- use PT_SUSPEND/PT_RESUME to control running threads in
resume/stop/singleStep
- change manual stop signal from SIGTRAP to SIGSTOP to make manual stop
handling simpler
- change (*nativeProcess).trapWaitInternal to suspend newly created
threads when we are stepping a thread
- change (*nativeProcess).trapWaitInternal to handle some unhandled
stop events
- remove misleading (*nativeProcess).waitFast which does not do
anything different from the normal wait variant
- rewrite (*nativeProcess).stop to only set breakpoints for threads of
which we have received SIGTRAP
- rewrite (*nativeThread).singleStep to actually execute a single
instruction and to properly route signals
This commit improves the handling of hardcoded breakpoints in Delve.
A hardcoded breakpoint is a breakpoint instruction hardcoded in the
text of the program, for example through runtime.Breakpoint.
1. hardcoded breakpoints are now indicated by setting the breakpoint
field on any thread stopped by a hardcoded breakpoint
2. if multiple hardcoded breakpoints are hit during a single stop all
will be notified to the user.
3. a debugger breakpoint with an unmet condition can't hide a hardcoded
breakpoint anymore.
Fix signal handling during thread single stepping so that signals that
are generated by executing the current instruction are immediately
propagated to the inferior, while signals other signals sent to the
thread are delayed until the full resume happens.
Fixes a bug where a breakpoint set on an instruction that causes a
SIGSEGV would make Delve hang and a bug where signals received during
single step would make it look like an instruction is executed twice.
Fixes#2801Fixes#2792
A RequestManualStop received while the target program is stopped can
induce a crash when the target is restarted.
This is caused by the phantom breakpoint detection that was introduced
in PR #2179 / commit e69d536.
Instead of always interpreting an unexplained SIGTRAP as a phantom
breakpoint memorize all possible unreported breakpoint hits and only
act on it when the thread hasn't moved from one.
Also clarifies the behavior of the halt command when it is received
while the target is stopped or in the process of stopping.
On linux we can not read memory if the thread we use to do it is
occupied doing certain system calls. The exact conditions when this
happens have never been clear.
This problem was worked around by using the Blocked method which
recognized the most common circumstances where this would happen.
However this is a hack: Blocked returning true doesn't mean that the
problem will manifest and Blocked returning false doesn't necessarily
mean the problem will not manifest. A side effect of this is issue
#2151 where sometimes we can't read the memory of a thread and find its
associated goroutine.
This commit fixes this problem by always reading memory using a thread
we know to be good for this, specifically the one returned by
ContinueOnce. In particular the changes are as follows:
1. Remove (ProcessInternal).CurrentThread and
(ProcessInternal).SetCurrentThread, the "current thread" becomes a
field of Target, CurrentThread becomes a (*Target) method and
(*Target).SwitchThread basically just sets a field Target.
2. The backends keep track of their own internal idea of what the
current thread is, to use it to read memory, this is the thread they
return from ContinueOnce as trapthread
3. The current thread in the backend and the current thread in Target
only ever get synchronized in two places: when the backend creates a
Target object the currentThread field of Target is initialized with the
backend's current thread and when (*Target).Restart gets called (when a
recording is rewound the currentThread used by Target might not exist
anymore).
4. We remove the MemoryReadWriter interface embedded in Thread and
instead add a Memory method to Process that returns a MemoryReadWriter.
The backends will return something here that will read memory using
the current thread saved by the backend.
5. The Thread.Blocked method is removed
One possible problem with this change is processes that have threads
with different memory maps. As far as I can determine this could happen
on old versions of linux but this option was removed in linux 2.5.
Fixes#2151
Since proc is supposed to work independently from the target
architecture it shouldn't use architecture-dependent types, like
uintptr. For example when reading a 64bit core file on a 32bit
architecture, uintptr will be 32bit but the addresses proc needs to
represent will be 64bit.
Changes implementations of proc.Registers interface and the
op.DwarfRegisters struct so that floating point registers can be loaded
only when they are needed.
Removes the floatingPoint parameter from proc.Thread.Registers.
This accomplishes three things:
1. it simplifies the proc.Thread.Registers interface
2. it makes it impossible to accidentally create a broken set of saved
registers or of op.DwarfRegisters by accidentally calling
Registers(false)
3. it improves general performance of Delve by avoiding to load
floating point registers as much as possible
Floating point registers are loaded under two circumstances:
1. When the Slice method is called with floatingPoint == true
2. When the Copy method is called
Benchmark before:
BenchmarkConditionalBreakpoints-4 1 4327350142 ns/op
Benchmark after:
BenchmarkConditionalBreakpoints-4 1 3852642917 ns/op
Updates #1549
* proc/native/linux: only set breakpoints on threads that receive SIGTRAP
* proc/native/linux: do not call (*Thread).Stopped inside (*Process).stop
(*Thread).Stopped is slow because it needs to open, read and parse a
file in /proc, we don't actually need to do that, we can just rely on
the value of Thread.os.running.
