From 3a659bb19898a2295ef10b7b72f87293e7351255 Mon Sep 17 00:00:00 2001 From: Max Filippov Date: Thu, 12 Nov 2015 06:11:58 +0300 Subject: binutils: add xtensa-specific patches for 2.25.1 Signed-off-by: Max Filippov diff --git a/patches/binutils/2.25.1/905-Fix-trampolines-search-code-for-conditional-branches.patch b/patches/binutils/2.25.1/905-Fix-trampolines-search-code-for-conditional-branches.patch new file mode 100644 index 0000000..8aeb064 --- /dev/null +++ b/patches/binutils/2.25.1/905-Fix-trampolines-search-code-for-conditional-branches.patch @@ -0,0 +1,90 @@ +From 415480d6471e67aef97c0241d451ef2423a1da9d Mon Sep 17 00:00:00 2001 +From: Max Filippov +Date: Tue, 25 Nov 2014 21:33:21 +0300 +Subject: [PATCH] Fix trampolines search code for conditional branches + +For conditional branches that need more than one trampoline to reach its +target assembler couldn't always find suitable trampoline because +post-loop condition check was placed inside the loop, resulting in +premature loop termination. Move check outside the loop. + +This fixes the following build errors seen when assembling huge files +produced by gcc: + Error: jump target out of range; no usable trampoline found + Error: operand 1 of 'j' has out of range value '307307' + +2014-11-25 Max Filippov + +gas/ + * config/tc-xtensa.c (search_trampolines): Move post-loop + condition check outside the search loop. + +gas/testsuite/ + * gas/xtensa/trampoline.d: Add expected output for branches. + * gas/xtensa/trampoline.s: Add test case for branches. + +Signed-off-by: Max Filippov +--- +Backported from: d92b6eece424f0ad35d96fdd85bf207295e8c4c3 +Changes to ChangeLogs are dropped. + + gas/config/tc-xtensa.c | 8 ++++---- + gas/testsuite/gas/xtensa/trampoline.d | 9 +++++++++ + gas/testsuite/gas/xtensa/trampoline.s | 7 +++++++ + 3 files changed, 20 insertions(+), 4 deletions(-) + +diff --git a/gas/config/tc-xtensa.c b/gas/config/tc-xtensa.c +index d11b0c7..f23ccf8 100644 +--- a/gas/config/tc-xtensa.c ++++ b/gas/config/tc-xtensa.c +@@ -9514,11 +9514,11 @@ search_trampolines (TInsn *tinsn, fragS *fragP, bfd_boolean unreachable_only) + if (next_addr == 0 || addr - next_addr > J_RANGE) + break; + } +- if (abs (addr - this_addr) < J_RANGE) +- return tf; +- +- return NULL; + } ++ if (abs (addr - this_addr) < J_RANGE) ++ return tf; ++ ++ return NULL; + } + for ( ; tf; tf = tf->next) + { +diff --git a/gas/testsuite/gas/xtensa/trampoline.d b/gas/testsuite/gas/xtensa/trampoline.d +index b4f65dc..5ae32a6 100644 +--- a/gas/testsuite/gas/xtensa/trampoline.d ++++ b/gas/testsuite/gas/xtensa/trampoline.d +@@ -24,3 +24,12 @@ + .*33462:.*j.0x49407 + #... + .*49407:.*j.0x49407 ++.*4940a:.*beqz.n.a2,.0x4940f ++.*4940c:.*j.0x693d1 ++#... ++.*693d1:.*j.0x7ddd4 ++#... ++.*7ddd4:.*j.0x927f5 ++#... ++.*927f5:.*j.0x927f5 ++#... +diff --git a/gas/testsuite/gas/xtensa/trampoline.s b/gas/testsuite/gas/xtensa/trampoline.s +index 259a3bb..4465786 100644 +--- a/gas/testsuite/gas/xtensa/trampoline.s ++++ b/gas/testsuite/gas/xtensa/trampoline.s +@@ -19,3 +19,10 @@ + .endr + 3: + j 3b ++ bnez a2, 4f ++ .rep 50000 ++ and a2, a2, a3 ++ _ret ++ .endr ++4: ++ j 4b +-- +1.8.1.4 + diff --git a/patches/binutils/2.25.1/906-xtensa-optimize-check_section_ebb_pcrels_fit.patch b/patches/binutils/2.25.1/906-xtensa-optimize-check_section_ebb_pcrels_fit.patch new file mode 100644 index 0000000..8a21100 --- /dev/null +++ b/patches/binutils/2.25.1/906-xtensa-optimize-check_section_ebb_pcrels_fit.patch @@ -0,0 +1,502 @@ +From 20c79baf82273a0b368587f761f152c4d3a593a4 Mon Sep 17 00:00:00 2001 +From: Max Filippov +Date: Fri, 27 Mar 2015 07:13:55 +0300 +Subject: [PATCH 1/4] xtensa: optimize check_section_ebb_pcrels_fit + +The original check_section_ebb_pcrels_fit algorithm checks that text +actions proposed for current EBB are OK for every relocation in a +section. There's no need to check every relocation, because text actions +for EBB can only change size of that EBB, thus only affecting +relocations that in any way cross that EBB. In addition EBBs are +iterated in ascending order of their VMA, making it easier to track +relevant relocations. + +Introduce a structure that can track relocations that cross the range of +VMAs of EBB and use it to only check relocations relevant to current EBB +in check_section_ebb_pcrels_fit. +It takes O(N log N) operations to build it and O(N) operations to move +current EBB VMA window through its entire range, where N is the number +of relocations in a section. The resulting complexity of +compute_text_actions is thus reduced from O(N^2) to O(N log N + N * M), +where M is the average number of relocations crossing each EBB. + +Original profile: + +% time self children called name +----------------------------------------- + 44.26 71.53 6429/6429 compute_text_actions + 50.2 44.26 71.53 6429 check_section_ebb_pcrels_fit + 1.16 20.12 347506666/347576152 pcrel_reloc_fits + 2.95 16.52 347506666/348104944 get_relocation_opnd + 2.01 9.74 347575100/361252208 r_reloc_init + 0.55 7.53 347575100/363381467 r_reloc_get_section + 5.76 0.02 695013332/695013332 