dwarf_loader: Fix for BTF id drift caused by adding unspecified types
Eduard Zingerman
maintainerPATCH 1/1 dwarf_loader: Fix for BTF id drift caused by adding unspecified types
Recent changes to handle unspecified types (see [1]) cause BTF ID drift.
Specifically, the intent of commits [2], [3] and [4] is to render references to unspecified types as void type.
However, as a consequence:
- in die__process_unit() call to cu__add_tag() allocates small_id for unspecified type tags and adds these tags to cu->types_table;
- btf_encoder__encode_tag() skips generation of BTF entries for DW_TAG_unspecified_type tags.
Such logic causes ID drift if unspecified type is not the last type processed for compilation unit. small_id of each type following unspecified type in the cu->types_table would have its BTF id off by -1. Thus, rendering references established on recode phase invalid.
This commit reverts unspecified_type id/tag tracking. Instead, the following is done:
- small_id for unspecified type tags is set to 0, thus reference to unspecified type tag would render BTF id of a void on recode phase;
- unspecified type tags are not added to cu->types_table.
This change also happens to fix issue reported in [5], the gist of that issue is that the field encoder->unspecified_type is set but not reset by function btf_encoder__encode_cu(). Thus, the following sequence of events might occur when BTF encoding is requested:
- CU with unspecified type is processed:
- unspecified type id is 42
- encoder->unspecified_type is set to 42
- CU without unspecified type is processed next using the same encoder object:
- some struct foo has id 42 in this CU
- the references to struct foo are set 0 by function
`btf_encoder__tag_type()`.
[1] https://lore.kernel.org/all/Y0R7uu3s%[email protected]/
[2] bcc648a10cbc ("btf_encoder: Encode DW_TAG_unspecified_type returning routines as void")
[3] cffe5e1f75e1 ("core: Record if a CU has a DW_TAG_unspecified_type")
[4] 75e0fe28bb02 ("core: Add DW_TAG_unspecified_type to tag__is_tag_type() set")
[5] https://lore.kernel.org/bpf/Y%[email protected]/
--- btf_encoder.c | 8 -------- dwarf_loader.c | 25 +++++++++++++++++++------ dwarves.h | 8 -------- 3 files changed, 19 insertions(+), 22 deletions(-)
btf_encoder.c +0 -8
@@ -69,7 +69,6 @@ struct btf_encoder {const char *filename;struct elf_symtab *symtab;uint32_t type_id_off; -uint32_t unspecified_type;int saved_func_cnt;bool has_index_type, need_index_type,
@@ -635,11 +634,6 @@ static int32_t btf_encoder__tag_type(struct btf_encoder *encoder, uint32_t tag_tif (tag_type ==0 )return 0 ; -if (encoder->unspecified_type && tag_type == encoder->unspecified_type) { -// No provision for encoding this, turn it into void. -return 0 ; - } -return encoder->type_id_off + tag_type; }
@@ -1746,8 +1740,6 @@ int btf_encoder__encode_cu(struct btf_encoder *encoder, struct cu *cu, struct co encoder->cu = cu; encoder->type_id_off =btf__type_cnt (encoder->btf) -1 ; -if (encoder->cu->unspecified_type.tag) - encoder->unspecified_type = encoder->cu->unspecified_type.type;if (!encoder->has_index_type) {/* cu__find_base_type_by_name() takes "type_id_t *id" */
dwarf_loader.c +19 -6
@@ -2155,8 +2155,7 @@ static struct tag *__die__process_tag(Dwarf_Die *die, struct cu *cu,case DW_TAG_atomic_type: tag =die__create_new_tag (die, cu);break ;case DW_TAG_unspecified_type: - cu->unspecified_type.tag = - tag =die__create_new_tag (die, cu);break ; + tag =die__create_new_tag (die, cu);break ;case DW_TAG_pointer_type: tag =die__create_new_pointer_tag (die, cu, conf);break ;case DW_TAG_ptr_to_member_type:
@@ -2219,13 +2218,27 @@ static int die__process_unit(Dwarf_Die *die, struct cu *cu, struct conf_load *cocontinue ; } -uint32_t id; -cu__add_tag (cu, tag, &id); +uint32_t id =0 ; +/* There is no BTF representation for unspecified types. +* Currently we want such types to be represented as `void` +* (and thus skip BTF encoding). +* +* As BTF encoding is skipped, such types must not be added to type table, +* otherwise an ID for a type would be allocated and we would be forced +* to put something in BTF at this ID. +* Thus avoid `cu__add_tag()` call for such types. +* +* On the other hand, there might be references to this type from other +* tags, so `dwarf_cu__find_tag_by_ref()` must return something. +* Thus call `cu__hash()` for such types. +* +* Note, that small_id of zero would be assigned to unspecified type entry. +*/ +if (tag->tag != DW_TAG_unspecified_type) +cu__add_tag (cu, tag, &id);cu__hash (cu, tag);struct dwarf_tag *dtag = tag->priv; dtag->small_id = id; -if (tag->tag == DW_TAG_unspecified_type) - cu->unspecified_type.type = id; }while (dwarf_siblingof (die, die) ==0 );return 0 ;
dwarves.h +0 -8
@@ -236,10 +236,6 @@ struct debug_fmt_ops {#define ARCH_MAX_REGISTER_PARAMS 8 -/* -* unspecified_type: If this CU has a DW_TAG_unspecified_type, as BTF doesn't have a representation for this -* and thus we need to check functions returning this to convert it to void. -*/ struct cu {struct list_head node;struct list_head tags;
@@ -248,10 +244,6 @@ struct cu {struct ptr_table functions_table;struct ptr_table tags_table;struct rb_root functions; -struct { -struct tag *tag; -uint32_t type; - } unspecified_type;char *name;char *filename;void *priv;
Signature · 1 line
2.39.1
Quoted 41 lines · Eduard Zingerman
Recent changes to handle unspecified types (see [1]) cause BTF ID drift.
Specifically, the intent of commits [2], [3] and [4] is to render references to unspecified types as void type.
However, as a consequence:
- indie__process_unit()call tocu__add_tag()allocatessmall_idfor unspecified type tags and adds these tags tocu->types_table;
-btf_encoder__encode_tag()skips generation of BTF entries forDW_TAG_unspecified_typetags.Such logic causes ID drift if unspecified type is not the last type processed for compilation unit.
small_idof each type following unspecified type in thecu->types_tablewould have its BTF id off by -1. Thus, rendering references established on recode phase invalid.This commit reverts
unspecified_typeid/tag tracking. Instead, the following is done:
-small_idfor unspecified type tags is set to 0, thus reference to unspecified type tag would render BTF id of avoidon recode phase;
- unspecified type tags are not added tocu->types_table.This change also happens to fix issue reported in [5], the gist of that issue is that the field
encoder->unspecified_typeis set but not reset by functionbtf_encoder__encode_cu(). Thus, the following sequence of events might occur when BTF encoding is requested:
- CU with unspecified type is processed:
- unspecified type id is 42
-encoder->unspecified_typeis set to 42
- CU without unspecified type is processed next using the sameencoderobject:
- somestruct foohas id 42 in this CU
- the references tostruct fooare set 0 by function`btf_encoder__tag_type()`.[1] https://lore.kernel.org/all/Y0R7uu3s%[email protected]/
[2]bcc648a10cbc("btf_encoder: EncodeDW_TAG_unspecified_typereturning routines as void")
[3]cffe5e1f75e1("core: Record if a CU has aDW_TAG_unspecified_type")
[4]75e0fe28bb02("core: AddDW_TAG_unspecified_typetotag__is_tag_type()set")
[5] https://lore.kernel.org/bpf/Y%[email protected]/
Quoted 6 lines
Fixes bcc648a10cbc ("btf_encoder: Encode DW_TAG_unspecified_type returning routines as void")Fixes 52b25808e44a ("btf_encoder: Store type_id_off, unspecified type in encoder")Tested-by KP Singh <[email protected]>Reported-by Matt Bobrowski <[email protected]>Signed-off-by Eduard Zingerman <[email protected]>---
Looks like a sensible approach to me.
Quoted 5 lines
btf_encoder.c | 8 -------- dwarf_loader.c | 25 +++++++++++++++++++------ dwarves.h | 8 -------- 3 files changed, 19 insertions(+), 22 deletions(-)
[...]
Quoted 40 lines · Eduard Zingerman
Recent changes to handle unspecified types (see [1]) cause BTF ID drift.
Specifically, the intent of commits [2], [3] and [4] is to render references to unspecified types as void type.
