From: "Metzger, Markus T" <markus.t.metzger@intel.com>
To: Simon Marchi <simark@simark.ca>
Cc: "Aktemur, Tankut Baris" <tankut.baris.aktemur@intel.com>,
"gdb-patches@sourceware.org" <gdb-patches@sourceware.org>
Subject: RE: [PATCH v3 10/44] gdb, gdbserver, rsp, ze: acknowledge libraries
Date: Fri, 12 Dec 2025 14:28:12 +0000 [thread overview]
Message-ID: <DM8PR11MB57493BD91628FE788BF25091DEAEA@DM8PR11MB5749.namprd11.prod.outlook.com> (raw)
In-Reply-To: <2e3e88ca-69c1-4299-ae6f-0ceab94d82a3@simark.ca>
Hello Simon,
>> From: Markus Metzger <markus.t.metzger@intel.com>
>>
>> On some accelerator devices, device shared libraries are loaded from a
>> host thread rather than from a device thread. The reporting entity may
>> not be the one that actually does the load.
>>
>> Intel GPU devices, for example, that are based on Level-Zero, will report
>> shared library events via the device's debug interface. This is triggered
>> from a host thread calling the run-time interface for loading a device
>> shared library.
>>
>> The Level-Zero run-time ensures that this host thread will not return
>> until the respective debug event has been acknowledged by the debugger.
>> This allows debuggers to set breakpoints before the new library is used.
>>
>> Add a mechanism that allows gdbserver to request acknowledgement of
>newly
>> reported shared libraries and GDB to acknowledge requested libraries after
>> placing breakpoints.
>
>I had trouble understanding the need for this until I remembered that
>the remote inferior only has the GPU threads. So the thread loading the
>library isn't even part of the inferior.
>
>Otherwise, I think it would just naturally work:
>
> - host thread hits magic breakpoint
> - gdbserver fetches event, notices new library loaded, returns "library
> loaded" event to gdb
> - gdb loads the new solibs, then resumes the host thread
> - gdbserver resumes the host thread, allowing the application to
> use the newly loaded library
Indeed, that's how it would work if we involved the host process.
In our case, however, the user-mode driver notifies the kernel-mode driver
about modules. The debugger will receive those as events.
To ensure that the debugger has had a chance to place breakpoints before
the host process submits those kernels, we require the debugger to acknowledge
such a module load event. The kernel mode driver will keep the host process
thread stopped in the module load ioctl() until the load has been acknowledged.
In your case, the breakpoint inside the host process loader does that.
Regards,
Markus.
Intel Deutschland GmbH
Registered Address: Dornacher Straße 1, 85622 Feldkirchen, Germany
Tel: +49 89 991 430, www.intel.de
Managing Directors: Harry Demas, Jeffrey Schneiderman, Yin Chong Sorrell
Chairperson of the Supervisory Board: Nicole Lau
Registered Seat: Munich
Commercial Register: Amtsgericht München HRB 186928
next prev parent reply other threads:[~2025-12-12 14:28 UTC|newest]
Thread overview: 92+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-08-01 9:37 [PATCH v3 00/44] A new target to debug Intel GPUs Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 01/44] gdb, intelgt: add intelgt as a basic machine Tankut Baris Aktemur
2025-12-09 20:44 ` Simon Marchi
2025-12-19 11:13 ` Aktemur, Tankut Baris
2025-08-01 9:37 ` [PATCH v3 02/44] bfd: add intelgt target to BFD Tankut Baris Aktemur
2025-08-01 12:20 ` Jan Beulich
2025-08-08 5:03 ` Metzger, Markus T
2025-12-09 21:05 ` Simon Marchi
2025-12-19 12:46 ` Aktemur, Tankut Baris
2025-08-01 9:37 ` [PATCH v3 03/44] ld: add intelgt as a target configuration Tankut Baris Aktemur
2025-08-01 12:06 ` Jan Beulich
2025-08-08 5:03 ` Metzger, Markus T
2025-08-01 9:37 ` [PATCH v3 04/44] opcodes: add intelgt as a configuration Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 05/44] gdb, gdbserver, gdbsupport: add 'device' tag to XML target description Tankut Baris Aktemur
2025-12-09 21:27 ` Simon Marchi
2025-12-15 21:03 ` Simon Marchi
2025-12-18 15:04 ` Aktemur, Tankut Baris
2026-01-09 19:12 ` Aktemur, Tankut Baris
2026-01-09 19:34 ` Simon Marchi
2025-08-01 9:37 ` [PATCH v3 06/44] gdb, arch, intelgt: add intelgt arch definitions Tankut Baris Aktemur
2025-12-09 21:48 ` Simon Marchi
2025-12-16 15:47 ` Metzger, Markus T
2025-08-01 9:37 ` [PATCH v3 07/44] gdb, intelgt: add the target-dependent definitions for the Intel GT architecture Tankut Baris Aktemur
2025-12-11 18:53 ` Simon Marchi
2025-12-19 16:01 ` Aktemur, Tankut Baris
2025-08-01 9:37 ` [PATCH v3 08/44] gdb, intelgt: add disassemble feature " Tankut Baris Aktemur
2025-12-11 19:37 ` Simon Marchi
2025-12-23 11:03 ` Aktemur, Tankut Baris
2025-08-01 9:37 ` [PATCH v3 09/44] gdb, gdbserver, ze: in-memory libraries Tankut Baris Aktemur
2025-12-12 4:13 ` Simon Marchi
2025-12-12 11:20 ` Metzger, Markus T
2025-12-12 19:34 ` Simon Marchi
2025-12-15 13:07 ` Metzger, Markus T
2025-12-15 21:25 ` Simon Marchi
2025-08-01 9:37 ` [PATCH v3 10/44] gdb, gdbserver, rsp, ze: acknowledge libraries Tankut Baris Aktemur
2025-12-12 4:41 ` Simon Marchi
