Add a minimal dlcall plugin that lets the guest invoke host functions
through magic system calls. The plugin registers a vCPU syscall filter
callback that intercepts a reserved syscall number and dispatches a set
of pass-through operations: querying host attributes, loading and freeing
shared libraries, resolving symbols, retrieving the last library error,
and invoking a host function through a common interface.
The magic syscall number defaults to 4096 and can be overridden at load
time with the "syscall_num=N" argument; values low enough to clash with a
real syscall are rejected.
Co-authored-by: Kailiang Xu <xukl2019@sjtu.edu.cn>
Co-authored-by: Mingyuan Xia <xiamy@ultrarisc.com>
Signed-off-by: Ziyang Zhang <functioner@sjtu.edu.cn>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@oss.qualcomm.com>
Link: https://lore.kernel.org/qemu-devel/20260711094523.622997-2-functioner@sjtu.edu.cn
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@oss.qualcomm.com>
We can now demonstrate what previous changes allow us to do. Since all
callbacks have a userdata pointer, we can use that mechanism to move an
object through all of them.
In other words, we can now have stateful plugins without resorting to
any global variable.
As an example, we implement tb counting plugin with our cpp plugin. It
produces an output similar to hotblocks, with same performance.
Reviewed-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
Link: https://lore.kernel.org/qemu-devel/20260615193526.2883349-27-pierrick.bouvier@oss.qualcomm.com
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@oss.qualcomm.com>
The qemu_plugin_{read,write} register API previously was inconsistent
with regard to its docstring (where a return value of both -1 and 0
would indicate an error) and to the memory read/write APIs, which
already return a boolean value to indicate success or failure.
Returning the number of bytes read or written is superfluous, as the
GByteArray* passed to the API functions already encodes the length.
See the linked thread for more details.
This patch moves from returning an int (number of bytes read/written) to
returning a bool from the register read/write API, bumps the plugin API
version, and adjusts plugins and tests accordingly.
Signed-off-by: Florian Hofhammer <florian.hofhammer@fhofhammer.de>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Link: https://lore.kernel.org/qemu-devel/f877dd79-1285-4752-811e-f0d430ff27fe@fhofhammer.de
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
We recently introduced new plugin API for registration of discontinuity
related callbacks. This change introduces a minimal plugin showcasing
the new API. It simply counts the occurances of interrupts, exceptions
and host calls per CPU and reports the counts when exitting.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Julian Ganz <neither@nut.email>
Message-ID: <20251027110344.2289945-11-alex.bennee@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Enhance uftrace_symbols.py to generate .dbg files, containing
source location for every symbol present in .sym file.
It allows to use `uftrace {replay,dump} --srcline` and show origin of
functions, connecting trace to original source code.
It was first implemented with pyelftools DWARF parser, which was way
too slow (~minutes) to get locations for every symbol in the linux
kernel. Thus, we use `addr2line` instead, which runs in seconds.
As well, there were some bugs with latest pyelftools release,
requiring to run master version, which is not installable with pip.
Thus, since we now require binutils (addr2line), we can ditch pyelftools
based implementation and simply rely on `nm` to get symbols information,
which is faster and better.
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-ID: <20251016150357.876415-8-alex.bennee@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
usage: contrib/plugins/uftrace_symbols.py \
--prefix-symbols \
arm-trusted-firmware/build/qemu/debug/bl1/bl1.elf \
arm-trusted-firmware/build/qemu/debug/bl2/bl2.elf \
arm-trusted-firmware/build/qemu/debug/bl31/bl31.elf \
u-boot/u-boot:0x60000000 \
u-boot/u-boot.relocated:0x000000023f6b6000 \
linux/vmlinux
Will generate symbols and memory mapping files for uftrace, allowing to
have an enhanced trace, instead of raw addresses.
It takes a collection of elf files, and automatically find all their
symbols, and generate an ordered memory map based on that.
This script uses the python (native) pyelftools module.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Acked-by: Alex Bennée <alex.bennee@linaro.org>
Message-ID: <20250902075042.223990-9-pierrick.bouvier@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-ID: <20250922093711.2768983-25-alex.bennee@linaro.org>
Beyond traces per cpu, uftrace expect to find some specific files.
- info: contains information about machine/program run
those values are not impacting uftrace behaviour (only reported by
uftrace info), and we simply added empty strings.
- memory mapping: how every binary is mapped in memory. For system mode,
we generate an empty mapping (uftrace_symbols.py, coming in future
commit, will take care of that). For user mode, we copy current
/proc/self/maps. We don't need to do any special filtering, as
reported addresses will necessarily concern guest program, and not
QEMU and its libraries.
- task: list of tasks. We present every vcpu/privilege level as a
separate process, as it's the best view we can have when generating a
(visual) chrome trace. Using threads is less convenient in terms of
UI.
Reviewed-by: Manos Pitsidianakis <manos.pitsidianakis@linaro.org>
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Message-ID: <20250902075042.223990-7-pierrick.bouvier@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-ID: <20250922093711.2768983-23-alex.bennee@linaro.org>
In insn_check_regs() we don't explicitly check whether
qemu_plugin_read_register() failed, which confuses Coverity into
thinking that sz can be -1 in the memcmp(). In fact the assertion
that sz == reg->last->len means this can't happen, but it's clearer
to both humans and Coverity if we explicitly assert that sz > 0, as
we already do in init_vcpu_register().
Coverity: CID 1611901, 1611902
Fixes: af6e4e0a22 ("contrib/plugins: extend execlog to track register changes")
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Message-ID: <20250710144543.1187715-1-peter.maydell@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-ID: <20250922093711.2768983-17-alex.bennee@linaro.org>
Currently, execlog searches for a space separator between the
instruction mnemonic and operands, but some disassemblers, e.g. Alpha's,
use a tab separator instead; this results in a null pointer being passed
as the haystack in g_strstr during a subsequent register search, i.e.
undefined behavior, because of a missing null check.
This patch adds tab to the separator search and a null check on the
result.
Also, an affected pointer is changed to const.
Lastly, a break statement was added to immediately terminate the
register search when a user-requested register is found in the current
instruction as a trivial optimization, because searching for the
remaining requested registers is unnecessary once one is found.
Suggested-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Yodel Eldar <yodel.eldar@gmail.com>
Message-ID: <20250630164124.26315-2-yodel.eldar@gmail.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-ID: <20250710104531.3099313-4-alex.bennee@linaro.org>
The build fails on Windows. Replace calls to Unix programs like ´cat´,
´sed´ and ´true´ with calls to ´python´ and wrap calls to
´os.path.relpath´ in try-except because it can fail when the two paths
are on different drives. Make sure to convert the Windows paths to Unix
paths to prevent warnings in generated files.
Signed-off-by: oltolm <oleg.tolmatcev@gmail.com>
Message-id: 20250612221521.1109-2-oleg.tolmatcev@gmail.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
On Emscripten, function pointer casts can result in runtime failures due to
strict function signature checks. This affects the use of g_list_sort and
g_slist_sort, which internally perform function pointer casts that are not
supported by Emscripten. To avoid these issues, g_list_sort_with_data and
g_slist_sort_with_data should be used instead, as they do not rely on
function pointer casting.
Signed-off-by: Kohei Tokunaga <ktokunaga.mail@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-ID: <0fcddfca16ca8da2bdaa7b2c114476f5b73d032b.1745295397.git.ktokunaga.mail@gmail.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