However, as a consequence:
- indie__process_unit()call tocu__add_tag()allocatessmall_idfor unspecified type tags and adds these tags tocu->types_table;
-btf_encoder__encode_tag()skips generation of BTF entries forDW_TAG_unspecified_typetags.Such logic causes ID drift if unspecified type is not the last type processed for compilation unit.
small_idof each type following unspecified type in thecu->types_tablewould have its BTF id off by -1. Thus, rendering references established on recode phase invalid.This commit reverts
unspecified_typeid/tag tracking. Instead, the following is done:
-small_idfor unspecified type tags is set to 0, thus reference to unspecified type tag would render BTF id of avoidon recode phase;
- unspecified type tags are not added tocu->types_table.This change also happens to fix issue reported in [5], the gist of that issue is that the field
encoder->unspecified_typeis set but not reset by functionbtf_encoder__encode_cu(). Thus, the following sequence of events might occur when BTF encoding is requested:
- CU with unspecified type is processed:
- unspecified type id is 42
-encoder->unspecified_typeis set to 42
- CU without unspecified type is processed next using the sameencoderobject:
- somestruct foohas id 42 in this CU
- the references tostruct fooare set 0 by function`btf_encoder__tag_type()`.[1] https://lore.kernel.org/all/Y0R7uu3s%[email protected]/
[2]bcc648a10cbc("btf_encoder: EncodeDW_TAG_unspecified_typereturning routines as void")
[3]cffe5e1f75e1("core: Record if a CU has aDW_TAG_unspecified_type")
[4]75e0fe28bb02("core: AddDW_TAG_unspecified_typetotag__is_tag_type()set")
[5] https://lore.kernel.org/bpf/Y%[email protected]/
Quoted 5 lines
Fixes bcc648a10cbc ("btf_encoder: Encode DW_TAG_unspecified_type returning routines as void")Fixes 52b25808e44a ("btf_encoder: Store type_id_off, unspecified type in encoder")Tested-by KP Singh <[email protected]>Reported-by Matt Bobrowski <[email protected]>Signed-off-by Eduard Zingerman <[email protected]>
lgtm, tested on top of the pahole next branch with bpf selftests
jirka
Quoted 114 lines
--- btf_encoder.c | 8 -------- dwarf_loader.c | 25 +++++++++++++++++++------ dwarves.h | 8 -------- 3 files changed, 19 insertions(+), 22 deletions(-)btf_encoder.c +0
-8@@ -69,7 +69,6 @@ struct btf_encoder {const char *filename;struct elf_symtab *symtab;uint32_t type_id_off; -uint32_t unspecified_type;int saved_func_cnt;bool has_index_type, need_index_type,@@ -635,11 +634,6 @@ static int32_t btf_encoder__tag_type(struct btf_encoder *encoder, uint32_t tag_tif (tag_type ==0 )return 0 ; -if (encoder->unspecified_type && tag_type == encoder->unspecified_type) { -// No provision for encoding this, turn it into void. -return 0 ; - } -return encoder->type_id_off + tag_type; }@@ -1746,8 +1740,6 @@ int btf_encoder__encode_cu(struct btf_encoder *encoder, struct cu *cu, struct co encoder->cu = cu; encoder->type_id_off =btf__type_cnt (encoder->btf) -1 ; -if (encoder->cu->unspecified_type.tag) - encoder->unspecified_type = encoder->cu->unspecified_type.type;if (!encoder->has_index_type) {/* cu__find_base_type_by_name() takes "type_id_t *id" */ dwarf_loader.c +19
-6@@ -2155,8 +2155,7 @@ static struct tag *__die__process_tag(Dwarf_Die *die, struct cu *cu,case DW_TAG_atomic_type: tag =die__create_new_tag (die, cu);break ;case DW_TAG_unspecified_type: - cu->unspecified_type.tag = - tag =die__create_new_tag (die, cu);break ; + tag =die__create_new_tag (die, cu);break ;case DW_TAG_pointer_type: tag =die__create_new_pointer_tag (die, cu, conf);break ;case DW_TAG_ptr_to_member_type:@@ -2219,13 +2218,27 @@ static int die__process_unit(Dwarf_Die *die, struct cu *cu, struct conf_load *cocontinue ; } -uint32_t id; -cu__add_tag (cu, tag, &id); +uint32_t id =0 ; +/* There is no BTF representation for unspecified types. +* Currently we want such types to be represented as `void` +* (and thus skip BTF