[PATCH v11 00/65] Fix DRM_USE_DYNAMIC_DEBUG=y

[PATCH v11 00/65] Fix DRM_USE_DYNAMIC_DEBUG=y

[Jim Cromie]

The drm subsystem has *lots* of debug statements, in 11 categories

 $> ack '\w*_dbg' drivers/gpu/drm/ | wc
    5532   29318  553806
 $> ack 'DRM_DEBUG\w*' drivers/gpu/drm/ | wc
    2208   12856  212035

All of these are bit-tests on __drm_debug, exposed to users as
/sys/module/drm/parameters/debug.  Many of these are done often,
vblank is done ~100/sec for some displays.  Over the uptime of many
boxes, this is a lot of cpu cycles, on bits that are almost always
off.

Dynamic-debug excels at replacing such tests with NOOPS.  Classmaps
was devised to bring that 0-off-cost to drm's categories.

Classmaps-v1 went into the kernel in Sept 2022, in 2 chunks:
  b7b4eebdba7b..6ea3bf466ac6    # core dyndbg changes
  0406faf25fb1..ee7d633f2dfb    # drm adoption

Sadly DRM-CI found a regression during init with drm.debug=<initval>;
the static-keys underneath the drm-dbgs in drm.ko got enabled, but
those in drivers & helpers did not.

Root Problem:

DECLARE_DYNDBG_CLASSMAP defined the classmap, but it repeated use in
both core and drivers violated a K&R rule "define once, refer
afterwards".  This flaw resulted in a regression; with drm.debug=0xFF
boot arg, drm-core got enabled, but drivers,etc did not.

So in feb 2023, this resulted in:
commit bb2ff6c27bc9 ("drm: Disable dynamic debug as broken")

This patchset replaces it with DYNDBG_CLASSMAP_DEFINE (invoked once in
drm-core) and DYNDBG_CLASSMAP_USE (invoked repeatedly, in drivers &
helpers).

_DEFINE exports the classmap it creates (in drm.ko), other modules
_USE the classmap.  The _USE adds a record ref'g the _DEFINEd (&
exported) classmap, in a 2nd __dyndbg_class_users section.

So now at modprobe, dyndbg scans the new section after the 1st
__dyndbg_class_maps section, follows the linkage to the _DEFINEr
module, finds the (optional) kernel-param controlling the classmap,
examines its drm.debug=<initval>, and applies it to the module being
initialized.

Savings effect:

No real effort was made to quantify the savings; bit-tests are almost
unmeasurable individually, and [1] DRM_USE_DYNAMIC_DEBUG=y purposely
depends upon JUMP_LABEL to avoid its use where it cannot help.

That said, Ive booted one box with drm.debug=0x1ff on the boot-line,
with a script run by systemd to turn it off once boot completes. It
issued ~25k messages in that time.

Status-v11:

Now rebased on drm-misc-next, to allow grinding on DRM-CI.
https://gitlab.freedesktop.org/jim.cromie/kernel-drm-next-dd/-/pipelines/1622778

DRM-CI revealed 2 bugs, fixes now included:
1. drm_buddy had 32bit truncation error, causing unexpected pass
2. drm_printer_debug_fn disregarded drm.debug, spewing msgs, causing timeout.

Also tuned vblank* to avoid drm_debug_enabled() bit-test.
Probably should defer this until [1]=y is default. It isnt yet.

Up to this point, classmaps-v2 has been stuck, straddled on the fence
between drm and dynamic-debug; it needs a little help towards one side
or the other.  ISTM that drm is the actual user, and the acid-test of
its proper function; the dyndbg selftest recaps the 2-module scenario,
but this cannot finally prove anything absent drm.

Verification: DRM-CI Pipeline #1622778
  - KUnit (arm32, arm64, x86_64): All 621 tests PASSED.
    Specifically confirms the 64-bit truncation fixes in drm_buddy.
  - i915-CML: 370 tests PASSED, 0 FAIL.
    Confirms the quelling of the UART storm in shared print callbacks.
  - Overall: No regressions found in dyndbg or buddy allocator logic.

Future:

Lukas Bartosik has been patiently waiting for classmaps to land, so he
can add dyndbg-to-ftrace functionality.  This new feature will allow
steering any/all drm.debug messages to a private ftrace buffer, and
also to mix in other pr_debug messages.  Other ftrace streams could
also be sent there (possibly w/o code changes, tbd).  It could be a
big step towards a featureful crash-buffer.

Ive also cobbled together a maple-tree based replacement for the
__dyndbg_descriptors linker section; it copies the section contents
into 3 maple trees, then drops the section in __init.  By eliminating
the redundant columns (module, file, function) it reduces the storage
needed by 40%.  Early measurements show no net runtime costs, maple
trees make good furniture.

Given that progress, I also hacked up a maple-tree based dynamic
prefix cache.  Currently the dynamic-prefix is re-written for every
prefixed callsite; this can get expensive.  By careful keying (flags
on top, descriptor address underneath), a single prefix string can
span an entire range (smallest of enabled module, file, function).

