diff mbox series

[v3,04/20] gendwarfksyms: Add address matching

Message ID 20240923181846.549877-26-samitolvanen@google.com (mailing list archive)
State Handled Elsewhere
Headers show
Series Implement DWARF modversions | expand

Commit Message

Sami Tolvanen Sept. 23, 2024, 6:18 p.m. UTC
The compiler may choose not to emit type information in DWARF for all
aliases, but it's possible for each alias to be exported separately.
To ensure we find type information for the aliases as well, read
{section, address} tuples from the symbol table and match symbols also
by address.

Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
---
 scripts/gendwarfksyms/gendwarfksyms.c |   2 +
 scripts/gendwarfksyms/gendwarfksyms.h |  13 +++
 scripts/gendwarfksyms/symbols.c       | 153 +++++++++++++++++++++++++-
 3 files changed, 165 insertions(+), 3 deletions(-)

Comments

Petr Pavlu Oct. 1, 2024, 2:06 p.m. UTC | #1
On 9/23/24 20:18, Sami Tolvanen wrote:
> The compiler may choose not to emit type information in DWARF for all
> aliases, but it's possible for each alias to be exported separately.
> To ensure we find type information for the aliases as well, read
> {section, address} tuples from the symbol table and match symbols also
> by address.
> 
> Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
> ---
>  scripts/gendwarfksyms/gendwarfksyms.c |   2 +
>  scripts/gendwarfksyms/gendwarfksyms.h |  13 +++
>  scripts/gendwarfksyms/symbols.c       | 153 +++++++++++++++++++++++++-
>  3 files changed, 165 insertions(+), 3 deletions(-)
> 
> diff --git a/scripts/gendwarfksyms/gendwarfksyms.c b/scripts/gendwarfksyms/gendwarfksyms.c
> index 096a334fa5b3..5032ec487626 100644
> --- a/scripts/gendwarfksyms/gendwarfksyms.c
> +++ b/scripts/gendwarfksyms/gendwarfksyms.c
> @@ -105,6 +105,8 @@ int main(int argc, char **argv)
>  			return -1;
>  		}
>  
> +		symbol_read_symtab(fd);
> +
>  		dwfl = dwfl_begin(&callbacks);
>  		if (!dwfl) {
>  			error("dwfl_begin failed for '%s': %s", argv[n],
> diff --git a/scripts/gendwarfksyms/gendwarfksyms.h b/scripts/gendwarfksyms/gendwarfksyms.h
> index 1a10d18f178e..a058647e2361 100644
> --- a/scripts/gendwarfksyms/gendwarfksyms.h
> +++ b/scripts/gendwarfksyms/gendwarfksyms.h
> @@ -66,14 +66,27 @@ extern int dump_dies;
>   * symbols.c
>   */
>  
> +static inline unsigned int addr_hash(uintptr_t addr)
> +{
> +	return hash_ptr((const void *)addr);
> +}
> +
> +struct symbol_addr {
> +	uint32_t section;
> +	Elf64_Addr address;
> +};
> +
>  struct symbol {
>  	const char *name;
> +	struct symbol_addr addr;
> +	struct hlist_node addr_hash;
>  	struct hlist_node name_hash;
>  };
>  
>  typedef void (*symbol_callback_t)(struct symbol *, void *arg);
>  
>  void symbol_read_exports(FILE *file);
> +void symbol_read_symtab(int fd);
>  struct symbol *symbol_get(const char *name);
>  
>  /*
> diff --git a/scripts/gendwarfksyms/symbols.c b/scripts/gendwarfksyms/symbols.c
> index 1809be93d18c..d84b46675dd1 100644
> --- a/scripts/gendwarfksyms/symbols.c
> +++ b/scripts/gendwarfksyms/symbols.c
> @@ -6,9 +6,41 @@
>  #include "gendwarfksyms.h"
>  
>  #define SYMBOL_HASH_BITS 15
> +
> +/* struct symbol_addr -> struct symbol */
> +static HASHTABLE_DEFINE(symbol_addrs, 1 << SYMBOL_HASH_BITS);
> +/* name -> struct symbol */
>  static HASHTABLE_DEFINE(symbol_names, 1 << SYMBOL_HASH_BITS);
>  
> -static int for_each(const char *name, symbol_callback_t func, void *data)
> +static inline unsigned int symbol_addr_hash(const struct symbol_addr *addr)
> +{
> +	return hash_32(addr->section ^ addr_hash(addr->address));
> +}
> +
> +static int __for_each_addr(struct symbol *sym, symbol_callback_t func,
> +			   void *data)
> +{
> +	struct hlist_node *tmp;
> +	struct symbol *match = NULL;
> +	int processed = 0;
> +
> +	hash_for_each_possible_safe(symbol_addrs, match, tmp, addr_hash,
> +				    symbol_addr_hash(&sym->addr)) {
> +		if (match == sym)
> +			continue; /* Already processed */
> +
> +		if (match->addr.section == sym->addr.section &&
> +		    match->addr.address == sym->addr.address) {
> +			func(match, data);
> +			++processed;
> +		}
> +	}
> +
> +	return processed;
> +}
> +
> +static int for_each(const char *name, bool name_only, symbol_callback_t func,
> +		    void *data)
>  {
>  	struct hlist_node *tmp;
>  	struct symbol *match;
> @@ -21,9 +53,13 @@ static int for_each(const char *name, symbol_callback_t func, void *data)
>  		if (strcmp(match->name, name))
>  			continue;
>  
> +		/* Call func for the match, and all address matches */
>  		if (func)
>  			func(match, data);
>  
> +		if (!name_only && match->addr.section != SHN_UNDEF)
> +			return checkp(__for_each_addr(match, func, data)) + 1;
> +
>  		return 1;
>  	}
>  
> @@ -32,7 +68,7 @@ static int for_each(const char *name, symbol_callback_t func, void *data)
>  
>  static bool is_exported(const char *name)
>  {
> -	return checkp(for_each(name, NULL, NULL)) > 0;
> +	return checkp(for_each(name, true, NULL, NULL)) > 0;
>  }
>  
>  void symbol_read_exports(FILE *file)
> @@ -55,6 +91,7 @@ void symbol_read_exports(FILE *file)
>  
>  		sym = xcalloc(1, sizeof(struct symbol));
>  		sym->name = name;
> +		sym->addr.section = SHN_UNDEF;
>  
>  		hash_add(symbol_names, &sym->name_hash, hash_str(sym->name));
>  		++nsym;
> @@ -77,6 +114,116 @@ struct symbol *symbol_get(const char *name)
>  {
>  	struct symbol *sym = NULL;
>  
> -	for_each(name, get_symbol, &sym);
> +	for_each(name, false, get_symbol, &sym);
>  	return sym;
>  }

What is the reason that the for_each() call in symbol_get() is invoked
with name_only=false?

> +
> +typedef void (*elf_symbol_callback_t)(const char *name, GElf_Sym *sym,
> +				      Elf32_Word xndx, void *arg);
> +
> +static void elf_for_each_global(int fd, elf_symbol_callback_t func, void *arg)
> +{
> +	size_t sym_size;
> +	GElf_Shdr shdr_mem;
> +	GElf_Shdr *shdr;
> +	Elf_Data *xndx_data = NULL;
> +	Elf_Scn *scn;
> +	Elf *elf;
> +
> +	if (elf_version(EV_CURRENT) != EV_CURRENT)
> +		error("elf_version failed: %s", elf_errmsg(-1));
> +
> +	elf = elf_begin(fd, ELF_C_READ_MMAP, NULL);
> +	if (!elf)
> +		error("elf_begin failed: %s", elf_errmsg(-1));
> +
> +	scn = elf_nextscn(elf, NULL);
> +
> +	while (scn) {
> +		shdr = gelf_getshdr(scn, &shdr_mem);
> +
> +		if (shdr && shdr->sh_type == SHT_SYMTAB_SHNDX) {
> +			xndx_data = elf_getdata(scn, NULL);
> +			break;
> +		}
> +
> +		scn = elf_nextscn(elf, scn);
> +	}
> +
> +	sym_size = gelf_fsize(elf, ELF_T_SYM, 1, EV_CURRENT);
> +	scn = elf_nextscn(elf, NULL);
> +
> +	while (scn) {
> +		shdr = gelf_getshdr(scn, &shdr_mem);
> +
> +		if (shdr && shdr->sh_type == SHT_SYMTAB) {
> +			Elf_Data *data = elf_getdata(scn, NULL);
> +			unsigned int nsyms;
> +			unsigned int n;
> +
> +			if (shdr->sh_entsize != sym_size)
> +				error("expected sh_entsize (%lu) to be %zu",
> +				      shdr->sh_entsize, sym_size);
> +
> +			nsyms = shdr->sh_size / shdr->sh_entsize;
> +
> +			for (n = 1; n < nsyms; ++n) {
> +				const char *name = NULL;
> +				Elf32_Word xndx = 0;
> +				GElf_Sym sym_mem;
> +				GElf_Sym *sym;
> +
> +				sym = gelf_getsymshndx(data, xndx_data, n,
> +						       &sym_mem, &xndx);

