From: "Madhavan T. Venkataraman" <[email protected]>
I have synced this patch series to v5.19-rc2.
I have also removed the following patch.
[PATCH v14 7/7] arm64: Select HAVE_RELIABLE_STACKTRACE
as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
yet. This patch will be added in the future once Objtool is enhanced to
provide stack validation in some form.
Split unwind_init()
===================
Unwind initialization has 3 cases. Accordingly, define 3 separate init
functions as follows:
- unwind_init_from_regs()
- unwind_init_from_current()
- unwind_init_from_task()
This makes it easier to understand and add specialized code to each case
in the future.
Copy task argument
==================
Copy the task argument passed to arch_stack_walk() to unwind_state so that
it can be passed to unwind functions via unwind_state rather than as a
separate argument. The task is a fundamental part of the unwind state.
Redefine the unwinder loop
==========================
Redefine the unwinder loop and make it simple and somewhat similar to other
architectures. Define the following:
while (unwind_continue(&state, consume_entry, cookie))
unwind_next(&state);
unwind_continue()
This new function implements checks to determine whether the
unwind should continue or terminate.
Reliability checks
==================
There are some kernel features and conditions that make a stack trace
unreliable. Callers may require the unwinder to detect these cases.
E.g., livepatch.
Introduce a new function called unwind_check_reliability() that will detect
these cases and set a boolean "reliable" in the stackframe. Call
unwind_check_reliability() for every frame.
Introduce the first reliability check in unwind_check_reliability() - If
a return PC is not a valid kernel text address, consider the stack
trace unreliable. It could be some generated code.
Other reliability checks will be added in the future.
Make unwind() return a boolean to indicate reliability of the stack trace.
SYM_CODE check
==============
This is the second reliability check implemented.
SYM_CODE functions do not follow normal calling conventions. They cannot
be unwound reliably using the frame pointer. Collect the address ranges
of these functions in a special section called "sym_code_functions".
In unwind_check_reliability(), check the return PC against these ranges. If
a match is found, then mark the stack trace unreliable.
Last stack frame
================
If a SYM_CODE function occurs in the very last frame in the stack trace,
then the stack trace is not considered unreliable. This is because there
is no more unwinding to do. Examples:
- EL0 exception stack traces end in the top level EL0 exception
handlers.
- All kernel thread stack traces end in ret_from_fork().
arch_stack_walk_reliable()
==========================
Introduce arch_stack_walk_reliable() for ARM64. This works like
arch_stack_walk() except that it returns an error if the stack trace is
found to be unreliable.
Until all of the reliability checks are in place in
unwind_check_reliability(), arch_stack_walk_reliable() may not be used by
livepatch. But it may be used by debug and test code.
HAVE_RELIABLE_STACKTRACE
========================
Select this config for arm64. However, make it conditional on
STACK_VALIDATION. When objtool is enhanced to implement stack
validation for arm64, STACK_VALIDATION will be defined.
---
Changelog:
v15:
From Mark Brown:
- Sync this patch series to v5.19-rc2.
- Reviewed-By: for:
[PATCH v14 1/7] arm64: Split unwind_init()
[PATCH v14 2/7] arm64: Copy the task argument to unwind_state
From Madhavan T. Venkataraman:
- Remove the following patch from the series:
[PATCH v14 7/7] arm64: Select HAVE_RELIABLE_STACKTRACE
as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is
not present yet. This patch will be added in the future once
Objtool is enhanced to provide stack validation in some form.
v14:
From Mark Rutland, Mark Brown:
- Add requirements for the three helper functions that init a stack
trace.
From Mark Rutland:
- Change the comment for the task field in struct stackframe.
- Hard code the task to current in unwind_init_from_regs(). Add a
sanity check task == current.
- Rename unwind_init_from_current() to unwind_init_from_caller().
- Remove task argument from unwind_init_from_caller().
From Mark Brown:
- Reviewed-By: for:
[PATCH v13 05/11] arm64: Copy the task argument to unwind_state
v13:
From Mark Brown:
- Reviewed-by for the following:
[PATCH v12 03/10] arm64: Rename stackframe to unwind_state
[PATCH v11 05/10] arm64: Copy unwind arguments to unwind_state
[PATCH v11 07/10] arm64: Introduce stack trace reliability checks
in the unwinder
[PATCH v11 5/5] arm64: Create a list of SYM_CODE functions, check
return PC against list
From Mark Rutland:
- Reviewed-by for the following:
[PATCH v12 01/10] arm64: Remove NULL task check from unwind_frame()
[PATCH v12 02/10] arm64: Rename unwinder functions
[PATCH v12 03/10] arm64: Rename stackframe to unwind_state
- For each of the 3 cases of unwind initialization, have a separate
init function. Call the common init from each of these init
functions rather than call it separately.
- Only copy the task argument to arch_stack_walk() into
unwind state. Pass the rest of the arguments as arguments to
unwind functions.
v12:
From Mark Brown:
- Reviewed-by for the following:
[PATCH v11 1/5] arm64: Call stack_backtrace() only from within
walk_stackframe()
[PATCH v11 2/5] arm64: Rename unwinder functions
[PATCH v11 3/5] arm64: Make the unwind loop in unwind() similar to
other architectures
[PATCH v11 5/5] arm64: Create a list of SYM_CODE functions, check
return PC against list
- Add an extra patch at the end to select HAVE_RELIABLE_STACKTRACE
just as a place holder for the review. I have added it and made
it conditional on STACK_VALIDATION which has not yet been
implemented.
- Mark had a concern about the code for the check for the final
frame being repeated in two places. I have now added a new
field called "final_fp" in struct stackframe which I compute
once in stacktrace initialization. I have added an explicit
comment that the stacktrace must terminate at the final_fp.
- Place the implementation of arch_stack_walk_reliable() in a
separate patch after all the reliability checks have been
implemented.
From Mark Rutland:
- Place the removal of the NULL task check in unwind_frame() in
a separate patch.
- Add a task field to struct stackframe so the task pointer can be
passed around via the frame instead of as a separate argument. I have
taken this a step further by copying all of the arguments to
arch_stack_walk() into struct stackframe so that only that
struct needs to be passed to unwind functions.
- Rename start_backtrace() to unwind_init() instead of unwind_start().
- Acked-by for the following:
[PATCH v11 2/5] arm64: Rename unwinder functions
- Rename "struct stackframe" to "struct unwind_state".
- Define separate inline functions for initializing the starting
FP and PC from regs, or caller, or blocked task. Don't merge
unwind_init() into unwind().
v11:
From Mark Rutland:
- Peter Zijlstra has submitted patches that make ARCH_STACKWALK
independent of STACKTRACE. Mark Rutland extracted some of the
patches from my v10 series and added his own patches and comments,
rebased it on top of Peter's changes and submitted the series.
So, I have rebased the rest of the patches from v10 on top of
Mark Rutland's changes.
- Split the renaming of the unwinder functions and annotating them
with notrace and NOKPROBE_SYMBOL(). Also, there is currently no
need to annotate unwind_start() as its caller is already annotated
properly. So, I am removing the annotation patch from the series.
This can be done separately later if deemed necessary. Similarly,
I have removed the annotations from unwind_check_reliability() and
unwind_continue().
From Nobuta Keiya:
- unwind_start() should check for final frame and not mark the
final frame unreliable.
v9, v10:
- v9 had a threading problem. So, I resent it as v10.
From me:
- Removed the word "RFC" from the subject line as I believe this
is mature enough to be a regular patch.
From Mark Brown, Mark Rutland:
- Split the patches into smaller, self-contained ones.
- Always enable STACKTRACE so that arch_stack_walk() is always
defined.
From Mark Rutland:
- Update callchain_trace() take the return value of
perf_callchain_store() into acount.
- Restore get_wchan() behavior to the original code.
- Simplify an if statement in dump_backtrace().
From Mark Brown:
- Do not abort the stack trace on the first unreliable frame.
v8:
- Synced to v5.14-rc5.
From Mark Rutland:
- Make the unwinder loop similar to other architectures.
- Keep details to within the unwinder functions and return a simple
boolean to the caller.
- Convert some of the current code that contains unwinder logic to
simply use arch_stack_walk(). I have converted all of them.
- Do not copy sym_code_functions[]. Just place it in rodata for now.
- Have the main loop check for termination conditions rather than
having unwind_frame() check for them. In other words, let
unwind_frame() assume that the fp is valid.