That last one doesn't help DRM, cuz it doesnt use dynamic-prefixes,
but with this optimization in place, it could be adapted to do so.

Lastly, I have an RFC jump-label patchset that dramatically cuts IPIs,
currently 1 per pr_debug (to ~1/140), by sorting the (existing) queue,
rather than flushing it when a non-increasing patch-addr is queued.

I'd like to move forward with all these, but I anticipate a "fix
classmaps 1st" response.

To: airlied@gmail.com
To: daniel@ffwll.ch
To: jbaron@akamai.com
To: gregkh@linuxfoundation.org
Cc: mripard@kernel.org
Cc: tzimmermann@suse.de
Cc: maarten.lankhorst@linux.intel.com
Cc: jani.nikula@intel.com
Cc: ville.syrjala@linux.intel.com
Cc: christian.koenig@amd.com
Cc: matthew.auld@intel.com
Cc: arunpravin.paneerselvam@amd.com
Cc: louis.chauvet@bootlin.com
Cc: skhan@linuxfoundation.org
Cc: pmladek@suse.com
Cc: ukaszb@chromium.org
Cc: dri-devel@lists.freedesktop.org
Cc: linux-kernel@vger.kernel.org
Cc: intel-gfx@lists.freedesktop.org
Cc: amd-gfx@lists.freedesktop.org

Jim Cromie (65):
  dyndbg: fix NULL ptr on i386 due to section alignment
  dyndbg: factor ddebug_match_desc out from ddebug_change
  dyndbg: add stub macro for DECLARE_DYNDBG_CLASSMAP
  docs/dyndbg: update examples \012 to \n
  docs/dyndbg: explain flags parse 1st
  test-dyndbg: fixup CLASSMAP usage error
  dyndbg: reword "class unknown," to "class:_UNKNOWN_"
  dyndbg: make ddebug_class_param union members same size
  dyndbg: drop NUM_TYPE_ARRAY
  dyndbg: tweak pr_fmt to avoid expansion conflicts
  dyndbg: reduce verbose/debug clutter
  dyndbg: refactor param_set_dyndbg_classes and below
  dyndbg: tighten fn-sig of ddebug_apply_class_bitmap
  dyndbg: replace classmap list with a vector
  dyndbg: macrofy a 2-index for-loop pattern
  dyndbg,module: make proper substructs in _ddebug_info
  dyndbg: hoist classmap-filter-by-modname up to ddebug_add_module
  dyndbg: move mod_name down from struct ddebug_table to _ddebug_info
  dyndbg-API: remove DD_CLASS_TYPE_(DISJOINT|LEVEL)_NAMES and code
  selftests-dyndbg: add a dynamic_debug run_tests target
  dyndbg: change __dynamic_func_call_cls* macros into expressions
  dyndbg-API: replace DECLARE_DYNDBG_CLASSMAP
  dyndbg: detect class_id reservation conflicts
  dyndbg: check DYNAMIC_DEBUG_CLASSMAP_DEFINE args at compile-time
  dyndbg-test: change do_prints testpoint to accept a loopct
  dyndbg-API: promote DYNAMIC_DEBUG_CLASSMAP_PARAM to API
  dyndbg: treat comma as a token separator
  dyndbg: split multi-query strings with %
  selftests-dyndbg: add test_mod_submod
  dyndbg: resolve "protection" of class'd pr_debug
  dyndbg: add DYNAMIC_DEBUG_CLASSMAP_USE_(dd_class_name, offset)
  dyndbg: Harden classmap and callsite validation
  docs/dyndbg: add classmap info to howto
  drm: use correct ccflags-y spelling
  drm-dyndbg: adapt drm core to use dyndbg classmaps-v2
  drm-dyndbg: adapt DRM to invoke DYNAMIC_DEBUG_CLASSMAP_PARAM
  drm-print: modernize an archaic comment
  drm-print: fix config-dependent unused variable
  drm-dyndbg: DRM_CLASSMAP_USE in amdgpu driver
  drm-dyndbg: DRM_CLASSMAP_USE in i915 driver
  drm-dyndbg: DRM_CLASSMAP_USE in drm_crtc_helper
  drm-dyndbg: DRM_CLASSMAP_USE in drm_dp_helper
  drm-dyndbg: DRM_CLASSMAP_USE in nouveau
  drm-dyndbg: add DRM_CLASSMAP_USE to Xe driver
  drm-dyndbg: add DRM_CLASSMAP_USE to virtio_gpu
  drm-dyndbg: add DRM_CLASSMAP_USE to simpledrm
  drm-dyndbg: add DRM_CLASSMAP_USE to bochs
  drm-dyndbg: add DRM_CLASSMAP_USE to etnaviv
  drm-dyndbg: add DRM_CLASSMAP_USE to gma500 driver
  drm-dyndbg: add DRM_CLASSMAP_USE to radeon
  drm-dyndbg: add DRM_CLASSMAP_USE to vmwgfx driver
  drm-dyndbg: add DRM_CLASSMAP_USE to vkms driver
  drm-dyndbg: add DRM_CLASSMAP_USE to udl driver
  drm-dyndbg: add DRM_CLASSMAP_USE to mgag200 driver
  drm-dyndbg: add DRM_CLASSMAP_USE to the gud driver
  drm-dyndbg: add DRM_CLASSMAP_USE to the qxl driver
  drm-dyndbg: add DRM_CLASSMAP_USE to the drm_gem_shmem_helper driver
  accel: add -DDYNAMIC_DEBUG_MODULE to subdir-ccflags
  accel/ivpu: implement IVPU_DBG_* as a dyndbg classmap
  drm: restore CONFIG_DRM_USE_DYNAMIC_DEBUG un-BROKEN
  accel/ethosu: enable drm.debug control
  accel/rocket: enable drm.debug control
  drm_buddy: fix 64-bit truncation in power-of-2 rounding
  drm_print: fix drm_printer dynamic debug bypass
  drm_vblank: use dyndbg's static-key to avoid flag-check