Please check for sym==NULL in case the file is malformed, e.g.
.symtab_shndx is truncated.

> +
> +				if (GELF_ST_BIND(sym->st_info) == STB_LOCAL)
> +					continue;
> +
> +				if (sym->st_shndx != SHN_XINDEX)
> +					xndx = sym->st_shndx;
> +
> +				name = elf_strptr(elf, shdr->sh_link,
> +						  sym->st_name);
> +
> +				/* Skip empty symbol names */
> +				if (name && *name)
> +					func(name, sym, xndx, arg);
> +			}
> +		}
> +
> +		scn = elf_nextscn(elf, scn);
> +	}
> +
> +	check(elf_end(elf));
> +}
> +
> +static void set_symbol_addr(struct symbol *sym, void *arg)
> +{
> +	struct symbol_addr *addr = arg;
> +
> +	if (sym->addr.section == SHN_UNDEF) {
> +		sym->addr = *addr;
> +		hash_add(symbol_addrs, &sym->addr_hash,
> +			 symbol_addr_hash(&sym->addr));
> +
> +		debug("%s -> { %u, %lx }", sym->name, sym->addr.section,
> +		      sym->addr.address);
> +	} else {
> +		warn("multiple addresses for symbol %s?", sym->name);
> +	}
> +}
> +
> +static void elf_set_symbol_addr(const char *name, GElf_Sym *sym,
> +				Elf32_Word xndx, void *arg)
> +{
> +	struct symbol_addr addr = { .section = xndx, .address = sym->st_value };
> +
> +	/* Set addresses for exported symbols */
> +	if (addr.section != SHN_UNDEF)
> +		for_each(name, true, set_symbol_addr, &addr);
> +}
> +
> +void symbol_read_symtab(int fd)
> +{
> +	elf_for_each_global(fd, elf_set_symbol_addr, NULL);
> +}
Sami Tolvanen Oct. 1, 2024, 8:06 p.m. UTC | #2
On Tue, Oct 1, 2024 at 2:06 PM Petr Pavlu <petr.pavlu@suse.com> wrote:
>
> On 9/23/24 20:18, Sami Tolvanen wrote:
> > -     for_each(name, get_symbol, &sym);
> > +     for_each(name, false, get_symbol, &sym);
> >       return sym;
> >  }
>
> What is the reason that the for_each() call in symbol_get() is invoked
> with name_only=false?

It was initially added to skip address checking when reading the
symbol list, but it's redundant since there are no addresses to check
at that point anyway. I think we can just drop the name_only argument
completely. I'll change this in v4.

> > +                     for (n = 1; n < nsyms; ++n) {
> > +                             const char *name = NULL;
> > +                             Elf32_Word xndx = 0;
> > +                             GElf_Sym sym_mem;
> > +                             GElf_Sym *sym;
> > +
> > +                             sym = gelf_getsymshndx(data, xndx_data, n,
> > +                                                    &sym_mem, &xndx);
>
> Please check for sym==NULL in case the file is malformed, e.g.
> .symtab_shndx is truncated.

Good catch, I'll add a check.

