]> git.ipfire.org Git - thirdparty/kernel/stable-queue.git/commitdiff
4.19-stable patches
authorGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 20 Jul 2020 12:51:01 +0000 (14:51 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 20 Jul 2020 12:51:01 +0000 (14:51 +0200)
added patches:
printk-queue-wake_up_klogd-irq_work-only-if-per-cpu-areas-are-ready.patch

queue-4.19/printk-queue-wake_up_klogd-irq_work-only-if-per-cpu-areas-are-ready.patch [new file with mode: 0644]
queue-4.19/series

diff --git a/queue-4.19/printk-queue-wake_up_klogd-irq_work-only-if-per-cpu-areas-are-ready.patch b/queue-4.19/printk-queue-wake_up_klogd-irq_work-only-if-per-cpu-areas-are-ready.patch
new file mode 100644 (file)
index 0000000..b005fe6
--- /dev/null
@@ -0,0 +1,208 @@
+From ab6f762f0f53162d41497708b33c9a3236d3609e Mon Sep 17 00:00:00 2001
+From: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
+Date: Tue, 3 Mar 2020 20:30:02 +0900
+Subject: printk: queue wake_up_klogd irq_work only if per-CPU areas are ready
+
+From: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
+
+commit ab6f762f0f53162d41497708b33c9a3236d3609e upstream.
+
+printk_deferred(), similarly to printk_safe/printk_nmi, does not
+immediately attempt to print a new message on the consoles, avoiding
+calls into non-reentrant kernel paths, e.g. scheduler or timekeeping,
+which potentially can deadlock the system.
+
+Those printk() flavors, instead, rely on per-CPU flush irq_work to print
+messages from safer contexts.  For same reasons (recursive scheduler or
+timekeeping calls) printk() uses per-CPU irq_work in order to wake up
+user space syslog/kmsg readers.
+
+However, only printk_safe/printk_nmi do make sure that per-CPU areas
+have been initialised and that it's safe to modify per-CPU irq_work.
+This means that, for instance, should printk_deferred() be invoked "too
+early", that is before per-CPU areas are initialised, printk_deferred()
+will perform illegal per-CPU access.
+
+Lech Perczak [0] reports that after commit 1b710b1b10ef ("char/random:
+silence a lockdep splat with printk()") user-space syslog/kmsg readers
+are not able to read new kernel messages.
+
+The reason is printk_deferred() being called too early (as was pointed
+out by Petr and John).
+
+Fix printk_deferred() and do not queue per-CPU irq_work before per-CPU
+areas are initialized.
+
+Link: https://lore.kernel.org/lkml/aa0732c6-5c4e-8a8b-a1c1-75ebe3dca05b@camlintechnologies.com/
+Reported-by: Lech Perczak <l.perczak@camlintechnologies.com>
+Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
+Tested-by: Jann Horn <jannh@google.com>
+Reviewed-by: Petr Mladek <pmladek@suse.com>
+Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+Cc: Theodore Ts'o <tytso@mit.edu>
+Cc: John Ogness <john.ogness@linutronix.de>
+Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ include/linux/printk.h      |    5 -----
+ init/main.c                 |    1 -
+ kernel/printk/internal.h    |    5 +++++
+ kernel/printk/printk.c      |   34 ++++++++++++++++++++++++++++++++++
+ kernel/printk/printk_safe.c |   11 +----------
+ 5 files changed, 40 insertions(+), 16 deletions(-)
+
+--- a/include/linux/printk.h
++++ b/include/linux/printk.h
+@@ -206,7 +206,6 @@ __printf(1, 2) void dump_stack_set_arch_
+ void dump_stack_print_info(const char *log_lvl);
+ void show_regs_print_info(const char *log_lvl);
+ extern asmlinkage void dump_stack(void) __cold;
+-extern void printk_safe_init(void);
+ extern void printk_safe_flush(void);
+ extern void printk_safe_flush_on_panic(void);
+ #else
+@@ -273,10 +272,6 @@ static inline asmlinkage void dump_stack
+ {
+ }
+-static inline void printk_safe_init(void)
+-{
+-}
+-
+ static inline void printk_safe_flush(void)
+ {
+ }
+--- a/init/main.c
++++ b/init/main.c
+@@ -641,7 +641,6 @@ asmlinkage __visible void __init start_k
+       softirq_init();
+       timekeeping_init();
+       time_init();
+-      printk_safe_init();
+       perf_event_init();
+       profile_init();
+       call_function_init();
+--- a/kernel/printk/internal.h
++++ b/kernel/printk/internal.h
+@@ -35,6 +35,9 @@ __printf(1, 0) int vprintk_func(const ch
+ void __printk_safe_enter(void);
+ void __printk_safe_exit(void);
++void printk_safe_init(void);
++bool printk_percpu_data_ready(void);
++
+ #define printk_safe_enter_irqsave(flags)      \
+       do {                                    \
+               local_irq_save(flags);          \
+@@ -76,4 +79,6 @@ __printf(1, 0) int vprintk_func(const ch
+ #define printk_safe_enter_irq() local_irq_disable()
+ #define printk_safe_exit_irq() local_irq_enable()
++static inline void printk_safe_init(void) { }
++static inline bool printk_percpu_data_ready(void) { return false; }
+ #endif /* CONFIG_PRINTK */
+--- a/kernel/printk/printk.c
++++ b/kernel/printk/printk.c
+@@ -443,6 +443,18 @@ static char __log_buf[__LOG_BUF_LEN] __a
+ static char *log_buf = __log_buf;
+ static u32 log_buf_len = __LOG_BUF_LEN;
++/*
++ * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
++ * per_cpu_areas are initialised. This variable is set to true when
++ * it's safe to access per-CPU data.