 .../admin-guide/dynamic-debug-howto.rst       | 184 +++-
 MAINTAINERS                                   |   3 +-
 drivers/accel/Makefile                        |   7 +-
 drivers/accel/ethosu/ethosu_drv.c             |   3 +
 drivers/accel/ivpu/ivpu_drv.c                 |  27 +-
 drivers/accel/ivpu/ivpu_drv.h                 |  45 +-
 drivers/accel/rocket/rocket_gem.c             |   2 +
 drivers/gpu/buddy.c                           |   4 +-
 drivers/gpu/drm/Kconfig.debug                 |   1 -
 drivers/gpu/drm/Makefile                      |   3 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c       |  12 +-
 drivers/gpu/drm/display/drm_dp_helper.c       |  12 +-
 drivers/gpu/drm/drm_crtc_helper.c             |  12 +-
 drivers/gpu/drm/drm_gem_shmem_helper.c        |   1 +
 drivers/gpu/drm/drm_print.c                   |  38 +-
 drivers/gpu/drm/drm_vblank.c                  |  10 +-
 drivers/gpu/drm/etnaviv/etnaviv_drv.c         |   2 +
 drivers/gpu/drm/gma500/psb_drv.c              |   2 +
 drivers/gpu/drm/gud/gud_drv.c                 |   2 +
 drivers/gpu/drm/i915/i915_params.c            |  12 +-
 drivers/gpu/drm/mgag200/mgag200_drv.c         |   2 +
 drivers/gpu/drm/nouveau/nouveau_drm.c         |  12 +-
 drivers/gpu/drm/qxl/qxl_drv.c                 |   2 +
 drivers/gpu/drm/radeon/radeon_drv.c           |   2 +
 drivers/gpu/drm/sysfb/simpledrm.c             |   2 +
 drivers/gpu/drm/tiny/bochs.c                  |   2 +
 drivers/gpu/drm/udl/udl_main.c                |   2 +
 drivers/gpu/drm/virtio/virtgpu_drv.c          |   2 +
 drivers/gpu/drm/vkms/vkms_drv.c               |   2 +
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c           |   2 +
 drivers/gpu/drm/xe/xe_drm_client.c            |   2 +
 include/asm-generic/vmlinux.lds.h             |  19 +-
 include/drm/drm_print.h                       |  22 +-
 include/linux/dynamic_debug.h                 | 304 +++++--
 kernel/module/main.c                          |  15 +-
 lib/Kconfig.debug                             |  24 +-
 lib/Makefile                                  |   5 +
 lib/dynamic_debug.c                           | 814 ++++++++++++------
 lib/test_dynamic_debug.c                      | 198 +++--
 lib/test_dynamic_debug_submod.c               |  21 +
 tools/testing/selftests/Makefile              |   1 +
 .../testing/selftests/dynamic_debug/Makefile  |   9 +
 tools/testing/selftests/dynamic_debug/config  |   7 +
 .../dynamic_debug/dyndbg_selftest.sh          | 373 ++++++++
 44 files changed, 1665 insertions(+), 561 deletions(-)
 create mode 100644 lib/test_dynamic_debug_submod.c
 create mode 100644 tools/testing/selftests/dynamic_debug/Makefile
 create mode 100644 tools/testing/selftests/dynamic_debug/config
 create mode 100755 tools/testing/selftests/dynamic_debug/dyndbg_selftest.sh

-- 
2.53.0

[PATCH v11 01/65] dyndbg: fix NULL ptr on i386 due to section alignment

[Jim Cromie]

When dyndbg classmaps get used (later in this series), the
__dyndbg_classes section (which has 28 byte structs on i386), causes
mis-alignment of the following __dyndbg section, resulting in a NULL
pointer deref in dynamic_debug_init().

To fix this, employ belt + suspenders:

1. move __dyndbg section above __dyndbg_classes.  This restores it to
its original position directly after the ALIGN(8), and fixes the
immediate problem.

2. add ALIGN(8) to the BOUNDED_SECTION* macros.  This aligns all
existing sections using the macro, and also fixes a future
dyndbg_class_* addition which would suffer the same misalignment on
i386.  Many of the existing macro uses already have a preceding ALIGN,
these are now redundant, but are harmless, so are left to avoid churn.

3. remove BOUNDED_SECTION* uses in ORC_UNWINDER sections.  These have
smaller alignments, *and* scripts/sorttable.c does not tolerate the
added ALIGN(8) padding.

Reported-by: kernel test robot <oliver.sang@intel.com>
Closes: https://lore.kernel.org/oe-lkp/202601211325.7e1f336-lkp@intel.com
Signed-off-by: Jim Cromie <jim.cromie@gmail.com>
---
 include/asm-generic/vmlinux.lds.h | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index eeb070f330bd..a2ba7e3d9994 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -212,11 +212,13 @@
 #endif
 
 #define BOUNDED_SECTION_PRE_LABEL(_sec_, _label_, _BEGIN_, _END_)	\
+	. = ALIGN(8);							\
 	_BEGIN_##_label_ = .;						\
 	KEEP(*(_sec_))							\
 	_END_##_label_ = .;
 
 #define BOUNDED_SECTION_POST_LABEL(_sec_, _label_, _BEGIN_, _END_)	\
+	. = ALIGN(8);							\
 	_label_##_BEGIN_ = .;						\
 	KEEP(*(_sec_))							\
 	_label_##_END_ = .;
@@ -383,8 +385,8 @@
 	*(__tracepoints)						\
 	/* implement dynamic printk debug */				\
 	. = ALIGN(8);							\
-	BOUNDED_SECTION_BY(__dyndbg_classes, ___dyndbg_classes)		\
 	BOUNDED_SECTION_BY(__dyndbg, ___dyndbg)				\
+	BOUNDED_SECTION_BY(__dyndbg_classes, ___dyndbg_classes)		\
 	CODETAG_SECTIONS()						\
 	LIKELY_PROFILE()		       				\
 	BRANCH_PROFILE()						\
@@ -867,15 +869,21 @@
 #ifdef CONFIG_UNWINDER_ORC
 #define ORC_UNWIND_TABLE						\
 	.orc_header : AT(ADDR(.orc_header) - LOAD_OFFSET) {		\
-		BOUNDED_SECTION_BY(.orc_header, _orc_header)		\
+		__start_orc_header = .;					\
+		KEEP(*(.orc_header))					\
+		__stop_orc_header = .;					\
 	}								\
 	. = ALIGN(4);							\
 	.orc_unwind_ip : AT(ADDR(.orc_unwind_ip) - LOAD_OFFSET) {	\
-		BOUNDED_SECTION_BY(.orc_unwind_ip, _orc_unwind_ip)	\
+		__start_orc_unwind_ip = .;				\
+		KEEP(*(.orc_unwind_ip))					\
+		__stop_orc_unwind_ip = .;				\
 	}								\
 	. = ALIGN(2);							\
 	.orc_unwind : AT(ADDR(.orc_unwind) - LOAD_OFFSET) {		\
-		BOUNDED_SECTION_BY(.orc_unwind, _orc_unwind)		\
+		__start_orc_unwind = .;					\
+		KEEP(*(.orc_unwind))					\
+		__stop_orc_unwind = .;					\
 	}								\
 	text_size = _etext - _stext;					\
 	. = ALIGN(4);							\
-- 
2.53.0

[PATCH v11 02/65] dyndbg: factor ddebug_match_desc out from ddebug_change

[Jim Cromie]

ddebug_change() is a big (~100 lines) function with a nested for loop.

The outer loop walks the per-module ddebug_tables list, and does
module stuff: it filters on a query's "module FOO*" and "class BAR",
failures here skip the entire inner loop.

The inner loop (60 lines) scans a module's descriptors.  It starts
with a long block of filters on function, line, format, and the
validated "BAR" class (or the legacy/_DPRINTK_CLASS_DFLT).

These filters "continue" past pr_debugs that don't match the query
criteria, before it falls through the code below that counts matches,
then adjusts the flags and static-keys.  This is unnecessarily hard to
think about.

So move the per-descriptor filter-block into a boolean function:
ddebug_match_desc(desc), and change each "continue" to "return false".
This puts a clear interface in place, so any future changes are either
inside, outside, or across this interface.

also fix checkpatch complaints about spaces and braces.

