Skip to contentSkip to conversation

Optimize USDT discovery a little

PATCH In review [email protected] 2 participants 10 messages Jun 25, 2025 – Jul 23, 2025

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 Optimize USDT discovery a little

From: Eugene Loh <[email protected]>

We want to reduce the performance cost of USDT discovery if possible.

Specifically, a number of statements -- with probe descriptions such as "dtrace:::BEGIN" that could never specify USDT probes -- will not get their clause flags set with DT_CLSFLAG_USDT_EXCLUDE. So these statements get considered unnecessarily during periodic probe discovery.

Therefore:

*) Expand ignore_clause(dtp, n, uprp) to support the case uprp==NULL.
This case is independent of any knowledge of a specific underlying probe.

*) During probe discovery, check ignore_clause(dtp, i, NULL). This
sets the DT_CLSFLAG_USDT_[INCLUDE|EXCLUDE] flag and allows faster exclusion of statements that do not need to be reconsidered.

To take advantage of this optimization, users should specify providers. E.g., instead of "BEGIN" (which could conceivably be a USDT probe), specify "dtrace:::BEGIN".

---
 libdtrace/dt_prov_uprobe.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)
libdtrace/dt_prov_uprobe.c +10 -8
index cf5cfd431..5ba50b678 100644
@@ -504,7 +504,8 @@ clean_usdt_probes(dtrace_hdl_t *dtp)
 
 /*
  * Judge whether clause "n" could ever be called as a USDT probe
- * for this underlying probe.
+ * for this underlying probe.  We can pass uprp==NULL to see if
+ * the clause can be excluded for every probe.
  */
 static int
 ignore_clause(dtrace_hdl_t *dtp, int n, const dt_probe_t *uprp)
@@ -512,6 +513,9 @@ ignore_clause(dtrace_hdl_t *dtp, int n, const dt_probe_t *uprp)
 	dtrace_stmtdesc_t	*stp = dtp->dt_stmts[n];
 	dtrace_probedesc_t	*pdp = &stp->dtsd_ecbdesc->dted_probe;
 
+	if (stp == NULL)
+		return 1;
+
 	/*
 	 * Some clauses could never be called for a USDT probe,
 	 * regardless of the underlying probe uprp.  Cache this
@@ -525,7 +529,7 @@ ignore_clause(dtrace_hdl_t *dtp, int n, const dt_probe_t *uprp)
 		 * neither '*' nor a digit, it cannot be a USDT probe.
 		 */
 		if (len > 1) {
-			char	lastchar = pdp->prv[len - 1];
+			char	lastchar = (pdp->prv[0] != '\0' ? pdp->prv[len - 1] : '*');
 
 			if (lastchar != '*' && !isdigit(lastchar)) {
 				dt_stmt_clsflag_set(stp, DT_CLSFLAG_USDT_EXCLUDE);
@@ -555,6 +559,8 @@ ignore_clause(dtrace_hdl_t *dtp, int n, const dt_probe_t *uprp)
 	}
 	if (dt_stmt_clsflag_test(stp, DT_CLSFLAG_USDT_EXCLUDE) == 1)
 		return 1;
+	if (uprp == NULL)
+		return 0;
 
 	/*
 	 * If we cannot ignore this statement, try to use uprp.
@@ -751,13 +757,9 @@ static int discover(dtrace_hdl_t *dtp)
 	 */
 	memset(&pcb, 0, sizeof(dt_pcb_t));
 	for (i = 0; i < dtp->dt_stmt_nextid; i++) {
-		dtrace_stmtdesc_t *stp;
-
-		stp = dtp->dt_stmts[i];
-		if (stp == NULL)
+		if (ignore_clause(dtp, i, NULL))
 			continue;
-		if (dt_stmt_clsflag_test(stp, DT_CLSFLAG_USDT_EXCLUDE) != 1)
-			dt_pid_create_usdt_probes(&stp->dtsd_ecbdesc->dted_probe, dtp, &pcb);
+		dt_pid_create_usdt_probes(&dtp->dt_stmts[i]->dtsd_ecbdesc->dted_probe, dtp, &pcb);
 	}
 
 	return 0;
Signature · 1 line
2.43.5
#1 lore raw

PATCH v2 3/4 Sync up the version numbers

From: Eugene Loh <[email protected]>

DTrace has many version numbers -- e.g., for the release, packaging, and API.

In reality, the variations in numbering have become nearly meaningless:

- Packaging numbers -- like the Version in the dtrace*spec RPM spec file and the VERSION in GNUmakefile -- have basically been tracking the DTrace release since 2.0 anyway.

- Stability attributes for idents are haphazard. Generally, they are 1.0 (or sometimes other 1.x), and all the BPFs are 2.0. While this is generally accurate, it is not exactly robust, and idents are sometimes introduced or modified without careful regard to the version number. Further, the stability of user D scripts is likely to depend more on kernel variations -- e.g., the contents of available_filter_functions -- than on D changes.

- Version number updates are susceptible to mistakes, and so there have been version mismatches.

Bring the version numbers into sync for 2.0.3. Clean up the descriptions in dt_version.h.

---
 GNUmakefile                                  |  5 +-
 dtrace.spec                                  |  1 +
 libdtrace/dt_open.c                          |  4 ++
 libdtrace/dt_version.h                       | 54 ++++++++++++--------
 test/unittest/dtrace-util/tst.APIVersion.r   |  2 +-
 test/unittest/dtrace-util/tst.APIVersion.r.p |  6 +++
 test/unittest/options/tst.version.r          |  2 +-
 test/unittest/options/tst.version.sh         |  4 +-
 8 files changed, 52 insertions(+), 26 deletions(-)
 create mode 100755 test/unittest/dtrace-util/tst.APIVersion.r.p
GNUmakefile +3 -2
index d1e18bb1b..00269785b 100644
@@ -3,7 +3,7 @@
 # Build files in subdirectories are included by this file.
 #
 # Oracle Linux DTrace.
-# Copyright (c) 2011, 2024, Oracle and/or its affiliates. All rights reserved.
+# Copyright (c) 2011, 2025, Oracle and/or its affiliates. All rights reserved.
 # Licensed under the Universal Permissive License v 1.0 as shown at
 # http://oss.oracle.com/licenses/upl.
 
@@ -14,7 +14,8 @@
 SHELL = /bin/bash
 
 PROJECT := dtrace
-VERSION := 2.0.1
+# When updating version, see comments in dt_version.h.
+VERSION := 2.0.3
 