++ */
++static bool __printk_percpu_data_ready __read_mostly;
++
++bool printk_percpu_data_ready(void)
++{
++      return __printk_percpu_data_ready;
++}
++
+ /* Return log buffer address */
+ char *log_buf_addr_get(void)
+ {
+@@ -1101,12 +1113,28 @@ static void __init log_buf_add_cpu(void)
+ static inline void log_buf_add_cpu(void) {}
+ #endif /* CONFIG_SMP */
++static void __init set_percpu_data_ready(void)
++{
++      printk_safe_init();
++      /* Make sure we set this flag only after printk_safe() init is done */
++      barrier();
++      __printk_percpu_data_ready = true;
++}
++
+ void __init setup_log_buf(int early)
+ {
+       unsigned long flags;
+       char *new_log_buf;
+       unsigned int free;
++      /*
++       * Some archs call setup_log_buf() multiple times - first is very
++       * early, e.g. from setup_arch(), and second - when percpu_areas
++       * are initialised.
++       */
++      if (!early)
++              set_percpu_data_ready();
++
+       if (log_buf != __log_buf)
+               return;
+@@ -2913,6 +2941,9 @@ static DEFINE_PER_CPU(struct irq_work, w
+ void wake_up_klogd(void)
+ {
++      if (!printk_percpu_data_ready())
++              return;
++
+       preempt_disable();
+       if (waitqueue_active(&log_wait)) {
+               this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
+@@ -2923,6 +2954,9 @@ void wake_up_klogd(void)
+ void defer_console_output(void)
+ {
++      if (!printk_percpu_data_ready())
++              return;
++
+       preempt_disable();
+       __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
+       irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
+--- a/kernel/printk/printk_safe.c
++++ b/kernel/printk/printk_safe.c
+@@ -39,7 +39,6 @@
+  * There are situations when we want to make sure that all buffers
+  * were handled or when IRQs are blocked.
+  */
+-static int printk_safe_irq_ready __read_mostly;
+ #define SAFE_LOG_BUF_LEN ((1 << CONFIG_PRINTK_SAFE_LOG_BUF_SHIFT) -   \
+                               sizeof(atomic_t) -                      \
+@@ -63,7 +62,7 @@ static DEFINE_PER_CPU(struct printk_safe
+ /* Get flushed in a more safe context. */
+ static void queue_flush_work(struct printk_safe_seq_buf *s)
+ {
+-      if (printk_safe_irq_ready)
++      if (printk_percpu_data_ready())
+               irq_work_queue(&s->work);
+ }
+@@ -414,14 +413,6 @@ void __init printk_safe_init(void)
+ #endif
+       }
+-      /*
+-       * In the highly unlikely event that a NMI were to trigger at
+-       * this moment. Make sure IRQ work is set up before this
+-       * variable is set.
+-       */
+-      barrier();
+-      printk_safe_irq_ready = 1;
+-
+       /* Flush pending messages that did not have scheduled IRQ works. */
+       printk_safe_flush();
+ }
index 77d5b52dc48cb4ccc428c78d944761268002ee6b..358cc067b0a8169911c5c9f955d0ca00689ff5ff 100644 (file)
@@ -129,3 +129,4 @@ arm64-compat-ensure-upper-32-bits-of-x0-are-zero-on-syscall-return.patch
 sched-fix-unreliable-rseq-cpu_id-for-new-tasks.patch
 sched-fair-handle-case-of-task_h_load-returning-0.patch
 genirq-affinity-handle-affinity-setting-on-inactive-interrupts-correctly.patch
+printk-queue-wake_up_klogd-irq_work-only-if-per-cpu-areas-are-ready.patch