kdb: properly synchronize vkdb_printf() calls with other CPUs
authorPetr Mladek <pmladek@suse.com>
Wed, 14 Dec 2016 23:05:55 +0000 (15:05 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 15 Dec 2016 00:04:08 +0000 (16:04 -0800)
kdb_printf_lock does not prevent other CPUs from entering the critical
section because it is ignored when KDB_STATE_PRINTF_LOCK is set.

The problematic situation might look like:

CPU0 CPU1

vkdb_printf()
  if (!KDB_STATE(PRINTF_LOCK))
    KDB_STATE_SET(PRINTF_LOCK);
    spin_lock_irqsave(&kdb_printf_lock, flags);

vkdb_printf()
  if (!KDB_STATE(PRINTF_LOCK))

BANG: The PRINTF_LOCK state is set and CPU1 is entering the critical
section without spinning on the lock.

The problem is that the code tries to implement locking using two state
variables that are not handled atomically.  Well, we need a custom
locking because we want to allow reentering the critical section on the
very same CPU.

Let's use solution from Petr Zijlstra that was proposed for a similar
scenario, see
https://lkml.kernel.org/r/20161018171513.734367391@infradead.org

This patch uses the same trick with cmpxchg().  The only difference is
that we want to handle only recursion from the same context and
therefore we disable interrupts.

In addition, KDB_STATE_PRINTF_LOCK is removed.  In fact, we are not able
to set it a non-racy way.

Link: http://lkml.kernel.org/r/1480412276-16690-3-git-send-email-pmladek@suse.com
Signed-off-by: Petr Mladek <pmladek@suse.com>
Reviewed-by: Daniel Thompson <daniel.thompson@linaro.org>
Cc: Jason Wessel <jason.wessel@windriver.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/debug/kdb/kdb_io.c
kernel/debug/kdb/kdb_private.h

index 46f477bebe0c155aefffb124d84a375824922043..daa76154fb1bb5f907934a4e0401370baa517b82 100644 (file)
@@ -555,16 +555,16 @@ int vkdb_printf(enum kdb_msgsrc src, const char *fmt, va_list ap)
        int colcount;
        int logging, saved_loglevel = 0;
        int saved_trap_printk;
-       int got_printf_lock = 0;
        int retlen = 0;
        int fnd, len;
+       int this_cpu, old_cpu;
+       static int kdb_printf_cpu = -1;
        char *cp, *cp2, *cphold = NULL, replaced_byte = ' ';
        char *moreprompt = "more> ";
        struct console *c = console_drivers;
-       static DEFINE_SPINLOCK(kdb_printf_lock);
        unsigned long uninitialized_var(flags);
 
-       preempt_disable();
+       local_irq_save(flags);
        saved_trap_printk = kdb_trap_printk;
        kdb_trap_printk = 0;
 
@@ -572,12 +572,13 @@ int vkdb_printf(enum kdb_msgsrc src, const char *fmt, va_list ap)
         * But if any cpu goes recursive in kdb, just print the output,
         * even if it is interleaved with any other text.
         */
-       if (!KDB_STATE(PRINTF_LOCK)) {
-               KDB_STATE_SET(PRINTF_LOCK);
-               spin_lock_irqsave(&kdb_printf_lock, flags);
-               got_printf_lock = 1;
-       } else {
-               __acquire(kdb_printf_lock);
+       this_cpu = smp_processor_id();
+       for (;;) {
+               old_cpu = cmpxchg(&kdb_printf_cpu, -1, this_cpu);
+               if (old_cpu == -1 || old_cpu == this_cpu)
+                       break;
+
+               cpu_relax();
        }
 
        diag = kdbgetintenv("LINES", &linecount);
@@ -846,15 +847,10 @@ kdb_print_out:
        suspend_grep = 0; /* end of what may have been a recursive call */
        if (logging)
                console_loglevel = saved_loglevel;
-       if (KDB_STATE(PRINTF_LOCK) && got_printf_lock) {
-               got_printf_lock = 0;
-               spin_unlock_irqrestore(&kdb_printf_lock, flags);
-               KDB_STATE_CLEAR(PRINTF_LOCK);
-       } else {
-               __release(kdb_printf_lock);
-       }
+       /* kdb_printf_cpu locked the code above. */
+       smp_store_release(&kdb_printf_cpu, old_cpu);
        kdb_trap_printk = saved_trap_printk;
-       preempt_enable();
+       local_irq_restore(flags);
        return retlen;
 }
 
index 75014d7f45681b1a75643fd06dbc119e8ea16cb7..fc224fbcf95492b83e49539038cf727571c2e315 100644 (file)
@@ -132,7 +132,6 @@ extern int kdb_state;
 #define KDB_STATE_PAGER                0x00000400      /* pager is available */
 #define KDB_STATE_GO_SWITCH    0x00000800      /* go is switching
                                                 * back to initial cpu */
-#define KDB_STATE_PRINTF_LOCK  0x00001000      /* Holds kdb_printf lock */
 #define KDB_STATE_WAIT_IPI     0x00002000      /* Waiting for kdb_ipi() NMI */
 #define KDB_STATE_RECURSE      0x00004000      /* Recursive entry to kdb */
 #define KDB_STATE_IP_ADJUSTED  0x00008000      /* Restart IP has been