 # Verify supported hardware.
 
dtrace.spec +1 -0
index 6bb792acb..dd44a5b67 100644
@@ -72,6 +72,7 @@ Requires:     libdtrace-ctf >= 1.1.0
 BuildRequires: libdtrace-ctf-devel >= 1.1.0
 %endif
 Summary:      DTrace user interface.
+# When updating version, see comments in dt_version.h.
 Version:      2.0.3
 Release:      1%{?dist}
 Source:       dtrace-%{version}.tar.bz2
libdtrace/dt_open.c +4 -0
index 7d47cab34..6cb797df6 100644
@@ -60,6 +60,10 @@ const dt_version_t _dtrace_versions[] = {
 	DT_VERS_1_6_3,	/* D API 1.6.3 */
 	DT_VERS_1_6_4,	/* D API 1.6.4 */
 	DT_VERS_2_0,	/* D API 2.0 */
+	DT_VERS_2_0_1,	/* D API 2.0.1 */
+	DT_VERS_2_0_2,	/* D API 2.0.2 */
+	DT_VERS_2_0_3,	/* D API 2.0.3 */
+	/* When updating version, see comments in dt_version.h. */
 	0
 };
 
libdtrace/dt_version.h +34 -20

74 lines not on this page · raw message

test/unittest/dtrace-util/tst.APIVersion.r +1 -1
index 6bc7b9d72..02bf03150 100644
@@ -1 +1 @@
-dtrace: Oracle D 2.0
+dtrace: Oracle D 2.0.x
test/unittest/dtrace-util/tst.APIVersion.r.p new file +6 -0
new file mode 100755
index 000000000..32ec94df4
@@ -0,0 +1,6 @@
+#!/usr/bin/gawk -f
+
+# The test allows the version string to vary in micro number as well as
+# other suffixes (like "Beta").  The .r.p and .r files still need to be
+# updated for each minor number change.
+{ sub("^dtrace: Oracle D 2\\.0\\..*$", "dtrace: Oracle D 2.0.x"); print }
test/unittest/options/tst.version.r +1 -1
index 15010b3db..882e208ed 100644
@@ -1,2 +1,2 @@
-version is 2.0
+version is 2.0.x
 
test/unittest/options/tst.version.sh +2 -2
index ffffcdd8b..684120af8 100755
@@ -1,7 +1,7 @@
 #!/bin/bash
 #
 # Oracle Linux DTrace.
-# Copyright (c) 2023, Oracle and/or its affiliates. All rights reserved.
+# Copyright (c) 2023, 2025, Oracle and/or its affiliates. All rights reserved.
 # Licensed under the Universal Permissive License v 1.0 as shown at
 # http://oss.oracle.com/licenses/upl.
 #
@@ -9,7 +9,7 @@
 dtrace=$1
 
 myversion=`$dtrace $dt_flags -V | gawk '{ print $NF }'`
-echo version is $myversion
+echo version is $myversion | sed 's:2.0.[0-9]:2.0.x:'
 
 $dtrace $dt_flags -xversion=$myversion -qn 'BEGIN { exit(0) }'
 exit $?
Signature · 1 line
2.43.5
Reviewed-by on v2 3/4
#2 lore raw

Re: [DTrace-devel] [PATCH v2 3/4] Sync up the version numbers

On 25 Jun 2025, eugene loh verbalised:

Quoted 6 lines

From: Eugene Loh <[email protected]>

DTrace has many version numbers -- e.g., for the release, packaging, and API.

In reality, the variations in numbering have become nearly meaningless:

Tell me about it :(

Quoted 12 lines

- Packaging numbers -- like the Version in the dtrace*spec RPM spec file and the VERSION in GNUmakefile -- have basically been tracking the DTrace release since 2.0 anyway.

- Stability attributes for idents are haphazard. Generally, they are 1.0 (or sometimes other 1.x), and all the BPFs are 2.0. While this is generally accurate, it is not exactly robust, and idents are sometimes introduced or modified without careful regard to the version number. Further, the stability of user D scripts is likely to depend more on kernel variations -- e.g., the contents of available_filter_functions -- than on D changes.

Quoted 5 lines

- Version number updates are susceptible to mistakes, and so there have been version mismatches.

Bring the version numbers into sync for 2.0.3. Clean up the descriptions in dt_version.h.

So... this means all the ident version numbers change to 2? I doubt it :) but at least we *have* 2.0.[23] stability-related version numbers now.

modulo the tiny nit below.

Quoted 27 lines
---
 GNUmakefile                                  |  5 +-
 dtrace.spec                                  |  1 +
 libdtrace/dt_open.c                          |  4 ++
 libdtrace/dt_version.h                       | 54 ++++++++++++--------
 test/unittest/dtrace-util/tst.APIVersion.r   |  2 +-
 test/unittest/dtrace-util/tst.APIVersion.r.p |  6 +++
 test/unittest/options/tst.version.r          |  2 +-
 test/unittest/options/tst.version.sh         |  4 +-
 8 files changed, 52 insertions(+), 26 deletions(-)
 create mode 100755 test/unittest/dtrace-util/tst.APIVersion.r.p
GNUmakefile +1 -1
index d1e18bb1b..00269785b 100644
@@ -3,7 +3,7 @@
 # Build files in subdirectories are included by this file.
 #
 # Oracle Linux DTrace.
-# Copyright (c) 2011, 2024, Oracle and/or its affiliates. All rights reserved.
+# Copyright (c) 2011, 2025, Oracle and/or its affiliates. All rights reserved.
 # Licensed under the Universal Permissive License v 1.0 as shown at
 # http://oss.oracle.com/licenses/upl.
 

GNUmakefile line 18

Quoted 6 lines
GNUmakefile +2 -1
@@ -14,7 +14,8 @@
 SHELL = /bin/bash
 
 PROJECT := dtrace
-VERSION := 2.0.1
+# When updating version, see comments in dt_version.h.
+VERSION := 2.0.3

Oh dear :( this one isn't invisible to users...

Quoted 21 lines
libdtrace/dt_version.h +5 -7
index 0f47ea70d..5cfcab69a 100644
@@ -37,32 +37,28 @@ extern "C" {
  *
  * These #defines are used in identifier tables to fill in the version fields
  * associated with each identifier.  The DT_VERS_* macros declare the encoded
- * integer values of all versions used so far.  DT_VERS_STRING must be an ASCII
- * string that contains the latest version within it along with any suffixes
- * (e.g. Beta).  You must update DT_VERS_STRING when adding a new version,
- * and then add the new version to the _dtrace_versions[] array declared in
- * dt_open.c.
+ * integer values of all versions used so far.
  *
- * Refer to the Solaris Dynamic Tracing Guide Versioning chapter for an
- * explanation of these DTrace features and their values.
+ * The major number should be incremented when a fundamental change has been
+ * made that would affect all consumers, and would reflect sweeping changes
+ * to DTrace or the D language.
+ *

libdtrace/dt_version.h extern "C" { · line 55

Quoted 10 lines
libdtrace/dt_version.h +10 -0
+ * The minor number should be incremented when a change is introduced that
+ * could break scripts that had previously worked;  for example, adding a
+ * new built-in variable could break a script which was already using that
+ * identifier.
+ *
+ * The micro number should be changed when introducing functionality changes
+ * or major bug fixes that do not affect backward compatibility -- this is
+ * merely to make capabilities easily determined from the version number.
+ *
+ * Minor bugs do not require any modification to the version number.

Nice general rules! Now we just have to follow them :)

libdtrace/dt_version.h extern "C" { · line 95

Quoted 13 lines
libdtrace/dt_version.h +13 -0
+ * When the version number is updated, the following must be kept in sync:
+ *
+ *   libdtrace/dt_version.h    DT_VERS_STRING, an ASCII string that contains
+ *                             the latest version within it along with any
+ *                             suffixes (e.g. Beta)
+ *
+ *   libdtrace/dt_open.c       _dtrace_versions[]
+ *
+ *   dtrace.spec               Version
+ *
+ *   libdtrace/Build           libdtrace_VERSION
+ *
+ *   GNUmakefile               VERSION

I don't think that's quite true -- in particular, libdtrace_VERSION should only be bumped if it experiences incompatible API or ABI changes, whether or not DTrace itself has seen e.g. major language changes (which is why libdtrace_VERSION is not simply $(VERSION) already).

Signature · 1 line
NULL && (void)
#3 lore raw

PATCH v2 4/4 test: Add test for predefined preprocessor definitions

In reply to [email protected]

From: Eugene Loh <[email protected]>

Orabug: 28763074

---
 COMMANDLINE-OPTIONS                          |  10 +-
 test/unittest/preprocessor/tst.predefined.r  |   1 +
 test/unittest/preprocessor/tst.predefined.sh | 119 +++++++++++++++++++
 3 files changed, 125 insertions(+), 5 deletions(-)
 create mode 100644 test/unittest/preprocessor/tst.predefined.r
 create mode 100755 test/unittest/preprocessor/tst.predefined.sh
COMMANDLINE-OPTIONS +5 -5
index 40561af91..73be89b1f 100644
@@ -321,12 +321,12 @@ definitions are always specified and valid in all modes:
     * __sparcv9 (on SPARC® systems only when 64–bit programs are compiled)
     * __i386 (on x86 systems only when 32–bit programs are compiled)
     * __amd64 (on x86 systems only when 64–bit programs are compiled)
-    * _`uname -s` (for example, __Linux)
+    * __`uname -s` (for example, __Linux)
     * __SUNW_D=1
-    * _SUNW_D_VERSION=0x_MMmmmuuu (where MM is the Major release value
-      in hexadecimal, mmm is the Minor release value in hexadecimal,
-      and uuu is the Micro release value in hexadecimal; see Chapter
-      41, Versioning for more information about DTrace versioning)
+    * _SUNW_D_VERSION=(MM << 24 | mmm << 12 | uuu), where
+      MM is the Major release value
+      mmm is the Minor release value
+      uuu is the Micro release value
 
 -Z
 Permit probe descriptions that match zero probes. If the -Z option is
test/unittest/preprocessor/tst.predefined.r new file +1 -0
new file mode 100644
index 000000000..2e9ba477f
@@ -0,0 +1 @@
+success
test/unittest/preprocessor/tst.predefined.sh new file +119 -0

119 lines not on this page · raw message

Signature · 1 line
2.43.5
#4 lore raw

PATCH v2 2/2 Extend the USDT bit mask to multiple words

In reply to [email protected]

From: Eugene Loh <[email protected]>

Currently, USDT is limited to 64 probe descriptions since the underlying probe uses a 64-bit mask to decide which probes to execute.

Change to a multi-word bit mask that can be extended to however many probe descriptions there are.

Also, change the mask words to be 32-bit rather than 64-bit. The reason is that, commonly, there will be fewer than 32 probe descriptions. In this case, we shorten the value of the "USDT prids" BPF map from 16 bytes

        uint32_t        prid;
        long long       mask[1];
down to 8 bytes
        uint32_t        prid;
        uint32_t        mask[1];

(The second member is smaller and no longer costs extra padding.)

We also add an

        extern int usdt_prids_map_val_extra_bytes;

to denote how many extra bytes will be needed for the extended mask. This value is computed by usdt_prids_map_val_extra_bytes_init(). Currently, this function is awkwardly called in gmap_create_usdt(), just before the value is needed. Such a call to a provider-specific function is clumsy, but there are no other calls to the provider between compilation (where the number of statements is determined) and this map creation.

---
 libdtrace/dt_bpf.c                            |   6 +-
 libdtrace/dt_bpf_maps.h                       |   5 +-
 libdtrace/dt_prov_uprobe.c                    |  81 ++++++++---
 .../unittest/usdt/tst.manyprobedescriptions.r |   1 +
 .../usdt/tst.manyprobedescriptions.sh         |  64 +++++++++
 .../usdt/tst.manyprobedescriptions2.r         |   1 +
 .../usdt/tst.manyprobedescriptions2.sh        | 127 ++++++++++++++++++
 7 files changed, 261 insertions(+), 24 deletions(-)
 create mode 100644 test/unittest/usdt/tst.manyprobedescriptions.r
 create mode 100755 test/unittest/usdt/tst.manyprobedescriptions.sh
 create mode 100644 test/unittest/usdt/tst.manyprobedescriptions2.r
 create mode 100755 test/unittest/usdt/tst.manyprobedescriptions2.sh
libdtrace/dt_bpf.c +5 -1
index ddd849d0b..029d5fcdb 100644
@@ -967,6 +967,7 @@ gmap_create_probes(dtrace_hdl_t *dtp)
 	return 0;
 }
 
+void usdt_prids_map_val_extra_bytes_init(dtrace_hdl_t *dtp);
 /*
  * Create the 'usdt_names' and 'usdt_prids' BPF maps.
  *
@@ -992,8 +993,11 @@ gmap_create_usdt(dtrace_hdl_t *dtp)
 	if (dtp->dt_usdt_namesmap_fd == -1)
 		return -1;
 
+	usdt_prids_map_val_extra_bytes_init(dtp);
+
 	dtp->dt_usdt_pridsmap_fd = create_gmap(dtp, "usdt_prids", BPF_MAP_TYPE_HASH,
-	    sizeof(usdt_prids_map_key_t), sizeof(usdt_prids_map_val_t), nusdtprobes);
+	    sizeof(usdt_prids_map_key_t),
+	    sizeof(usdt_prids_map_val_t) + usdt_prids_map_val_extra_bytes, nusdtprobes);
 	if (dtp->dt_usdt_pridsmap_fd == -1)
 		return -1;
 
libdtrace/dt_bpf_maps.h +3 -2
index 884dc3983..bdb20c9f2 100644
@@ -1,6 +1,6 @@
 /*
  * Oracle Linux DTrace.
- * Copyright (c) 2019, 2024, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2019, 2025, Oracle and/or its affiliates. All rights reserved.
  * Licensed under the Universal Permissive License v 1.0 as shown at
  * http://oss.oracle.com/licenses/upl.
  */
@@ -48,8 +48,9 @@ typedef struct usdt_prids_map_key {
 } usdt_prids_map_key_t;
 typedef struct usdt_prids_map_val {
 	uint32_t	prid;		/* should be dtrace_id_t, sys/dtrace_types.h */
-	long long	mask;
+	uint32_t	mask[1];
 } usdt_prids_map_val_t;
+extern int usdt_prids_map_val_extra_bytes;
 
 #ifdef  __cplusplus
 }
libdtrace/dt_prov_uprobe.c +60 -21

188 lines not on this page · raw message

test/unittest/usdt/tst.manyprobedescriptions.r new file +1 -0
new file mode 100644
index 000000000..2e9ba477f
@@ -0,0 +1 @@
+success
test/unittest/usdt/tst.manyprobedescriptions.sh new file +64 -0

64 lines not on this page · raw message

test/unittest/usdt/tst.manyprobedescriptions2.r new file +1 -0
new file mode 100644
index 000000000..2e9ba477f
@@ -0,0 +1 @@
+success
test/unittest/usdt/tst.manyprobedescriptions2.sh new file +127 -0

127 lines not on this page · raw message

Signature · 1 line
2.43.5
#5 lore raw

Re: [PATCH v2 2/2] Extend the USDT bit mask to multiple words

On 25 Jun 2025, eugene loh outgrape:

Quoted 7 lines

From: Eugene Loh <[email protected]>

Currently, USDT is limited to 64 probe descriptions since the underlying probe uses a 64-bit mask to decide which probes to execute.

Change to a multi-word bit mask that can be extended to however many probe descriptions there are.

Oh yeah!!!!

Not quite happy enough with this one to r-b it, but it's sooo close... :)

Quoted 9 lines

Also, change the mask words to be 32-bit rather than 64-bit. The reason is that, commonly, there will be fewer than 32 probe descriptions. In this case, we shorten the value of the "USDT prids" BPF map from 16 bytes

        uint32_t        prid;
        long long       mask[1];
down to 8 bytes
        uint32_t        prid;
        uint32_t        mask[1];

(The second member is smaller and no longer costs extra padding.)

Nice!

Quoted 3 lines

We also add an

        extern int usdt_prids_map_val_extra_bytes;

to denote how many extra bytes will be needed for the extended mask.

Ugh. Is this really the best we can do? A global variable referenced via extern? Can't we put it in the dtp or anything? This will fail as soon as we have more than one dtrace handle... and yes I now that's not exactly tested and probably doesn't work, but we shouldn't break it worse when keeping it working as well as it is now is so simple.

libdtrace/dt_bpf_maps.h typedef struct usdt_prids_map_key { · line 53

Quoted 8 lines
libdtrace/dt_bpf_maps.h +2 -1
@@ -48,8 +48,9 @@ typedef struct usdt_prids_map_key {
 } usdt_prids_map_key_t;
 typedef struct usdt_prids_map_val {
 	uint32_t	prid;		/* should be dtrace_id_t, sys/dtrace_types.h */
-	long long	mask;
+	uint32_t	mask[1];
 } usdt_prids_map_val_t;
+extern int usdt_prids_map_val_extra_bytes;

Note that this is not how you're supposed to implement VLAs: [1] is extremely obsolete (as in, pre-C99) and very deprecated at this point. [] is the recommended approach, if you can live with its restrictions (and I think we can here: all we need to do is remove the extra in extra_bytes and *always* allocate at least one byte, which actually simplifies the code in usdt_prids_map_val_extra_bytes(). [] does include the size of necessary padding: mask will start at precisely sizeof(usdt_prids_map_val_t).)

libdtrace/dt_prov_uprobe.c typedef struct list_key { · line 306

Quoted 9 lines
libdtrace/dt_prov_uprobe.c +1 -0
index 5ba50b678..65413cba6 100644
@@ -303,6 +303,8 @@ typedef struct list_key {
 	usdt_prids_map_key_t	key;
 } list_key_t;
 
+int usdt_prids_map_val_extra_bytes;

Ugh!

libdtrace/dt_prov_uprobe.c clean_usdt_probes(dtrace_hdl_t *dtp) · line 408

Quoted 9 lines
libdtrace/dt_prov_uprobe.c +1 -1
 static const dtrace_pattr_t	pattr = {
 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA },
 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
@@ -403,7 +405,7 @@ clean_usdt_probes(dtrace_hdl_t *dtp)
 	int			fdprids = dtp->dt_usdt_pridsmap_fd;
 	int			fdnames = dtp->dt_usdt_namesmap_fd;
 	usdt_prids_map_key_t	key, nxt;
-	usdt_prids_map_val_t	val;
+	usdt_prids_map_val_t	*val = alloca(sizeof(usdt_prids_map_val_t) + usdt_prids_map_val_extra_bytes);

(this bit will need changing slightly if you use [].)

libdtrace/dt_prov_uprobe.c static void usdt_error(dt_pcb_t *pcb, const char *fmt, ...) · line 594

Quoted 2 lines
libdtrace/dt_prov_uprobe.c +1 -0
+void usdt_prids_map_val_extra_bytes_init(dtrace_hdl_t *dtp) {

Wrongly-placed {.

libdtrace/dt_prov_uprobe.c static void usdt_error(dt_pcb_t *pcb, const char *fmt, ...) · line 603

Quoted 9 lines
libdtrace/dt_prov_uprobe.c +9 -0
+	int i, n = 0, w = sizeof(((usdt_prids_map_val_t *)0)->mask[0]);
+
+	/* Count how many statements cannot be ignored, regardless of uprp. */
+	for (i = 0; i < dtp->dt_stmt_nextid; i++) {
+		dtrace_stmtdesc_t *stp;
+
+		stp = dtp->dt_stmts[i];
+		if (stp == NULL || ignore_clause(dtp, i, NULL))
+			continue;

I was about to complain about the cost of all these ignore_clause()s, but of course this is only invoked once per run...

libdtrace/dt_prov_uprobe.c static void usdt_error(dt_pcb_t *pcb, const char *fmt, ...) · line 612

Quoted 8 lines
libdtrace/dt_prov_uprobe.c +8 -0
+		n++;
+	}
+
+	/* Determine how many bytes are needed for this many bits. */
+	n = (n + CHAR_BIT - 1) / CHAR_BIT;
+
+	/* Determine how many words are needed for this many bytes. */
+	n = (n + w - 1) / w;

Thank God you commented this -- it took me ages to realise how it works. It feels like it should be a standard idiom, but I've never encountered it before...

libdtrace/dt_prov_uprobe.c static void usdt_error(dt_pcb_t *pcb, const char *fmt, ...) · line 618

Quoted 5 lines
libdtrace/dt_prov_uprobe.c +5 -0
+	/* Determine how many extra bytes are needed. */
+	if (n > 1)
+		usdt_prids_map_val_extra_bytes = (n - 1) * w;
+	else
+		usdt_prids_map_val_extra_bytes = 0;

If you use a [] array, this can simply become

usdt_prids_map_val_bytes = n * w;

Quoted 49 lines
libdtrace/dt_prov_uprobe.c +12 -5
+}
+
 static int add_probe_uprobe(dtrace_hdl_t *dtp, dt_probe_t *prp)
 {
 	dtrace_difo_t   *dp;
@@ -650,6 +679,7 @@ static int add_probe_usdt(dtrace_hdl_t *dtp, dt_probe_t *prp)
 	int				fd = dtp->dt_usdt_namesmap_fd;
 	pid_t				pid;
 	list_probe_t			*pup;
+	usdt_prids_map_val_t		*val;
 
 	/* Add probe name elements to usdt_names map. */
 	p = probnam;
@@ -685,11 +715,11 @@ static int add_probe_usdt(dtrace_hdl_t *dtp, dt_probe_t *prp)
 	}
 
 	/* Add prid and bit mask to usdt_prids map. */
+	val = alloca(sizeof(usdt_prids_map_val_t) + usdt_prids_map_val_extra_bytes);
 	for (pup = prp->prv_data; pup != NULL; pup = dt_list_next(pup)) {
 		dt_probe_t		*uprp = pup->probe;
-		long long		mask = 0, bit = 1;
+		uint32_t		iword = 0, mask = 0, bit = 1;
 		usdt_prids_map_key_t	key;
-		usdt_prids_map_val_t	val;
 		dt_uprobe_t		*upp = uprp->prv_data;
 
 		/*
@@ -707,11 +737,15 @@ static int add_probe_usdt(dtrace_hdl_t *dtp, dt_probe_t *prp)
 				dtrace_stmtdesc_t *stp;
 
 				stp = dtp->dt_stmts[n];
-				if (stp == NULL)
+				if (stp == NULL || ignore_clause(dtp, n, uprp))
 					continue;
 
-				if (ignore_clause(dtp, n, uprp))
-					continue;
+				if (bit == 0) {
+					val->mask[iword] = mask;
+					mask = 0;
+					iword++;
+					bit = 1;
+				}
 
 				if (dt_gmatch(prp->desc->prv, stp->dtsd_ecbdesc->dted_probe.prv) &&
 				    dt_gmatch(prp->desc->mod, stp->dtsd_ecbdesc->dted_probe.mod) &&
@@ -726,11 +760,11 @@ static int add_probe_usdt(dtrace_hdl_t *dtp, dt_probe_t *prp)
 		key.pid = pid;
 		key.uprid = uprp->desc->id;

libdtrace/dt_prov_uprobe.c static int add_probe_usdt(dtrace_hdl_t *dtp, dt_probe_t *prp) · line 768

Quoted 10 lines
libdtrace/dt_prov_uprobe.c +3 -3
 
-		val.prid = prp->desc->id;
-		val.mask = mask;
+		val->prid = prp->desc->id;
+		val->mask[iword] = mask;
 
 		// FIXME Check return value, but how should errors be handled?
-		dt_bpf_map_update(dtp->dt_usdt_pridsmap_fd, &key, &val);
+		dt_bpf_map_update(dtp->dt_usdt_pridsmap_fd, &key, val);
 	}

... am I missing something, or do you never modify "mask" (the local variable) anywhere in this function? So it ends up always zero...

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl) · line 1488

Quoted 6 lines
libdtrace/dt_prov_uprobe.c +1 -1
@@ -1451,7 +1485,7 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
 	const list_probe_t	*pop;
 	uint_t			lbl_exit = pcb->pcb_exitlbl;
 	dt_ident_t		*usdt_prids = dt_dlib_get_map(dtp, "usdt_prids");
-	int			n;
+	int			n, ibit, w = CHAR_BIT * sizeof(((usdt_prids_map_val_t *)0)->mask[0]);

We could really do with better variable names than "n" and "w". w seems to be the bit-width of a single word: word_width? width?

Quoted 11 lines
libdtrace/dt_prov_uprobe.c +2 -1
@@ -1538,7 +1572,8 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
 	 */
 	assert(sizeof(usdt_prids_map_key_t) <= DT_STK_SLOT_SZ);
 	emit(dlp,  BPF_STORE(BPF_W, BPF_REG_FP, DT_TRAMP_SP_SLOT(0), BPF_REG_0));
-	emit(dlp,  BPF_STORE_IMM(BPF_W, BPF_REG_FP, DT_TRAMP_SP_SLOT(0) + (int)sizeof(pid_t), uprp->desc->id));
+	emit(dlp,  BPF_STORE_IMM(BPF_W, BPF_REG_FP,
+		   DT_TRAMP_SP_SLOT(0) + (int)sizeof(pid_t), uprp->desc->id));
 	dt_cg_xsetx(dlp, usdt_prids, DT_LBL_NONE, BPF_REG_1, usdt_prids->di_id);
 	emit(dlp,  BPF_MOV_REG(BPF_REG_2, BPF_REG_FP));
 	emit(dlp,  BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, DT_TRAMP_SP_SLOT(0)));

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl) · line 1611

Quoted 7 lines
libdtrace/dt_prov_uprobe.c +2 -2
@@ -1572,8 +1607,8 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
 	emit(dlp,  BPF_LOAD(BPF_W, BPF_REG_1, BPF_REG_0, 0));
 	emit(dlp,  BPF_STORE(BPF_W, BPF_REG_7, DMST_PRID, BPF_REG_1));
 
-	/* Read the bit mask from the table lookup in %r6. */    // FIXME someday, extend this past 64 bits
-	emit(dlp,  BPF_LOAD(BPF_DW, BPF_REG_6, BPF_REG_0, offsetof(usdt_prids_map_val_t, mask)));
+	/* Store the value key for reuse. */
+	emit(dlp,  BPF_STORE(BPF_DW, BPF_REG_FP, DT_TRAMP_SP_SLOT(0), BPF_REG_0));

This all makes sense...

Quoted 10 lines
libdtrace/dt_prov_uprobe.c +1 -1
@@ -1587,21 +1622,24 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
 	/*
 	 * Hold the bit mask in %r6 between clause calls.
 	 */
-	for (n = 0; n < dtp->dt_stmt_nextid; n++) {
+	for (ibit = n = 0; n < dtp->dt_stmt_nextid; n++) {
 		dtrace_stmtdesc_t *stp;
 		dt_ident_t	*idp;
 		uint_t		lbl_next;
 

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl) · line 1635

Quoted 10 lines
libdtrace/dt_prov_uprobe.c +1 -4
 		stp = dtp->dt_stmts[n];
-		if (stp == NULL)
-			continue;
-
-		if (ignore_clause(dtp, n, uprp))
+		if (stp == NULL || ignore_clause(dtp, n, uprp))
 			continue;
 
 		idp = stp->dtsd_clause;
 		lbl_next = dt_irlist_label(dlp);

... as does this (although we are now calling ignore_clause() at least twice for every clause, I think three times)...

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl) · line 1640

Quoted 4 lines
libdtrace/dt_prov_uprobe.c +4 -0
+		/* Load the next word of the bit mask into %r6. */
+		if (ibit % w == 0) {
+			emit(dlp,  BPF_LOAD(BPF_DW, BPF_REG_0, BPF_REG_FP, DT_TRAMP_SP_SLOT(0)));
+			emit(dlp,  BPF_LOAD(BPF_W, BPF_REG_6, BPF_REG_0, offsetof(usdt_prids_map_val_t, mask[ibit / w])));

ok, so this uses the 'w' thing to get an array offset out of a bit count, and then you can just use the existing repated-shift code. Should work.

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl) · line 1670

Quoted 10 lines
libdtrace/dt_prov_uprobe.c +3 -0
+		}
+
 		/* If the lowest %r6 bit is 0, skip over this clause. */
 		emit(dlp,  BPF_MOV_REG(BPF_REG_1, BPF_REG_6));
 		emit(dlp,  BPF_ALU64_IMM(BPF_AND, BPF_REG_1, 1));
@@ -1629,6 +1667,7 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
 
 		/* Right-shift %r6. */
 		emit(dlp,  BPF_ALU64_IMM(BPF_RSH, BPF_REG_6, 1));
+		ibit++;

... so, this isn't in the loop conditionl because the values are only filled for non-ignored, non-NULL statements. This assumption is duplicated in two places (here and in usdt_prids_map_val_extra_bytes()) and if it ever drifts we get uninitialized rubbish or buffer overruns. Should it be centralized somewhere, somehow? At least it needs a comment.

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)

Quoted 4 lines
libdtrace/dt_prov_uprobe.c +3 -0
+# We stick 80 probe descriptions in each of 3 scripts to test
+# USDT's ability to handle hundreds of probe descriptions.
 
+++ b/test/unittest/usdt/tst.manyprobedescriptions2.sh

... maybe manyprobedescriptions-bytesize.sh :)

Quoted 20 lines
libdtrace/dt_prov_uprobe.c +19 -0
@@ -0,0 +1,127 @@
+#!/bin/bash
+#
+# Oracle Linux DTrace.
+# Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+# Licensed under the Universal Permissive License v 1.0 as shown at
+# http://oss.oracle.com/licenses/upl.
+
+# This test uses many probes and probe descriptions.  Therefore, the
+# number of BPF programs to load into the kernel -- dt_bpf_load_prog()
+# calling prp->prov->impl->load_prog(), which is dt_bpf_prog_load() --
+# and the duration of each load are both increasing.
+# @@timeout: 400
+
+dtrace=$1
+
+DIRNAME="$tmpdir/usdt-many_probe_descriptions2.$$.$RANDOM"
+mkdir -p $DIRNAME
+cd $DIRNAME
+

libdtrace/dt_prov_uprobe.c line 29

Quoted 10 lines
libdtrace/dt_prov_uprobe.c +10 -0
+# Set the lists.
+# - The probes will be foo$x$y.
+# - The probe descriptions will be foo$x* and foo*$y, for each $d.
+# So if there are nx items in xlist, ny in ylist, and nd in dlist,
+# - there will be roughly nx*ny probes
+# - there will be roughly (nx+ny)*nd probe descriptions
+
+xlist="a b c d e f g h i j k l m"
+ylist="n o p q r s t u v w x y z"
+dlist="0 1 2 3 4 5 6 7 8"

... and with 8... maybe we should also test 0, 1, 7 and 9, checking all the boundary conditions (and refactor things so we don't need to copy all of this every time).

(You'll only really see things go wrong under valgrind, but that's fine, valgrinding should be routine before releases anyway.)

Signature · 1 line
NULL && (void)
author
#6 lore raw

Re: [PATCH v2 2/2] Extend the USDT bit mask to multiple words

Quoted 24 lines · Nick Alcock

On 25 Jun 2025, eugene loh outgrape:

Quoted 7 lines

From: Eugene Loh <[email protected]>

Currently, USDT is limited to 64 probe descriptions since the underlying probe uses a 64-bit mask to decide which probes to execute.

Change to a multi-word bit mask that can be extended to however many probe descriptions there are.

Oh yeah!!!!

Not quite happy enough with this one to r-b it, but it's sooo close... :)

Quoted 9 lines

Also, change the mask words to be 32-bit rather than 64-bit. The reason is that, commonly, there will be fewer than 32 probe descriptions. In this case, we shorten the value of the "USDT prids" BPF map from 16 bytes

         uint32_t        prid;
         long long       mask[1];
down to 8 bytes
         uint32_t        prid;
         uint32_t        mask[1];

(The second member is smaller and no longer costs extra padding.)

Nice!

Quoted 8 lines
Quoted 3 lines

We also add an

         extern int usdt_prids_map_val_extra_bytes;

to denote how many extra bytes will be needed for the extended mask.

Ugh. Is this really the best we can do? A global variable referenced via extern? Can't we put it in the dtp or anything? This will fail as soon as we have more than one dtrace handle... and yes I now that's not exactly tested and probably doesn't work, but we shouldn't break it worse when keeping it working as well as it is now is so simple.

Okay, changed.

Quoted 8 lines

libdtrace/dt_bpf_maps.h typedef struct usdt_prids_map_key { · line 53

Quoted 8 lines
libdtrace/dt_bpf_maps.h +2 -1
@@ -48,8 +48,9 @@ typedef struct usdt_prids_map_key {
  } usdt_prids_map_key_t;
  typedef struct usdt_prids_map_val {
  	uint32_t	prid;		/* should be dtrace_id_t, sys/dtrace_types.h */
-	long long	mask;
+	uint32_t	mask[1];
  } usdt_prids_map_val_t;
+extern int usdt_prids_map_val_extra_bytes;
Quoted 8 lines

Note that this is not how you're supposed to implement VLAs: [1] is extremely obsolete (as in, pre-C99) and very deprecated at this point. [] is the recommended approach, if you can live with its restrictions (and I think we can here: all we need to do is remove the extra in extra_bytes and *always* allocate at least one byte, which actually simplifies the code in usdt_prids_map_val_extra_bytes(). [] does include the size of necessary padding: mask will start at precisely sizeof(usdt_prids_map_val_t).)

Also changed.

Quoted 10 lines

libdtrace/dt_prov_uprobe.c typedef struct list_key { · line 306

Quoted 9 lines
libdtrace/dt_prov_uprobe.c +1 -0
index 5ba50b678..65413cba6 100644
@@ -303,6 +303,8 @@ typedef struct list_key {
  	usdt_prids_map_key_t	key;
  } list_key_t;
 
+int usdt_prids_map_val_extra_bytes;

Ugh!

Ditto.

Quoted 12 lines

libdtrace/dt_prov_uprobe.c clean_usdt_probes(dtrace_hdl_t *dtp) · line 408

Quoted 9 lines
libdtrace/dt_prov_uprobe.c +1 -1
  static const dtrace_pattr_t	pattr = {
  { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA },
  { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
@@ -403,7 +405,7 @@ clean_usdt_probes(dtrace_hdl_t *dtp)
  	int			fdprids = dtp->dt_usdt_pridsmap_fd;
  	int			fdnames = dtp->dt_usdt_namesmap_fd;
  	usdt_prids_map_key_t	key, nxt;
-	usdt_prids_map_val_t	val;
+	usdt_prids_map_val_t	*val = alloca(sizeof(usdt_prids_map_val_t) + usdt_prids_map_val_extra_bytes);

(this bit will need changing slightly if you use [].)

Quoted 1 line
Quoted 2 lines
Quoted 1 line

+void usdt_prids_map_val_extra_bytes_init(dtrace_hdl_t *dtp) {

Wrongly-placed {.

Got it.

Quoted 24 lines

libdtrace/dt_prov_uprobe.c static void usdt_error(dt_pcb_t *pcb, const char *fmt, ...) · line 603

Quoted 9 lines
libdtrace/dt_prov_uprobe.c +9 -0
+	int i, n = 0, w = sizeof(((usdt_prids_map_val_t *)0)->mask[0]);
+
+	/* Count how many statements cannot be ignored, regardless of uprp. */
+	for (i = 0; i < dtp->dt_stmt_nextid; i++) {
+		dtrace_stmtdesc_t *stp;
+
+		stp = dtp->dt_stmts[i];
+		if (stp == NULL || ignore_clause(dtp, i, NULL))
+			continue;

I was about to complain about the cost of all these ignore_clause()s, but of course this is only invoked once per run...

libdtrace/dt_prov_uprobe.c static void usdt_error(dt_pcb_t *pcb, const char *fmt, ...) · line 612

Quoted 8 lines
libdtrace/dt_prov_uprobe.c +8 -0
+		n++;
+	}
+
+	/* Determine how many bytes are needed for this many bits. */
+	n = (n + CHAR_BIT - 1) / CHAR_BIT;
+
+	/* Determine how many words are needed for this many bytes. */
+	n = (n + w - 1) / w;

Thank God you commented this -- it took me ages to realise how it works. It feels like it should be a standard idiom, but I've never encountered it before...

Quoted 8 lines

libdtrace/dt_prov_uprobe.c static void usdt_error(dt_pcb_t *pcb, const char *fmt, ...) · line 618

Quoted 5 lines
libdtrace/dt_prov_uprobe.c +5 -0
+	/* Determine how many extra bytes are needed. */
+	if (n > 1)
+		usdt_prids_map_val_extra_bytes = (n - 1) * w;
+	else
+		usdt_prids_map_val_extra_bytes = 0;

If you use a [] array, this can simply become

usdt_prids_map_val_bytes = n * w;

Earlier you wrote, "*always* allocate at least one byte, which actually simplifies the code in usdt_prids_map_val_extra_bytes()." So I'm missing what you mean in the case n=0.

