From 44dc5c41b5b1267d4dd037d26afc0c4d3a568acb Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Google)" Date: Fri, 9 Feb 2024 06:36:22 -0500 Subject: tracing: Fix wasted memory in saved_cmdlines logic While looking at improving the saved_cmdlines cache I found a huge amount of wasted memory that should be used for the cmdlines. The tracing data saves pids during the trace. At sched switch, if a trace occurred, it will save the comm of the task that did the trace. This is saved in a "cache" that maps pids to comms and exposed to user space via the /sys/kernel/tracing/saved_cmdlines file. Currently it only caches by default 128 comms. The structure that uses this creates an array to store the pids using PID_MAX_DEFAULT (which is usually set to 32768). This causes the structure to be of the size of 131104 bytes on 64 bit machines. In hex: 131104 = 0x20020, and since the kernel allocates generic memory in powers of two, the kernel would allocate 0x40000 or 262144 bytes to store this structure. That leaves 131040 bytes of wasted space. Worse, the structure points to an allocated array to store the comm names, which is 16 bytes times the amount of names to save (currently 128), which is 2048 bytes. Instead of allocating a separate array, make the structure end with a variable length string and use the extra space for that. This is similar to a recommendation that Linus had made about eventfs_inode names: https://lore.kernel.org/all/20240130190355.11486-5-torvalds@linux-foundation.org/ Instead of allocating a separate string array to hold the saved comms, have the structure end with: char saved_cmdlines[]; and round up to the next power of two over sizeof(struct saved_cmdline_buffers) + num_cmdlines * TASK_COMM_LEN It will use this extra space for the saved_cmdline portion. Now, instead of saving only 128 comms by default, by using this wasted space at the end of the structure it can save over 8000 comms and even saves space by removing the need for allocating the other array. Link: https://lore.kernel.org/linux-trace-kernel/20240209063622.1f7b6d5f@rorschach.local.home Cc: stable@vger.kernel.org Cc: Masami Hiramatsu Cc: Mathieu Desnoyers Cc: Vincent Donnefort Cc: Sven Schnelle Cc: Mete Durlu Fixes: 939c7a4f04fcd ("tracing: Introduce saved_cmdlines_size file") Signed-off-by: Steven Rostedt (Google) --- kernel/trace/trace.c | 75 ++++++++++++++++++++++++++-------------------------- 1 file changed, 37 insertions(+), 38 deletions(-) (limited to 'kernel/trace/trace.c') diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 2a7c6fd934e9..9ff8a439d674 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -2320,7 +2320,7 @@ struct saved_cmdlines_buffer { unsigned *map_cmdline_to_pid; unsigned cmdline_num; int cmdline_idx; - char *saved_cmdlines; + char saved_cmdlines[]; }; static struct saved_cmdlines_buffer *savedcmd; @@ -2334,47 +2334,58 @@ static inline void set_cmdline(int idx, const char *cmdline) strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN); } -static int allocate_cmdlines_buffer(unsigned int val, - struct saved_cmdlines_buffer *s) +static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s) +{ + int order = get_order(sizeof(*s) + s->cmdline_num * TASK_COMM_LEN); + + kfree(s->map_cmdline_to_pid); + free_pages((unsigned long)s, order); +} + +static struct saved_cmdlines_buffer *allocate_cmdlines_buffer(unsigned int val) { + struct saved_cmdlines_buffer *s; + struct page *page; + int orig_size, size; + int order; + + /* Figure out how much is needed to hold the given number of cmdlines */ + orig_size = sizeof(*s) + val * TASK_COMM_LEN; + order = get_order(orig_size); + size = 1 << (order + PAGE_SHIFT); + page = alloc_pages(GFP_KERNEL, order); + if (!page) + return NULL; + + s = page_address(page); + memset(s, 0, sizeof(*s)); + + /* Round up to actual allocation */ + val = (size - sizeof(*s)) / TASK_COMM_LEN; + s->cmdline_num = val; + s->map_cmdline_to_pid = kmalloc_array(val, sizeof(*s->map_cmdline_to_pid), GFP_KERNEL); - if (!s->map_cmdline_to_pid) - return -ENOMEM; - - s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL); - if (!s->saved_cmdlines) { - kfree(s->map_cmdline_to_pid); - return -ENOMEM; + if (!s->map_cmdline_to_pid) { + free_saved_cmdlines_buffer(s); + return NULL; } s->cmdline_idx = 0; - s->cmdline_num = val; memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(s->map_pid_to_cmdline)); memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP, val * sizeof(*s->map_cmdline_to_pid)); - return 0; + return s; } static int trace_create_savedcmd(void) { - int ret; - - savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL); - if (!savedcmd) - return -ENOMEM; - - ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd); - if (ret < 0) { - kfree(savedcmd); - savedcmd = NULL; - return -ENOMEM; - } + savedcmd = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT); - return 0; + return savedcmd ? 0 : -ENOMEM; } int is_tracing_stopped(void) @@ -6056,26 +6067,14 @@ tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf, return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); } -static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s) -{ - kfree(s->saved_cmdlines); - kfree(s->map_cmdline_to_pid); - kfree(s); -} - static int tracing_resize_saved_cmdlines(unsigned int val) { struct saved_cmdlines_buffer *s, *savedcmd_temp; - s = kmalloc(sizeof(*s), GFP_KERNEL); + s = allocate_cmdlines_buffer(val); if (!s) return -ENOMEM; - if (allocate_cmdlines_buffer(val, s) < 0) { - kfree(s); - return -ENOMEM; - } - preempt_disable(); arch_spin_lock(&trace_cmdline_lock); savedcmd_temp = savedcmd; -- cgit v1.2.3-70-g09d2 From a6eaa24f1cc2c7aecec6047556bdfe32042094c3 Mon Sep 17 00:00:00 2001 From: Sven Schnelle Date: Mon, 5 Feb 2024 07:53:40 +0100 Subject: tracing: Use ring_buffer_record_is_set_on() in tracer_tracing_is_on() tracer_tracing_is_on() checks whether record_disabled is not zero. This checks both the record_disabled counter and the RB_BUFFER_OFF flag. Reading the source it looks like this function should only check for the RB_BUFFER_OFF flag. Therefore use ring_buffer_record_is_set_on(). This fixes spurious fails in the 'test for function traceon/off triggers' test from the ftrace testsuite when the system is under load. Link: https://lore.kernel.org/linux-trace-kernel/20240205065340.2848065-1-svens@linux.ibm.com Cc: Masami Hiramatsu Cc: Mathieu Desnoyers Tested-By: Mete Durlu Signed-off-by: Sven Schnelle Signed-off-by: Steven Rostedt (Google) --- kernel/trace/trace.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel/trace/trace.c') diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 9ff8a439d674..aa54810e8b56 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -1532,7 +1532,7 @@ void disable_trace_on_warning(void) bool tracer_tracing_is_on(struct trace_array *tr) { if (tr->array_buffer.buffer) - return ring_buffer_record_is_on(tr->array_buffer.buffer); + return ring_buffer_record_is_set_on(tr->array_buffer.buffer); return !tr->buffer_disabled; } -- cgit v1.2.3-70-g09d2 From 2394ac4145ea91b92271e675a09af2a9ea6840b7 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Google)" Date: Wed, 14 Feb 2024 11:20:46 -0500 Subject: tracing: Inform kmemleak of saved_cmdlines allocation The allocation of the struct saved_cmdlines_buffer structure changed from: s = kmalloc(sizeof(*s), GFP_KERNEL); s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL); to: orig_size = sizeof(*s) + val * TASK_COMM_LEN; order = get_order(orig_size); size = 1 << (order + PAGE_SHIFT); page = alloc_pages(GFP_KERNEL, order); if (!page) return NULL; s = page_address(page); memset(s, 0, sizeof(*s)); s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL); Where that s->saved_cmdlines allocation looks to be a dangling allocation to kmemleak. That's because kmemleak only keeps track of kmalloc() allocations. For allocations that use page_alloc() directly, the kmemleak needs to be explicitly informed about it. Add kmemleak_alloc() and kmemleak_free() around the page allocation so that it doesn't give the following false positive: unreferenced object 0xffff8881010c8000 (size 32760): comm "swapper", pid 0, jiffies 4294667296 hex dump (first 32 bytes): ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ................ ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ................ backtrace (crc ae6ec1b9): [] kmemleak_alloc+0x45/0x80 [] __kmalloc_large_node+0x10d/0x190 [] __kmalloc+0x3b1/0x4c0 [] allocate_cmdlines_buffer+0x113/0x230 [] tracer_alloc_buffers.isra.0+0x124/0x460 [] early_trace_init+0x14/0xa0 [] start_kernel+0x12e/0x3c0 [] x86_64_start_reservations+0x18/0x30 [] x86_64_start_kernel+0x7b/0x80 [] secondary_startup_64_no_verify+0x15e/0x16b Link: https://lore.kernel.org/linux-trace-kernel/87r0hfnr9r.fsf@kernel.org/ Link: https://lore.kernel.org/linux-trace-kernel/20240214112046.09a322d6@gandalf.local.home Cc: Masami Hiramatsu Cc: Mathieu Desnoyers Cc: Catalin Marinas Fixes: 44dc5c41b5b1 ("tracing: Fix wasted memory in saved_cmdlines logic") Reported-by: Kalle Valo Tested-by: Kalle Valo Signed-off-by: Steven Rostedt (Google) --- kernel/trace/trace.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'kernel/trace/trace.c') diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index aa54810e8b56..8198bfc54b58 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -39,6 +39,7 @@ #include #include #include +#include #include #include #include @@ -2339,6 +2340,7 @@ static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s) int order = get_order(sizeof(*s) + s->cmdline_num * TASK_COMM_LEN); kfree(s->map_cmdline_to_pid); + kmemleak_free(s); free_pages((unsigned long)s, order); } @@ -2358,6 +2360,7 @@ static struct saved_cmdlines_buffer *allocate_cmdlines_buffer(unsigned int val) return NULL; s = page_address(page); + kmemleak_alloc(s, size, 1, GFP_KERNEL); memset(s, 0, sizeof(*s)); /* Round up to actual allocation */ -- cgit v1.2.3-70-g09d2