[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20251015105435.527088-1-johannes.thumshirn@wdc.com>
Date: Wed, 15 Oct 2025 12:54:19 +0200
From: Johannes Thumshirn <johannes.thumshirn@....com>
To: axboe@...nel.dk
Cc: chaitanyak@...dia.com,
dlemoal@...nel.org,
hare@...e.de,
hch@....de,
john.g.garry@...cle.com,
linux-block@...r.kernel.org,
linux-btrace@...r.kernel.org,
linux-kernel@...r.kernel.org,
linux-trace-kernel@...r.kernel.org,
martin.petersen@...cle.com,
mathieu.desnoyers@...icios.com,
mhiramat@...nel.org,
naohiro.aota@....com,
rostedt@...dmis.org,
shinichiro.kawasaki@....com,
Johannes Thumshirn <johannes.thumshirn@....com>
Subject: [PATCH v3 00/16] block: add blktrace support for zoned block devi
This patch series extends the kernel blktrace infrastructure to support
tracing zoned block device commands. Zoned block devices (e.g., ZAC/ZBC and
ZNS) introduce command types such as zone open, close, reset, finish, and zone
append. These are currently not visible in blktrace, making it difficult to
debug and analyze I/O workloads on zoned devices.
The patches in this series utilize the new trace points for these zone
management operations, and propagate the necessary context to the blktrace
logging path. These additions are designed to be backward-compatible, and are
only active when zoned devices are in use.
In order to not break the user-space ABI, a new ioctl was introduced to
request the new version of the blk_io_trace with extended 'action' field.
The user-space tools in the blktrace suite are updated in a separate patch
series to interpret and display the new trace events.This patch series extends
the kernel blktrace infrastructure to support tracing zoned block device
commands. Zoned block devices (e.g., ZAC/ZBC and ZNS) introduce command types
such as zone open, close, reset, finish, and zone append. These are currently
not visible in blktrace, making it difficult to debug and analyze I/O
workloads on zoned devices.
The patches in this series utilize the new trace points for these zone
management operations, and propagate the necessary context to the blktrace
logging path. These additions are designed to be backward-compatible, and are
only active when zoned devices are in use.
In order to not break the user-space ABI, a new ioctl was introduced to
request the new version of the blk_io_trace with extended 'action' field.
The user-space tools in the blktrace suite are updated in a separate patch
series to interpret and display the new trace events.
I've tested on SMR drives, TCMU emulated SMR drives and zloop, with both XFS
and BTRFS as filesystems on top of the hardwdare. Testing of different
hardware/setups is highly encouraged.
Changes to v2:
- Collect Reviewed-bys
- Convert trace_note and ftrace's blk_io_tracer to blk_io_trace2
- Don't play games with the layout of the 'action' field
- Fix structure alignments
- Drop Zone Management trace action, it's a command not an action
Johannes Thumshirn (16):
blktrace: only calculate trace length once
blktrace: factor out recording a blktrace event
blktrace: split out relaying a blktrace event
blktrace: untangle if/else sequence in __blk_add_trace
blktrace: change the internal action to 64bit
blktrace: split do_blk_trace_setup into two functions
blktrace: add definitions for blk_user_trace_setup2
blktrace: pass blk_user_trace2 to setup functions
blktrace: add definitions for struct blk_io_trace2
blktrace: differentiate between blk_io_trace versions
blktrace: move trace_note to blk_io_trace2
blktrace: move ftrace blk_io_tracer to blk_io_trace2
blktrace: add block trace commands for zone operations
blktrace: expose ZONE APPEND completions to blktrace
blktrace: trace zone write plugging operations
blktrace: handle BLKTRACESETUP2 ioctl
block/ioctl.c | 1 +
include/linux/blktrace_api.h | 3 +-
include/uapi/linux/blktrace_api.h | 53 +++-
include/uapi/linux/fs.h | 1 +
kernel/trace/blktrace.c | 463 ++++++++++++++++++++++--------
5 files changed, 400 insertions(+), 121 deletions(-)
--
2.51.0
Powered by blists - more mailing lists