perf/x86: Fix 32-bit perf user callgraph collection

A basic perf callgraph record operation causes an immediate panic on a
32-bit kernel compiled with CONFIG_CC_STACKPROTECTOR=y:

  $ perf record -g ls
  Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in: c0404fbd

  CPU: 0 PID: 998 Comm: ls Not tainted 4.7.0-rc5+ #1
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.9.1-1.fc24 04/01/2014
   c0dd5967 ff7afe1c 00000086 f41dbc2c c07445a0 464c457f f41dbca8 f41dbc44
   c05646f4 f41dbca8 464c457f f41dbca8 464c457f f41dbc54 c04625be c0ce56fc
   c0404fbd f41dbc88 c0404fbd b74668f0 f41dc000 00000000 c0000000 00000000
  Call Trace:
   [<c07445a0>] dump_stack+0x58/0x78
   [<c05646f4>] panic+0x8e/0x1c6
   [<c04625be>] __stack_chk_fail+0x1e/0x30
   [<c0404fbd>] ? perf_callchain_user+0x22d/0x230
   [<c0404fbd>] perf_callchain_user+0x22d/0x230
   [<c055f89f>] get_perf_callchain+0x1ff/0x270
   [<c055f988>] perf_callchain+0x78/0x90
   [<c055c7eb>] perf_prepare_sample+0x24b/0x370
   [<c055c934>] perf_event_output_forward+0x24/0x70
   [<c05531c0>] __perf_event_overflow+0xa0/0x210
   [<c0550a93>] ? cpu_clock_event_read+0x43/0x50
   [<c0553431>] perf_swevent_hrtimer+0x101/0x180
   [<c0456235>] ? kmap_atomic_prot+0x35/0x140
   [<c056dc69>] ? get_page_from_freelist+0x279/0x950
   [<c058fdd8>] ? vma_interval_tree_remove+0x158/0x230
   [<c05939f4>] ? wp_page_copy.isra.82+0x2f4/0x630
   [<c05a050d>] ? page_add_file_rmap+0x1d/0x50
   [<c0565611>] ? unlock_page+0x61/0x80
   [<c0566755>] ? filemap_map_pages+0x305/0x320
   [<c059769f>] ? handle_mm_fault+0xb7f/0x1560
   [<c074cbeb>] ? timerqueue_del+0x1b/0x70
   [<c04cfefe>] ? __remove_hrtimer+0x2e/0x60
   [<c04d017b>] __hrtimer_run_queues+0xcb/0x2a0
   [<c0553330>] ? __perf_event_overflow+0x210/0x210
   [<c04d0a2a>] hrtimer_interrupt+0x8a/0x180
   [<c043ecc2>] local_apic_timer_interrupt+0x32/0x60
   [<c043f643>] smp_apic_timer_interrupt+0x33/0x50
   [<c0b0cd38>] apic_timer_interrupt+0x34/0x3c
  Kernel Offset: disabled
  ---[ end Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in: c0404fbd

The panic is caused by the fact that perf_callchain_user() mistakenly
assumes it's 64-bit only and ends up corrupting the stack.
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
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: stable@vger.kernel.org # v4.5+
Fixes: 75925e1a ("perf/x86: Optimize stack walk user accesses")
Link: http://lkml.kernel.org/r/1a547f5077ec30f75f9b57074837c3c80df86e5e.1467432113.git.jpoimboe@redhat.comSigned-off-by: Ingo Molnar <mingo@kernel.org>

arch/x86/events/core.c

@@ -2319,7 +2319,7 @@ void
 perf_callchain_user(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs)
 {
 	struct stack_frame frame;
-	const void __user *fp;
+	const unsigned long __user *fp;
 
 	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
 		/* TODO: We don't support guest os callchain now */
@@ -2332,7 +2332,7 @@ perf_callchain_user(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs
 	if (regs->flags & (X86_VM_MASK | PERF_EFLAGS_VM))
 		return;
 
-	fp = (void __user *)regs->bp;
+	fp = (unsigned long __user *)regs->bp;
 
 	perf_callchain_store(entry, regs->ip);
 
@@ -2345,16 +2345,17 @@ perf_callchain_user(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs
 	pagefault_disable();
 	while (entry->nr < entry->max_stack) {
 		unsigned long bytes;
+
 		frame.next_frame	     = NULL;
 		frame.return_address = 0;
 
-		if (!access_ok(VERIFY_READ, fp, 16))
+		if (!access_ok(VERIFY_READ, fp, sizeof(*fp) * 2))
 			break;
 
-		bytes = __copy_from_user_nmi(&frame.next_frame, fp, 8);
+		bytes = __copy_from_user_nmi(&frame.next_frame, fp, sizeof(*fp));
 		if (bytes != 0)
 			break;
-		bytes = __copy_from_user_nmi(&frame.return_address, fp+8, 8);
+		bytes = __copy_from_user_nmi(&frame.return_address, fp + 1, sizeof(*fp));
 		if (bytes != 0)
 			break;