Benchmark before:
BenchmarkConditionalBreakpoints-4 1 12476166303 ns/op
Benchmark after:
BenchmarkConditionalBreakpoints-4 1 10403533675 ns/op
Conditional breakpoint evaluation: 1.24ms -> 1ms
Updates #1549
This change adds `ProcessVmRead` and `ProcessVmWrite` wrappers around
the syscalls `process_vm_readv` and `process_vm_writev`, available since
Linux 3.2. These follow the same permission model as `ptrace`, but they
don't actually require being attached, which means they can be called
directly from any thread in the debugger. They also use `iovec` to write
entire blocks at once, rather than having to peek/poke each `uintptr`.
These wrappers are used in `Thread.ReadMemory` and `WriteMemory`, still
falling back to `ptrace` if that fails for any reason. Notably,
`process_vm_writev` respects memory protection, so it can't modify
read-only memory like `ptrace`. This frequently occurs when writing
breakpoints in read-only `.text`, so to avoid a lot of wasted `EFAULT`
calls, we only try `process_vm_writev` for larger writes.
As proc/native is arch related, it should move some functions to arch-relate file. And this patch can help us to separate the architecture code, make code tidy. So that the merge of arm64 code later will not cause chaos.(#118)
* proc/linux: do not route signals to threads while stopping
While we are trying to stop the process we should not route signals
sent to threads because that will result in threads being resumed.
Also keep better track of which threads are stopped.
This fixes an incompatibility with Go 1.14, which sends a lot of
signals to its threads to implement non-cooperative preemption,
resulting in Delve hanging waiting for an already-stopped thread to
stop.
In principle however this bug has nothing to do with Go 1.14 and could
manifest in any instance of high signal pressure.
* Makefile: discard stderr of "go list"
In module mode "go" will print messages about downloading modules to
stderr, we shouldn't confuse them for the real command output.
RestoreRegisters on linux would also restore FS_BASE and GS_BASE, if
the target goroutine migrated to a different thread during the call
injection this would result in two threads of the target process
pointing to the same TLS area which would greatly confuse the target
runtime, leading to fatal panics with nonsensical stack traces.
Other backends are unaffected:
- native/windows doesn't store the TLS in the same CONTEXT struct as
the other register values.
- native/darwin doesn't support function calls (and wouldn't store the
TLS value in the same struct)
- gdbserial/rr doesn't support function calls (because it's a
recording)
- gsdbserial/lldb extracts the value of TLS by executing code in the
target process.
The repository is being switched from the personal account
github.com/derekparker/delve to the organization account
github.com/go-delve/delve. This patch updates imports and docs, while
preserving things which should not be changed such as my name in the
CHANGELOG and in TODO comments.
The linux version of proc/native and proc/core contained largely
overlapping implementations of the register handling code, deduplicate
it by moving it into proc/linutil.
Fncall.go was written with the assumption that the object returned by
proc.Thread.Registers does not change after we call
proc.Thread.SetPC/etc.
This is true for the native backend but not for gdbserial. I had
anticipated this problem and introduced the Save/SavedRegisters
mechanism during the first implementation of fncall.go but that's
insufficient.
Instead:
1. clarify that the object returned by proc.Thread.Registers could
change when the CPU registers are modified.
2. add a Copy method to Registers that returns a copy of the registers
that are guaranteed not to change when the CPU registers change.
3. remove the Save/SavedRegisters mechanism.
This solution leaves us the option, in the future, to cache the output
of proc.(Thread).Registers, avoiding a system call every time it's
called.
Implements the function call injection protocol introduced in go 1.11
by https://go-review.googlesource.com/c/go/+/109699.
This is only the basic support, see TODO comments in pkg/proc/fncall.go
for a list of missing features.
Updates #119
If a breakpoint is hit close to process death on a thread that isn't
the group leader the process could die while we are trying to stop it.
This can be easily reproduced by having the goroutine that's executing
main.main (which will almost always run on the thread group leader)
wait for a second goroutine before exiting, then setting a breakpoint
on the second goroutine and stepping through it (see TestIssue1101 in
proc_test.go).
When stepping over the return instruction of main.f the deferred
wg.Done() call will be executed which will cause the main goroutine to
resume and proceed to exit. Both the temporary breakpoint on wg.Done
and the temporary breakpoint on the return address of main.f will be in
close proximity to main.main calling os.Exit() and causing the death of
the thread group leader.
Under these circumstances the call to native.(*Thread).waitFast in
native.(*Thread).halt can hang forever due to a bug similar to
https://sourceware.org/bugzilla/show_bug.cgi?id=12702 (see comment in
native.(*Thread).wait for an explanation).
Replacing waitFast with a normal wait work in most circumstances,
however, besides the performance hit, it looks like in this
circumstances trapWait sometimes receives a spurious SIGTRAP on the
dying group leader which would cause the subsequent call to wait in
halt to accidentally reap the process without noting that it did exit.
Instead this patch removes the call to wait from halt and instead calls
trapWait in a loop in setCurrentBreakpoints until all threads are set
to running=false. This is also a better fix than the workaround to
ESRCH error while setting current breakpoints implemented in 94b50d.
Fixes#1101