Signed-off-by: Jim Cromie <jim.cromie@gmail.com>
---
 lib/dynamic_debug.c | 83 +++++++++++++++++++++++++--------------------
 1 file changed, 47 insertions(+), 36 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index 18a71a9108d3..6b1e983cfedc 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -172,6 +172,52 @@ static struct ddebug_class_map *ddebug_find_valid_class(struct ddebug_table cons
  * callsites, normally the same as number of changes.  If verbose,
  * logs the changes.  Takes ddebug_lock.
  */
+static bool ddebug_match_desc(const struct ddebug_query *query,
+			      struct _ddebug *dp,
+			      int valid_class)
+{
+	/* match site against query-class */
+	if (dp->class_id != valid_class)
+		return false;
+
+	/* match against the source filename */
+	if (query->filename &&
+	    !match_wildcard(query->filename, dp->filename) &&
+	    !match_wildcard(query->filename,
+			    kbasename(dp->filename)) &&
+	    !match_wildcard(query->filename,
+			    trim_prefix(dp->filename)))
+		return false;
+
+	/* match against the function */
+	if (query->function &&
+	    !match_wildcard(query->function, dp->function))
+		return false;
+
+	/* match against the format */
+	if (query->format) {
+		if (*query->format == '^') {
+			char *p;
+			/* anchored search. match must be at beginning */
+			p = strstr(dp->format, query->format + 1);
+			if (p != dp->format)
+				return false;
+		} else if (!strstr(dp->format, query->format)) {
+			return false;
+		}
+	}
+
+	/* match against the line number range */
+	if (query->first_lineno &&
+	    dp->lineno < query->first_lineno)
+		return false;
+	if (query->last_lineno &&
+	    dp->lineno > query->last_lineno)
+		return false;
+
+	return true;
+}
+
 static int ddebug_change(const struct ddebug_query *query,
 			 struct flag_settings *modifiers)
 {
@@ -204,42 +250,7 @@ static int ddebug_change(const struct ddebug_query *query,
 		for (i = 0; i < dt->num_ddebugs; i++) {
 			struct _ddebug *dp = &dt->ddebugs[i];
 
-			/* match site against query-class */
-			if (dp->class_id != valid_class)
-				continue;
-
-			/* match against the source filename */
-			if (query->filename &&
-			    !match_wildcard(query->filename, dp->filename) &&
-			    !match_wildcard(query->filename,
-					   kbasename(dp->filename)) &&
-			    !match_wildcard(query->filename,
-					   trim_prefix(dp->filename)))
-				continue;
-
-			/* match against the function */
-			if (query->function &&
-			    !match_wildcard(query->function, dp->function))
-				continue;
-
-			/* match against the format */
-			if (query->format) {
-				if (*query->format == '^') {
-					char *p;
-					/* anchored search. match must be at beginning */
-					p = strstr(dp->format, query->format+1);
-					if (p != dp->format)
-						continue;
-				} else if (!strstr(dp->format, query->format))
-					continue;
-			}
-
-			/* match against the line number range */
-			if (query->first_lineno &&
-			    dp->lineno < query->first_lineno)
-				continue;
-			if (query->last_lineno &&
-			    dp->lineno > query->last_lineno)
+			if (!ddebug_match_desc(query, dp, valid_class))
 				continue;
 
 			nfound++;
-- 
2.53.0

[PATCH v11 00/65] Fix DRM_USE_DYNAMIC_DEBUG=y

[Jim Cromie]

resend - apologies, faf'g with --cc-cover & to: lists

The drm subsystem has *lots* of debug statements, in 11 categories

 $> ack '\w*_dbg' drivers/gpu/drm/ | wc
    5532   29318  553806
 $> ack 'DRM_DEBUG\w*' drivers/gpu/drm/ | wc
    2208   12856  212035

All of these are bit-tests on __drm_debug, exposed to users as
/sys/module/drm/parameters/debug.  Many of these are done often,
vblank is done ~100/sec for some displays.  Over the uptime of many
boxes, this is a lot of cpu cycles, on bits that are almost always
off.

Dynamic-debug excels at replacing such tests with NOOPS.  Classmaps
was devised to bring that 0-off-cost to drm's categories.

Classmaps-v1 went into the kernel in Sept 2022, in 2 chunks:
  b7b4eebdba7b..6ea3bf466ac6    # core dyndbg changes
  0406faf25fb1..ee7d633f2dfb    # drm adoption

Sadly DRM-CI found a regression during init with drm.debug=<initval>;
the static-keys underneath the drm-dbgs in drm.ko got enabled, but
those in drivers & helpers did not.

Root Problem:

DECLARE_DYNDBG_CLASSMAP defined the classmap, but it repeated use in
both core and drivers violated a K&R rule "define once, refer
afterwards".  This flaw resulted in a regression; with drm.debug=0xFF
boot arg, drm-core got enabled, but drivers,etc did not.

So in feb 2023, this resulted in:
commit bb2ff6c27bc9 ("drm: Disable dynamic debug as broken")

This patchset replaces it with DYNDBG_CLASSMAP_DEFINE (invoked once in
drm-core) and DYNDBG_CLASSMAP_USE (invoked repeatedly, in drivers &
helpers).

_DEFINE exports the classmap it creates (in drm.ko), other modules
_USE the classmap.  The _USE adds a record ref'g the _DEFINEd (&
exported) classmap, in a 2nd __dyndbg_class_users section.

So now at modprobe, dyndbg scans the new section after the 1st
__dyndbg_class_maps section, follows the linkage to the _DEFINEr
module, finds the (optional) kernel-param controlling the classmap,
examines its drm.debug=<initval>, and applies it to the module being
initialized.