- Replace the big comment for SYM_CODE functions with a shorter
comment.
/*
* As SYM_CODE functions don't follow the usual calling
* conventions, we assume by default that any SYM_CODE function
* cannot be unwound reliably.
*
* Note that this includes:
*
* - Exception handlers and entry assembly
* - Trampoline assembly (e.g., ftrace, kprobes)
* - Hypervisor-related assembly
* - Hibernation-related assembly
* - CPU start-stop, suspend-resume assembly
* - Kernel relocation assembly
*/
v7:
The Mailer screwed up the threading on this. So, I have resent this
same series as version 8 with proper threading to avoid confusion.
v6:
From Mark Rutland:
- The per-frame reliability concept and flag are acceptable. But more
work is needed to make the per-frame checks more accurate and more
complete. E.g., some code reorg is being worked on that will help.
I have now removed the frame->reliable flag and deleted the whole
concept of per-frame status. This is orthogonal to this patch series.
Instead, I have improved the unwinder to return proper return codes
so a caller can take appropriate action without needing per-frame
status.
- Remove the mention of PLTs and update the comment.
I have replaced the comment above the call to __kernel_text_address()
with the comment suggested by Mark Rutland.
Other comments:
- Other comments on the per-frame stuff are not relevant because
that approach is not there anymore.
v5:
From Keiya Nobuta:
- The term blacklist(ed) is not to be used anymore. I have changed it
to unreliable. So, the function unwinder_blacklisted() has been
changed to unwinder_is_unreliable().
From Mark Brown:
- Add a comment for the "reliable" flag in struct stackframe. The
reliability attribute is not complete until all the checks are
in place. Added a comment above struct stackframe.
- Include some of the comments in the cover letter in the actual
code so that we can compare it with the reliable stack trace
requirements document for completeness. I have added a comment:
- above unwinder_is_unreliable() that lists the requirements
that are addressed by the function.
- above the __kernel_text_address() call about all the cases
the call covers.
v4:
From Mark Brown:
- I was checking the return PC with __kernel_text_address() before
the Function Graph trace handling. Mark Brown felt that all the
reliability checks should be performed on the original return PC
once that is obtained. So, I have moved all the reliability checks
to after the Function Graph Trace handling code in the unwinder.
Basically, the unwinder should perform PC translations first (for
rhe return trampoline for Function Graph Tracing, Kretprobes, etc).
Then, the reliability checks should be applied to the resulting
PC.
- Mark said to improve the naming of the new functions so they don't
collide with existing ones. I have used a prefix "unwinder_" for
all the new functions.
From Josh Poimboeuf:
- In the error scenarios in the unwinder, the reliable flag in the
stack frame should be set. Implemented this.
- Some of the other comments are not relevant to the new code as
I have taken a different approach in the new code. That is why
I have not made those changes. E.g., Ard wanted me to add the
"const" keyword to the global section array. That array does not
exist in v4. Similarly, Mark Brown said to use ARRAY_SIZE() for
the same array in a for loop.
Other changes:
- Add a new definition for SYM_CODE_END() that adds the address
range of the function to a special section called
"sym_code_functions".
- Include the new section under initdata in vmlinux.lds.S.
- Define an early_initcall() to copy the contents of the
"sym_code_functions" section to an array by the same name.
- Define a function unwinder_blacklisted() that compares a return
PC against sym_code_sections[]. If there is a match, mark the
stack trace unreliable. Call this from unwind_frame().
v3:
- Implemented a sym_code_ranges[] array to contains sections bounds
for text sections that contain SYM_CODE_*() functions. The unwinder
checks each return PC against the sections. If it falls in any of
the sections, the stack trace is marked unreliable.
- Moved SYM_CODE functions from .text and .init.text into a new
text section called ".code.text". Added this section to
vmlinux.lds.S and sym_code_ranges[].
- Fixed the logic in the unwinder that handles Function Graph
Tracer return trampoline.
- Removed all the previous code that handles:
- ftrace entry code for traced function
- special_functions[] array that lists individual functions
- kretprobe_trampoline() special case
v2
- Removed the terminating entry { 0, 0 } in special_functions[]
and replaced it with the idiom { /* sentinel */ }.
- Change the ftrace trampoline entry ftrace_graph_call in
special_functions[] to ftrace_call + 4 and added explanatory
comments.
- Unnested #ifdefs in special_functions[] for FTRACE.
v1
- Define a bool field in struct stackframe. This will indicate if
a stack trace is reliable.
- Implement a special_functions[] array that will be populated
with special functions in which the stack trace is considered
unreliable.
- Using kallsyms_lookup(), get the address ranges for the special
functions and record them.
- Implement an is_reliable_function(pc). This function will check
if a given return PC falls in any of the special functions. If
it does, the stack trace is unreliable.
- Implement check_reliability() function that will check if a
stack frame is reliable. Call is_reliable_function() from
check_reliability().
- Before a return PC is checked against special_funtions[], it
must be validates as a proper kernel text address. Call
__kernel_text_address() from check_reliability().
- Finally, call check_reliability() from unwind_frame() for
each stack frame.
- Add EL1 exception handlers to special_functions[].
el1_sync();
el1_irq();
el1_error();
el1_sync_invalid();
el1_irq_invalid();
el1_fiq_invalid();
el1_error_invalid();
- The above functions are currently defined as LOCAL symbols.
Make them global so that they can be referenced from the
unwinder code.
- Add FTRACE trampolines to special_functions[]:
ftrace_graph_call()
ftrace_graph_caller()
return_to_handler()
- Add the kretprobe trampoline to special functions[]:
kretprobe_trampoline()
Previous versions and discussion
================================
v14: https://lore.kernel.org/linux-arm-kernel/[email protected]/T/#t
v13: https://lore.kernel.org/linux-arm-kernel/[email protected]/T/#t
v12: https://lore.kernel.org/linux-arm-kernel/[email protected]/T/#m21e86eecb9b8f0831196568f0bf62c3b56f65bf0
v11: https://lore.kernel.org/linux-arm-kernel/[email protected]/T/#t
v10: https://lore.kernel.org/linux-arm-kernel/[email protected]/T/#t
v9: Mailer screwed up the threading. Sent the same as v10 with proper threading.
v8: https://lore.kernel.org/linux-arm-kernel/[email protected]/
v7: Mailer screwed up the threading. Sent the same as v8 with proper threading.
v6: https://lore.kernel.org/linux-arm-kernel/[email protected]/
v5: https://lore.kernel.org/linux-arm-kernel/[email protected]/
v4: https://lore.kernel.org/linux-arm-kernel/[email protected]/
v3: https://lore.kernel.org/linux-arm-kernel/[email protected]/
v2: https://lore.kernel.org/linux-arm-kernel/[email protected]/
v1: https://lore.kernel.org/linux-arm-kernel/[email protected]/
Madhavan T. Venkataraman (6):
arm64: Split unwind_init()
arm64: Copy the task argument to unwind_state
arm64: Make the unwind loop in unwind() similar to other architectures
arm64: Introduce stack trace reliability checks in the unwinder
arm64: Create a list of SYM_CODE functions, check return PC against
list
arm64: Introduce arch_stack_walk_reliable()
arch/arm64/include/asm/linkage.h | 11 ++
arch/arm64/include/asm/sections.h | 1 +
arch/arm64/kernel/stacktrace.c | 266 +++++++++++++++++++++++++-----
arch/arm64/kernel/vmlinux.lds.S | 10 ++
4 files changed, 246 insertions(+), 42 deletions(-)
base-commit: b13baccc3850ca8b8cccbf8ed9912dbaa0fdf7f3
--
2.25.1
From: "Madhavan T. Venkataraman" <[email protected]>
Introduce arch_stack_walk_reliable() for ARM64. This works like
arch_stack_walk() except that it returns -EINVAL if the stack trace is not
reliable.
Until all the reliability checks are in place, arch_stack_walk_reliable()
may not be used by livepatch. But it may be used by debug and test code.
