From 1e0cb399c7653462d9dadf8ab9425337c355d358 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Google)" Date: Fri, 29 Sep 2023 18:01:13 -0400 Subject: [PATCH 1/4] ring-buffer: Update "shortest_full" in polling It was discovered that the ring buffer polling was incorrectly stating that read would not block, but that's because polling did not take into account that reads will block if the "buffer-percent" was set. Instead, the ring buffer polling would say reads would not block if there was any data in the ring buffer. This was incorrect behavior from a user space point of view. This was fixed by commit 42fb0a1e84ff by having the polling code check if the ring buffer had more data than what the user specified "buffer percent" had. The problem now is that the polling code did not register itself to the writer that it wanted to wait for a specific "full" value of the ring buffer. The result was that the writer would wake the polling waiter whenever there was a new event. The polling waiter would then wake up, see that there's not enough data in the ring buffer to notify user space and then go back to sleep. The next event would wake it up again. Before the polling fix was added, the code would wake up around 100 times for a hackbench 30 benchmark. After the "fix", due to the constant waking of the writer, it would wake up over 11,0000 times! It would never leave the kernel, so the user space behavior was still "correct", but this definitely is not the desired effect. To fix this, have the polling code add what it's waiting for to the "shortest_full" variable, to tell the writer not to wake it up if the buffer is not as full as it expects to be. Note, after this fix, it appears that the waiter is now woken up around 2x the times it was before (~200). This is a tremendous improvement from the 11,000 times, but I will need to spend some time to see why polling is more aggressive in its wakeups than the read blocking code. Link: https://lore.kernel.org/linux-trace-kernel/20230929180113.01c2cae3@rorschach.local.home Cc: stable@vger.kernel.org Cc: Masami Hiramatsu Cc: Mark Rutland Fixes: 42fb0a1e84ff ("tracing/ring-buffer: Have polling block on watermark") Reported-by: Julia Lawall Tested-by: Julia Lawall Signed-off-by: Steven Rostedt (Google) --- kernel/trace/ring_buffer.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index 28daf0ce95c51..515cafdb18d98 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -1137,6 +1137,9 @@ __poll_t ring_buffer_poll_wait(struct trace_buffer *buffer, int cpu, if (full) { poll_wait(filp, &work->full_waiters, poll_table); work->full_waiters_pending = true; + if (!cpu_buffer->shortest_full || + cpu_buffer->shortest_full > full) + cpu_buffer->shortest_full = full; } else { poll_wait(filp, &work->waiters, poll_table); work->waiters_pending = true; From 23cce5f25491968b23fb9c399bbfb25f13870cd9 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Cl=C3=A9ment=20L=C3=A9ger?= Date: Fri, 29 Sep 2023 21:16:37 +0200 Subject: [PATCH 2/4] tracing: relax trace_event_eval_update() execution with cond_resched() MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit When kernel is compiled without preemption, the eval_map_work_func() (which calls trace_event_eval_update()) will not be preempted up to its complete execution. This can actually cause a problem since if another CPU call stop_machine(), the call will have to wait for the eval_map_work_func() function to finish executing in the workqueue before being able to be scheduled. This problem was observe on a SMP system at boot time, when the CPU calling the initcalls executed clocksource_done_booting() which in the end calls stop_machine(). We observed a 1 second delay because one CPU was executing eval_map_work_func() and was not preempted by the stop_machine() task. Adding a call to cond_resched() in trace_event_eval_update() allows other tasks to be executed and thus continue working asynchronously like before without