- 20 Dec, 2018 1 commit
-
-
Steven Rostedt (VMware) authored
Since commit 79922b80 ("ftrace: Optimize function graph to be called directly"), dynamic trampolines should not be calling the function graph tracer at the end. If they do, it could cause the function graph tracer to trace functions that it filtered out. Right now it does not cause a problem because there's a test to check if the function graph tracer is attached to the same function as the function tracer, which for now is true. But the function graph tracer is undergoing changes that can make this no longer true which will cause the function graph tracer to trace other functions. For example: # cd /sys/kernel/tracing/ # echo do_IRQ > set_ftrace_filter # mkdir instances/foo # echo ip_rcv > instances/foo/set_ftrace_filter # echo function_graph > current_tracer # echo function > instances/foo/current_tracer Would cause the function graph tracer to trace both do_IRQ and ip_rcv, if the current tests change. As the current tests prevent this from being a problem, this code does not need to be backported. But it does make the code cleaner. Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Borislav Petkov <bp@alien8.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: x86@kernel.org Signed-off-by:
Steven Rostedt (VMware) <rostedt@goodmis.org>
-
- 10 Dec, 2018 6 commits
-
-
Steven Rostedt (VMware) authored
The ret_stack should not be accessed directly via the curr_ret_stack variable on the task_struct. This is because the ret_stack is going to be converted into a series of longs and not an array of ret_stack structures. The way that a ret_stack should be retrieved is via the ftrace_graph_get_ret_stack structure, but it needs to be documented on how to use it. Signed-off-by: -
Steven Rostedt (VMware) authored
It has been reported that ftrace_replace_code() which is called by ftrace_modify_all_code() can cause a soft lockup warning for an allmodconfig kernel. This is because all the debug options enabled causes the loop in ftrace_replace_code() (which loops over all the functions being enabled where there can be 10s of thousands), is too slow, and never schedules out. To solve this, setting FTRACE_MAY_SLEEP to the command passed into ftrace_replace_code() will make it call cond_resched() in the loop, which prevents the soft lockup warning from triggering. Link: http://lkml.kernel.org/r/20181204192903.8193-1-anders.roxell@linaro.org Link: http://lkml.kernel.org/r/20181205183304.000714627@goodmis.orgAcked-by:
Will Deacon <will.deacon@arm.com> Reported-by:
Anders Roxell <anders.roxell@linaro.org> Tested-by:
-
Steven Rostedt (VMware) authored
The function ftrace_replace_code() is the ftrace engine that does the work to modify all the nops into the calls to the function callback in all the functions being traced. The generic version which is normally called from stop machine, but an architecture can implement a non stop machine version and still use the generic ftrace_replace_code(). When an architecture does this, ftrace_replace_code() may be called from a schedulable context, where it can allow the code to be preemptible, and schedule out. In order to allow an architecture to make ftrace_replace_code() schedulable, a new command flag is added called: FTRACE_MAY_SLEEP Which can be or'd to the command that is passed to ftrace_modify_all_code() that calls ftrace_replace_code() and will have it call cond_resched() in the loop that modifies the nops into the calls to the ftrace trampolines. Link: http://lkml.kernel.org/r/20181204192903.8193-1-anders.roxell@linaro.org Link: http://lkml.kernel.org/r/20181205183303.828422192@goodmis.orgReported-by:
-
Masami Hiramatsu authored
Add common testcases for dynamic_events interface. - Add/remove kprobe events via dynamic_events - Add/remove synthetic events via dynamic_events - Selective clear events (clear events other interfaces) - Genelic clear events ("!LINE" syntax) Link: http://lkml.kernel.org/r/154140872590.17322.10394440849261743052.stgit@devboxReviewed-by:Tom Zanussi <tom.zanussi@linux.intel.com> Tested-by:
Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by:
-
Masami Hiramatsu authored
Add a generic method to remove event from dynamic event list. This is same as other system under ftrace. You just need to pass the event name with '!', e.g. # echo p:new_grp/new_event _do_fork > dynamic_events This creates an event, and # echo '!p:new_grp/new_event _do_fork' > dynamic_events Or, # echo '!p:new_grp/new_event' > dynamic_events will remove new_grp/new_event event. Note that this doesn't check the event prefix (e.g. "p:") strictly, because the "group/event" name must be unique. Link: http://lkml.kernel.org/r/154140869774.17322.8887303560398645347.stgit@devboxReviewed-by:
-
Steven Rostedt (VMware) authored
The trace_add/remove_event_call_nolock() functions were added to allow the tace_add/remove_event_call() code be called when the event_mutex lock was already taken. Now that all callers are done within the event_mutex, there's no reason to have two different interfaces. Remove the current wrapper trace_add/remove_event_call()s and rename the _nolock versions back to the original names. Link: http://lkml.kernel.org/r/154140866955.17322.2081425494660638846.stgit@devboxAcked-by:
-
- 09 Dec, 2018 23 commits
-
-
Masami Hiramatsu authored
Rmove unneeded synth_event_mutex. This mutex protects the reference count in synth_event, however, those operational points are already protected by event_mutex. 1. In __create_synth_event() and create_or_delete_synth_event(), those synth_event_mutex clearly obtained right after event_mutex. 2. event_hist_trigger_func() is trigger_hist_cmd.func() which is called by trigger_process_regex(), which is a part of event_trigger_regex_write() and this function takes event_mutex. 3. hist_unreg_all() is trigger_hist_cmd.unreg_all() which is called by event_trigger_regex_open() and it takes event_mutex. 4. onmatch_destroy() and onmatch_create() have long call tree, but both are finally invoked from event_trigger_regex_write() and event_trace_del_tracer(), former takes event_mutex, and latter ensures called under event_mutex locked. Finally, I ensured there is no resource conflict. For safety, I added lockdep_assert_held(&event_mutex) for each function. Link: http://lkml.kernel.org/r/154140864134.17322.4796059721306031894.stgit@devboxReviewed-by:
-
Masami Hiramatsu authored
Use dyn_event framework for synthetic events. This shows synthetic events on "tracing/dynamic_events" file in addition to tracing/synthetic_events interface. User can also define new events via tracing/dynamic_events with "s:" prefix. So, the new syntax is below; s:[synthetic/]EVENT_NAME TYPE ARG; [TYPE ARG;]... To remove events via tracing/dynamic_events, you can use "-:" prefix as same as other events. Link: http://lkml.kernel.org/r/154140861301.17322.15454611233735614508.stgit@devboxReviewed-by:
-
Masami Hiramatsu authored
Use dyn_event framework for uprobe events. This shows uprobe events on "dynamic_events" file. User can also define new uprobe events via dynamic_events. Link: http://lkml.kernel.org/r/154140858481.17322.9091293846515154065.stgit@devboxReviewed-by:
-
Masami Hiramatsu authored
Use dyn_event framework for kprobe events. This shows kprobe events on "tracing/dynamic_events" file. User can also define new events via tracing/dynamic_events. Link: http://lkml.kernel.org/r/154140855646.17322.6619219995865980392.stgit@devboxReviewed-by:
-
Masami Hiramatsu authored
Add unified dynamic event framework for ftrace kprobes, uprobes and synthetic events. Those dynamic events can be co-exist on same file because those syntax doesn't overlap. This introduces a framework part which provides a unified tracefs interface and operations. Link: http://lkml.kernel.org/r/154140852824.17322.12250362185969352095.stgit@devboxReviewed-by:
-
Masami Hiramatsu authored
Integrate similar argument parsers for kprobes and uprobes events into traceprobe_parse_probe_arg(). Link: http://lkml.kernel.org/r/154140850016.17322.9836787731210512176.stgit@devboxReviewed-by:
-
Masami Hiramatsu authored
Since the event_mutex and synth_event_mutex ordering issue is gone, we can skip existing event check when adding or deleting events, and some redundant code in error path. This changes release_all_synth_events() to abort the process when it hits any error and returns the error code. It succeeds only if it has no error. Link: http://lkml.kernel.org/r/154140847194.17322.17960275728005067803.stgit@devboxReviewed-by:
-
Masami Hiramatsu authored
synthetic event is using synth_event_mutex for protecting synth_event_list, and event_trigger_write() path acquires locks as below order. event_trigger_write(event_mutex) ->trigger_process_regex(trigger_cmd_mutex) ->event_hist_trigger_func(synth_event_mutex) On the other hand, synthetic event creation and deletion paths call trace_add_event_call() and trace_remove_event_call() which acquires event_mutex. In that case, if we keep the synth_event_mutex locked while registering/unregistering synthetic events, its dependency will be inversed. To avoid this issue, current synthetic event is using a 2 phase process to create/delete events. For example, it searches existing events under synth_event_mutex to check for event-name conflicts, and unlocks synth_event_mutex, then registers a new event under event_mutex locked. Finally, it locks synth_event_mutex and tries to add the new event to the list. But it can introduce complexity and a chance for name conflicts. To solve this simpler, this introduces trace_add_event_call_nolock() and trace_remove_event_call_nolock() which don't acquire event_mutex inside. synthetic event can lock event_mutex before synth_event_mutex to solve the lock dependency issue simpler. Link: http://lkml.kernel.org/r/154140844377.17322.13781091165954002713.stgit@devboxReviewed-by: -