Signed-off-by: Madhavan T. Venkataraman <[email protected]>
Reviewed-by: Mark Brown <[email protected]>
---
arch/arm64/kernel/stacktrace.c | 23 +++++++++++++++++++++++
1 file changed, 23 insertions(+)
diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
index eda8581f7dbe..8016ba0e2c96 100644
--- a/arch/arm64/kernel/stacktrace.c
+++ b/arch/arm64/kernel/stacktrace.c
@@ -383,3 +383,26 @@ noinline notrace void arch_stack_walk(stack_trace_consume_fn consume_entry,
unwind(&state, consume_entry, cookie);
}
+
+/*
+ * arch_stack_walk_reliable() may not be used for livepatch until all of
+ * the reliability checks are in place in unwind_consume(). However,
+ * debug and test code can choose to use it even if all the checks are not
+ * in place.
+ */
+noinline int notrace arch_stack_walk_reliable(
+ stack_trace_consume_fn consume_entry,
+ void *cookie,
+ struct task_struct *task)
+{
+ struct unwind_state state;
+ bool reliable;
+
+ if (task == current)
+ unwind_init_from_caller(&state);
+ else
+ unwind_init_from_task(&state, task);
+
+ reliable = unwind(&state, consume_entry, cookie);
+ return reliable ? 0 : -EINVAL;
+}
--
2.25.1
From: "Madhavan T. Venkataraman" <[email protected]>
unwind_init() is currently a single function that initializes all of the
unwind state. Split it into the following functions and call them
appropriately:
- unwind_init_from_regs() - initialize from regs passed by caller.
- unwind_init_from_caller() - initialize for the current task
from the caller of arch_stack_walk().
- unwind_init_from_task() - initialize from the saved state of a
task other than the current task. In this case, the other
task must not be running.
This is done for two reasons:
- the different ways of initializing are clear
- specialized code can be added to each initializer in the future.
Signed-off-by: Madhavan T. Venkataraman <[email protected]>
Reviewed-by: Mark Brown <[email protected]>
---
arch/arm64/kernel/stacktrace.c | 66 ++++++++++++++++++++++++++++------
1 file changed, 55 insertions(+), 11 deletions(-)
diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
index 0467cb79f080..e44f93ff25f0 100644
--- a/arch/arm64/kernel/stacktrace.c
+++ b/arch/arm64/kernel/stacktrace.c
@@ -50,11 +50,8 @@ struct unwind_state {
#endif
};
-static notrace void unwind_init(struct unwind_state *state, unsigned long fp,
- unsigned long pc)
+static void unwind_init_common(struct unwind_state *state)
{
- state->fp = fp;
- state->pc = pc;
#ifdef CONFIG_KRETPROBES
state->kr_cur = NULL;
#endif
@@ -72,7 +69,57 @@ static notrace void unwind_init(struct unwind_state *state, unsigned long fp,
state->prev_fp = 0;
state->prev_type = STACK_TYPE_UNKNOWN;
}
-NOKPROBE_SYMBOL(unwind_init);
+
+/*
+ * Start an unwind from a pt_regs.
+ *
+ * The unwind will begin at the PC within the regs.
+ *
+ * The regs must be on a stack currently owned by the calling task.
+ */
+static inline void unwind_init_from_regs(struct unwind_state *state,
+ struct pt_regs *regs)
+{
+ unwind_init_common(state);
+
+ state->fp = regs->regs[29];
+ state->pc = regs->pc;
+}
+
+/*
+ * Start an unwind from a caller.
+ *
+ * The unwind will begin at the caller of whichever function this is inlined
+ * into.
+ *
+ * The function which invokes this must be noinline.
+ */
+static __always_inline void unwind_init_from_caller(struct unwind_state *state)
+{
+ unwind_init_common(state);
+
+ state->fp = (unsigned long)__builtin_frame_address(1);
+ state->pc = (unsigned long)__builtin_return_address(0);
+}
+
+/*
+ * Start an unwind from a blocked task.
+ *
+ * The unwind will begin at the blocked tasks saved PC (i.e. the caller of
+ * cpu_switch_to()).
+ *
+ * The caller should ensure the task is blocked in cpu_switch_to() for the
+ * duration of the unwind, or the unwind will be bogus. It is never valid to
+ * call this for the current task.
+ */
+static inline void unwind_init_from_task(struct unwind_state *state,
+ struct task_struct *task)
+{
+ unwind_init_common(state);
+
+ state->fp = thread_saved_fp(task);
+ state->pc = thread_saved_pc(task);
+}
/*
* Unwind from one frame record (A) to the next frame record (B).
@@ -213,14 +260,11 @@ noinline notrace void arch_stack_walk(stack_trace_consume_fn consume_entry,
struct unwind_state state;
if (regs)
- unwind_init(&state, regs->regs[29], regs->pc);
+ unwind_init_from_regs(&state, regs);
else if (task == current)
- unwind_init(&state,
- (unsigned long)__builtin_frame_address(1),
- (unsigned long)__builtin_return_address(0));
+ unwind_init_from_caller(&state);
else
- unwind_init(&state, thread_saved_fp(task),
- thread_saved_pc(task));
+ unwind_init_from_task(&state, task);
unwind(task, &state, consume_entry, cookie);
}
--
2.25.1
From: "Madhavan T. Venkataraman" <[email protected]>
Change the loop in unwind()
===========================
Change the unwind loop in unwind() to:
while (unwind_continue(state, consume_entry, cookie))
unwind_next(state);
This is easy to understand and maintain.
New function unwind_continue()
==============================
Define a new function unwind_continue() that is used in the unwind loop
to check for conditions that terminate a stack trace.
The conditions checked are:
- If the bottom of the stack (final frame) has been reached,
terminate.
- If the consume_entry() function returns false, the caller of
unwind has asked to terminate the stack trace. So, terminate.
- If unwind_next() failed for some reason (like stack corruption),
terminate.
Do not return an error value from unwind_next()
===============================================
We want to check for terminating conditions only in unwind_continue() from
the unwinder loop. So, do not return an error value from unwind_next().
Simply set a flag in unwind_state and check the flag in unwind_continue().
Final FP
========
Introduce a new field "final_fp" in "struct unwind_state". Initialize this
to the final frame of the stack trace:
task_pt_regs(task)->stackframe
This is where the stacktrace must terminate if it is successful. Add an
explicit comment to that effect.
Signed-off-by: Madhavan T. Venkataraman <[email protected]>
Reviewed-by: Mark Brown <[email protected]>
---
arch/arm64/kernel/stacktrace.c | 78 ++++++++++++++++++++++------------
1 file changed, 52 insertions(+), 26 deletions(-)
diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
index 8e43444d50e2..c749129aba5a 100644
--- a/arch/arm64/kernel/stacktrace.c
+++ b/arch/arm64/kernel/stacktrace.c
@@ -40,6 +40,10 @@
* value.
*
* @task: The task being unwound.
+ *
+ * @final_fp: Pointer to the final frame.
+ *
+ * @failed: Unwind failed.
*/
struct unwind_state {
unsigned long fp;
@@ -51,6 +55,8 @@ struct unwind_state {
struct llist_node *kr_cur;
#endif
struct task_struct *task;
+ unsigned long final_fp;
+ bool failed;
};
static void unwind_init_common(struct unwind_state *state,
@@ -73,6 +79,10 @@ static void unwind_init_common(struct unwind_state *state,
bitmap_zero(state->stacks_done, __NR_STACK_TYPES);
state->prev_fp = 0;
state->prev_type = STACK_TYPE_UNKNOWN;
+ state->failed = false;
+
+ /* Stack trace terminates here. */
+ state->final_fp = (unsigned long)task_pt_regs(task)->stackframe;
}
/*
@@ -126,6 +136,25 @@ static inline void unwind_init_from_task(struct unwind_state *state,
state->pc = thread_saved_pc(task);
}
+static bool notrace unwind_continue(struct unwind_state *state,
+ stack_trace_consume_fn consume_entry,
+ void *cookie)
+{
+ if (state->failed) {
+ /* PC is suspect. Cannot consume it. */
+ return false;
+ }
+
+ if (!consume_entry(cookie, state->pc)) {
+ /* Caller terminated the unwind. */
+ state->failed = true;
+ return false;
+ }
+
+ return state->fp != state->final_fp;
+}
+NOKPROBE_SYMBOL(unwind_continue);
+
/*
* Unwind from one frame record (A) to the next frame record (B).
*
@@ -133,24 +162,26 @@ static inline void unwind_init_from_task(struct unwind_state *state,
* records (e.g. a cycle), determined based on the location and fp value of A
* and the location (but not the fp value) of B.