Savings effect:

No real effort was made to quantify the savings; bit-tests are almost
unmeasurable individually, and [1] DRM_USE_DYNAMIC_DEBUG=y purposely
depends upon JUMP_LABEL to avoid its use where it cannot help.

That said, Ive booted one box with drm.debug=0x1ff on the boot-line,
with a script run by systemd to turn it off once boot completes. It
issued ~25k messages in that time.

Status-v11:

Now rebased on drm-misc-next, to allow grinding on DRM-CI.
https://gitlab.freedesktop.org/jim.cromie/kernel-drm-next-dd/-/pipelines/1622778

DRM-CI revealed 2 bugs, fixes now included:
1. drm_buddy had 32bit truncation error, causing unexpected pass
2. drm_printer_debug_fn disregarded drm.debug, spewing msgs, causing timeout.

Also tuned vblank* to avoid drm_debug_enabled() bit-test.
Probably should defer this until [1]=y is default. It isnt yet.

Up to this point, classmaps-v2 has been stuck, straddled on the fence
between drm and dynamic-debug; it needs a little help towards one side
or the other.  ISTM that drm is the actual user, and the acid-test of
its proper function; the dyndbg selftest recaps the 2-module scenario,
but this cannot finally prove anything absent drm.

Verification: DRM-CI Pipeline #1622778
  - KUnit (arm32, arm64, x86_64): All 621 tests PASSED.
    Specifically confirms the 64-bit truncation fixes in drm_buddy.
  - i915-CML: 370 tests PASSED, 0 FAIL.
    Confirms the quelling of the UART storm in shared print callbacks.
  - Overall: No regressions found in dyndbg or buddy allocator logic.

Future:

Lukas Bartosik has been patiently waiting for classmaps to land, so he
can add dyndbg-to-ftrace functionality.  This new feature will allow
steering any/all drm.debug messages to a private ftrace buffer, and
also to mix in other pr_debug messages.  Other ftrace streams could
also be sent there (possibly w/o code changes, tbd).  It could be a
big step towards a featureful crash-buffer.

Ive also cobbled together a maple-tree based replacement for the
__dyndbg_descriptors linker section; it copies the section contents
into 3 maple trees, then drops the section in __init.  By eliminating
the redundant columns (module, file, function) it reduces the storage
needed by 40%.  Early measurements show no net runtime costs, maple
trees make good furniture.

Given that progress, I also hacked up a maple-tree based dynamic
prefix cache.  Currently the dynamic-prefix is re-written for every
prefixed callsite; this can get expensive.  By careful keying (flags
on top, descriptor address underneath), a single prefix string can
span an entire range (smallest of enabled module, file, function).

That last one doesn't help DRM, cuz it doesnt use dynamic-prefixes,
but with this optimization in place, it could be adapted to do so.

Lastly, I have an RFC jump-label patchset that dramatically cuts IPIs,
currently 1 per pr_debug (to ~1/140), by sorting the (existing) queue,
rather than flushing it when a non-increasing patch-addr is queued.

I'd like to move forward with all these, but I anticipate a "fix
classmaps 1st" response.

