perf/x86: Restore TASK_SIZE check on frame pointer

The following commit:

  75925e1a ("perf/x86: Optimize stack walk user accesses")

... switched from copy_from_user_nmi() to __copy_from_user_nmi() with a manual
access_ok() check.

Unfortunately, copy_from_user_nmi() does an explicit check against TASK_SIZE,
whereas the access_ok() uses whatever the current address limit of the task is.

We are getting NMIs when __probe_kernel_read() has switched to KERNEL_DS, and
then see vmalloc faults when we access what looks like pointers into vmalloc
space:

  [] WARNING: CPU: 3 PID: 3685731 at arch/x86/mm/fault.c:435 vmalloc_fault+0x289/0x290
  [] CPU: 3 PID: 3685731 Comm: sh Tainted: G        W       4.6.0-5_fbk1_223_gdbf0f40 #1
  [] Call Trace:
  []  <NMI>  [<ffffffff814717d1>] dump_stack+0x4d/0x6c
  []  [<ffffffff81076e43>] __warn+0xd3/0xf0
  []  [<ffffffff81076f2d>] warn_slowpath_null+0x1d/0x20
  []  [<ffffffff8104a899>] vmalloc_fault+0x289/0x290
  []  [<ffffffff8104b5a0>] __do_page_fault+0x330/0x490
  []  [<ffffffff8104b70c>] do_page_fault+0xc/0x10
  []  [<ffffffff81794e82>] page_fault+0x22/0x30
  []  [<ffffffff81006280>] ? perf_callchain_user+0x100/0x2a0
  []  [<ffffffff8115124f>] get_perf_callchain+0x17f/0x190
  []  [<ffffffff811512c7>] perf_callchain+0x67/0x80
  []  [<ffffffff8114e750>] perf_prepare_sample+0x2a0/0x370
  []  [<ffffffff8114e840>] perf_event_output+0x20/0x60
  []  [<ffffffff8114aee7>] ? perf_event_update_userpage+0xc7/0x130
  []  [<ffffffff8114ea01>] __perf_event_overflow+0x181/0x1d0
  []  [<ffffffff8114f484>] perf_event_overflow+0x14/0x20
  []  [<ffffffff8100a6e3>] intel_pmu_handle_irq+0x1d3/0x490
  []  [<ffffffff8147daf7>] ? copy_user_enhanced_fast_string+0x7/0x10
  []  [<ffffffff81197191>] ? vunmap_page_range+0x1a1/0x2f0
  []  [<ffffffff811972f1>] ? unmap_kernel_range_noflush+0x11/0x20
  []  [<ffffffff814f2056>] ? ghes_copy_tofrom_phys+0x116/0x1f0
  []  [<ffffffff81040d1d>] ? x2apic_send_IPI_self+0x1d/0x20
  []  [<ffffffff8100411d>] perf_event_nmi_handler+0x2d/0x50
  []  [<ffffffff8101ea31>] nmi_handle+0x61/0x110
  []  [<ffffffff8101ef94>] default_do_nmi+0x44/0x110
  []  [<ffffffff8101f13b>] do_nmi+0xdb/0x150
  []  [<ffffffff81795187>] end_repeat_nmi+0x1a/0x1e
  []  [<ffffffff8147daf7>] ? copy_user_enhanced_fast_string+0x7/0x10
  []  [<ffffffff8147daf7>] ? copy_user_enhanced_fast_string+0x7/0x10
  []  [<ffffffff8147daf7>] ? copy_user_enhanced_fast_string+0x7/0x10
  []  <<EOE>>  <IRQ>  [<ffffffff8115d05e>] ? __probe_kernel_read+0x3e/0xa0

Fix this by moving the valid_user_frame() check to before the uaccess
that loads the return address and the pointer to the next frame.
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Signed-off-by: NPeter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Fixes: 75925e1a ("perf/x86: Optimize stack walk user accesses")
Signed-off-by: NIngo Molnar <mingo@kernel.org>

arch/x86/events/core.c

@@ -2352,7 +2352,7 @@ perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry_ctx *ent
 		frame.next_frame     = 0;
 		frame.return_address = 0;
 
-		if (!access_ok(VERIFY_READ, fp, 8))
+		if (!valid_user_frame(fp, sizeof(frame)))
 			break;
 
 		bytes = __copy_from_user_nmi(&frame.next_frame, fp, 4);
@@ -2362,9 +2362,6 @@ perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry_ctx *ent
 		if (bytes != 0)
 			break;
 
-		if (!valid_user_frame(fp, sizeof(frame)))
-			break;
-
 		perf_callchain_store(entry, cs_base + frame.return_address);
 		fp = compat_ptr(ss_base + frame.next_frame);
 	}
@@ -2413,7 +2410,7 @@ perf_callchain_user(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs
 		frame.next_frame	     = NULL;
 		frame.return_address = 0;
 
-		if (!access_ok(VERIFY_READ, fp, sizeof(*fp) * 2))
+		if (!valid_user_frame(fp, sizeof(frame)))
 			break;
 
 		bytes = __copy_from_user_nmi(&frame.next_frame, fp, sizeof(*fp));
@@ -2423,9 +2420,6 @@ perf_callchain_user(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs
 		if (bytes != 0)
 			break;
 
-		if (!valid_user_frame(fp, sizeof(frame)))
-			break;
-
 		perf_callchain_store(entry, frame.return_address);
 		fp = (void __user *)frame.next_frame;
 	}