*/
-static int notrace unwind_next(struct unwind_state *state)
+static void notrace unwind_next(struct unwind_state *state)
{
struct task_struct *tsk = state->task;
unsigned long fp = state->fp;
struct stack_info info;
- /* Final frame; nothing to unwind */
- if (fp == (unsigned long)task_pt_regs(tsk)->stackframe)
- return -ENOENT;
-
- if (fp & 0x7)
- return -EINVAL;
+ if (fp & 0x7) {
+ state->failed = true;
+ return;
+ }
- if (!on_accessible_stack(tsk, fp, 16, &info))
- return -EINVAL;
+ if (!on_accessible_stack(tsk, fp, 16, &info)) {
+ state->failed = true;
+ return;
+ }
- if (test_bit(info.type, state->stacks_done))
- return -EINVAL;
+ if (test_bit(info.type, state->stacks_done)) {
+ state->failed = true;
+ return;
+ }
/*
* As stacks grow downward, any valid record on the same stack must be
@@ -166,8 +197,10 @@ static int notrace unwind_next(struct unwind_state *state)
* stack.
*/
if (info.type == state->prev_type) {
- if (fp <= state->prev_fp)
- return -EINVAL;
+ if (fp <= state->prev_fp) {
+ state->failed = true;
+ return;
+ }
} else {
set_bit(state->prev_type, state->stacks_done);
}
@@ -195,8 +228,10 @@ static int notrace unwind_next(struct unwind_state *state)
*/
orig_pc = ftrace_graph_ret_addr(tsk, NULL, state->pc,
(void *)state->fp);
- if (WARN_ON_ONCE(state->pc == orig_pc))
- return -EINVAL;
+ if (WARN_ON_ONCE(state->pc == orig_pc)) {
+ state->failed = true;
+ return;
+ }
state->pc = orig_pc;
}
#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
@@ -204,23 +239,14 @@ static int notrace unwind_next(struct unwind_state *state)
if (is_kretprobe_trampoline(state->pc))
state->pc = kretprobe_find_ret_addr(tsk, (void *)state->fp, &state->kr_cur);
#endif
-
- return 0;
}
NOKPROBE_SYMBOL(unwind_next);
static void notrace unwind(struct unwind_state *state,
stack_trace_consume_fn consume_entry, void *cookie)
{
- while (1) {
- int ret;
-
- if (!consume_entry(cookie, state->pc))
- break;
- ret = unwind_next(state);
- if (ret < 0)
- break;
- }
+ while (unwind_continue(state, consume_entry, cookie))
+ unwind_next(state);
}
NOKPROBE_SYMBOL(unwind);
--
2.25.1
On Fri, Jun 17, 2022 at 04:07:11PM -0500, [email protected] wrote:
> From: "Madhavan T. Venkataraman" <[email protected]>
>
> I have synced this patch series to v5.19-rc2.
> I have also removed the following patch.
>
> [PATCH v14 7/7] arm64: Select HAVE_RELIABLE_STACKTRACE
>
> as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
> yet. This patch will be added in the future once Objtool is enhanced to
> provide stack validation in some form.
Given that it's not at all obvious that we're going to end up using objtool
for arm64, does this patch series gain us anything in isolation?
Will
On 6/23/22 12:32, Will Deacon wrote:
> On Fri, Jun 17, 2022 at 04:07:11PM -0500, [email protected] wrote:
>> From: "Madhavan T. Venkataraman" <[email protected]>
>>
>> I have synced this patch series to v5.19-rc2.
>> I have also removed the following patch.
>>
>> [PATCH v14 7/7] arm64: Select HAVE_RELIABLE_STACKTRACE
>>
>> as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
>> yet. This patch will be added in the future once Objtool is enhanced to
>> provide stack validation in some form.
>
> Given that it's not at all obvious that we're going to end up using objtool
> for arm64, does this patch series gain us anything in isolation?
>
BTW, I have synced my patchset to 5.19-rc2 and sent it as v15.
So, to answer your question, patches 1 thru 3 in v15 are still useful even if we don't
consider reliable stacktrace. These patches reorganize the unwinder code based on
comments from both Mark Rutland and Mark Brown. Mark Brown has already OKed them.
If Mark Rutland OKes them, we should upstream them.
I can drop patches 4 thru 6. Actually, the objtool patch series that I have
sent separately for supporting livepatch already addresses reliability. So, if that
gets reviewed and accepted, we don't even need patches 4 thru 6.
If you are OK with that, I can resend v16 with just patches 1 thru 3. Let me know.
Madhavan
On 6/24/22 00:19, Madhavan T. Venkataraman wrote:
>
>
> On 6/23/22 12:32, Will Deacon wrote:
>> On Fri, Jun 17, 2022 at 04:07:11PM -0500, [email protected] wrote:
>>> From: "Madhavan T. Venkataraman" <[email protected]>
>>>
>>> I have synced this patch series to v5.19-rc2.
>>> I have also removed the following patch.
>>>
>>> [PATCH v14 7/7] arm64: Select HAVE_RELIABLE_STACKTRACE
>>>
>>> as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
>>> yet. This patch will be added in the future once Objtool is enhanced to
>>> provide stack validation in some form.
>>
>> Given that it's not at all obvious that we're going to end up using objtool
>> for arm64, does this patch series gain us anything in isolation?
>>
>
> BTW, I have synced my patchset to 5.19-rc2 and sent it as v15.
Sorry. What I wanted to say was that in v15 I have removed the patch titled:
arm64: Select HAVE_RELIABLE_STACKTRACE
since objtool changes are not in place.
Apologies.
Madhavan
>
> So, to answer your question, patches 1 thru 3 in v15 are still useful even if we don't
> consider reliable stacktrace. These patches reorganize the unwinder code based on
> comments from both Mark Rutland and Mark Brown. Mark Brown has already OKed them.
> If Mark Rutland OKes them, we should upstream them.
>
> I can drop patches 4 thru 6. Actually, the objtool patch series that I have
> sent separately for supporting livepatch already addresses reliability. So, if that
> gets reviewed and accepted, we don't even need patches 4 thru 6.
>
> If you are OK with that, I can resend v16 with just patches 1 thru 3. Let me know.
>
> Madhavan
>
On Thu, Jun 23, 2022 at 06:32:24PM +0100, Will Deacon wrote:
> On Fri, Jun 17, 2022 at 04:07:11PM -0500, [email protected] wrote:
> > as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
> > yet. This patch will be added in the future once Objtool is enhanced to
> > provide stack validation in some form.
> Given that it's not at all obvious that we're going to end up using objtool
> for arm64, does this patch series gain us anything in isolation?
Having the reliability information seems like it should be useful in
general even without doing live patching - we can use it to annotate
stack traces to warn people about anything that might be suspect in
there. For live patching it's probably something we'll want regardless
of the use of objtool, it's one more robustness check which always
helps.
On 6/24/22 06:42, Mark Brown wrote:
> On Thu, Jun 23, 2022 at 06:32:24PM +0100, Will Deacon wrote:
>> On Fri, Jun 17, 2022 at 04:07:11PM -0500, [email protected] wrote:
>
>>> as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
>>> yet. This patch will be added in the future once Objtool is enhanced to
>>> provide stack validation in some form.
>
>> Given that it's not at all obvious that we're going to end up using objtool
>> for arm64, does this patch series gain us anything in isolation?
>
> Having the reliability information seems like it should be useful in
> general even without doing live patching - we can use it to annotate
> stack traces to warn people about anything that might be suspect in
> there. For live patching it's probably something we'll want regardless
> of the use of objtool, it's one more robustness check which always
> helps.
Hi Mark, Will,
Your comments got me to thinking about the Objtool patch series I have sent earlier.
Since the general feeling is that Objtool is unlikely to be our path to livepatch on ARM64, I think that I can implement what I want in a simpler way as a kernel-only solution. The kernel already has a decoder. I don't need
to provide one. In the kernel-only solution, I don't have to worry about relocations, alternatives, etc, etc.
The number of patches would be about half of the original series with simpler code in many of the patches.
The amount of memory consumed by the CFI entries will most likely be just a fraction of the original series.
I will investigate this. If it works and turns out to be a lot simpler, I will send this as v3 of the livepatch
patch series. Also, if this works, we can replace the various reliability checks with just a single fp validation
check in the unwinder.
Thanks for the input.
