runtime_loader: Add support for DT_GNU_HASH.

This is an alternative to DT_HASH (SystemV/SVR4 hash tables.) Notably,
it uses a Bloom filter to allow an entire image to be skipped
at once rather than searching the actual symbol hash.

We don't currently build anything with DT_GNU_HASH support.
You can test this by adding -Wl,--hash-style=both to HAIKU_LINKFLAGS.
(It seems to increase image sizes by not too much: libroot goes
from 1347139 to 1367691 bytes (20.55 KB) in my build.)

Change-Id: I4a91276490fcd136db175833ee48b36e06ceed47
Reviewed-on: https://review.haiku-os.org/c/haiku/+/7855
Reviewed-by: waddlesplash <waddlesplash@gmail.com>

headers/os/kernel/elf.h

@@ -544,6 +544,8 @@ typedef struct {
 #define DT_PREINIT_ARRAY	32	/* preinitialization array */
 #define DT_PREINIT_ARRAYSZ	33	/* preinitialization array size */
 
+#define	DT_GNU_HASH		0x6ffffef5	/* GNU-style hash table */
+
 #define DT_VERSYM       0x6ffffff0	/* symbol version table */
 #define DT_VERDEF		0x6ffffffc	/* version definition table */
 #define DT_VERDEFNUM	0x6ffffffd	/* number of version definitions */

headers/private/runtime_loader/runtime_loader.h

@@ -107,16 +107,32 @@ typedef struct image_t {
 	addr_t				term_routine;
 	addr_t 				dynamic_ptr; 	// pointer to the dynamic section
 
-	// pointer to symbol participation data structures
+	// needed images
+	uint32				num_needed;
+	struct image_t		**needed;
+
+	// symbol participation data structures
 	uint32				*symhash;
 	elf_sym				*syms;
 	char				*strtab;
+	struct {
+		uint32			mask_words_count_mask;
+		uint32			shift2;
+		uint32			bucket_count;
+		elf_addr		*bloom;
+		uint32			*buckets;
+		uint32			*chain0;
+	} gnuhash;
+
+	// relocation information
 	elf_rel				*rel;
 	int					rel_len;
 	elf_rela			*rela;
 	int					rela_len;
 	elf_rel				*pltrel;
 	int					pltrel_len;
+
+	// init/term functions
 	addr_t				*init_array;
 	int					init_array_len;
 	addr_t				*preinit_array;
@@ -124,11 +140,6 @@ typedef struct image_t {
 	addr_t				*term_array;
 	int					term_array_len;
 
-	unsigned			dso_tls_id;
-
-	uint32				num_needed;
-	struct image_t		**needed;
-
 	// versioning related structures
 	uint32				num_version_definitions;
 	elf_verdef			*version_definitions;
@@ -138,6 +149,9 @@ typedef struct image_t {
 	elf_version_info	*versions;
 	uint32				num_versions;
 
+	// thread-local storage
+	unsigned			dso_tls_id;
+
 #ifdef __cplusplus
 	elf_sym*			(*find_undefined_symbol)(struct image_t* rootImage,
 							struct image_t* image,

src/system/runtime_loader/elf_load_image.cpp

@@ -321,6 +321,23 @@ parse_dynamic_segment(image_t* image)
 			case DT_SONAME:
 				sonameOffset = d[i].d_un.d_val;
 				break;
+			case DT_GNU_HASH:
+			{
+				uint32* gnuhash = (uint32*)
+					(d[i].d_un.d_ptr + image->regions[0].delta);
+				const uint32 bucketCount = gnuhash[0];
+				const uint32 symIndex = gnuhash[1];
+				const uint32 maskWordsCount = gnuhash[2];
+				const uint32 bloomSize = maskWordsCount * (sizeof(elf_addr) / 4);
+
+				image->gnuhash.mask_words_count_mask = maskWordsCount - 1;
+				image->gnuhash.shift2 = gnuhash[3];
+				image->gnuhash.bucket_count = bucketCount;
+				image->gnuhash.bloom = (elf_addr*)(gnuhash + 4);
+				image->gnuhash.buckets = gnuhash + 4 + bloomSize;
+				image->gnuhash.chain0 = image->gnuhash.buckets + bucketCount - symIndex;
+				break;
+			}
 			case DT_VERSYM:
 				image->symbol_versions = (elf_versym*)
 					(d[i].d_un.d_ptr + image->regions[0].delta);

src/system/runtime_loader/elf_symbol_lookup.cpp

@@ -31,7 +31,7 @@
 	\return \c true, iff \a name is non-NULL and matches the name of \a image.
 */
 static bool
-equals_image_name(image_t* image, const char* name)
+equals_image_name(const image_t* image, const char* name)
 {
 	if (name == NULL)
 		return false;
@@ -62,6 +62,19 @@ elf_hash(const char* _name)
 }
 