blocking any pending task at boot time. Link: https://lore.kernel.org/linux-trace-kernel/20230929191637.416931-1-cleger@rivosinc.com Cc: Masami Hiramatsu Signed-off-by: Clément Léger Tested-by: Atish Patra Reviewed-by: Atish Patra Signed-off-by: Steven Rostedt (Google) --- kernel/trace/trace_events.c | 1 + 1 file changed, 1 insertion(+) diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c index 91951d038ba49..f49d6ddb63425 100644 --- a/kernel/trace/trace_events.c +++ b/kernel/trace/trace_events.c @@ -2770,6 +2770,7 @@ void trace_event_eval_update(struct trace_eval_map **map, int len) update_event_fields(call, map[i]); } } + cond_resched(); } up_write(&trace_event_sem); } From 2de9ee94054263940122aee8720e902b30c27930 Mon Sep 17 00:00:00 2001 From: Beau Belgrave Date: Mon, 25 Sep 2023 23:08:28 +0000 Subject: [PATCH 3/4] tracing/user_events: Align set_bit() address for all archs MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit All architectures should use a long aligned address passed to set_bit(). User processes can pass either a 32-bit or 64-bit sized value to be updated when tracing is enabled when on a 64-bit kernel. Both cases are ensured to be naturally aligned, however, that is not enough. The address must be long aligned without affecting checks on the value within the user process which require different adjustments for the bit for little and big endian CPUs. Add a compat flag to user_event_enabler that indicates when a 32-bit value is being used on a 64-bit kernel. Long align addresses and correct the bit to be used by set_bit() to account for this alignment. Ensure compat flags are copied during forks and used during deletion clears. Link: https://lore.kernel.org/linux-trace-kernel/20230925230829.341-2-beaub@linux.microsoft.com Link: https://lore.kernel.org/linux-trace-kernel/20230914131102.179100-1-cleger@rivosinc.com/ Cc: stable@vger.kernel.org Fixes: 7235759084a4 ("tracing/user_events: Use remote writes for event enablement") Reported-by: Clément Léger Suggested-by: Clément Léger Signed-off-by: Beau Belgrave Signed-off-by: Steven Rostedt (Google) --- kernel/trace/trace_events_user.c | 58 ++++++++++++++++++++++++++++---- 1 file changed, 51 insertions(+), 7 deletions(-) diff --git a/kernel/trace/trace_events_user.c b/kernel/trace/trace_events_user.c index 6f046650e5271..b87f41187c6a9 100644 --- a/kernel/trace/trace_events_user.c +++ b/kernel/trace/trace_events_user.c @@ -127,8 +127,13 @@ struct user_event_enabler { /* Bit 7 is for freeing status of enablement */ #define ENABLE_VAL_FREEING_BIT 7 -/* Only duplicate the bit value */ -#define ENABLE_VAL_DUP_MASK ENABLE_VAL_BIT_MASK +/* Bit 8 is for marking 32-bit on 64-bit */ +#define ENABLE_VAL_32_ON_64_BIT 8 + +#define ENABLE_VAL_COMPAT_MASK (1 << ENABLE_VAL_32_ON_64_BIT) + +/* Only duplicate the bit and compat values */ +#define ENABLE_VAL_DUP_MASK (ENABLE_VAL_BIT_MASK | ENABLE_VAL_COMPAT_MASK) #define ENABLE_BITOPS(e) (&(e)->values) @@ -174,6 +179,30 @@ struct user_event_validator { int flags; }; +static inline void align_addr_bit(unsigned long *addr, int *bit, + unsigned long *flags) +{ + if (IS_ALIGNED(*addr, sizeof(long))) { +#ifdef __BIG_ENDIAN + /* 32 bit on BE 64 bit requires a 32 bit offset when aligned. */ + if (test_bit(ENABLE_VAL_32_ON_64_BIT, flags)) + *bit += 32; +#endif + return; + } + + *addr = ALIGN_DOWN(*addr, sizeof(long)); + + /* + * We only support 32 and 64 bit values. The only time we need + * to align is a 32 bit value on a 64 bit kernel, which on LE + * is always 32 bits, and on BE requires no change when unaligned. + */ +#ifdef __LITTLE_ENDIAN + *bit += 32; +#endif +} + typedef void (*user_event_func_t) (struct user_event *user, struct iov_iter *i, void *tpdata, bool *faulted); @@ -482,6 +511,7 @@ static int user_event_enabler_write(struct user_event_mm *mm, unsigned long *ptr; struct page *page; void *kaddr; + int bit = ENABLE_BIT(enabler); int ret; lockdep_assert_held(&event_mutex); @@ -497,6 +527,8 @@ static int user_event_enabler_write(struct user_event_mm *mm, test_bit(ENABLE_VAL_FREEING_BIT, ENABLE_BITOPS(enabler)))) return -EBUSY; + align_addr_bit(&uaddr, &bit, ENABLE_BITOPS(enabler)); + ret = pin_user_pages_remote(mm->mm, uaddr, 1, FOLL_WRITE | FOLL_NOFAULT, &page, NULL); @@ -515,9 +547,9 @@ static int user_event_enabler_write(struct user_event_mm *mm, /* Update bit atomically, user tracers must be atomic as well */ if (enabler->event && enabler->event->status) - set_bit(ENABLE_BIT(enabler), ptr); + set_bit(bit, ptr); else - clear_bit(ENABLE_BIT(enabler), ptr); + clear_bit(bit, ptr); kunmap_local(kaddr); unpin_user_pages_dirty_lock(&page, 1, true); @@ -849,6 +881,12 @@ static struct user_event_enabler enabler->event = user; enabler->addr = uaddr; enabler->values = reg->enable_bit; + +#if BITS_PER_LONG >= 64 + if (reg->enable_size == 4) + set_bit(ENABLE_VAL_32_ON_64_BIT, ENABLE_BITOPS(enabler)); +#endif + retry: /* Prevents state changes from racing with new enablers */ mutex_lock(&event_mutex); @@ -2377,7 +2415,8 @@ static long user_unreg_get(struct user_unreg __user *ureg, } static int user_event_mm_clear_bit(struct user_event_mm *user_mm, - unsigned long uaddr, unsigned char bit) + unsigned long uaddr, unsigned char bit, + unsigned long flags) { struct user_event_enabler enabler; int result; @@ -2385,7 +2424,7 @@ static int user_event_mm_clear_bit(struct user_event_mm *user_mm, memset(&enabler, 0, sizeof(enabler)); enabler.addr = uaddr; - enabler.values = bit; + enabler.values = bit | flags; retry: /* Prevents state changes from racing with new enablers */ mutex_lock(&event_mutex); @@ -2415,6 +2454,7 @@ static long user_events_ioctl_unreg(unsigned long uarg) struct user_event_mm *mm = current->user_event_mm; struct user_event_enabler *enabler, *next; struct user_unreg reg; + unsigned long flags; long ret; ret = user_unreg_get(ureg, ®); @@ -2425,6 +2465,7 @@ static long user_events_ioctl_unreg(unsigned long uarg) if (!mm) return -ENOENT; + flags = 0; ret = -ENOENT; /* @@ -2441,6 +2482,9 @@ static long user_events_ioctl_unreg(unsigned long uarg) ENABLE_BIT(enabler) == reg.disable_bit) { set_bit(ENABLE_VAL_FREEING_BIT, ENABLE_BITOPS(enabler)); + /* We must keep compat flags for the clear */ + flags |= enabler->values & ENABLE_VAL_COMPAT_MASK; + if (!test_bit(ENABLE_VAL_FAULTING_BIT, ENABLE_BITOPS(enabler))) user_event_enabler_destroy(enabler, true); @@ -2454,7 +2498,7 @@ static long user_events_ioctl_unreg(unsigned long uarg) /* Ensure bit is now cleared for user, regardless of event status */ if (!ret) ret = user_event_mm_clear_bit(mm, reg.disable_addr, - reg.disable_bit); + reg.disable_bit, flags); return ret; } From 2598bd3ca8dcf5bbca1161ee5b271b432398da37 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Google)" Date: Sat, 30 Sep 2023 09:01:06 -0400 Subject: [PATCH 4/4] eventfs: Test for dentries array allocated in eventfs_release() The dcache_dir_open_wrapper() could be called when a dynamic event is being deleted leaving a dentry with no children. In this case the dlist->dentries array will never be allocated. This needs to be checked for in eventfs_release(), otherwise it will trigger a NULL pointer dereference. Link: https://lore.kernel.org/linux-trace-kernel/20230930090106.1c3164e9@rorschach.local.home Cc: Mark Rutland Acked-by: Masami Hiramatsu (Google) Fixes: ef36b4f92868 ("eventfs: Remember what dentries were created on dir open") Signed-off-by: Steven Rostedt (Google) --- fs/tracefs/event_inode.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fs/tracefs/event_inode.c b/fs/tracefs/event_inode.c index 5f17140898844..8c8d64e76103e 100644 --- a/fs/tracefs/event_inode.c +++ b/fs/tracefs/event_inode.c @@ -421,7 +421,7 @@ static int eventfs_release(struct inode *inode, struct file *file) if (WARN_ON_ONCE(!dlist)) return -EINVAL; - for (i = 0; dlist->dentries[i]; i++) { + for (i = 0; dlist->dentries && dlist->dentries[i]; i++) { dput(dlist->dentries[i]); }