Madhavan
On Fri, Jun 17, 2022 at 04:07:12PM -0500, [email protected] wrote:
> From: "Madhavan T. Venkataraman" <[email protected]>
>
> unwind_init() is currently a single function that initializes all of the
> unwind state. Split it into the following functions and call them
> appropriately:
>
> - unwind_init_from_regs() - initialize from regs passed by caller.
>
> - unwind_init_from_caller() - initialize for the current task
> from the caller of arch_stack_walk().
>
> - unwind_init_from_task() - initialize from the saved state of a
> task other than the current task. In this case, the other
> task must not be running.
>
> This is done for two reasons:
>
> - the different ways of initializing are clear
>
> - specialized code can be added to each initializer in the future.
>
> Signed-off-by: Madhavan T. Venkataraman <[email protected]>
> Reviewed-by: Mark Brown <[email protected]>
Acked-by: Mark Rutland <[email protected]>
Mark.
> ---
> arch/arm64/kernel/stacktrace.c | 66 ++++++++++++++++++++++++++++------
> 1 file changed, 55 insertions(+), 11 deletions(-)
>
> diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
> index 0467cb79f080..e44f93ff25f0 100644
> --- a/arch/arm64/kernel/stacktrace.c
> +++ b/arch/arm64/kernel/stacktrace.c
> @@ -50,11 +50,8 @@ struct unwind_state {
> #endif
> };
>
> -static notrace void unwind_init(struct unwind_state *state, unsigned long fp,
> - unsigned long pc)
> +static void unwind_init_common(struct unwind_state *state)
> {
> - state->fp = fp;
> - state->pc = pc;
> #ifdef CONFIG_KRETPROBES
> state->kr_cur = NULL;
> #endif
> @@ -72,7 +69,57 @@ static notrace void unwind_init(struct unwind_state *state, unsigned long fp,
> state->prev_fp = 0;
> state->prev_type = STACK_TYPE_UNKNOWN;
> }
> -NOKPROBE_SYMBOL(unwind_init);
> +
> +/*
> + * Start an unwind from a pt_regs.
> + *
> + * The unwind will begin at the PC within the regs.
> + *
> + * The regs must be on a stack currently owned by the calling task.
> + */
> +static inline void unwind_init_from_regs(struct unwind_state *state,
> + struct pt_regs *regs)
> +{
> + unwind_init_common(state);
> +
> + state->fp = regs->regs[29];
> + state->pc = regs->pc;
> +}
> +
> +/*
> + * Start an unwind from a caller.
> + *
> + * The unwind will begin at the caller of whichever function this is inlined
> + * into.
> + *
> + * The function which invokes this must be noinline.
> + */
> +static __always_inline void unwind_init_from_caller(struct unwind_state *state)
> +{
> + unwind_init_common(state);
> +
> + state->fp = (unsigned long)__builtin_frame_address(1);
> + state->pc = (unsigned long)__builtin_return_address(0);
> +}
> +
> +/*
> + * Start an unwind from a blocked task.
> + *
> + * The unwind will begin at the blocked tasks saved PC (i.e. the caller of
> + * cpu_switch_to()).
> + *
> + * The caller should ensure the task is blocked in cpu_switch_to() for the
> + * duration of the unwind, or the unwind will be bogus. It is never valid to
> + * call this for the current task.
> + */
> +static inline void unwind_init_from_task(struct unwind_state *state,
> + struct task_struct *task)
> +{
> + unwind_init_common(state);
> +
> + state->fp = thread_saved_fp(task);
> + state->pc = thread_saved_pc(task);
> +}
>
> /*
> * Unwind from one frame record (A) to the next frame record (B).
> @@ -213,14 +260,11 @@ noinline notrace void arch_stack_walk(stack_trace_consume_fn consume_entry,
> struct unwind_state state;
>
> if (regs)
> - unwind_init(&state, regs->regs[29], regs->pc);
> + unwind_init_from_regs(&state, regs);
> else if (task == current)
> - unwind_init(&state,
> - (unsigned long)__builtin_frame_address(1),
> - (unsigned long)__builtin_return_address(0));
> + unwind_init_from_caller(&state);
> else
> - unwind_init(&state, thread_saved_fp(task),
> - thread_saved_pc(task));
> + unwind_init_from_task(&state, task);
>
> unwind(task, &state, consume_entry, cookie);
> }
> --
> 2.25.1
>
On Fri, Jun 17, 2022 at 04:07:14PM -0500, [email protected] wrote:
> From: "Madhavan T. Venkataraman" <[email protected]>
>
> Change the loop in unwind()
> ===========================
>
> Change the unwind loop in unwind() to:
>
> while (unwind_continue(state, consume_entry, cookie))
> unwind_next(state);
>
> This is easy to understand and maintain.
> New function unwind_continue()
> ==============================
>
> Define a new function unwind_continue() that is used in the unwind loop
> to check for conditions that terminate a stack trace.
>
> The conditions checked are:
>
> - If the bottom of the stack (final frame) has been reached,
> terminate.
>
> - If the consume_entry() function returns false, the caller of
> unwind has asked to terminate the stack trace. So, terminate.
>
> - If unwind_next() failed for some reason (like stack corruption),
> terminate.
I'm a bit confused as to why this structure, since AFAICT this doesn't match
other architectures (looking at x86, powerpc, and s390). I note that x86 has:
* In arch_stack_walk():
for (unwind_start(&state, task, regs, NULL); !unwind_done(&state);
unwind_next_frame(&state)) {
...
if (!consume_entry(...))
break;
...
}
* In arch_stack_walk_reliable():
for (unwind_start(&state, task, NULL, NULL);
!unwind_done(&state) && !unwind_error(&state);
unwind_next_frame(&state)) {
...
if (!consume_entry(...)
return -EINVAL;
}
... and back in v6 I suggeted exactly that shape:
https://lore.kernel.org/linux-arm-kernel/[email protected]/
>
> Do not return an error value from unwind_next()
> ===============================================
>
> We want to check for terminating conditions only in unwind_continue() from
> the unwinder loop. So, do not return an error value from unwind_next().
> Simply set a flag in unwind_state and check the flag in unwind_continue().
I'm fine with the concept of moving ghe return value out of unwind_next() (e.g.
if we go with an x86-like structure), but I don't think that we should
centralize the other checks *and* the consumption within unwind_continue(), as
I think those are two separate things.
>
> Final FP
> ========
>
> Introduce a new field "final_fp" in "struct unwind_state". Initialize this
> to the final frame of the stack trace:
>
> task_pt_regs(task)->stackframe
>
> This is where the stacktrace must terminate if it is successful. Add an
> explicit comment to that effect.
Can we please make this change as a preparatory step, as with the 'task' field?
We can wrap this in a helper like:
static bool is_final_frame(struct unwind state *state)
{
return state->fp == state->final_fp;
}
... and use that in the main loop.
Thanks,
Mark.
>
> Signed-off-by: Madhavan T. Venkataraman <[email protected]>
> Reviewed-by: Mark Brown <[email protected]>
> ---
> arch/arm64/kernel/stacktrace.c | 78 ++++++++++++++++++++++------------
> 1 file changed, 52 insertions(+), 26 deletions(-)
>
> diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
> index 8e43444d50e2..c749129aba5a 100644
> --- a/arch/arm64/kernel/stacktrace.c
> +++ b/arch/arm64/kernel/stacktrace.c
> @@ -40,6 +40,10 @@
> * value.
> *
> * @task: The task being unwound.
> + *
> + * @final_fp: Pointer to the final frame.
> + *
> + * @failed: Unwind failed.
> */
> struct unwind_state {
> unsigned long fp;
> @@ -51,6 +55,8 @@ struct unwind_state {
> struct llist_node *kr_cur;
> #endif
> struct task_struct *task;
> + unsigned long final_fp;
> + bool failed;
> };
>
> static void unwind_init_common(struct unwind_state *state,
> @@ -73,6 +79,10 @@ static void unwind_init_common(struct unwind_state *state,
> bitmap_zero(state->stacks_done, __NR_STACK_TYPES);
> state->prev_fp = 0;
> state->prev_type = STACK_TYPE_UNKNOWN;
> + state->failed = false;
> +
> + /* Stack trace terminates here. */
> + state->final_fp = (unsigned long)task_pt_regs(task)->stackframe;
> }
>
> /*
> @@ -126,6 +136,25 @@ static inline void unwind_init_from_task(struct unwind_state *state,
> state->pc = thread_saved_pc(task);
> }
>
> +static bool notrace unwind_continue(struct unwind_state *state,
> + stack_trace_consume_fn consume_entry,
> + void *cookie)
> +{
> + if (state->failed) {
> + /* PC is suspect. Cannot consume it. */
> + return false;
> + }
> +
> + if (!consume_entry(cookie, state->pc)) {
> + /* Caller terminated the unwind. */
> + state->failed = true;
> + return false;
> + }
> +
> + return state->fp != state->final_fp;
> +}
> +NOKPROBE_SYMBOL(unwind_continue);
> +
> /*
> * Unwind from one frame record (A) to the next frame record (B).