encoding). +* +* As BTF encoding is skipped, such types must not be added to type table, +* otherwise an ID for a type would be allocated and we would be forced +* to put something in BTF at this ID. +* Thus avoid `cu__add_tag()` call for such types. +* +* On the other hand, there might be references to this type from other +* tags, so `dwarf_cu__find_tag_by_ref()` must return something. +* Thus call `cu__hash()` for such types. +* +* Note, that small_id of zero would be assigned to unspecified type entry. +*/ +if (tag->tag != DW_TAG_unspecified_type) +cu__add_tag (cu, tag, &id);cu__hash (cu, tag);struct dwarf_tag *dtag = tag->priv; dtag->small_id = id; -if (tag->tag == DW_TAG_unspecified_type) - cu->unspecified_type.type = id; }while (dwarf_siblingof (die, die) ==0 );return 0 ;dwarves.h +0
-8@@ -236,10 +236,6 @@ struct debug_fmt_ops {#define ARCH_MAX_REGISTER_PARAMS 8 -/* -* unspecified_type: If this CU has a DW_TAG_unspecified_type, as BTF doesn't have a representation for this -* and thus we need to check functions returning this to convert it to void. -*/ struct cu {struct list_head node;struct list_head tags;@@ -248,10 +244,6 @@ struct cu {struct ptr_table functions_table;struct ptr_table tags_table;struct rb_root functions; -struct { -struct tag *tag; -uint32_t type; - } unspecified_type;char *name;char *filename;void *priv;Signature · 1 line
2.39.1
Arnaldo Carvalho de Melo
AppliedEm Wed, Mar 01, 2023 at 10:19:56PM +0100, Jiri Olsa escreveu:
Quoted 47 lines · Eduard Zingerman
Quoted 40 lines
Recent changes to handle unspecified types (see [1]) cause BTF ID drift.
Specifically, the intent of commits [2], [3] and [4] is to render references to unspecified types as void type.
However, as a consequence:
- indie__process_unit()call tocu__add_tag()allocatessmall_idfor unspecified type tags and adds these tags tocu->types_table;
-btf_encoder__encode_tag()skips generation of BTF entries forDW_TAG_unspecified_typetags.Such logic causes ID drift if unspecified type is not the last type processed for compilation unit.
small_idof each type following unspecified type in thecu->types_tablewould have its BTF id off by -1. Thus, rendering references established on recode phase invalid.This commit reverts
unspecified_typeid/tag tracking. Instead, the following is done:
-small_idfor unspecified type tags is set to 0, thus reference to unspecified type tag would render BTF id of avoidon recode phase;
- unspecified type tags are not added tocu->types_table.This change also happens to fix issue reported in [5], the gist of that issue is that the field
encoder->unspecified_typeis set but not reset by functionbtf_encoder__encode_cu(). Thus, the following sequence of events might occur when BTF encoding is requested:
- CU with unspecified type is processed:
- unspecified type id is 42
-encoder->unspecified_typeis set to 42
- CU without unspecified type is processed next using the sameencoderobject:
- somestruct foohas id 42 in this CU
- the references tostruct fooare set 0 by function`btf_encoder__tag_type()`.[1] https://lore.kernel.org/all/Y0R7uu3s%[email protected]/
[2]bcc648a10cbc("btf_encoder: EncodeDW_TAG_unspecified_typereturning routines as void")
[3]cffe5e1f75e1("core: Record if a CU has aDW_TAG_unspecified_type")
[4]75e0fe28bb02("core: AddDW_TAG_unspecified_typetotag__is_tag_type()set")
[5] https://lore.kernel.org/bpf/Y%[email protected]/Quoted 5 lines
Fixes bcc648a10cbc ("btf_encoder: Encode DW_TAG_unspecified_type returning routines as void")Fixes 52b25808e44a ("btf_encoder: Store type_id_off, unspecified type in encoder")Tested-by KP Singh <[email protected]>Reported-by Matt Bobrowski <[email protected]>Signed-off-by Eduard Zingerman <[email protected]>
Quoted 3 lines
lgtm, tested on top of the pahole next branch with bpf selftests
Tested-by Jiri Olsa <[email protected]>
Looks good to me as well, and way more elegant, thanks!
Applied.
- Arnaldo