Jim Cromie (65):
  dyndbg: fix NULL ptr on i386 due to section alignment
  dyndbg: factor ddebug_match_desc out from ddebug_change
  dyndbg: add stub macro for DECLARE_DYNDBG_CLASSMAP
  docs/dyndbg: update examples \012 to \n
  docs/dyndbg: explain flags parse 1st
  test-dyndbg: fixup CLASSMAP usage error
  dyndbg: reword "class unknown," to "class:_UNKNOWN_"
  dyndbg: make ddebug_class_param union members same size
  dyndbg: drop NUM_TYPE_ARRAY
  dyndbg: tweak pr_fmt to avoid expansion conflicts
  dyndbg: reduce verbose/debug clutter
  dyndbg: refactor param_set_dyndbg_classes and below
  dyndbg: tighten fn-sig of ddebug_apply_class_bitmap
  dyndbg: replace classmap list with a vector
  dyndbg: macrofy a 2-index for-loop pattern
  dyndbg,module: make proper substructs in _ddebug_info
  dyndbg: hoist classmap-filter-by-modname up to ddebug_add_module
  dyndbg: move mod_name down from struct ddebug_table to _ddebug_info
  dyndbg-API: remove DD_CLASS_TYPE_(DISJOINT|LEVEL)_NAMES and code
  selftests-dyndbg: add a dynamic_debug run_tests target
  dyndbg: change __dynamic_func_call_cls* macros into expressions
  dyndbg-API: replace DECLARE_DYNDBG_CLASSMAP
  dyndbg: detect class_id reservation conflicts
  dyndbg: check DYNAMIC_DEBUG_CLASSMAP_DEFINE args at compile-time
  dyndbg-test: change do_prints testpoint to accept a loopct
  dyndbg-API: promote DYNAMIC_DEBUG_CLASSMAP_PARAM to API
  dyndbg: treat comma as a token separator
  dyndbg: split multi-query strings with %
  selftests-dyndbg: add test_mod_submod
  dyndbg: resolve "protection" of class'd pr_debug
  dyndbg: add DYNAMIC_DEBUG_CLASSMAP_USE_(dd_class_name, offset)
  dyndbg: Harden classmap and callsite validation
  docs/dyndbg: add classmap info to howto
  drm: use correct ccflags-y spelling
  drm-dyndbg: adapt drm core to use dyndbg classmaps-v2
  drm-dyndbg: adapt DRM to invoke DYNAMIC_DEBUG_CLASSMAP_PARAM
  drm-print: modernize an archaic comment
  drm-print: fix config-dependent unused variable
  drm-dyndbg: DRM_CLASSMAP_USE in amdgpu driver
  drm-dyndbg: DRM_CLASSMAP_USE in i915 driver
  drm-dyndbg: DRM_CLASSMAP_USE in drm_crtc_helper
  drm-dyndbg: DRM_CLASSMAP_USE in drm_dp_helper
  drm-dyndbg: DRM_CLASSMAP_USE in nouveau
  drm-dyndbg: add DRM_CLASSMAP_USE to Xe driver
  drm-dyndbg: add DRM_CLASSMAP_USE to virtio_gpu
  drm-dyndbg: add DRM_CLASSMAP_USE to simpledrm
  drm-dyndbg: add DRM_CLASSMAP_USE to bochs
  drm-dyndbg: add DRM_CLASSMAP_USE to etnaviv
  drm-dyndbg: add DRM_CLASSMAP_USE to gma500 driver
  drm-dyndbg: add DRM_CLASSMAP_USE to radeon
  drm-dyndbg: add DRM_CLASSMAP_USE to vmwgfx driver
  drm-dyndbg: add DRM_CLASSMAP_USE to vkms driver
  drm-dyndbg: add DRM_CLASSMAP_USE to udl driver
  drm-dyndbg: add DRM_CLASSMAP_USE to mgag200 driver
  drm-dyndbg: add DRM_CLASSMAP_USE to the gud driver
  drm-dyndbg: add DRM_CLASSMAP_USE to the qxl driver
  drm-dyndbg: add DRM_CLASSMAP_USE to the drm_gem_shmem_helper driver
  accel: add -DDYNAMIC_DEBUG_MODULE to subdir-ccflags
  accel/ivpu: implement IVPU_DBG_* as a dyndbg classmap
  drm: restore CONFIG_DRM_USE_DYNAMIC_DEBUG un-BROKEN
  accel/ethosu: enable drm.debug control
  accel/rocket: enable drm.debug control
  drm_buddy: fix 64-bit truncation in power-of-2 rounding
  drm_print: fix drm_printer dynamic debug bypass
  drm_vblank: use dyndbg's static-key to avoid flag-check

 .../admin-guide/dynamic-debug-howto.rst       | 184 +++-
 MAINTAINERS                                   |   3 +-
 drivers/accel/Makefile                        |   7 +-
 drivers/accel/ethosu/ethosu_drv.c             |   3 +
 drivers/accel/ivpu/ivpu_drv.c                 |  27 +-
 drivers/accel/ivpu/ivpu_drv.h                 |  45 +-
 drivers/accel/rocket/rocket_gem.c             |   2 +
 drivers/gpu/buddy.c                           |   4 +-
 drivers/gpu/drm/Kconfig.debug                 |   1 -
 drivers/gpu/drm/Makefile                      |   3 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c       |  12 +-
 drivers/gpu/drm/display/drm_dp_helper.c       |  12 +-
 drivers/gpu/drm/drm_crtc_helper.c             |  12 +-
 drivers/gpu/drm/drm_gem_shmem_helper.c        |   1 +
 drivers/gpu/drm/drm_print.c                   |  38 +-
 drivers/gpu/drm/drm_vblank.c                  |  10 +-
 drivers/gpu/drm/etnaviv/etnaviv_drv.c         |   2 +
 drivers/gpu/drm/gma500/psb_drv.c              |   2 +
 drivers/gpu/drm/gud/gud_drv.c                 |   2 +
 drivers/gpu/drm/i915/i915_params.c            |  12 +-
 drivers/gpu/drm/mgag200/mgag200_drv.c         |   2 +
 drivers/gpu/drm/nouveau/nouveau_drm.c         |  12 +-
 drivers/gpu/drm/qxl/qxl_drv.c                 |   2 +
 drivers/gpu/drm/radeon/radeon_drv.c           |   2 +
 drivers/gpu/drm/sysfb/simpledrm.c             |   2 +
 drivers/gpu/drm/tiny/bochs.c                  |   2 +
 drivers/gpu/drm/udl/udl_main.c                |   2 +
 drivers/gpu/drm/virtio/virtgpu_drv.c          |   2 +
 drivers/gpu/drm/vkms/vkms_drv.c               |   2 +
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c           |   2 +
 drivers/gpu/drm/xe/xe_drm_client.c            |   2 +
 include/asm-generic/vmlinux.lds.h             |  19 +-
 include/drm/drm_print.h                       |  22 +-
 include/linux/dynamic_debug.h                 | 304 +++++--
 kernel/module/main.c                          |  15 +-
 lib/Kconfig.debug                             |  24 +-
 lib/Makefile                                  |   5 +
 lib/dynamic_debug.c                           | 814 ++++++++++++------
 lib/test_dynamic_debug.c                      | 198 +++--
 lib/test_dynamic_debug_submod.c               |  21 +
 tools/testing/selftests/Makefile              |   1 +
 .../testing/selftests/dynamic_debug/Makefile  |   9 +
 tools/testing/selftests/dynamic_debug/config  |   7 +
 .../dynamic_debug/dyndbg_selftest.sh          | 373 ++++++++
 44 files changed, 1665 insertions(+), 561 deletions(-)
 create mode 100644 lib/test_dynamic_debug_submod.c
 create mode 100644 tools/testing/selftests/dynamic_debug/Makefile
 create mode 100644 tools/testing/selftests/dynamic_debug/config
 create mode 100755 tools/testing/selftests/dynamic_debug/dyndbg_selftest.sh