2025-12-12 14:28 ` Metzger, Markus T [this message]
2025-08-01 9:37 ` [PATCH v3 11/44] gdb, solib, ze: update target_solib_ops::bfd_open_from_target_memory Tankut Baris Aktemur
2025-12-12 4:43 ` Simon Marchi
2025-12-12 14:33 ` Metzger, Markus T
2025-08-01 9:37 ` [PATCH v3 12/44] gdb, infrun, ze: allow saving process events Tankut Baris Aktemur
2025-12-12 4:57 ` Simon Marchi
2025-12-15 13:13 ` Metzger, Markus T
2025-12-16 21:10 ` Simon Marchi
2025-12-17 9:30 ` Metzger, Markus T
2025-12-17 20:44 ` Simon Marchi
2025-12-18 7:20 ` Metzger, Markus T
2025-08-01 9:37 ` [PATCH v3 13/44] gdb, ze: add TARGET_WAITKIND_UNAVAILABLE Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 14/44] gdb, infrun, ze: handle stopping unavailable threads Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 15/44] gdb, infrun, ze: allow resuming " Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 16/44] gdb, gdbserver, ze: add U stop reply Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 17/44] gdb, gdbserver, ze: add library notification to " Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 18/44] gdbserver, ze: report TARGET_WAITKIND_UNAVAILABLE events Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 19/44] gdb, ze: handle TARGET_WAITKIND_UNAVAILABLE in stop_all_threads Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 20/44] gdb, remote: handle thread unavailability in print_one_stopped_thread Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 21/44] gdb, remote: do 'remote_add_inferior' in 'remote_notice_new_inferior' earlier Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 22/44] gdb, remote: handle a generic process PID in remote_notice_new_inferior Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 23/44] gdb, remote: handle a generic process PID in process_stop_reply Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 24/44] gdb: use the pid from inferior in setup_inferior Tankut Baris Aktemur
2025-12-12 19:51 ` Simon Marchi
2025-12-13 12:40 ` Aktemur, Tankut Baris
2025-08-01 9:37 ` [PATCH v3 25/44] gdb: revise the pid_to_exec_file target op Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 26/44] gdb: load solibs if the target does not have the notion of an exec file Tankut Baris Aktemur
2025-12-12 20:30 ` Simon Marchi
2026-01-09 19:10 ` Aktemur, Tankut Baris
2025-08-01 9:37 ` [PATCH v3 27/44] gdbserver: import AC_LIB_HAVE_LINKFLAGS macro into the autoconf script Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 28/44] gdbserver: add a pointer to the owner thread in regcache Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 29/44] gdbserver: wait for stopped threads in queue_stop_reply_callback Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 30/44] gdbserver: adjust pid after the target attaches Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 31/44] gdb: do not create a thread after a process event Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 32/44] gdb, ze: on a whole process stop, mark all threads as not_resumed Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 33/44] gdb, dwarf, ze: add DW_OP_INTEL_regval_bits Tankut Baris Aktemur
2025-08-01 12:02 ` Jan Beulich
2025-08-01 12:31 ` Metzger, Markus T
2025-08-01 12:50 ` Jan Beulich
2025-08-08 5:25 ` Metzger, Markus T
2025-08-01 9:37 ` [PATCH v3 34/44] gdbserver: allow configuring for a heterogeneous target Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 35/44] gdbserver, ze, intelgt: introduce ze-low and intel-ze-low targets Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 36/44] testsuite, sycl: add SYCL support Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 37/44] testsuite, sycl: add test for backtracing inside a kernel Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 38/44] testsuite, sycl: add test for 'info locals' and 'info args' Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 39/44] testsuite, sycl: add tests for stepping and accessing data elements Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 40/44] testsuite, sycl: add test for 1-D and 2-D parallel_for kernels Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 41/44] testsuite, sycl: add test for scheduler-locking Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 42/44] testsuite, arch, intelgt: add a disassembly test Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 43/44] testsuite, arch, intelgt: add intelgt-program-bp.exp Tankut Baris Aktemur
2025-08-01 9:37 ` [PATCH v3 44/44] testsuite, sycl: test canceling a stepping flow Tankut Baris Aktemur
2025-09-17 12:43 ` [PATCH v3 00/44] A new target to debug Intel GPUs Aktemur, Tankut Baris
2025-10-14 6:34 ` Aktemur, Tankut Baris
2025-12-08 11:32 ` Aktemur, Tankut Baris
2025-12-09 21:30 ` Simon Marchi
2025-12-19 12:52 ` Aktemur, Tankut Baris
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=DM8PR11MB57493BD91628FE788BF25091DEAEA@DM8PR11MB5749.namprd11.prod.outlook.com \
--to=markus.t.metzger@intel.com \
--cc=gdb-patches@sourceware.org \
--cc=simark@simark.ca \
--cc=tankut.baris.aktemur@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox