Skip to content

Commit e8f5ff8

Browse files
guidosarducciguidosarducci
committed
tools/dwarves: update to v1.30
Update to include the latest upstream improvements and bugfixes, and add a pending patch to fix broken BTF encoding of functions on 32-bit systems. Release Notes: https://lore.kernel.org/dwarves/0f31b6bf-7622-4a32-85cc-b69c4bff2b0f@oracle.com/ Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
1 parent 81eaa16 commit e8f5ff8

2 files changed

Lines changed: 186 additions & 2 deletions

File tree

‎tools/dwarves/Makefile‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -3,12 +3,12 @@
33
include $(TOPDIR)/rules.mk
44

55
PKG_NAME:=dwarves
6-
PKG_VERSION:=1.29
6+
PKG_VERSION:=1.30
77
PKG_RELEASE:=1
88

99
PKG_SOURCE:=$(PKG_NAME)-$(PKG_VERSION).tar.bz2
1010
PKG_SOURCE_URL:=https://fedorapeople.org/~acme/dwarves/
11-
PKG_HASH:=59c597d4e953c714d6f3ff36aeed2ac30cba85c1d7b94d0c87ca91d611d98a56
11+
PKG_HASH:=d76724ba71fbf17008c79d0c1cab540d96e9613b4831fe739796e11705ec2046
1212

1313
PKG_MAINTAINER:=Tony Ambardar <itugrok@yahoo.com>
1414
PKG_LICENSE:=GPL-2.0-only

‎tools/dwarves/patches/100-fix-func-btf-encoding-32bit.patch‎

Lines changed: 184 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,184 @@
1+
From 73919b285fab9d52983a4585de60104e05dddb30 Mon Sep 17 00:00:00 2001
2+
In-Reply-To: <20250502070318.1561924-1-tony.ambardar@gmail.com>
3+
References: <20250502070318.1561924-1-tony.ambardar@gmail.com>
4+
From: Tony Ambardar <tony.ambardar@gmail.com>
5+
Date: Tue, 20 May 2025 19:13:54 -0700
6+
Subject: [PATCH dwarves v3] dwarf_loader: Fix skipped encoding of function BTF
7+
on 32-bit systems
8+
MIME-Version: 1.0
9+
Content-Type: text/plain; charset=UTF-8
10+
Content-Transfer-Encoding: 8bit
11+
To: dwarves@vger.kernel.org,
12+
bpf@vger.kernel.org
13+
Cc: Alan Maguire <alan.maguire@oracle.com>,
14+
Arnaldo Carvalho de Melo <acme@kernel.org>,
15+
Andrii Nakryiko <andrii@kernel.org>,
16+
Alexei Starovoitov <ast@kernel.org>,
17+
Daniel Borkmann <daniel@iogearbox.net>
18+
19+
I encountered an issue building BTF kernels for 32-bit armhf, where many
20+
functions are missing in BTF data:
21+
22+
LD vmlinux
23+
BTFIDS vmlinux
24+
WARN: resolve_btfids: unresolved symbol vfs_truncate
25+
WARN: resolve_btfids: unresolved symbol vfs_fallocate
26+
WARN: resolve_btfids: unresolved symbol scx_bpf_select_cpu_dfl
27+
WARN: resolve_btfids: unresolved symbol scx_bpf_pick_idle_cpu_node
28+
WARN: resolve_btfids: unresolved symbol scx_bpf_pick_idle_cpu
29+
WARN: resolve_btfids: unresolved symbol scx_bpf_pick_any_cpu_node
30+
WARN: resolve_btfids: unresolved symbol scx_bpf_pick_any_cpu
31+
WARN: resolve_btfids: unresolved symbol scx_bpf_kick_cpu
32+
WARN: resolve_btfids: unresolved symbol scx_bpf_exit_bstr
33+
WARN: resolve_btfids: unresolved symbol scx_bpf_dsq_nr_queued
34+
WARN: resolve_btfids: unresolved symbol scx_bpf_dsq_move_vtime
35+
WARN: resolve_btfids: unresolved symbol scx_bpf_dsq_move_to_local
36+
WARN: resolve_btfids: unresolved symbol scx_bpf_dsq_move
37+
WARN: resolve_btfids: unresolved symbol scx_bpf_dsq_insert_vtime
38+
WARN: resolve_btfids: unresolved symbol scx_bpf_dsq_insert
39+
WARN: resolve_btfids: unresolved symbol scx_bpf_dispatch_vtime_from_dsq
40+
WARN: resolve_btfids: unresolved symbol scx_bpf_dispatch_vtime
41+
WARN: resolve_btfids: unresolved symbol scx_bpf_dispatch_from_dsq_set_vtime
42+
WARN: resolve_btfids: unresolved symbol scx_bpf_dispatch_from_dsq_set_slice
43+
WARN: resolve_btfids: unresolved symbol scx_bpf_dispatch_from_dsq
44+
WARN: resolve_btfids: unresolved symbol scx_bpf_dispatch
45+
WARN: resolve_btfids: unresolved symbol scx_bpf_destroy_dsq
46+
WARN: resolve_btfids: unresolved symbol scx_bpf_create_dsq
47+
WARN: resolve_btfids: unresolved symbol scx_bpf_consume
48+
WARN: resolve_btfids: unresolved symbol bpf_throw
49+
WARN: resolve_btfids: unresolved symbol bpf_sock_ops_enable_tx_tstamp
50+
WARN: resolve_btfids: unresolved symbol bpf_percpu_obj_new_impl
51+
WARN: resolve_btfids: unresolved symbol bpf_obj_new_impl
52+
WARN: resolve_btfids: unresolved symbol bpf_lookup_user_key
53+
WARN: resolve_btfids: unresolved symbol bpf_lookup_system_key
54+
WARN: resolve_btfids: unresolved symbol bpf_iter_task_vma_new
55+
WARN: resolve_btfids: unresolved symbol bpf_iter_scx_dsq_new
56+
WARN: resolve_btfids: unresolved symbol bpf_get_kmem_cache
57+
WARN: resolve_btfids: unresolved symbol bpf_dynptr_from_xdp
58+
WARN: resolve_btfids: unresolved symbol bpf_dynptr_from_skb
59+
WARN: resolve_btfids: unresolved symbol bpf_cgroup_from_id
60+
NM System.map
61+
62+
After further debugging this can be reproduced more simply:
63+
64+
$ pahole -J -j --btf_features=decl_tag,consistent_func,decl_tag_kfuncs .tmp_vmlinux_armhf
65+
btf_encoder__tag_kfunc: failed to find kfunc 'scx_bpf_select_cpu_dfl' in BTF
66+
btf_encoder__tag_kfuncs: failed to tag kfunc 'scx_bpf_select_cpu_dfl'
67+
68+
$ pfunct -Fbtf -E -f scx_bpf_select_cpu_dfl .tmp_vmlinux_armhf
69+
<nothing>
70+
71+
$ pfunct -Fdwarf -E -f scx_bpf_select_cpu_dfl .tmp_vmlinux_armhf
72+
s32 scx_bpf_select_cpu_dfl(struct task_struct * p, s32 prev_cpu, u64 wake_flags, bool * is_idle);
73+
74+
$ pahole -J -j --btf_features=decl_tag,decl_tag_kfuncs .tmp_vmlinux_armhf
75+
76+
$ pfunct -Fbtf -E -f scx_bpf_select_cpu_dfl .tmp_vmlinux_armhf
77+
bpf_kfunc s32 scx_bpf_select_cpu_dfl(struct task_struct * p, s32 prev_cpu, u64 wake_flags, bool * is_idle);
78+
79+
The key things to note are the pahole 'consistent_func' feature and the u64
80+
'wake_flags' parameter vs. arm 32-bit registers. These point to existing
81+
code handling arguments larger than register-size, allowing them to be
82+
BTF encoded but only if structs.
83+
84+
Generalize the code for any argument type larger than register size (i.e.
85+
size > cu->addr_size). This should work for integral or aggregate types,
86+
and also avoids a bug in the current code where a register-sized struct
87+
could be mistaken for larger. Note that zero-sized arguments will still
88+
be marked as inconsistent and not encoded.
89+
90+
Fixes: a53c58158b76 ("dwarf_loader: Mark functions that do not use expected registers for params")
91+
Tested-by: Alexis Lothoré <alexis.lothore@bootlin.com>
92+
Tested-by: Alan Maguire <alan.maguire@oracle.com>
93+
Signed-off-by: Tony Ambardar <tony.ambardar@gmail.com>
94+
---
95+
v2 -> v3:
96+
- Added Tested-by: from Alexis and Alan.
97+
- Revert support for encoding 0-sized structs (as v1) after discussion:
98+
https://lore.kernel.org/dwarves/9a41b21f-c0ae-4298-bf95-09d0cdc3f3ab@oracle.com/
99+
- Inline param__is_wide() and clarify some naming/wording.
100+
101+
v1 -> v2:
102+
- Update to preserve existing behaviour where zero-sized struct params
103+
still permit the function to be encoded, as noted by Alan.
104+
105+
---
106+
dwarf_loader.c | 37 ++++++++++++-------------------------
107+
1 file changed, 12 insertions(+), 25 deletions(-)
108+
109+
diff --git a/dwarf_loader.c b/dwarf_loader.c
110+
index e1ba7bc..134a76b 100644
111+
--- a/dwarf_loader.c
112+
+++ b/dwarf_loader.c
113+
@@ -2914,23 +2914,9 @@ out:
114+
return 0;
115+
}
116+
117+
-static bool param__is_struct(struct cu *cu, struct tag *tag)
118+
+static inline bool param__is_wide(struct cu *cu, struct tag *tag)
119+
{
120+
- struct tag *type = cu__type(cu, tag->type);
121+
-
122+
- if (!type)
123+
- return false;
124+
-
125+
- switch (type->tag) {
126+
- case DW_TAG_structure_type:
127+
- return true;
128+
- case DW_TAG_const_type:
129+
- case DW_TAG_typedef:
130+
- /* handle "typedef struct", const parameter */
131+
- return param__is_struct(cu, type);
132+
- default:
133+
- return false;
134+
- }
135+
+ return tag__size(tag, cu) > cu->addr_size;
136+
}
137+
138+
static int cu__resolve_func_ret_types_optimized(struct cu *cu)
139+
@@ -2942,9 +2928,9 @@ static int cu__resolve_func_ret_types_optimized(struct cu *cu)
140+
struct tag *tag = pt->entries[i];
141+
struct parameter *pos;
142+
struct function *fn = tag__function(tag);
143+
- bool has_unexpected_reg = false, has_struct_param = false;
144+
+ bool has_unexpected_reg = false, has_wide_param = false;
145+
146+
- /* mark function as optimized if parameter is, or
147+
+ /* Mark function as optimized if parameter is, or
148+
* if parameter does not have a location; at this
149+
* point location presence has been marked in
150+
* abstract origins for cases where a parameter
151+
@@ -2953,10 +2939,11 @@ static int cu__resolve_func_ret_types_optimized(struct cu *cu)
152+
*
153+
* Also mark functions which, due to optimization,
154+
* use an unexpected register for a parameter.
155+
- * Exception is functions which have a struct
156+
- * as a parameter, as multiple registers may
157+
- * be used to represent it, throwing off register
158+
- * to parameter mapping.
159+
+ * Exception is functions with a wide parameter,
160+
+ * as single register won't be used to represent
161+
+ * it, throwing off register to parameter mapping.
162+
+ * Examples include large structs or 64-bit types
163+
+ * on a 32-bit arch.
164+
*/
165+
ftype__for_each_parameter(&fn->proto, pos) {
166+
if (pos->optimized || !pos->has_loc)
167+
@@ -2967,11 +2954,11 @@ static int cu__resolve_func_ret_types_optimized(struct cu *cu)
168+
}
169+
if (has_unexpected_reg) {
170+
ftype__for_each_parameter(&fn->proto, pos) {
171+
- has_struct_param = param__is_struct(cu, &pos->tag);
172+
- if (has_struct_param)
173+
+ has_wide_param = param__is_wide(cu, &pos->tag);
174+
+ if (has_wide_param)
175+
break;
176+
}
177+
- if (!has_struct_param)
178+
+ if (!has_wide_param)
179+
fn->proto.unexpected_reg = 1;
180+
}
181+
182+
--
183+
2.34.1
184+

0 commit comments

Comments
 (0)