Quoted 55 lines
Quoted 55 lines
libdtrace/dt_prov_uprobe.c +14 -7
+}
+
  static int add_probe_uprobe(dtrace_hdl_t *dtp, dt_probe_t *prp)
  {
  	dtrace_difo_t   *dp;
@@ -650,6 +679,7 @@ static int add_probe_usdt(dtrace_hdl_t *dtp, dt_probe_t *prp)
  	int				fd = dtp->dt_usdt_namesmap_fd;
  	pid_t				pid;
  	list_probe_t			*pup;
+	usdt_prids_map_val_t		*val;
 
  	/* Add probe name elements to usdt_names map. */
  	p = probnam;
@@ -685,11 +715,11 @@ static int add_probe_usdt(dtrace_hdl_t *dtp, dt_probe_t *prp)
  	}
 
  	/* Add prid and bit mask to usdt_prids map. */
+	val = alloca(sizeof(usdt_prids_map_val_t) + usdt_prids_map_val_extra_bytes);
  	for (pup = prp->prv_data; pup != NULL; pup = dt_list_next(pup)) {
  		dt_probe_t		*uprp = pup->probe;
-		long long		mask = 0, bit = 1;
+		uint32_t		iword = 0, mask = 0, bit = 1;
  		usdt_prids_map_key_t	key;
-		usdt_prids_map_val_t	val;
  		dt_uprobe_t		*upp = uprp->prv_data;
 
  		/*
@@ -707,11 +737,15 @@ static int add_probe_usdt(dtrace_hdl_t *dtp, dt_probe_t *prp)
  				dtrace_stmtdesc_t *stp;
 
  				stp = dtp->dt_stmts[n];
-				if (stp == NULL)
+				if (stp == NULL || ignore_clause(dtp, n, uprp))
  					continue;
 
-				if (ignore_clause(dtp, n, uprp))
-					continue;
+				if (bit == 0) {
+					val->mask[iword] = mask;
+					mask = 0;
+					iword++;
+					bit = 1;
+				}
 
  				if (dt_gmatch(prp->desc->prv, stp->dtsd_ecbdesc->dted_probe.prv) &&
  				    dt_gmatch(prp->desc->mod, stp->dtsd_ecbdesc->dted_probe.mod) &&
@@ -726,11 +760,11 @@ static int add_probe_usdt(dtrace_hdl_t *dtp, dt_probe_t *prp)
  		key.pid = pid;
  		key.uprid = uprp->desc->id;
 
-		val.prid = prp->desc->id;
-		val.mask = mask;
+		val->prid = prp->desc->id;
+		val->mask[iword] = mask;
 
Quoted 6 lines

libdtrace/dt_prov_uprobe.c static int add_probe_usdt(dtrace_hdl_t *dtp, dt_probe_t *prp) · line 768

Quoted 4 lines
libdtrace/dt_prov_uprobe.c +1 -1
  		// FIXME Check return value, but how should errors be handled?
-		dt_bpf_map_update(dtp->dt_usdt_pridsmap_fd, &key, &val);
+		dt_bpf_map_update(dtp->dt_usdt_pridsmap_fd, &key, val);
  	}

... am I missing something, or do you never modify "mask" (the local variable) anywhere in this function? So it ends up always zero...

Perhaps you're missing something?  mask |= bit

Quoted 8 lines

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl) · line 1488

Quoted 6 lines
libdtrace/dt_prov_uprobe.c +1 -1
@@ -1451,7 +1485,7 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
  	const list_probe_t	*pop;
  	uint_t			lbl_exit = pcb->pcb_exitlbl;
  	dt_ident_t		*usdt_prids = dt_dlib_get_map(dtp, "usdt_prids");
-	int			n;
+	int			n, ibit, w = CHAR_BIT * sizeof(((usdt_prids_map_val_t *)0)->mask[0]);

We could really do with better variable names than "n" and "w". w seems to be the bit-width of a single word: word_width? width?

Hopefully, this is good enough:  "w" for "word width" or "width"? Mimic other similar code closely.  Seemed like a reasonable choice to me.

Quoted 58 lines
Quoted 11 lines
libdtrace/dt_prov_uprobe.c +2 -1
@@ -1538,7 +1572,8 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
  	 */
  	assert(sizeof(usdt_prids_map_key_t) <= DT_STK_SLOT_SZ);
  	emit(dlp,  BPF_STORE(BPF_W, BPF_REG_FP, DT_TRAMP_SP_SLOT(0), BPF_REG_0));
-	emit(dlp,  BPF_STORE_IMM(BPF_W, BPF_REG_FP, DT_TRAMP_SP_SLOT(0) + (int)sizeof(pid_t), uprp->desc->id));
+	emit(dlp,  BPF_STORE_IMM(BPF_W, BPF_REG_FP,
+		   DT_TRAMP_SP_SLOT(0) + (int)sizeof(pid_t), uprp->desc->id));
  	dt_cg_xsetx(dlp, usdt_prids, DT_LBL_NONE, BPF_REG_1, usdt_prids->di_id);
  	emit(dlp,  BPF_MOV_REG(BPF_REG_2, BPF_REG_FP));
  	emit(dlp,  BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, DT_TRAMP_SP_SLOT(0)));

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl) · line 1611

Quoted 7 lines
libdtrace/dt_prov_uprobe.c +2 -2
@@ -1572,8 +1607,8 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
  	emit(dlp,  BPF_LOAD(BPF_W, BPF_REG_1, BPF_REG_0, 0));
  	emit(dlp,  BPF_STORE(BPF_W, BPF_REG_7, DMST_PRID, BPF_REG_1));
 
-	/* Read the bit mask from the table lookup in %r6. */    // FIXME someday, extend this past 64 bits
-	emit(dlp,  BPF_LOAD(BPF_DW, BPF_REG_6, BPF_REG_0, offsetof(usdt_prids_map_val_t, mask)));
+	/* Store the value key for reuse. */
+	emit(dlp,  BPF_STORE(BPF_DW, BPF_REG_FP, DT_TRAMP_SP_SLOT(0), BPF_REG_0));

This all makes sense...

Quoted 10 lines
libdtrace/dt_prov_uprobe.c +1 -1
@@ -1587,21 +1622,24 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
  	/*
  	 * Hold the bit mask in %r6 between clause calls.
  	 */
-	for (n = 0; n < dtp->dt_stmt_nextid; n++) {
+	for (ibit = n = 0; n < dtp->dt_stmt_nextid; n++) {
  		dtrace_stmtdesc_t *stp;
  		dt_ident_t	*idp;
  		uint_t		lbl_next;
 

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl) · line 1635

Quoted 10 lines
libdtrace/dt_prov_uprobe.c +1 -4
  		stp = dtp->dt_stmts[n];
-		if (stp == NULL)
-			continue;
-
-		if (ignore_clause(dtp, n, uprp))
+		if (stp == NULL || ignore_clause(dtp, n, uprp))
  			continue;
 
  		idp = stp->dtsd_clause;
  		lbl_next = dt_irlist_label(dlp);

... as does this (although we are now calling ignore_clause() at least twice for every clause, I think three times)...

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl) · line 1640

Quoted 4 lines
libdtrace/dt_prov_uprobe.c +4 -0
+		/* Load the next word of the bit mask into %r6. */
+		if (ibit % w == 0) {
+			emit(dlp,  BPF_LOAD(BPF_DW, BPF_REG_0, BPF_REG_FP, DT_TRAMP_SP_SLOT(0)));
+			emit(dlp,  BPF_LOAD(BPF_W, BPF_REG_6, BPF_REG_0, offsetof(usdt_prids_map_val_t, mask[ibit / w])));

ok, so this uses the 'w' thing to get an array offset out of a bit count, and then you can just use the existing repated-shift code. Should work.

Quoted 7 lines
libdtrace/dt_prov_uprobe.c +2 -0
+		}
+
  		/* If the lowest %r6 bit is 0, skip over this clause. */
  		emit(dlp,  BPF_MOV_REG(BPF_REG_1, BPF_REG_6));
  		emit(dlp,  BPF_ALU64_IMM(BPF_AND, BPF_REG_1, 1));
@@ -1629,6 +1667,7 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
 
Quoted 9 lines

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl) · line 1670

Quoted 3 lines
libdtrace/dt_prov_uprobe.c +1 -0
  		/* Right-shift %r6. */
  		emit(dlp,  BPF_ALU64_IMM(BPF_RSH, BPF_REG_6, 1));
+		ibit++;

... so, this isn't in the loop conditionl because the values are only filled for non-ignored, non-NULL statements. This assumption is duplicated in two places (here and in usdt_prids_map_val_extra_bytes()) and if it ever drifts we get uninitialized rubbish or buffer overruns. Should it be centralized somewhere, somehow? At least it needs a comment.