Sami
diff mbox series

Patch

diff --git a/scripts/gendwarfksyms/gendwarfksyms.c b/scripts/gendwarfksyms/gendwarfksyms.c
index 096a334fa5b3..5032ec487626 100644
--- a/scripts/gendwarfksyms/gendwarfksyms.c
+++ b/scripts/gendwarfksyms/gendwarfksyms.c
@@ -105,6 +105,8 @@  int main(int argc, char **argv)
 			return -1;
 		}
 
+		symbol_read_symtab(fd);
+
 		dwfl = dwfl_begin(&callbacks);
 		if (!dwfl) {
 			error("dwfl_begin failed for '%s': %s", argv[n],
diff --git a/scripts/gendwarfksyms/gendwarfksyms.h b/scripts/gendwarfksyms/gendwarfksyms.h
index 1a10d18f178e..a058647e2361 100644
--- a/scripts/gendwarfksyms/gendwarfksyms.h
+++ b/scripts/gendwarfksyms/gendwarfksyms.h
@@ -66,14 +66,27 @@  extern int dump_dies;
  * symbols.c
  */
 
+static inline unsigned int addr_hash(uintptr_t addr)
+{
+	return hash_ptr((const void *)addr);
+}
+
+struct symbol_addr {
+	uint32_t section;
+	Elf64_Addr address;
+};
+
 struct symbol {
 	const char *name;
+	struct symbol_addr addr;
+	struct hlist_node addr_hash;
 	struct hlist_node name_hash;
 };
 
 typedef void (*symbol_callback_t)(struct symbol *, void *arg);
 
 void symbol_read_exports(FILE *file);
+void symbol_read_symtab(int fd);
 struct symbol *symbol_get(const char *name);
 
 /*
diff --git a/scripts/gendwarfksyms/symbols.c b/scripts/gendwarfksyms/symbols.c
index 1809be93d18c..d84b46675dd1 100644
--- a/scripts/gendwarfksyms/symbols.c
+++ b/scripts/gendwarfksyms/symbols.c
@@ -6,9 +6,41 @@ 
 #include "gendwarfksyms.h"
 
 #define SYMBOL_HASH_BITS 15
+
+/* struct symbol_addr -> struct symbol */
+static HASHTABLE_DEFINE(symbol_addrs, 1 << SYMBOL_HASH_BITS);
+/* name -> struct symbol */
 static HASHTABLE_DEFINE(symbol_names, 1 << SYMBOL_HASH_BITS);
 
-static int for_each(const char *name, symbol_callback_t func, void *data)
+static inline unsigned int symbol_addr_hash(const struct symbol_addr *addr)
+{
+	return hash_32(addr->section ^ addr_hash(addr->address));
+}
+
+static int __for_each_addr(struct symbol *sym, symbol_callback_t func,
+			   void *data)
+{
+	struct hlist_node *tmp;
+	struct symbol *match = NULL;
+	int processed = 0;
+
+	hash_for_each_possible_safe(symbol_addrs, match, tmp, addr_hash,
+				    symbol_addr_hash(&sym->addr)) {
+		if (match == sym)
+			continue; /* Already processed */
+
+		if (match->addr.section == sym->addr.section &&
+		    match->addr.address == sym->addr.address) {
+			func(match, data);
+			++processed;
+		}
+	}
+
+	return processed;
+}
+
+static int for_each(const char *name, bool name_only, symbol_callback_t func,
+		    void *data)
 {
 	struct hlist_node *tmp;
 	struct symbol *match;
@@ -21,9 +53,13 @@  static int for_each(const char *name, symbol_callback_t func, void *data)
 		if (strcmp(match->name, name))
 			continue;
 
+		/* Call func for the match, and all address matches */
 		if (func)
 			func(match, data);
 
+		if (!name_only && match->addr.section != SHN_UNDEF)
+			return checkp(__for_each_addr(match, func, data)) + 1;
+
 		return 1;
 	}
 
@@ -32,7 +68,7 @@  static int for_each(const char *name, symbol_callback_t func, void *data)
 
 static bool is_exported(const char *name)
 {
-	return checkp(for_each(name, NULL, NULL)) > 0;
+	return checkp(for_each(name, true, NULL, NULL)) > 0;
 }
 
 void symbol_read_exports(FILE *file)
@@ -55,6 +91,7 @@  void symbol_read_exports(FILE *file)
 
 		sym = xcalloc(1, sizeof(struct symbol));
 		sym->name = name;
+		sym->addr.section = SHN_UNDEF;
 
 		hash_add(symbol_names, &sym->name_hash, hash_str(sym->name));
 		++nsym;
@@ -77,6 +114,116 @@  struct symbol *symbol_get(const char *name)
 {
 	struct symbol *sym = NULL;
 
