
+	testb	$3, CS(%rsp)
+	jz	.Lfrom_kernel_no_context_tracking_\@
+	CALL_enter_from_user_mode
+.Lfrom_kernel_no_context_tracking_\@:
 	.endif
 
 	movq	%rsp, %rdi			/* pt_regs pointer */
@@ -923,6 +932,7 @@ apicinterrupt IRQ_WORK_VECTOR			irq_work_interrupt		smp_irq_work_interrupt
  *			fresh stack.  (This is for #DB, which has a nasty habit
  *			of recursing.)
  * @create_gap:		create a 6-word stack gap when coming from kernel mode.
+ * @read_cr2:		load CR2 into the 3rd argument; done before calling any C code
  *
  * idtentry generates an IDT stub that sets up a usable kernel context,
  * creates struct pt_regs, and calls @do_sym.  The stub has the following
@@ -947,7 +957,7 @@ apicinterrupt IRQ_WORK_VECTOR			irq_work_interrupt		smp_irq_work_interrupt
  * @paranoid == 2 is special: the stub will never switch stacks.  This is for
  * #DF: if the thread stack is somehow unusable, we'll still get a useful OOPS.
  */
-.macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1 ist_offset=0 create_gap=0
+.macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1 ist_offset=0 create_gap=0 read_cr2=0
 ENTRY(\sym)
 	UNWIND_HINT_IRET_REGS offset=\has_error_code*8
 
@@ -985,7 +995,7 @@ ENTRY(\sym)
 .Lfrom_usermode_no_gap_\@:
 	.endif
 
-	idtentry_part \do_sym, \has_error_code, \paranoid, \shift_ist, \ist_offset
+	idtentry_part \do_sym, \has_error_code, \read_cr2, \paranoid, \shift_ist, \ist_offset
 
 	.if \paranoid == 1
 	/*
@@ -994,7 +1004,7 @@ ENTRY(\sym)
 	 * run in real process context if user_mode(regs).
 	 */
 .Lfrom_usermode_switch_stack_\@:
-	idtentry_part \do_sym, \has_error_code, paranoid=0
+	idtentry_part \do_sym, \has_error_code, \read_cr2, paranoid=0
 	.endif
 
 _ASM_NOKPROBE(\sym)
@@ -1006,7 +1016,7 @@ idtentry overflow			do_overflow			has_error_code=0
 idtentry bounds				do_bounds			has_error_code=0
 idtentry invalid_op			do_invalid_op			has_error_code=0
 idtentry device_not_available		do_device_not_available		has_error_code=0
-idtentry double_fault			do_double_fault			has_error_code=1 paranoid=2
+idtentry double_fault			do_double_fault			has_error_code=1 paranoid=2 read_cr2=1
 idtentry coprocessor_segment_overrun	do_coprocessor_segment_overrun	has_error_code=0
 idtentry invalid_TSS			do_invalid_TSS			has_error_code=1
 idtentry segment_not_present		do_segment_not_present		has_error_code=1
@@ -1179,10 +1189,10 @@ idtentry xenint3		do_int3			has_error_code=0
 #endif
 
 idtentry general_protection	do_general_protection	has_error_code=1
-idtentry page_fault		do_page_fault		has_error_code=1
+idtentry page_fault		do_page_fault		has_error_code=1	read_cr2=1
 
 #ifdef CONFIG_KVM_GUEST
-idtentry async_page_fault	do_async_page_fault	has_error_code=1
+idtentry async_page_fault	do_async_page_fault	has_error_code=1	read_cr2=1
 #endif
 
 #ifdef CONFIG_X86_MCE
@@ -1281,18 +1291,9 @@ ENTRY(error_entry)
 	movq	%rax, %rsp			/* switch stack */
 	ENCODE_FRAME_POINTER
 	pushq	%r12
-
-	/*
-	 * We need to tell lockdep that IRQs are off.  We can't do this until
-	 * we fix gsbase, and we should do it before enter_from_user_mode
-	 * (which can take locks).
-	 */
-	TRACE_IRQS_OFF
-	CALL_enter_from_user_mode
 	ret
 
 .Lerror_entry_done:
-	TRACE_IRQS_OFF
 	ret
 
 	/*