> *
> @@ -133,24 +162,26 @@ static inline void unwind_init_from_task(struct unwind_state *state,
> * records (e.g. a cycle), determined based on the location and fp value of A
> * and the location (but not the fp value) of B.
> */
> -static int notrace unwind_next(struct unwind_state *state)
> +static void notrace unwind_next(struct unwind_state *state)
> {
> struct task_struct *tsk = state->task;
> unsigned long fp = state->fp;
> struct stack_info info;
>
> - /* Final frame; nothing to unwind */
> - if (fp == (unsigned long)task_pt_regs(tsk)->stackframe)
> - return -ENOENT;
> -
> - if (fp & 0x7)
> - return -EINVAL;
> + if (fp & 0x7) {
> + state->failed = true;
> + return;
> + }
>
> - if (!on_accessible_stack(tsk, fp, 16, &info))
> - return -EINVAL;
> + if (!on_accessible_stack(tsk, fp, 16, &info)) {
> + state->failed = true;
> + return;
> + }
>
> - if (test_bit(info.type, state->stacks_done))
> - return -EINVAL;
> + if (test_bit(info.type, state->stacks_done)) {
> + state->failed = true;
> + return;
> + }
>
> /*
> * As stacks grow downward, any valid record on the same stack must be
> @@ -166,8 +197,10 @@ static int notrace unwind_next(struct unwind_state *state)
> * stack.
> */
> if (info.type == state->prev_type) {
> - if (fp <= state->prev_fp)
> - return -EINVAL;
> + if (fp <= state->prev_fp) {
> + state->failed = true;
> + return;
> + }
> } else {
> set_bit(state->prev_type, state->stacks_done);
> }
> @@ -195,8 +228,10 @@ static int notrace unwind_next(struct unwind_state *state)
> */
> orig_pc = ftrace_graph_ret_addr(tsk, NULL, state->pc,
> (void *)state->fp);
> - if (WARN_ON_ONCE(state->pc == orig_pc))
> - return -EINVAL;
> + if (WARN_ON_ONCE(state->pc == orig_pc)) {
> + state->failed = true;
> + return;
> + }
> state->pc = orig_pc;
> }
> #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
> @@ -204,23 +239,14 @@ static int notrace unwind_next(struct unwind_state *state)
> if (is_kretprobe_trampoline(state->pc))
> state->pc = kretprobe_find_ret_addr(tsk, (void *)state->fp, &state->kr_cur);
> #endif
> -
> - return 0;
> }
> NOKPROBE_SYMBOL(unwind_next);
>
> static void notrace unwind(struct unwind_state *state,
> stack_trace_consume_fn consume_entry, void *cookie)
> {
> - while (1) {
> - int ret;
> -
> - if (!consume_entry(cookie, state->pc))
> - break;
> - ret = unwind_next(state);
> - if (ret < 0)
> - break;
> - }
> + while (unwind_continue(state, consume_entry, cookie))
> + unwind_next(state);
> }
> NOKPROBE_SYMBOL(unwind);
>
> --
> 2.25.1
>
On Fri, Jun 17, 2022 at 04:07:17PM -0500, [email protected] wrote:
> From: "Madhavan T. Venkataraman" <[email protected]>
>
> Introduce arch_stack_walk_reliable() for ARM64. This works like
> arch_stack_walk() except that it returns -EINVAL if the stack trace is not
> reliable.
>
> Until all the reliability checks are in place, arch_stack_walk_reliable()
> may not be used by livepatch. But it may be used by debug and test code.
For the moment I would strongly perfer *not* to add this until we have the
missing bits and pieces sorted out.
Until then, I'd like to ensure that any infrastructure we add is immediately
useful and tested. One way to do that would be to enhance the stack dumping
code (i.e. dump_backtrace()) to log some metadata.
As an end-goal, I'd like to get to a point where we can do:
* Explicit logging when trace terminate at the final frame, e.g.
stacktrace:
function_c+offset/total
function_b+offset/total
function_a+offset/total
<unwind successful>
* Explicit logging of early termination, e.g.
stacktrace:
function_c+offset/total
<unwind terminated early (bad FP)>
* Unreliability on individual elements, e.g.
stacktrace:
function_c+offset/total
function_b+offset/total (?)
function_a+offset/total
* Annotations for special unwinding, e.g.
stacktrace:
function_c+offset/total (K) // kretprobes trampoline
function_b+offset/total (F) // ftrace trampoline
function_a+offset/total (FK) // ftrace and kretprobes
other_function+offset/total (P) // from pt_regs::pc
another_function+offset/total (L?) // from pt_regs::lr, unreliable
something_else+offset/total
Note: the comments here are just to explain the idea, I don't expect those in
the actual output.
That'll justify some of the infrastructure we need for reliable unwinding, and
ensure that it is tested, well before we actually enable reliable stacktracing.
Thanks,
Mark.
>
> Signed-off-by: Madhavan T. Venkataraman <[email protected]>
> Reviewed-by: Mark Brown <[email protected]>
> ---
> arch/arm64/kernel/stacktrace.c | 23 +++++++++++++++++++++++
> 1 file changed, 23 insertions(+)
>
> diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
> index eda8581f7dbe..8016ba0e2c96 100644
> --- a/arch/arm64/kernel/stacktrace.c
> +++ b/arch/arm64/kernel/stacktrace.c
> @@ -383,3 +383,26 @@ noinline notrace void arch_stack_walk(stack_trace_consume_fn consume_entry,
>
> unwind(&state, consume_entry, cookie);
> }
> +
> +/*
> + * arch_stack_walk_reliable() may not be used for livepatch until all of
> + * the reliability checks are in place in unwind_consume(). However,
> + * debug and test code can choose to use it even if all the checks are not
> + * in place.
> + */
> +noinline int notrace arch_stack_walk_reliable(
> + stack_trace_consume_fn consume_entry,
> + void *cookie,
> + struct task_struct *task)
> +{
> + struct unwind_state state;
> + bool reliable;
> +
> + if (task == current)
> + unwind_init_from_caller(&state);
> + else
> + unwind_init_from_task(&state, task);
> +
> + reliable = unwind(&state, consume_entry, cookie);
> + return reliable ? 0 : -EINVAL;
> +}
> --
> 2.25.1
>
On Fri, Jun 24, 2022 at 12:19:01AM -0500, Madhavan T. Venkataraman wrote:
>
>
> On 6/23/22 12:32, Will Deacon wrote:
> > On Fri, Jun 17, 2022 at 04:07:11PM -0500, [email protected] wrote:
> >> From: "Madhavan T. Venkataraman" <[email protected]>
> >>
> >> I have synced this patch series to v5.19-rc2.
> >> I have also removed the following patch.
> >>
> >> [PATCH v14 7/7] arm64: Select HAVE_RELIABLE_STACKTRACE
> >>
> >> as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
> >> yet. This patch will be added in the future once Objtool is enhanced to
> >> provide stack validation in some form.
> >
> > Given that it's not at all obvious that we're going to end up using objtool
> > for arm64, does this patch series gain us anything in isolation?
> >
>
> BTW, I have synced my patchset to 5.19-rc2 and sent it as v15.
>
> So, to answer your question, patches 1 thru 3 in v15 are still useful even if we don't
> consider reliable stacktrace. These patches reorganize the unwinder code based on
> comments from both Mark Rutland and Mark Brown. Mark Brown has already OKed them.
> If Mark Rutland OKes them, we should upstream them.
Sorry for the delay; I have been rather swamped recently and haven't had the
time to give this the time it needs.
I'm happy with patches 1 and 2, and I've acked those in case Will wants to pick
them.
Kalesh (cc'd) is working to share the unwinder code with hyp, and I think that
we need to take a step back and consider how we can make the design work
cleanly with that. I'd had a go at prototyping making the unwinder more data
driven, but I haven't come up with something satisfactory so far.