The loop bodies differ.  Okay, though, I added comments.

Quoted 4 lines

libdtrace/dt_prov_uprobe.c static int trampoline(dt_pcb_t *pcb, uint_t exitlbl) · line 1673

Quoted 3 lines
libdtrace/dt_prov_uprobe.c +3 -0
+# We stick 80 probe descriptions in each of 3 scripts to test
+# USDT's ability to handle hundreds of probe descriptions.
+++ b/test/unittest/usdt/tst.manyprobedescriptions2.sh

... maybe manyprobedescriptions-bytesize.sh :)

What is "byte size"?  Anyhow, the point is that there are many probe descriptions.

Quoted 25 lines
Quoted 25 lines
libdtrace/dt_prov_uprobe.c +24 -0
@@ -0,0 +1,127 @@
+#!/bin/bash
+#
+# Oracle Linux DTrace.
+# Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+# Licensed under the Universal Permissive License v 1.0 as shown at
+# http://oss.oracle.com/licenses/upl.
+
+# This test uses many probes and probe descriptions.  Therefore, the
+# number of BPF programs to load into the kernel -- dt_bpf_load_prog()
+# calling prp->prov->impl->load_prog(), which is dt_bpf_prog_load() --
+# and the duration of each load are both increasing.
+# @@timeout: 400
+
+dtrace=$1
+
+DIRNAME="$tmpdir/usdt-many_probe_descriptions2.$$.$RANDOM"
+mkdir -p $DIRNAME
+cd $DIRNAME
+
+# Set the lists.
+# - The probes will be foo$x$y.
+# - The probe descriptions will be foo$x* and foo*$y, for each $d.
+# So if there are nx items in xlist, ny in ylist, and nd in dlist,
+# - there will be roughly nx*ny probes
Quoted 8 lines

libdtrace/dt_prov_uprobe.c line 29

Quoted 5 lines
libdtrace/dt_prov_uprobe.c +5 -0
+# - there will be roughly (nx+ny)*nd probe descriptions
+
+xlist="a b c d e f g h i j k l m"
+ylist="n o p q r s t u v w x y z"
+dlist="0 1 2 3 4 5 6 7 8"

... and with 8... maybe we should also test 0, 1, 7 and 9, checking all the boundary conditions (and refactor things so we don't need to copy all of this every time).

But this is not a case of filling some byte up to 8 bits.  The test has nested loops over these lists, appending more probes or probe descriptions, as appropriate.  There are 13x13 probes, one after another.  There are (13+13)*9 probe descriptions, one after another.  There are 13x13x9 probe firings, one after another.  It just so happens that none of those numbers is a multiple of 8;  the only point is that each of those numbers is "big."

Thanks for the review.  Next version momentarily.

Quoted 2 lines

(You'll only really see things go wrong under valgrind, but that's fine, valgrinding should be routine before releases anyway.)

#7 lore raw

PATCH 1/1 test: Extend timeout for many-pids test

In reply to [email protected]

From: Eugene Loh <[email protected]>

---
 test/unittest/pid/tst.manypids.sh | 1 +
 1 file changed, 1 insertion(+)
test/unittest/pid/tst.manypids.sh +1 -0
index 4f77564f9..38757d0e4 100755
@@ -5,6 +5,7 @@
 # Licensed under the Universal Permissive License v 1.0 as shown at
 # http://oss.oracle.com/licenses/upl.
 #
+# @@timeout: 80
 
 if [ $# != 1 ]; then
 	echo expected one argument: '<'dtrace-path'>'
Signature · 1 line
2.43.5
Reviewed-by on v2 1/1
#9 lore raw
In reply to [email protected]

On 25 Jun 2025, eugene loh outgrape:

Quoted 11 lines

From: Eugene Loh <[email protected]>

We want to reduce the performance cost of USDT discovery if possible.

Specifically, a number of statements -- with probe descriptions such as "dtrace:::BEGIN" that could never specify USDT probes -- will not get their clause flags set with DT_CLSFLAG_USDT_EXCLUDE. So these statements get considered unnecessarily during periodic probe discovery.

Therefore:

Quoted 7 lines

*) Expand ignore_clause(dtp, n, uprp) to support the case uprp==NULL.
This case is independent of any knowledge of a specific underlying probe.

*) During probe discovery, check ignore_clause(dtp, i, NULL). This
sets the DT_CLSFLAG_USDT_[INCLUDE|EXCLUDE] flag and allows faster exclusion of statements that do not need to be reconsidered.

I'm pretty amazed you can even detect this optimization, but the code change feels like a good idea anyway, if just on code reuse grounds.

Quoted 3 lines

To take advantage of this optimization, users should specify providers. E.g., instead of "BEGIN" (which could conceivably be a USDT probe), specify "dtrace:::BEGIN".

It's a shame this is necessary though. (In practice, nobody is ever going to do it for BEGIN, at least -- but it's common enough with other probes to be worth taking advantage of.)

Quoted 1 line

modulo the tiny nit below.

libdtrace/dt_prov_uprobe.c ignore_clause(dtrace_hdl_t *dtp, int n, const dt_probe_t *uprp) · line 532

Quoted 2 lines
libdtrace/dt_prov_uprobe.c +1 -1
-			char	lastchar = pdp->prv[len - 1];
+			char	lastchar = (pdp->prv[0] != '\0' ? pdp->prv[len - 1] : '*');

I'd be a bit more belt-and-braces here, given that if this goes wrong it's a buffer overrun:

libdtrace/dt_prov_uprobe.c ignore_clause(dtrace_hdl_t *dtp, int n, const dt_probe_t *uprp) · line 563

Quoted 8 lines
libdtrace/dt_prov_uprobe.c +3 -0
+			char	lastchar = ((pdp->prv[0] != '\0' && len > 0) ? pdp->prv[len - 1] : '*');
@@ -555,6 +559,8 @@ ignore_clause(dtrace_hdl_t *dtp, int n, const dt_probe_t *uprp)
 	}
 	if (dt_stmt_clsflag_test(stp, DT_CLSFLAG_USDT_EXCLUDE) == 1)
 		return 1;
+	if (uprp == NULL)
+		return 0;

This could do with a blank line above the first if in this hunk, I think?

Quoted 6 lines
libdtrace/dt_prov_uprobe.c +0 -0
 	/*
 	 * If we cannot ignore this statement, try to use uprp.
@@ -751,13 +757,9 @@ static int discover(dtrace_hdl_t *dtp)
 	 */
 	memset(&pcb, 0, sizeof(dt_pcb_t));

libdtrace/dt_prov_uprobe.c static int discover(dtrace_hdl_t *dtp) · line 762

Quoted 10 lines
libdtrace/dt_prov_uprobe.c +2 -6
 	for (i = 0; i < dtp->dt_stmt_nextid; i++) {
-		dtrace_stmtdesc_t *stp;
-
-		stp = dtp->dt_stmts[i];
-		if (stp == NULL)
+		if (ignore_clause(dtp, i, NULL))
 			continue;
-		if (dt_stmt_clsflag_test(stp, DT_CLSFLAG_USDT_EXCLUDE) != 1)
-			dt_pid_create_usdt_probes(&stp->dtsd_ecbdesc->dted_probe, dtp, &pcb);
+		dt_pid_create_usdt_probes(&dtp->dt_stmts[i]->dtsd_ecbdesc->dted_probe, dtp, &pcb);

(straight moves into the callee, ok.)

Signature · 1 line
NULL && (void)