-- 
2.53.0

Re: [PATCH v11 00/65] Fix DRM_USE_DYNAMIC_DEBUG=y

[Louis Chauvet]

Hi Jim,

Thanks for this new iteration, the whole series is in good shape, and I am
also interested in the "future work", you can count on me for
tests/reviews.

This series is very long, but early patches are trivials or tiny fixes
(4..11), can you apply them to reduce the overall size?

I will add this series on my kernel to test it on "everyday life", I will
report issues if I have.

In addition, can you move the accel and class offset implementation after
the rest? I think the "basic" part is good and could be merged soon (my
comments are minor). This way Lukas can start to work on a stable basis.

Thanks,
Louis Chauvet

-- 
Louis Chauvet <louis.chauvet@bootlin.com>

Re: [PATCH v11 00/65] Fix DRM_USE_DYNAMIC_DEBUG=y

[jim.cromie]

On Fri, Mar 20, 2026 at 10:41 AM Louis Chauvet
<louis.chauvet@bootlin.com> wrote:
>
> Hi Jim,
>
> Thanks for this new iteration, the whole series is in good shape, and I am
> also interested in the "future work", you can count on me for
> tests/reviews.
>
> This series is very long, but early patches are trivials or tiny fixes
> (4..11), can you apply them to reduce the overall size?
>

lib/ "belongs" to GregKH, so its his call.
IIRC, he has applied low-risk things case-by-case up until rc3,
I dont know his criteria for doing so, or his assessment of
how low-risk the initial patches are, but we're past that milepost,
so its a more-than-usual ask.


> I will add this series on my kernel to test it on "everyday life", I will
> report issues if I have.

yay. that should be helpful

> In addition, can you move the accel and class offset implementation after
> the rest? I think the "basic" part is good and could be merged soon (my
> comments are minor). This way Lukas can start to work on a stable basis.
>

They are already quite late in the stack, do you mean _USE_ after all
the drm patches ?

Im testing v12 now, it has a few noteworthy changes:
(I will work your feedback in soon).

1- Ive found a way to address this TODO-ish item:

    drm: restore CONFIG_DRM_USE_DYNAMIC_DEBUG un-BROKEN

    Time for some thorough CI.

    Also, the previous 18 patches could perhaps be replaced by a single
    invocation of DYNDBG_CLASSMAP_USE, from a C-file linked into all drm
    drivers & helpers.  I didn't find such a file, nor a drm-client
    linkage item in the Makefile.

+subdir-ccflags-$(CONFIG_DRM_USE_DYNAMIC_DEBUG)  +=
-DDYNAMIC_DEBUG_MODULE -include
$(srctree)/drivers/gpu/drm/drm_dyndbg_classmap_mgr.h

injects that new file into every compilation unit,
the file has a single call: DYNAMIC_DEBUG_CLASSMAP_USE(drm_debug_classes)
with a few #ifdefs.

the approach has the advantages of:
a- no 18 1-liner patches needed
b- no need for future drivers to call the macro explicitly (since its injected)

disadvantages:
c- redundant class-user records (1 per object), 315 extras in i915
d- mysterious, action-at-a-distance

a- is fixable, by introducing a MODULE_MAIN, added once per driver/helper
but thats not much different than adding the DRM_CLASSMAP_USE() itself.

2- Ive split up patch 1, the BOUNDED_SECTION_* patch
a- split BOUNDED_* macros to include/asm-generic/bounded_sections.lds.h
b- include it from vmlinux.lds.h
c- imitate include/asm-generic/codetag.lds.h --> dyndbg.lds.h
d- scripts/module.lds.S includes c), calls MOD_DYNDBG_SECTIONS()

this addresses some lost sections in DRM-drivers, the problem was exposed by 1

> Thanks,
> Louis Chauvet
>
> --
> Louis Chauvet <louis.chauvet@bootlin.com>