It would be good if you could look at / comment on each others series.
Thanks,
Mark.
>
> I can drop patches 4 thru 6. Actually, the objtool patch series that I have
> sent separately for supporting livepatch already addresses reliability. So, if that
> gets reviewed and accepted, we don't even need patches 4 thru 6.
>
> If you are OK with that, I can resend v16 with just patches 1 thru 3. Let me know.
>
> Madhavan
>
On 6/26/22 04:18, Mark Rutland wrote:
> On Fri, Jun 24, 2022 at 12:19:01AM -0500, Madhavan T. Venkataraman wrote:
>>
>>
>> On 6/23/22 12:32, Will Deacon wrote:
>>> On Fri, Jun 17, 2022 at 04:07:11PM -0500, [email protected] wrote:
>>>> From: "Madhavan T. Venkataraman" <[email protected]>
>>>>
>>>> I have synced this patch series to v5.19-rc2.
>>>> I have also removed the following patch.
>>>>
>>>> [PATCH v14 7/7] arm64: Select HAVE_RELIABLE_STACKTRACE
>>>>
>>>> as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
>>>> yet. This patch will be added in the future once Objtool is enhanced to
>>>> provide stack validation in some form.
>>>
>>> Given that it's not at all obvious that we're going to end up using objtool
>>> for arm64, does this patch series gain us anything in isolation?
>>>
>>
>> BTW, I have synced my patchset to 5.19-rc2 and sent it as v15.
>>
>> So, to answer your question, patches 1 thru 3 in v15 are still useful even if we don't
>> consider reliable stacktrace. These patches reorganize the unwinder code based on
>> comments from both Mark Rutland and Mark Brown. Mark Brown has already OKed them.
>> If Mark Rutland OKes them, we should upstream them.
>
> Sorry for the delay; I have been rather swamped recently and haven't had the
> time to give this the time it needs.
>
> I'm happy with patches 1 and 2, and I've acked those in case Will wants to pick
> them.
>
> Kalesh (cc'd) is working to share the unwinder code with hyp, and I think that
> we need to take a step back and consider how we can make the design work
> cleanly with that. I'd had a go at prototyping making the unwinder more data
> driven, but I haven't come up with something satisfactory so far.
>
> It would be good if you could look at / comment on each others series.
>
I will review Kalesh's unwinder changes.
Thanks.
Madhavan
On 6/26/22 04:18, Mark Rutland wrote:
> On Fri, Jun 24, 2022 at 12:19:01AM -0500, Madhavan T. Venkataraman wrote:
>>
>>
>> On 6/23/22 12:32, Will Deacon wrote:
>>> On Fri, Jun 17, 2022 at 04:07:11PM -0500, [email protected] wrote:
>>>> From: "Madhavan T. Venkataraman" <[email protected]>
>>>>
>>>> I have synced this patch series to v5.19-rc2.
>>>> I have also removed the following patch.
>>>>
>>>> [PATCH v14 7/7] arm64: Select HAVE_RELIABLE_STACKTRACE
>>>>
>>>> as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
>>>> yet. This patch will be added in the future once Objtool is enhanced to
>>>> provide stack validation in some form.
>>>
>>> Given that it's not at all obvious that we're going to end up using objtool
>>> for arm64, does this patch series gain us anything in isolation?
>>>
>>
>> BTW, I have synced my patchset to 5.19-rc2 and sent it as v15.
>>
>> So, to answer your question, patches 1 thru 3 in v15 are still useful even if we don't
>> consider reliable stacktrace. These patches reorganize the unwinder code based on
>> comments from both Mark Rutland and Mark Brown. Mark Brown has already OKed them.
>> If Mark Rutland OKes them, we should upstream them.
>
> Sorry for the delay; I have been rather swamped recently and haven't had the
> time to give this the time it needs.
>
> I'm happy with patches 1 and 2, and I've acked those in case Will wants to pick
> them.
>
Thanks for the review.
Will,
Are you fine with picking up patches 1 and 2?
For the other patches, I have responded separately.
Madhavan
On 6/26/22 03:21, Mark Rutland wrote:
> On Fri, Jun 17, 2022 at 04:07:14PM -0500, [email protected] wrote:
>> From: "Madhavan T. Venkataraman" <[email protected]>
>>
>> Change the loop in unwind()
>> ===========================
>>
>> Change the unwind loop in unwind() to:
>>
>> while (unwind_continue(state, consume_entry, cookie))
>> unwind_next(state);
>>
>> This is easy to understand and maintain.
>> New function unwind_continue()
>> ==============================
>>
>> Define a new function unwind_continue() that is used in the unwind loop
>> to check for conditions that terminate a stack trace.
>>
>> The conditions checked are:
>>
>> - If the bottom of the stack (final frame) has been reached,
>> terminate.
>>
>> - If the consume_entry() function returns false, the caller of
>> unwind has asked to terminate the stack trace. So, terminate.
>>
>> - If unwind_next() failed for some reason (like stack corruption),
>> terminate.
>
> I'm a bit confused as to why this structure, since AFAICT this doesn't match
> other architectures (looking at x86, powerpc, and s390). I note that x86 has:
>
> * In arch_stack_walk():
>
> for (unwind_start(&state, task, regs, NULL); !unwind_done(&state);
> unwind_next_frame(&state)) {
> ...
> if (!consume_entry(...))
> break;
> ...
> }
>
> * In arch_stack_walk_reliable():
>
> for (unwind_start(&state, task, NULL, NULL);
> !unwind_done(&state) && !unwind_error(&state);
> unwind_next_frame(&state)) {
> ...
> if (!consume_entry(...)
> return -EINVAL;
> }
>
> ... and back in v6 I suggeted exactly that shape:
>
> https://lore.kernel.org/linux-arm-kernel/[email protected]/
>
OK. I will take a look at your suggestion and resend this patch.
>>
>> Do not return an error value from unwind_next()
>> ===============================================
>>
>> We want to check for terminating conditions only in unwind_continue() from
>> the unwinder loop. So, do not return an error value from unwind_next().
>> Simply set a flag in unwind_state and check the flag in unwind_continue().
>
> I'm fine with the concept of moving ghe return value out of unwind_next() (e.g.
> if we go with an x86-like structure), but I don't think that we should
> centralize the other checks *and* the consumption within unwind_continue(), as
> I think those are two separate things.
>
OK. I will address this in the next version.
>>
>> Final FP
>> ========
>>
>> Introduce a new field "final_fp" in "struct unwind_state". Initialize this
>> to the final frame of the stack trace:
>>
>> task_pt_regs(task)->stackframe
>>
>> This is where the stacktrace must terminate if it is successful. Add an
>> explicit comment to that effect.
>
> Can we please make this change as a preparatory step, as with the 'task' field?
>
> We can wrap this in a helper like:
>
> static bool is_final_frame(struct unwind state *state)
> {
> return state->fp == state->final_fp;
> }
>
> ... and use that in the main loop.
>
OK. I will make these changes.
Thanks.
Madhavan
On 6/26/22 03:57, Mark Rutland wrote:
> On Fri, Jun 17, 2022 at 04:07:17PM -0500, [email protected] wrote:
>> From: "Madhavan T. Venkataraman" <[email protected]>
>>
>> Introduce arch_stack_walk_reliable() for ARM64. This works like
>> arch_stack_walk() except that it returns -EINVAL if the stack trace is not
>> reliable.
>>
>> Until all the reliability checks are in place, arch_stack_walk_reliable()
>> may not be used by livepatch. But it may be used by debug and test code.
>
> For the moment I would strongly perfer *not* to add this until we have the
> missing bits and pieces sorted out.
>
Yes. I am removing this from the patch series.
> Until then, I'd like to ensure that any infrastructure we add is immediately
> useful and tested. One way to do that would be to enhance the stack dumping
> code (i.e. dump_backtrace()) to log some metadata.
>
> As an end-goal, I'd like to get to a point where we can do:
>
> * Explicit logging when trace terminate at the final frame, e.g.
>
> stacktrace:
> function_c+offset/total
> function_b+offset/total
> function_a+offset/total
> <unwind successful>
>
> * Explicit logging of early termination, e.g.
>
> stacktrace:
> function_c+offset/total
> <unwind terminated early (bad FP)>
>
> * Unreliability on individual elements, e.g.
>
> stacktrace:
> function_c+offset/total
> function_b+offset/total (?)
> function_a+offset/total
>
> * Annotations for special unwinding, e.g.
>
> stacktrace:
> function_c+offset/total (K) // kretprobes trampoline
> function_b+offset/total (F) // ftrace trampoline
> function_a+offset/total (FK) // ftrace and kretprobes
> other_function+offset/total (P) // from pt_regs::pc
> another_function+offset/total (L?) // from pt_regs::lr, unreliable
> something_else+offset/total
>
> Note: the comments here are just to explain the idea, I don't expect those in
> the actual output.
>
> That'll justify some of the infrastructure we need for reliable unwinding, and
> ensure that it is tested, well before we actually enable reliable stacktracing.
>
In the current code structure, the annotations are a problem.
The printing of the entry along with the annotations and metadata cannot be done in
the unwind functions themselves as the caller may not even want anything printed.
The printing has to be done in consume_entry() if the caller wants to do it. But
consume_entry() only gets the PC as the argument (apart from the cookie passed by
the caller). It currently has no way of figuring out where the PC was obtained from
(ftrace, kretprobe, pt_regs, etc) or if the PC is reliable.
We need to replace the PC argument with a pointer to a structure that contains the
PC as well as other information about the PC. unwind_init() and unwind_next() need
to update that for each frame.
If this approach is acceptable, I will submit a patch series for that. Please let
me know.
Thanks.
Madhavan
On Sun, Jun 26, 2022 at 11:48:36PM -0500, Madhavan T. Venkataraman wrote:
>
>
> On 6/26/22 04:18, Mark Rutland wrote:
> > On Fri, Jun 24, 2022 at 12:19:01AM -0500, Madhavan T. Venkataraman wrote:
> >>
> >>
> >> On 6/23/22 12:32, Will Deacon wrote:
> >>> On Fri, Jun 17, 2022 at 04:07:11PM -0500, [email protected] wrote:
> >>>> From: "Madhavan T. Venkataraman" <[email protected]>
> >>>>
> >>>> I have synced this patch series to v5.19-rc2.
> >>>> I have also removed the following patch.
> >>>>
> >>>> [PATCH v14 7/7] arm64: Select HAVE_RELIABLE_STACKTRACE
> >>>>
> >>>> as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
> >>>> yet. This patch will be added in the future once Objtool is enhanced to
> >>>> provide stack validation in some form.
> >>>
> >>> Given that it's not at all obvious that we're going to end up using objtool
> >>> for arm64, does this patch series gain us anything in isolation?
> >>>
> >>
> >> BTW, I have synced my patchset to 5.19-rc2 and sent it as v15.
> >>
> >> So, to answer your question, patches 1 thru 3 in v15 are still useful even if we don't
> >> consider reliable stacktrace. These patches reorganize the unwinder code based on
> >> comments from both Mark Rutland and Mark Brown. Mark Brown has already OKed them.
> >> If Mark Rutland OKes them, we should upstream them.
> >
> > Sorry for the delay; I have been rather swamped recently and haven't had the
> > time to give this the time it needs.
> >
> > I'm happy with patches 1 and 2, and I've acked those in case Will wants to pick
> > them.
> >
>
> Thanks for the review.
>
> Will,
>
> Are you fine with picking up patches 1 and 2?
>
> For the other patches, I have responded separately.
Sure thing, I'll do that today. Thanks for persevering with this.
Will
On Sun, Jun 26, 2022 at 9:33 PM Madhavan T. Venkataraman
<[email protected]> wrote:
>
>
>
> On 6/26/22 04:18, Mark Rutland wrote:
> > On Fri, Jun 24, 2022 at 12:19:01AM -0500, Madhavan T. Venkataraman wrote:
> >>
> >>
> >> On 6/23/22 12:32, Will Deacon wrote:
> >>> On Fri, Jun 17, 2022 at 04:07:11PM -0500, [email protected] wrote:
> >>>> From: "Madhavan T. Venkataraman" <[email protected]>
> >>>>
> >>>> I have synced this patch series to v5.19-rc2.
> >>>> I have also removed the following patch.
> >>>>
> >>>> [PATCH v14 7/7] arm64: Select HAVE_RELIABLE_STACKTRACE
> >>>>
> >>>> as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
> >>>> yet. This patch will be added in the future once Objtool is enhanced to
> >>>> provide stack validation in some form.
> >>>
> >>> Given that it's not at all obvious that we're going to end up using objtool
> >>> for arm64, does this patch series gain us anything in isolation?
> >>>
> >>
> >> BTW, I have synced my patchset to 5.19-rc2 and sent it as v15.
> >>
> >> So, to answer your question, patches 1 thru 3 in v15 are still useful even if we don't
> >> consider reliable stacktrace. These patches reorganize the unwinder code based on
> >> comments from both Mark Rutland and Mark Brown. Mark Brown has already OKed them.
> >> If Mark Rutland OKes them, we should upstream them.
> >
> > Sorry for the delay; I have been rather swamped recently and haven't had the
> > time to give this the time it needs.
> >
> > I'm happy with patches 1 and 2, and I've acked those in case Will wants to pick
> > them.
> >
> > Kalesh (cc'd) is working to share the unwinder code with hyp, and I think that
> > we need to take a step back and consider how we can make the design work
> > cleanly with that. I'd had a go at prototyping making the unwinder more data
> > driven, but I haven't come up with something satisfactory so far.
> >
> > It would be good if you could look at / comment on each others series.
> >
>
> I will review Kalesh's unwinder changes.
Thanks Mark, I'll take a look.
Madhavan, I'm in the process of preparing a new version. Let me rebase
on your first 2 patches and resend, so you can look at that version
instead.
Thanks,
Kalesh
>
> Thanks.
>
> Madhavan
On 6/27/22 11:32, Kalesh Singh wrote:
> On Sun, Jun 26, 2022 at 9:33 PM Madhavan T. Venkataraman
> <[email protected]> wrote:
>>
>>
>>
>> On 6/26/22 04:18, Mark Rutland wrote:
>>> On Fri, Jun 24, 2022 at 12:19:01AM -0500, Madhavan T. Venkataraman wrote:
>>>>
>>>>
>>>> On 6/23/22 12:32, Will Deacon wrote:
>>>>> On Fri, Jun 17, 2022 at 04:07:11PM -0500, [email protected] wrote:
>>>>>> From: "Madhavan T. Venkataraman" <[email protected]>
>>>>>>
>>>>>> I have synced this patch series to v5.19-rc2.
>>>>>> I have also removed the following patch.
>>>>>>
>>>>>> [PATCH v14 7/7] arm64: Select HAVE_RELIABLE_STACKTRACE
>>>>>>
>>>>>> as HAVE_RELIABLE_STACKTRACE depends on STACK_VALIDATION which is not present
>>>>>> yet. This patch will be added in the future once Objtool is enhanced to
>>>>>> provide stack validation in some form.
>>>>>
>>>>> Given that it's not at all obvious that we're going to end up using objtool
>>>>> for arm64, does this patch series gain us anything in isolation?
>>>>>
>>>>
>>>> BTW, I have synced my patchset to 5.19-rc2 and sent it as v15.
>>>>
>>>> So, to answer your question, patches 1 thru 3 in v15 are still useful even if we don't
>>>> consider reliable stacktrace. These patches reorganize the unwinder code based on
>>>> comments from both Mark Rutland and Mark Brown. Mark Brown has already OKed them.
>>>> If Mark Rutland OKes them, we should upstream them.
>>>
>>> Sorry for the delay; I have been rather swamped recently and haven't had the
>>> time to give this the time it needs.
>>>
>>> I'm happy with patches 1 and 2, and I've acked those in case Will wants to pick
>>> them.
>>>
>>> Kalesh (cc'd) is working to share the unwinder code with hyp, and I think that
>>> we need to take a step back and consider how we can make the design work
>>> cleanly with that. I'd had a go at prototyping making the unwinder more data
>>> driven, but I haven't come up with something satisfactory so far.
>>>
>>> It would be good if you could look at / comment on each others series.
>>>
>>
>> I will review Kalesh's unwinder changes.
>
> Thanks Mark, I'll take a look.
>
> Madhavan, I'm in the process of preparing a new version. Let me rebase
> on your first 2 patches and resend, so you can look at that version
> instead.
>
Sure thing.
Thanks.
Madhavan