-	for_each(name, get_symbol, &sym);
+	for_each(name, false, get_symbol, &sym);
 	return sym;
 }
+
+typedef void (*elf_symbol_callback_t)(const char *name, GElf_Sym *sym,
+				      Elf32_Word xndx, void *arg);
+
+static void elf_for_each_global(int fd, elf_symbol_callback_t func, void *arg)
+{
+	size_t sym_size;
+	GElf_Shdr shdr_mem;
+	GElf_Shdr *shdr;
+	Elf_Data *xndx_data = NULL;
+	Elf_Scn *scn;
+	Elf *elf;
+
+	if (elf_version(EV_CURRENT) != EV_CURRENT)
+		error("elf_version failed: %s", elf_errmsg(-1));
+
+	elf = elf_begin(fd, ELF_C_READ_MMAP, NULL);
+	if (!elf)
+		error("elf_begin failed: %s", elf_errmsg(-1));
+
+	scn = elf_nextscn(elf, NULL);
+
+	while (scn) {
+		shdr = gelf_getshdr(scn, &shdr_mem);
+
+		if (shdr && shdr->sh_type == SHT_SYMTAB_SHNDX) {
+			xndx_data = elf_getdata(scn, NULL);
+			break;
+		}
+
+		scn = elf_nextscn(elf, scn);
+	}
+
+	sym_size = gelf_fsize(elf, ELF_T_SYM, 1, EV_CURRENT);
+	scn = elf_nextscn(elf, NULL);
+
+	while (scn) {
+		shdr = gelf_getshdr(scn, &shdr_mem);
+
+		if (shdr && shdr->sh_type == SHT_SYMTAB) {
+			Elf_Data *data = elf_getdata(scn, NULL);
+			unsigned int nsyms;
+			unsigned int n;
+
+			if (shdr->sh_entsize != sym_size)
+				error("expected sh_entsize (%lu) to be %zu",
+				      shdr->sh_entsize, sym_size);
+
+			nsyms = shdr->sh_size / shdr->sh_entsize;
+
+			for (n = 1; n < nsyms; ++n) {
+				const char *name = NULL;
+				Elf32_Word xndx = 0;
+				GElf_Sym sym_mem;
+				GElf_Sym *sym;
+
+				sym = gelf_getsymshndx(data, xndx_data, n,
+						       &sym_mem, &xndx);
+
+				if (GELF_ST_BIND(sym->st_info) == STB_LOCAL)
+					continue;
+
+				if (sym->st_shndx != SHN_XINDEX)
+					xndx = sym->st_shndx;
+
+				name = elf_strptr(elf, shdr->sh_link,
+						  sym->st_name);
+
+				/* Skip empty symbol names */
+				if (name && *name)
+					func(name, sym, xndx, arg);
+			}
+		}
+
+		scn = elf_nextscn(elf, scn);
+	}
+
+	check(elf_end(elf));
+}
+
+static void set_symbol_addr(struct symbol *sym, void *arg)
+{
+	struct symbol_addr *addr = arg;
+
+	if (sym->addr.section == SHN_UNDEF) {
+		sym->addr = *addr;
+		hash_add(symbol_addrs, &sym->addr_hash,
+			 symbol_addr_hash(&sym->addr));
+
+		debug("%s -> { %u, %lx }", sym->name, sym->addr.section,
+		      sym->addr.address);
+	} else {
+		warn("multiple addresses for symbol %s?", sym->name);
+	}
+}
+
+static void elf_set_symbol_addr(const char *name, GElf_Sym *sym,
+				Elf32_Word xndx, void *arg)
+{
+	struct symbol_addr addr = { .section = xndx, .address = sym->st_value };
+
+	/* Set addresses for exported symbols */
+	if (addr.section != SHN_UNDEF)
+		for_each(name, true, set_symbol_addr, &addr);
+}
+
+void symbol_read_symtab(int fd)
+{
+	elf_for_each_global(fd, elf_set_symbol_addr, NULL);
+}