 
+uint32
+elf_gnuhash(const char* _name)
+{
+	const uint8* name = (const uint8*)_name;
+
+	uint32 h = 5381;
+	for (uint8 c = *name; c != '\0'; c = *++name)
+		h = (h * 33) + c;
+
+	return h;
+}
+
+
 struct match_result {
 	elf_sym* symbol;
 	elf_sym* versioned_symbol;
@@ -72,7 +85,7 @@ struct match_result {
 
 
 static bool
-match_symbol(image_t* image, const SymbolLookupInfo& lookupInfo, uint32 symIdx,
+match_symbol(const image_t* image, const SymbolLookupInfo& lookupInfo, uint32 symIdx,
 	match_result& result)
 {
 	elf_sym* symbol = &image->syms[symIdx];
@@ -188,8 +201,43 @@ match_symbol(image_t* image, const SymbolLookupInfo& lookupInfo, uint32 symIdx,
 }
 
 
-elf_sym*
-find_symbol(image_t* image, const SymbolLookupInfo& lookupInfo)
+static elf_sym*
+find_symbol_gnuhash(const image_t* image, const SymbolLookupInfo& lookupInfo)
+{
+	// Test against the Bloom filter.
+	const uint32 wordSize = sizeof(elf_addr) * 8;
+	const uint32 firstHash = lookupInfo.gnuhash & (wordSize - 1);
+	const uint32 secondHash = lookupInfo.gnuhash >> image->gnuhash.shift2;
+	const uint32 index = (lookupInfo.gnuhash / wordSize) & image->gnuhash.mask_words_count_mask;
+	const elf_addr bloomWord = image->gnuhash.bloom[index];
+	if (((bloomWord >> firstHash) & (bloomWord >> secondHash) & 1) == 0)
+		return NULL;
+
+	// Locate hash chain and corresponding value element.
+	const uint32 bucket = image->gnuhash.buckets[lookupInfo.gnuhash % image->gnuhash.bucket_count];
+	if (bucket == 0)
+		return NULL;
+
+	match_result result;
+	const uint32* chain0 = image->gnuhash.chain0;
+	const uint32* hashValue = &chain0[bucket];
+	do {
+		if (((*hashValue ^ lookupInfo.gnuhash) >> 1) != 0)
+			continue;
+
+		uint32 symIndex = hashValue - chain0;
+		if (match_symbol(image, lookupInfo, symIndex, result))
+			return result.symbol;
+	} while ((*hashValue++ & 1) == 0);
+
+	if (result.versioned_symbol_count == 1)
+		return result.versioned_symbol;
+	return NULL;
+}
+
+
+static elf_sym*
+find_symbol_sysv(const image_t* image, const SymbolLookupInfo& lookupInfo)
 {
 	if (image->dynamic_ptr == 0)
 		return NULL;
@@ -209,6 +257,21 @@ find_symbol(image_t* image, const SymbolLookupInfo& lookupInfo)
 }
 
 
+elf_sym*
+find_symbol(image_t* image, const SymbolLookupInfo& lookupInfo)
+{
+	if (image->gnuhash.buckets != NULL) {
+		if (lookupInfo.gnuhash == 0)
+			const_cast<uint32&>(lookupInfo.gnuhash) = elf_gnuhash(lookupInfo.name);
+		return find_symbol_gnuhash(image, lookupInfo);
+	}
+
+	if (lookupInfo.hash == 0)
+		const_cast<uint32&>(lookupInfo.hash) = elf_hash(lookupInfo.name);
+	return find_symbol_sysv(image, lookupInfo);
+}
+
+
 void
 patch_defined_symbol(image_t* image, const char* name, void** symbol,
 	int32* type)

src/system/runtime_loader/elf_symbol_lookup.h

@@ -17,12 +17,13 @@
 
 
 uint32 elf_hash(const char* name);
+uint32 elf_gnuhash(const char* name);
 
 
 struct SymbolLookupInfo {
 	const char*				name;
 	int32					type;
-	uint32					hash;
+	uint32					hash, gnuhash;
 	uint32					flags;
 	const elf_version_info*	version;
 	elf_sym*				requestingSymbol;
@@ -33,7 +34,8 @@ struct SymbolLookupInfo {
 		:
 		name(name),
 		type(type),
-		hash(elf_hash(name)),
+		hash(0),
+		gnuhash(0),
 		flags(flags),
 		version(version),
 		requestingSymbol(requestingSymbol)