Masami Hiramatsu authored
Add a busy check loop in cleanup_all_probes() before trying to remove all events in uprobe_events, the same way that kprobe_events does. Without this change, writing null to uprobe_events will try to remove events but if one of them is enabled, it will stop there leaving some events cleared and others not clceared. With this change, writing null to uprobe_events makes sure all events are not enabled before removing events. So, it clears all events, or returns an error (-EBUSY) with keeping all events. Link: http://lkml.kernel.org/r/154140841557.17322.12653952888762532401.stgit@devboxReviewed-by:
-
Steven Rostedt (VMware) authored
After running several tests, it appears that having the reader wait till half the buffer is full before starting to read (and causing its own events to fill up the ring buffer constantly), works well. It keeps trace-cmd (the main user of this interface) from dominating the traces it records. Signed-off-by: -
Steven Rostedt (VMware) authored
Add a "buffer_percentage" file, that allows users to specify how much of the buffer (percentage of pages) need to be filled before waking up a task blocked on a per cpu trace_pipe_raw file. Signed-off-by: -
Steven Rostedt (VMware) authored
Instead of just waiting for a page to be full before waking up a pending reader, allow the reader to pass in a "percentage" of pages that have content before waking up a reader. This should help keep the process of reading the events not cause wake ups that constantly cause reading of the buffer. Signed-off-by: -
Joe Lawrence authored
When building with -ffunction-sections, the compiler will place each function into its own ELF section, prefixed with ".text". For example, a simple test module with functions test_module_do_work() and test_module_wq_func(): % objdump --section-headers test_module.o | awk '/\.text/{print $2}' .text .text.test_module_do_work .text.test_module_wq_func .init.text .exit.text Adjust the recordmcount scripts to look for ".text" as a section name prefix. This will ensure that those functions will be included in the __mcount_loc relocations: % objdump --reloc --section __mcount_loc test_module.o OFFSET TYPE VALUE 0000000000000000 R_X86_64_64 .text.test_module_do_work 0000000000000008 R_X86_64_64 .text.test_module_wq_func 0000000000000010 R_X86_64_64 .init.text Link: http://lkml.kernel.org/r/1542745158-25392-2-git-send-email-joe.lawrence@redhat.comSigned-off-by:Joe Lawrence <joe.lawrence@redhat.com> Signed-off-by:
-
Dan Carpenter authored
Dan Carpenter reviewed the trace_stack.c code and figured he found an off by one bug. "From reviewing the code, it seems possible for stack_trace_max.nr_entries to be set to .max_entries and in that case we would be reading one element beyond the end of the stack_dump_trace[] array. If it's not set to .max_entries then the bug doesn't affect runtime." Although it looks to be the case, it is not. Because we have: static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] = { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX }; struct stack_trace stack_trace_max = { .max_entries = STACK_TRACE_ENTRIES - 1, .entries = &stack_dump_trace[0], }; And: stack_trace_max.nr_entries = x; for (; x < i; x++) stack_dump_trace[x] = ULONG_MAX; Even if nr_entries equals max_entries, indexing with it into the stack_dump_trace[] array will not overflow the array. But if it is the case, the second part of the conditional that tests stack_dump_trace[nr_entries] to ULONG_MAX will always be true. By applying Dan's patch, it removes the subtle aspect of it and makes the if conditional slightly more efficient. Link: http://lkml.kernel.org/r/20180620110758.crunhd5bfep7zuiz@kili.mountainSigned-off-by:Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by:
-
Steven Rostedt (VMware) authored
The ret_stack processing is going to change, and that is going to break anything that is accessing the ret_stack directly. One user is the function graph profiler. By using the ftrace_graph_get_ret_stack() helper function, the profiler can access the ret_stack entry without relying on the implementation details of the stack itself. Reviewed-by:
Joel Fernandes (Google) <joel@joelfernandes.org> Signed-off-by:
-
Steven Rostedt (VMware) authored
Move the function function_graph_ret_addr() to fgraph.c, as the management of the curr_ret_stack is going to change, and all the accesses to ret_stack needs to be done in fgraph.c. Reviewed-by:
-
Steven Rostedt (VMware) authored
Currently the registering of function graph is to pass in a entry and return function. We need to have a way to associate those functions together where the entry can determine to run the return hook. Having a structure that contains both functions will facilitate the process of converting the code to be able to do such. This is similar to the way function hooks are enabled (it passes in ftrace_ops). Instead of passing in the functions to use, a single structure is passed in to the registering function. The unregister function is now passed in the fgraph_ops handle. When we allow more than one callback to the function graph hooks, this will let the system know which one to remove. Reviewed-by:
-
Steven Rostedt (VMware) authored
Rearrange the functions in trace_sched_wakeup.c so that there are fewer #ifdef CONFIG_FUNCTION_TRACER and #ifdef CONFIG_FUNCTION_GRAPH_TRACER, instead of having the #ifdefs spread all over. No functional change is made. Signed-off-by: -
Steven Rostedt (VMware) authored
To make the function graph infrastructure more managable, the code needs to be in its own file (fgraph.c). Move the code that is specific for managing the function graph infrastructure out of ftrace.c and into fgraph.c Reviewed-by:
-
Steven Rostedt (VMware) authored
When the function profiler is not configured, the "graph_time" option is meaningless, as the function profiler is the only thing that makes use of it. Do not expose it if the profiler is not configured. Link: http://lkml.kernel.org/r/20181123061133.GA195223@google.comReported-by:
-
Steven Rostedt (VMware) authored
In order to move function graph infrastructure into its own file (fgraph.h) it needs to access various functions and variables in ftrace.c that are currently static. Create a new file called ftrace-internal.h that holds the function prototypes and the extern declarations of the variables needed by fgraph.c as well, and make them global in ftrace.c such that they can be used outside that file. Reviewed-by:
-
Steven Rostedt (VMware) authored
The curr_ret_stack is no longer set to a negative value when a function is not to be traced by the function graph tracer. Remove the usage of FTRACE_NOTRACE_DEPTH, as it is no longer needed. Reviewed-by:
-
Steven Rostedt (VMware) authored
Functions in the set_graph_notrace no longer subtract FTRACE_NOTRACE_DEPTH from curr_ret_stack, as that is now implemented via the trace_recursion flags. Access to curr_ret_stack no longer needs to worry about checking for this. curr_ret_stack is still initialized to -1, when there's not a shadow stack allocated. Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: linux-arm-kernel@lists.infradead.org Acked-by:
-
- 30 Nov, 2018 6 commits
-
-
Steven Rostedt (VMware) authored
In order to make the function graph infrastructure more generic, there can not be code specific for the function_graph tracer in the generic code. This includes the set_graph_notrace logic, that stops all graph calls when a function in the set_graph_notrace is hit. By using the trace_recursion mask, we can use a bit in the current task_struct to implement the notrace code, and move the logic out of fgraph.c and into trace_functions_graph.c and keeps it affecting only the tracer and not all call graph callbacks. Acked-by:
Namhyung Kim <namhyung@kernel.org> Reviewed-by:
-
Steven Rostedt (VMware) authored
As the function graph infrastructure can be used by thing other than tracing, moving the code to its own file out of the trace_functions_graph.c code makes more sense. The fgraph.c file will only contain the infrastructure required to hook into functions and their return code. Reviewed-by:
-
Steven Rostedt (VMware) authored
Commit 588ca1786f2dd ("function_graph: Use new curr_ret_depth to manage depth instead of curr_ret_stack") removed a parameter from the call ftrace_push_return_trace() that made it so that the entire call was under 80 characters, but it did not remove the line break. There's no reason to break that line up, so make it a single line. Link: http://lkml.kernel.org/r/20181122100322.GN2131@hirez.programming.kicks-ass.netReported-by:Peter Zijlstra <peterz@infradead.org> Signed-off-by:
-
Steven Rostedt (VMware) authored
The static inline function task_curr_ret_stack() is unused, remove it. Reviewed-by:
-
Steven Rostedt (VMware) authored
The tracefs file set_graph_function is used to only function graph functions that are listed in that file (or all functions if the file is empty). The way this is implemented is that the function graph tracer looks at every function, and if the current depth is zero and the function matches something in the file then it will trace that function. When other functions are called, the depth will be greater than zero (because the original function will be at depth zero), and all functions will be traced where the depth is greater than zero. The issue is that when a function is first entered, and the handler that checks this logic is called, the depth is set to zero. If an interrupt comes in and a function in the interrupt handler is traced, its depth will be greater than zero and it will automatically be traced, even if the original function was not. But because the logic only looks at depth it may trace interrupts when it should not be. The recent design change of the function graph tracer to fix other bugs caused the depth to be zero while the function graph callback handler is being called for a longer time, widening the race of this happening. This bug was actually there for a longer time, but because the race window was so small it seldom happened. The Fixes tag below is for the commit that widen the race window, because that commit belongs to a series that will also help fix the original bug. Cc: stable@kernel.org Fixes: 39eb456d ("function_graph: Use new curr_ret_depth to manage depth instead of curr_ret_stack") Reported-by:
-
Zenghui Yu authored
After enabling KVM event tracing, almost all of trace_kvm_exit()'s printk shows "kvm_exit: IRQ: ..." even if the actual exception_type is NOT IRQ. More specifically, trace_kvm_exit() is defined in virt/kvm/arm/trace.h by TRACE_EVENT. This slight problem may have existed after commit e6753f23 ("tracepoint: Make rcuidle tracepoint callers use SRCU"). There are two variables in trace_kvm_exit() and __DO_TRACE() which have the same name, *idx*. Thus the actual value of *idx* will be overwritten when tracing. Fix it by adding a simple prefix. Cc: Joel Fernandes <joel@joelfernandes.org> Cc: Wang Haibin <wanghaibin.wang@huawei.com> Cc: linux-trace-devel@vger.kernel.org Cc: stable@vger.kernel.org Fixes: e6753f23 ("tracepoint: Make rcuidle tracepoint callers use SRCU") Reviewed-by:
Zenghui Yu <yuzenghui@huawei.com> Signed-off-by:
-
- 28 Nov, 2018 4 commits
-
-
Pavankumar Kondeti authored
commit 3f5fe9fe ("sched/debug: Fix task state recording/printout") tried to fix the problem introduced by a previous commit efb40f58 ("sched/tracing: Fix trace_sched_switch task-state printing"). However the prev_state output in sched_switch is still broken. task_state_index() uses fls() which considers the LSB as 1. Left shifting 1 by this value gives an incorrect mapping to the task state. Fix this by decrementing the value returned by __get_task_state() before shifting. Link: http://lkml.kernel.org/r/1540882473-1103-1-git-send-email-pkondeti@codeaurora.org Cc: stable@vger.kernel.org Fixes: 3f5fe9fe ("sched/debug: Fix task state recording/printout") Signed-off-by:
Pavankumar Kondeti <pkondeti@codeaurora.org> Signed-off-by:
-
Steven Rostedt (VMware) authored
The profiler uses trace->depth to find its entry on the ret_stack, but the depth may not match the actual location of where its entry is (if an interrupt were to preempt the processing of the profiler for another function, the depth and the curr_ret_stack will be different). Have it use the curr_ret_stack as the index to find its ret_stack entry instead of using the depth variable, as that is no longer guaranteed to be the same. Cc: stable@kernel.org Fixes: 03274a3f ("tracing/fgraph: Adjust fgraph depth before calling trace return callback") Reviewed-by:
-
Steven Rostedt (VMware) authored
The function graph profiler uses the ret_stack to store the "subtime" and reuse it by nested functions and also on the return. But the current logic has the profiler callback called before the ret_stack is updated, and it is just modifying the ret_stack that will later be allocated (it's just lucky that the "subtime" is not touched when it is allocated). This could also cause a crash if we are at the end of the ret_stack when this happens. By reversing the order of the allocating the ret_stack and then calling the callbacks attached to a function being traced, the ret_stack entry is no longer used before it is allocated. Cc: stable@kernel.org Fixes: 03274a3f ("tracing/fgraph: Adjust fgraph depth before calling trace return callback") Reviewed-by:
-
Steven Rostedt (VMware) authored
In the past, curr_ret_stack had two functions. One was to denote the depth of the call graph, the other is to keep track of where on the ret_stack the data is used. Although they may be slightly related, there are two cases where they need to be used differently. The one case is that it keeps the ret_stack data from being corrupted by an interrupt coming in and overwriting the data still in use. The other is just to know where the depth of the stack currently is. The function profiler uses the ret_stack to save a "subtime" variable that is part of the data on the ret_stack. If curr_ret_stack is modified too early, then this variable can be corrupted. The "max_depth" option, when set to 1, will record the first functions going into the kernel. To see all top functions (when dealing with timings), the depth variable needs to be lowered before calling the return hook. But by lowering the curr_ret_stack, it makes the data on the ret_stack still being used by the return hook susceptible to being overwritten. Now that there's two variables to handle both cases (curr_ret_depth), we can move them to the locations where they can handle both cases. Cc: stable@kernel.org Fixes: 03274a3f ("tracing/fgraph: Adjust fgraph depth before calling trace return callback") Reviewed-by:
-