xlate_offset_with_removed_text + 0.68 3.89 347575100/363483827 bfd_octets_per_byte + 0.32 0.00 347506666/349910253 is_alt_relocation + 0.18 0.11 6391/6391 build_xlate_map + 0.00 0.00 6429/19417168 get_xtensa_relax_info + 0.00 0.00 6391/6391 free_xlate_map +----------------------------------------- + +Same data, after optimization: + +% time self children called name +----------------------------------------- + 2.56 3.08 6429/6429 compute_text_actions + 8.2 2.56 3.08 6429 check_section_ebb_pcrels_fit + 0.08 0.91 17721075/17790561 pcrel_reloc_fits + 0.17 0.47 17721075/31685977 r_reloc_init + 0.43 0.00 35442150/35442150 xlate_offset_with_removed_text + 0.02 0.37 17721075/33815236 r_reloc_get_section + 0.22 0.11 6391/6391 build_xlate_map + 0.05 0.22 17721075/33917596 bfd_octets_per_byte + 0.03 0.00 17721075/20405299 is_alt_relocation + 0.01 0.00 6429/6429 reloc_range_list_update_range + 0.00 0.00 6429/19417168 get_xtensa_relax_info + 0.00 0.00 6391/6391 free_xlate_map +----------------------------------------- + +2015-04-01 Max Filippov +bfd/ + * elf32-xtensa.c (reloc_range_list, reloc_range_list_entry, + reloc_range): new typedef. + (reloc_range_list_struct, reloc_range_list_entry_struct, + reloc_range_struct): new structures. + (reloc_range_compare, build_reloc_ranges, + reloc_range_list_append, reloc_range_list_remove, + reloc_range_list_update_range, free_reloc_range_list): new + functions. + (compute_text_actions): precompute relocation opcodes before the + loop. Add relevant_relocs variable, initialize it before the + loop, pass it to the check_section_ebb_pcrels_fit. + (check_section_ebb_pcrels_fit): add new parameter: + relevant_relocs. Update address range in the relevant_relocs if + it's non-NULL and iterate only over relevant relocations. + +Backported from: b2b326d246f839ee218192ac88da2384d929a072 +Signed-off-by: Max Filippov +--- + bfd/elf32-xtensa.c | 321 +++++++++++++++++++++++++++++++++++++++++++++++++---- + 1 file changed, 298 insertions(+), 23 deletions(-) + +diff --git a/bfd/elf32-xtensa.c b/bfd/elf32-xtensa.c +index 0b6f584..872370b 100644 +--- a/bfd/elf32-xtensa.c ++++ b/bfd/elf32-xtensa.c +@@ -6619,8 +6619,10 @@ static bfd_boolean compute_text_actions + (bfd *, asection *, struct bfd_link_info *); + static bfd_boolean compute_ebb_proposed_actions (ebb_constraint *); + static bfd_boolean compute_ebb_actions (ebb_constraint *); ++typedef struct reloc_range_list_struct reloc_range_list; + static bfd_boolean check_section_ebb_pcrels_fit +- (bfd *, asection *, bfd_byte *, Elf_Internal_Rela *, const ebb_constraint *, ++ (bfd *, asection *, bfd_byte *, Elf_Internal_Rela *, ++ reloc_range_list *, const ebb_constraint *, + const xtensa_opcode *); + static bfd_boolean check_section_ebb_reduces (const ebb_constraint *); + static void text_action_add_proposed +@@ -7219,6 +7221,221 @@ build_reloc_opcodes (bfd *abfd, + return reloc_opcodes; + } + ++struct reloc_range_struct ++{ ++ bfd_vma addr; ++ bfd_boolean add; /* TRUE if start of a range, FALSE otherwise. */ ++ /* Original irel index in the array of relocations for a section. */ ++ unsigned irel_index; ++}; ++typedef struct reloc_range_struct reloc_range; ++ ++typedef struct reloc_range_list_entry_struct reloc_range_list_entry; ++struct reloc_range_list_entry_struct ++{ ++ reloc_range_list_entry *next; ++ reloc_range_list_entry *prev; ++ Elf_Internal_Rela *irel; ++ xtensa_opcode opcode; ++ int opnum; ++}; ++ ++struct reloc_range_list_struct ++{ ++ /* The rest of the structure is only meaningful when ok is TRUE. */ ++ bfd_boolean ok; ++ ++ unsigned n_range; /* Number of range markers. */ ++ reloc_range *range; /* Sorted range markers. */ ++ ++ unsigned first; /* Index of a first range element in the list. */ ++ unsigned last; /* One past index of a last range element in the list. */ ++ ++ unsigned n_list; /* Number of list elements. */ ++ reloc_range_list_entry *reloc; /* */ ++ reloc_range_list_entry list_root; ++}; ++ ++static int ++reloc_range_compare (const void *a, const void *b) ++{ ++ const reloc_range *ra = a; ++ const reloc_range *rb = b; ++ ++ if (ra->addr != rb->addr) ++ return ra->addr < rb->addr ? -1 : 1; ++ if (ra->add != rb->add) ++ return ra->add ? -1 : 1; ++ return 0; ++} ++ ++static void ++build_reloc_ranges (bfd *abfd, asection *sec, ++ bfd_byte *contents, ++ Elf_Internal_Rela *internal_relocs, ++ xtensa_opcode *reloc_opcodes, ++ reloc_range_list *list) ++{ ++ unsigned i; ++ size_t n = 0; ++ size_t max_n = 0; ++ reloc_range *ranges = NULL; ++ reloc_range_list_entry *reloc = ++ bfd_malloc (sec->reloc_count * sizeof (*reloc)); ++ ++ memset (list, 0, sizeof (*list)); ++ list->ok = TRUE; ++ ++ for (i = 0; i < sec->reloc_count; i++) ++ { ++ Elf_Internal_Rela *irel = &internal_relocs[i]; ++ int r_type = ELF32_R_TYPE (irel->r_info); ++ reloc_howto_type *howto = &elf_howto_table[r_type]; ++ r_reloc r_rel; ++ ++ if (r_type == R_XTENSA_ASM_SIMPLIFY ++ || r_type == R_XTENSA_32_PCREL ++ || !howto->pc_relative) ++ continue; ++ ++ r_reloc_init (&r_rel, abfd, irel, contents, ++ bfd_get_section_limit (abfd, sec)); ++ ++ if (r_reloc_get_section (&r_rel) != sec) ++ continue; ++ ++ if (n + 2 > max_n) ++ { ++ max_n = (max_n + 2) * 2; ++ ranges = bfd_realloc (ranges, max_n * sizeof (*ranges)); ++ } ++ ++ ranges[n].addr = irel->r_offset; ++ ranges[n + 1].addr = r_rel.target_offset; ++ ++ ranges[n].add = ranges[n].addr < ranges[n + 1].addr; ++ ranges[n + 1].add = !ranges[n].add; ++ ++ ranges[n].irel_index = i; ++ ranges[n + 1].irel_index = i; ++ ++ n += 2; ++ ++ reloc[i].irel = irel; ++ ++ /* Every relocation won't possibly be checked in the optimized version of ++ check_section_ebb_pcrels_fit, so this needs to be done here. */ ++ if (is_alt_relocation (ELF32_R_TYPE (irel->r_info))) ++ { ++ /* None of the current alternate relocs are PC-relative, ++ and only PC-relative relocs matter here. */ ++ } ++ else ++ { ++ xtensa_opcode opcode; ++ int opnum; ++ ++ if (reloc_opcodes) ++ opcode = reloc_opcodes[i]; ++ else ++ opcode = get_relocation_opcode (abfd, sec, contents, irel); ++ ++ if (opcode == XTENSA_UNDEFINED) ++ { ++ list->ok = FALSE; ++ break; ++ } ++ ++ opnum = get_relocation_opnd (opcode, ELF32_R_TYPE (irel->r_info)); ++ if (opnum == XTENSA_UNDEFINED) ++ { ++ list->ok = FALSE; ++ break; ++ } ++ ++ /* Record relocation opcode and opnum as we've calculated them ++ anyway and they won't change. */ ++ reloc[i].opcode = opcode; ++ reloc[i].opnum = opnum; ++ } ++ } ++ ++ if (list->ok) ++ { ++ ranges = bfd_realloc (ranges, n * sizeof (*ranges)); ++ qsort (ranges, n, sizeof (*ranges), reloc_range_compare); ++ ++ list->n_range = n; ++ list->range = ranges; ++ list->reloc = reloc; ++ list->list_root.prev = &list->list_root; ++ list->list_root.next = &list->list_root; ++ } ++ else ++ { ++ free (ranges); ++ free (reloc); ++ } ++} ++ ++static void reloc_range_list_append (reloc_range_list *list, ++ unsigned irel_index) ++{ ++ reloc_range_list_entry *entry = list->reloc + irel_index; ++ ++ entry->prev = list->list_root.prev; ++ entry->next = &list->list_root; ++ entry->prev->next = entry; ++ entry->next->prev = entry; ++ ++list->n_list; ++} ++ ++static void reloc_range_list_remove (reloc_range_list *list, ++ unsigned irel_index) ++{ ++ reloc_range_list_entry *entry = list->reloc + irel_index; ++ ++ entry->next->prev = entry->prev; ++ entry->prev->next = entry->next; ++ --list->n_list; ++} ++ ++/* Update relocation list object so that it lists all relocations that cross ++ [first; last] range. Range bounds should not decrease with successive ++ invocations. */ ++static void reloc_range_list_update_range (reloc_range_list *list, ++ bfd_vma first, bfd_vma last) ++{ ++ /* This should not happen: EBBs are iterated from lower addresses to higher. ++ But even if that happens there's no need to break: just flush current list ++ and start from scratch. */ ++ if ((list->last > 0 && list->range[list->last - 1].addr > last) || ++ (list->first > 0 && list->range[list->first - 1].addr >= first)) ++ { ++ list->first = 0; ++ list->last = 0; ++ list->n_list = 0; ++ list->list_root.next = &list->list_root; ++ list->list_root.prev = &list->list_root; ++ fprintf (stderr, "%s: move backwards requested\n", __func__); ++ } ++ ++ for (; list->last < list->n_range && ++ list->range[list->last].addr <= last; ++list->last) ++ if (list->range[list->last].add) ++ reloc_range_list_append (list, list->range[list->last].irel_index); ++ ++ for (; list->first < list->n_range && ++ list->range[list->first].addr < first; ++list->first) ++ if (!list->range[list->first].add) ++ reloc_range_list_remove (list, list->range[list->first].irel_index); ++} ++ ++static void free_reloc_range_list (reloc_range_list *list) ++{ ++ free (list->range); ++ free (list->reloc); ++} + + /* The compute_text_actions function will build a list of potential + transformation actions for code in the extended basic block of each +@@ -7245,6 +7462,7 @@ compute_text_actions (bfd *abfd, + property_table_entry *prop_table = 0; + int ptblsize = 0; + bfd_size_type sec_size; ++ reloc_range_list relevant_relocs; + + relax_info = get_xtensa_relax_info (sec); + BFD_ASSERT (relax_info); +@@ -7277,6 +7495,12 @@ compute_text_actions (bfd *abfd, + goto error_return; + } + ++ /* Precompute the opcode for each relocation. */ ++ reloc_opcodes = build_reloc_opcodes (abfd, sec, contents, internal_relocs); ++ ++ build_reloc_ranges (abfd, sec, contents, internal_relocs, reloc_opcodes, ++ &relevant_relocs); ++ + for (i = 0; i < sec->reloc_count; i++) + { + Elf_Internal_Rela *irel = &internal_relocs[i]; +@@ -7340,17 +7564,13 @@ compute_text_actions (bfd *abfd, + ebb->start_reloc_idx = i; + ebb->end_reloc_idx = i; + +- /* Precompute the opcode for each relocation. */ +- if (reloc_opcodes == NULL) +- reloc_opcodes = build_reloc_opcodes (abfd, sec, contents, +- internal_relocs); +- + if (!extend_ebb_bounds (ebb) + || !compute_ebb_proposed_actions (&ebb_table) + || !compute_ebb_actions (&ebb_table) + || !check_section_ebb_pcrels_fit (abfd, sec, contents, +- internal_relocs, &ebb_table, +- reloc_opcodes) ++ internal_relocs, ++ &relevant_relocs, ++ &ebb_table, reloc_opcodes) + || !check_section_ebb_reduces (&ebb_table)) + { + /* If anything goes wrong or we get unlucky and something does +@@ -7372,6 +7592,8 @@ compute_text_actions (bfd *abfd, + free_ebb_constraint (&ebb_table); + } + ++ free_reloc_range_list (&relevant_relocs); ++ + #if DEBUG + if (relax_info->action_list.head) + print_action_list (stderr, &relax_info->action_list); +@@ -7974,14 +8196,17 @@ check_section_ebb_pcrels_fit (bfd *abfd, + asection *sec, + bfd_byte *contents, + Elf_Internal_Rela *internal_relocs, ++ reloc_range_list *relevant_relocs, + const ebb_constraint *constraint, + const xtensa_opcode *reloc_opcodes) + { + unsigned i, j; ++ unsigned n = sec->reloc_count; + Elf_Internal_Rela *irel; + xlate_map_t *xmap = NULL; + bfd_boolean ok = TRUE; + xtensa_relax_info *relax_info; ++ reloc_range_list_entry *entry = NULL; + + relax_info = get_xtensa_relax_info (sec); + +@@ -7992,7 +8217,40 @@ check_section_ebb_pcrels_fit (bfd *abfd, + can still be used. */ + } + +- for (i = 0; i < sec->reloc_count; i++) ++ if (relevant_relocs && constraint->action_count) ++ { ++ if (!relevant_relocs->ok) ++ { ++ ok = FALSE; ++ n = 0; ++ } ++ else ++ { ++ bfd_vma min_offset, max_offset; ++ min_offset = max_offset = constraint->actions[0].offset; ++ ++ for (i = 1; i < constraint->action_count; ++i) ++ { ++ proposed_action *action = &constraint->actions[i]; ++ bfd_vma offset = action->offset; ++ ++ if (offset < min_offset) ++ min_offset = offset; ++ if (offset > max_offset) ++ max_offset = offset; ++ } ++ reloc_range_list_update_range (relevant_relocs, min_offset, ++ max_offset); ++ n = relevant_relocs->n_list; ++ entry = &relevant_relocs->list_root; ++ } ++ } ++ else ++ { ++ relevant_relocs = NULL; ++ } ++ ++ for (i = 0; i < n; i++) + { + r_reloc r_rel; + bfd_vma orig_self_offset, orig_target_offset; +@@ -8001,7 +8259,15 @@ check_section_ebb_pcrels_fit (bfd *abfd, + reloc_howto_type *howto; + int self_removed_bytes, target_removed_bytes; + +- irel = &internal_relocs[i]; ++ if (relevant_relocs) ++ { ++ entry = entry->next; ++ irel = entry->irel; ++ } ++ else ++ { ++ irel = internal_relocs + i; ++ } + r_type = ELF32_R_TYPE (irel->r_info); + + howto = &elf_howto_table[r_type]; +@@ -8067,21 +8333,30 @@ check_section_ebb_pcrels_fit (bfd *abfd, + xtensa_opcode opcode; + int opnum; + +- if (reloc_opcodes) +- opcode = reloc_opcodes[i]; +- else +- opcode = get_relocation_opcode (abfd, sec, contents, irel); +- if (opcode == XTENSA_UNDEFINED) ++ if (relevant_relocs) + { +- ok = FALSE; +- break; ++ opcode = entry->opcode; ++ opnum = entry->opnum; + } +- +- opnum = get_relocation_opnd (opcode, ELF32_R_TYPE (irel->r_info)); +- if (opnum == XTENSA_UNDEFINED) ++ else + { +- ok = FALSE; +- break; ++ if (reloc_opcodes) ++ opcode = reloc_opcodes[relevant_relocs ? ++ (unsigned)(entry - relevant_relocs->reloc) : i]; ++ else ++ opcode = get_relocation_opcode (abfd, sec, contents, irel); ++ if (opcode == XTENSA_UNDEFINED) ++ { ++ ok = FALSE; ++ break; ++ } ++ ++ opnum = get_relocation_opnd (opcode, ELF32_R_TYPE (irel->r_info)); ++ if (opnum == XTENSA_UNDEFINED) ++ { ++ ok = FALSE; ++ break; ++ } + } + + if (!pcrel_reloc_fits (opcode, opnum, self_offset, target_offset)) +@@ -8778,7 +9053,7 @@ move_shared_literal (asection *sec, + /* Check all of the PC-relative relocations to make sure they still fit. */ + relocs_fit = check_section_ebb_pcrels_fit (target_sec->owner, target_sec, + target_sec_cache->contents, +- target_sec_cache->relocs, ++ target_sec_cache->relocs, NULL, + &ebb_table, NULL); + + if (!relocs_fit) +-- +1.8.1.4 + diff --git a/patches/binutils/2.25.1/907-xtensa-optimize-removed_by_actions.patch b/patches/binutils/2.25.1/907-xtensa-optimize-removed_by_actions.patch new file mode 100644 index 0000000..9df8065 --- /dev/null +++ b/patches/binutils/2.25.1/907-xtensa-optimize-removed_by_actions.patch @@ -0,0 +1,356 @@ +From 3e3f60207399ab29dd55af109e5ae9facc7d8e83 Mon Sep 17 00:00:00 2001 +From: Max Filippov +Date: Sat, 28 Mar 2015 08:46:28 +0300 +Subject: [PATCH 2/4] xtensa: optimize removed_by_actions + +The function removed_by_actions iterates through text actions to +calculate an offset applied by text actions to a given VMA. Although it +has a parameter p_start_action that allows for incremental offset +calculation, in many places it's used with p_start_action explicitly set +to the first action. After the first relaxation pass when the list of +text actions is finalized, an array of offsets sorted by VMA may be used +to speed up this function. + +Original profile: + +% time self children called name +----------------------------------------- + 0.35 0.00 33872/4808961 relax_section_symbols + 3.32 0.00 326022/4808961 relax_property_section + 12.83 0.00 1259379/4808961 offset_with_removed_text + 32.50 0.00 3189688/4808961 translate_reloc + 71.5 49.00 0.00 4808961 removed_by_actions +----------------------------------------- + +Same data, after optimization: + +% time self children called name +----------------------------------------- + 0.00 0.00 33872/4808537 relax_section_symbols + 0.01 0.00 326022/4808537 relax_property_section + 0.05 0.00 1258955/4808537 offset_with_removed_text_map + 0.13 0.00 3189688/4808537 translate_reloc + 1.0 0.20 0.00 4808537 removed_by_actions_map + 0.00 0.00 120/120 map_removal_by_action +----------------------------------------- + +2015-04-01 Max Filippov +bfd/ + * elf32-xtensa.c (removal_by_action_entry_struct, + removal_by_action_map_struct): new structures. + (removal_by_action_entry, removal_by_action_map): new typedefs. + (text_action_list_struct): add new field: map. + (map_removal_by_action, removed_by_actions_map, + offset_with_removed_text_map): new functions. + (relax_section): replace offset_with_removed_text with + offset_with_removed_text_map. + (translate_reloc, relax_property_section, relax_section_symbols): + replace removed_by_actions with removed_by_actions_map. + +Backported from: 071aa5c98a31c966f5fbfc573fcee61350fd1936 +Signed-off-by: Max Filippov +--- + bfd/elf32-xtensa.c | 181 +++++++++++++++++++++++++++++++++++++++++++++-------- + 1 file changed, 156 insertions(+), 25 deletions(-) + +diff --git a/bfd/elf32-xtensa.c b/bfd/elf32-xtensa.c +index 872370b..21b2871 100644 +--- a/bfd/elf32-xtensa.c ++++ b/bfd/elf32-xtensa.c +@@ -5420,11 +5420,28 @@ struct text_action_struct + text_action *next; + }; + ++struct removal_by_action_entry_struct ++{ ++ bfd_vma offset; ++ int removed; ++ int eq_removed; ++ int eq_removed_before_fill; ++}; ++typedef struct removal_by_action_entry_struct removal_by_action_entry; ++ ++struct removal_by_action_map_struct ++{ ++ unsigned n_entries; ++ removal_by_action_entry *entry; ++}; ++typedef struct removal_by_action_map_struct removal_by_action_map; ++ + + /* List of all of the actions taken on a text section. */ + struct text_action_list_struct + { + text_action *head; ++ removal_by_action_map map; + }; + + +@@ -5636,6 +5653,101 @@ action_list_count (text_action_list *action_list) + return count; + } + ++static void ++map_removal_by_action (text_action_list *action_list) ++{ ++ text_action *r; ++ int removed = 0; ++ removal_by_action_map map; ++ bfd_boolean eq_complete; ++ ++ map.n_entries = 0; ++ map.entry = bfd_malloc (action_list_count (action_list) * ++ sizeof (removal_by_action_entry)); ++ eq_complete = FALSE; ++ ++ for (r = action_list->head; r;) ++ { ++ removal_by_action_entry *ientry = map.entry + map.n_entries; ++ ++ if (map.n_entries && (ientry - 1)->offset == r->offset) ++ { ++ --ientry; ++ } ++ else ++ { ++ ++map.n_entries; ++ eq_complete = FALSE; ++ ientry->offset = r->offset; ++ ientry->eq_removed_before_fill = removed; ++ } ++ ++ if (!eq_complete) ++ { ++ if (r->action != ta_fill || r->removed_bytes >= 0) ++ { ++ ientry->eq_removed = removed; ++ eq_complete = TRUE; ++ } ++ else ++ ientry->eq_removed = removed + r->removed_bytes; ++ } ++ ++ removed += r->removed_bytes; ++ ientry->removed = removed; ++ r = r->next; ++ } ++ action_list->map = map; ++} ++ ++static int ++removed_by_actions_map (text_action_list *action_list, bfd_vma offset, ++ bfd_boolean before_fill) ++{ ++ unsigned a, b; ++ ++ if (!action_list->map.entry) ++ map_removal_by_action (action_list); ++ ++ if (!action_list->map.n_entries) ++ return 0; ++ ++ a = 0; ++ b = action_list->map.n_entries; ++ ++ while (b - a > 1) ++ { ++ unsigned c = (a + b) / 2; ++ ++ if (action_list->map.entry[c].offset <= offset) ++ a = c; ++ else ++ b = c; ++ } ++ ++ if (action_list->map.entry[a].offset < offset) ++ { ++ return action_list->map.entry[a].removed; ++ } ++ else if (action_list->map.entry[a].offset == offset) ++ { ++ return before_fill ? ++ action_list->map.entry[a].eq_removed_before_fill : ++ action_list->map.entry[a].eq_removed; ++ } ++ else ++ { ++ return 0; ++ } ++} ++ ++static bfd_vma ++offset_with_removed_text_map (text_action_list *action_list, bfd_vma offset) ++{ ++ int removed = removed_by_actions_map (action_list, offset, FALSE); ++ return offset - removed; ++} ++ + + /* The find_insn_action routine will only find non-fill actions. */ + +@@ -5909,6 +6021,9 @@ init_xtensa_relax_info (asection *sec) + + relax_info->action_list.head = NULL; + ++ relax_info->action_list.map.n_entries = 0; ++ relax_info->action_list.map.entry = NULL; ++ + relax_info->fix_list = NULL; + relax_info->fix_array = NULL; + relax_info->fix_array_count = 0; +@@ -9218,7 +9333,7 @@ relax_section (bfd *abfd, asection *sec, struct bfd_link_info *link_info) + if (elf_hash_table (link_info)->dynamic_sections_created) + shrink_dynamic_reloc_sections (link_info, abfd, sec, irel); + irel->r_info = ELF32_R_INFO (0, R_XTENSA_NONE); +- irel->r_offset = offset_with_removed_text ++ irel->r_offset = offset_with_removed_text_map + (&relax_info->action_list, irel->r_offset); + continue; + } +@@ -9255,7 +9370,7 @@ relax_section (bfd *abfd, asection *sec, struct bfd_link_info *link_info) + } + } + +- source_offset = offset_with_removed_text ++ source_offset = offset_with_removed_text_map + (&relax_info->action_list, irel->r_offset); + irel->r_offset = source_offset; + } +@@ -9352,7 +9467,7 @@ relax_section (bfd *abfd, asection *sec, struct bfd_link_info *link_info) + break; + } + +- new_end_offset = offset_with_removed_text ++ new_end_offset = offset_with_removed_text_map + (&target_relax_info->action_list, + r_rel.target_offset + diff_value); + diff_value = new_end_offset - new_reloc.target_offset; +@@ -9750,7 +9865,6 @@ translate_reloc (const r_reloc *orig_rel, r_reloc *new_rel, asection *sec) + xtensa_relax_info *relax_info; + removed_literal *removed; + bfd_vma target_offset, base_offset; +- text_action *act; + + *new_rel = *orig_rel; + +@@ -9803,19 +9917,26 @@ translate_reloc (const r_reloc *orig_rel, r_reloc *new_rel, asection *sec) + offset. */ + + base_offset = r_reloc_get_target_offset (new_rel) - new_rel->rela.r_addend; +- act = relax_info->action_list.head; + if (base_offset <= target_offset) + { +- int base_removed = removed_by_actions (&act, base_offset, FALSE); +- int addend_removed = removed_by_actions (&act, target_offset, FALSE); ++ int base_removed = removed_by_actions_map (&relax_info->action_list, ++ base_offset, FALSE); ++ int addend_removed = removed_by_actions_map (&relax_info->action_list, ++ target_offset, FALSE) - ++ base_removed; ++ + new_rel->target_offset = target_offset - base_removed - addend_removed; + new_rel->rela.r_addend -= addend_removed; + } + else + { + /* Handle a negative addend. The base offset comes first. */ +- int tgt_removed = removed_by_actions (&act, target_offset, FALSE); +- int addend_removed = removed_by_actions (&act, base_offset, FALSE); ++ int tgt_removed = removed_by_actions_map (&relax_info->action_list, ++ target_offset, FALSE); ++ int addend_removed = removed_by_actions_map (&relax_info->action_list, ++ base_offset, FALSE) - ++ tgt_removed; ++ + new_rel->target_offset = target_offset - tgt_removed; + new_rel->rela.r_addend += addend_removed; + } +@@ -10138,9 +10259,10 @@ relax_property_section (bfd *abfd, + bfd_vma old_offset = val.r_rel.target_offset; + bfd_vma new_offset; + long old_size, new_size; +- text_action *act = target_relax_info->action_list.head; +- new_offset = old_offset - +- removed_by_actions (&act, old_offset, FALSE); ++ int removed_by_old_offset = ++ removed_by_actions_map (&target_relax_info->action_list, ++ old_offset, FALSE); ++ new_offset = old_offset - removed_by_old_offset; + + /* Assert that we are not out of bounds. */ + old_size = bfd_get_32 (abfd, size_p); +@@ -10164,9 +10286,10 @@ relax_property_section (bfd *abfd, + + /* Recompute the new_offset, but this time don't + include any fill inserted by relaxation. */ +- act = target_relax_info->action_list.head; +- new_offset = old_offset - +- removed_by_actions (&act, old_offset, TRUE); ++ removed_by_old_offset = ++ removed_by_actions_map (&target_relax_info->action_list, ++ old_offset, TRUE); ++ new_offset = old_offset - removed_by_old_offset; + + /* If it is not unreachable and we have not yet + seen an unreachable at this address, place it +@@ -10182,8 +10305,12 @@ relax_property_section (bfd *abfd, + } + } + else +- new_size -= +- removed_by_actions (&act, old_offset + old_size, TRUE); ++ { ++ int removed_by_old_offset_size = ++ removed_by_actions_map (&target_relax_info->action_list, ++ old_offset + old_size, TRUE); ++ new_size -= removed_by_old_offset_size - removed_by_old_offset; ++ } + + if (new_size != old_size) + { +@@ -10441,14 +10568,16 @@ relax_section_symbols (bfd *abfd, asection *sec) + + if (isym->st_shndx == sec_shndx) + { +- text_action *act = relax_info->action_list.head; + bfd_vma orig_addr = isym->st_value; ++ int removed = removed_by_actions_map (&relax_info->action_list, ++ orig_addr, FALSE); + +- isym->st_value -= removed_by_actions (&act, orig_addr, FALSE); +- ++ isym->st_value -= removed; + if (ELF32_ST_TYPE (isym->st_info) == STT_FUNC) + isym->st_size -= +- removed_by_actions (&act, orig_addr + isym->st_size, FALSE); ++ removed_by_actions_map (&relax_info->action_list, ++ orig_addr + isym->st_size, FALSE) - ++ removed; + } + } + +@@ -10466,15 +10595,17 @@ relax_section_symbols (bfd *abfd, asection *sec) + || sym_hash->root.type == bfd_link_hash_defweak) + && sym_hash->root.u.def.section == sec) + { +- text_action *act = relax_info->action_list.head; + bfd_vma orig_addr = sym_hash->root.u.def.value; ++ int removed = removed_by_actions_map (&relax_info->action_list, ++ orig_addr, FALSE); + +- sym_hash->root.u.def.value -= +- removed_by_actions (&act, orig_addr, FALSE); ++ sym_hash->root.u.def.value -= removed; + + if (sym_hash->type == STT_FUNC) + sym_hash->size -= +- removed_by_actions (&act, orig_addr + sym_hash->size, FALSE); ++ removed_by_actions_map (&relax_info->action_list, ++ orig_addr + sym_hash->size, FALSE) - ++ removed; + } + } + +-- +1.8.1.4 + diff --git a/patches/binutils/2.25.1/908-xtensa-optimize-find_removed_literal.patch b/patches/binutils/2.25.1/908-xtensa-optimize-find_removed_literal.patch new file mode 100644 index 0000000..96d526f --- /dev/null +++ b/patches/binutils/2.25.1/908-xtensa-optimize-find_removed_literal.patch @@ -0,0 +1,146 @@ +From 288c2b709e5e6841484e1a129eaccd299db36877 Mon Sep 17 00:00:00 2001 +From: Max Filippov +Date: Sat, 4 Apr 2015 14:49:42 +0300 +Subject: [PATCH 3/4] xtensa: optimize find_removed_literal + +find_removed_literal uses linear search to find removed literal by its +VMA. The list of literals is fixed at that point, build an ordered index +array and use binary search instead. + +Original profile: + +% time self children called name +----------------------------------------- + 56.72 0.00 297578/669392 translate_reloc + 70.86 0.00 371814/669392 relax_section + 67.9 127.58 0.00 669392 find_removed_literal +----------------------------------------- + +Same data, after optimization: + +% time self children called name +----------------------------------------- + 0.00 0.00 297578/669392 translate_reloc + 0.00 0.00 371814/669392 relax_section + 0.0 0.00 0.00 669392 find_removed_literal + 0.00 0.00 23838/23838 map_removed_literal +----------------------------------------- + +2015-04-03 Max Filippov +bfd/ + * elf32-xtensa.c (removed_literal_map_entry): new typedef. + (removed_literal_map_entry_struct): new structure. + (removed_literal_list_struct): add new fields: n_map and map. + (map_removed_literal, removed_literal_compare): new functions. + (find_removed_literal): build index array for literals ordered + by VMA, use binary search to find removed literal. + +Backported from: 3439c466273378021821473d3fc84990e089ae34 +Signed-off-by: Max Filippov +--- + bfd/elf32-xtensa.c | 64 +++++++++++++++++++++++++++++++++++++++++++++++++----- + 1 file changed, 58 insertions(+), 6 deletions(-) + +diff --git a/bfd/elf32-xtensa.c b/bfd/elf32-xtensa.c +index 21b2871..51733ad 100644 +--- a/bfd/elf32-xtensa.c ++++ b/bfd/elf32-xtensa.c +@@ -5832,6 +5832,7 @@ print_action_list (FILE *fp, text_action_list *action_list) + by the "from" offset field. */ + + typedef struct removed_literal_struct removed_literal; ++typedef struct removed_literal_map_entry_struct removed_literal_map_entry; + typedef struct removed_literal_list_struct removed_literal_list; + + struct removed_literal_struct +@@ -5841,10 +5842,19 @@ struct removed_literal_struct + removed_literal *next; + }; + ++struct removed_literal_map_entry_struct ++{ ++ bfd_vma addr; ++ removed_literal *literal; ++}; ++ + struct removed_literal_list_struct + { + removed_literal *head; + removed_literal *tail; ++ ++ unsigned n_map; ++ removed_literal_map_entry *map; + }; + + +@@ -5893,6 +5903,39 @@ add_removed_literal (removed_literal_list *removed_list, + } + } + ++static void ++map_removed_literal (removed_literal_list *removed_list) ++{ ++ unsigned n_map = 0; ++ unsigned i; ++ removed_literal_map_entry *map = NULL; ++ removed_literal *r = removed_list->head; ++ ++ for (i = 0; r; ++i, r = r->next) ++ { ++ if (i == n_map) ++ { ++ n_map = (n_map * 2) + 2; ++ map = bfd_realloc (map, n_map * sizeof (*map)); ++ } ++ map[i].addr = r->from.target_offset; ++ map[i].literal = r; ++ } ++ removed_list->map = map; ++ removed_list->n_map = i; ++} ++ ++static int ++removed_literal_compare (const void *a, const void *b) ++{ ++ const removed_literal_map_entry *pa = a; ++ const removed_literal_map_entry *pb = b; ++ ++ if (pa->addr == pb->addr) ++ return 0; ++ else ++ return pa->addr < pb->addr ? -1 : 1; ++} + + /* Check if the list of removed literals contains an entry for the + given address. Return the entry if found. */ +@@ -5900,12 +5943,21 @@ add_removed_literal (removed_literal_list *removed_list, + static removed_literal * + find_removed_literal (removed_literal_list *removed_list, bfd_vma addr) + { +- removed_literal *r = removed_list->head; +- while (r && r->from.target_offset < addr) +- r = r->next; +- if (r && r->from.target_offset == addr) +- return r; +- return NULL; ++ removed_literal_map_entry *p; ++ removed_literal *r = NULL; ++ ++ if (removed_list->map == NULL) ++ map_removed_literal (removed_list); ++ ++ p = bsearch (&addr, removed_list->map, removed_list->n_map, ++ sizeof (*removed_list->map), removed_literal_compare); ++ if (p) ++ { ++ while (p != removed_list->map && (p - 1)->addr == addr) ++ --p; ++ r = p->literal; ++ } ++ return r; + } + + +-- +1.8.1.4 + diff --git a/patches/binutils/2.25.1/909-xtensa-replace-action-list-with-splay-tree.patch b/patches/binutils/2.25.1/909-xtensa-replace-action-list-with-splay-tree.patch new file mode 100644 index 0000000..3090cc2 --- /dev/null +++ b/patches/binutils/2.25.1/909-xtensa-replace-action-list-with-splay-tree.patch @@ -0,0 +1,826 @@ +From e5409aedd3ee2192855018a564650ffb75c26e60 Mon Sep 17 00:00:00 2001 +From: Max Filippov +Date: Sun, 5 Apr 2015 17:04:22 +0300 +Subject: [PATCH 4/4] xtensa: replace action list with splay tree + +text_action_add uses linear list search to order text actions list by +action VMA. The list is used at the first relaxation pass, when it's not +fixed yet. +Replace the list with splay tree from libiberty. + +Original profile: + +% time self children called name +----------------------------------------- + 0.00 0.00 14/158225 compute_text_actions + 3.62 0.00 25211/158225 remove_dead_literal + 8.42 0.00 58645/158225 coalesce_shared_literal + 10.68 0.00 74355/158225 text_action_add_proposed + 38.8 22.73 0.00 158225 text_action_add + 0.00 0.00 144527/293246 bfd_zmalloc +----------------------------------------- + +Same data, after optimization: + +% time self children called name +----------------------------------------- + 0.00 0.00 14/158225 compute_text_actions + 0.00 0.00 25211/158225 remove_dead_literal + 0.00 0.01 58645/158225 coalesce_shared_literal + 0.00 0.01 74355/158225 text_action_add_proposed + 0.1 0.00 0.02 158225 text_action_add + 0.01 0.00 144527/144527 splay_tree_insert + 0.00 0.00 144527/195130 splay_tree_lookup + 0.00 0.00 144527/293246 bfd_zmalloc +----------------------------------------- + +2015-04-03 Max Filippov +bfd/ + * elf32-xtensa.c (splay-tree.h): include header. + (text_action_struct): drop next pointer. + (text_action_list_struct): drop head pointer, add count and + tree fields. + (find_fill_action): instead of linear search in text_action_list + search in the tree. + (text_action_compare, action_first, action_next): new functions. + (text_action_add, text_action_add_literal): instead of linear + search and insertion insert new node into the tree. + (removed_by_actions): pass additional parameter: action_list, + use it to traverse the tree. + (offset_with_removed_text): pass additional action_list parameter + to removed_by_actions. + (map_action_fn_context): new typedef. + (map_action_fn_context_struct): new structure. + (map_action_fn): new function. + (map_removal_by_action): use splay_tree_foreach to build map. + (find_insn_action): replace linear search in text_action_list + with series of splay_tree_lookups. + (print_action, print_action_list_fn): new functions. + (print_action_list): use splay_tree_foreach. + (init_xtensa_relax_info): drop action_list.head initialization. + Initialize the tree. + (compute_text_actions): use non-zero action_list_count instead of + non-NULL action list. + (xlate_map_context): new typedef. + (xlate_map_context_struct): new structure. + (xlate_map_fn): new function. + (build_xlate_map): use splay_tree_foreach to build map. + (action_remove_bytes_fn): new function. + (relax_section): use zero action_list_count instead of NULL + action list. Use splay_tree_foreach to count final section size. + Drop unused variable 'removed'. + +Backported from: 4c2af04fe8b4452bf51d2debf1bb467fafcd0f08 +Signed-off-by: Max Filippov +--- + bfd/elf32-xtensa.c | 488 +++++++++++++++++++++++++++++++---------------------- + 1 file changed, 282 insertions(+), 206 deletions(-) + +diff --git a/bfd/elf32-xtensa.c b/bfd/elf32-xtensa.c +index 51733ad..53af1c6 100644 +--- a/bfd/elf32-xtensa.c ++++ b/bfd/elf32-xtensa.c +@@ -28,6 +28,7 @@ + #include "libbfd.h" + #include "elf-bfd.h" + #include "elf/xtensa.h" ++#include "splay-tree.h" + #include "xtensa-isa.h" + #include "xtensa-config.h" + +@@ -5416,8 +5417,6 @@ struct text_action_struct + bfd_vma virtual_offset; /* Zero except for adding literals. */ + int removed_bytes; + literal_value value; /* Only valid when adding literals. */ +- +- text_action *next; + }; + + struct removal_by_action_entry_struct +@@ -5440,7 +5439,8 @@ typedef struct removal_by_action_map_struct removal_by_action_map; + /* List of all of the actions taken on a text section. */ + struct text_action_list_struct + { +- text_action *head; ++ unsigned count; ++ splay_tree tree; + removal_by_action_map map; + }; + +@@ -5448,20 +5448,18 @@ struct text_action_list_struct + static text_action * + find_fill_action (text_action_list *l, asection *sec, bfd_vma offset) + { +- text_action **m_p; ++ text_action a; + + /* It is not necessary to fill at the end of a section. */ + if (sec->size == offset) + return NULL; + +- for (m_p = &l->head; *m_p && (*m_p)->offset <= offset; m_p = &(*m_p)->next) +- { +- text_action *t = *m_p; +- /* When the action is another fill at the same address, +- just increase the size. */ +- if (t->offset == offset && t->action == ta_fill) +- return t; +- } ++ a.offset = offset; ++ a.action = ta_fill; ++ ++ splay_tree_node node = splay_tree_lookup (l->tree, (splay_tree_key)&a); ++ if (node) ++ return (text_action *)node->value; + return NULL; + } + +@@ -5509,6 +5507,49 @@ adjust_fill_action (text_action *ta, int fill_diff) + } + + ++static int ++text_action_compare (splay_tree_key a, splay_tree_key b) ++{ ++ text_action *pa = (text_action *)a; ++ text_action *pb = (text_action *)b; ++ static const int action_priority[] = ++ { ++ [ta_fill] = 0, ++ [ta_none] = 1, ++ [ta_convert_longcall] = 2, ++ [ta_narrow_insn] = 3, ++ [ta_remove_insn] = 4, ++ [ta_remove_longcall] = 5, ++ [ta_remove_literal] = 6, ++ [ta_widen_insn] = 7, ++ [ta_add_literal] = 8, ++ }; ++ ++ if (pa->offset == pb->offset) ++ { ++ if (pa->action == pb->action) ++ return 0; ++ return action_priority[pa->action] - action_priority[pb->action]; ++ } ++ else ++ return pa->offset < pb->offset ? -1 : 1; ++} ++ ++static text_action * ++action_first (text_action_list *action_list) ++{ ++ splay_tree_node node = splay_tree_min (action_list->tree); ++ return node ? (text_action *)node->value : NULL; ++} ++ ++static text_action * ++action_next (text_action_list *action_list, text_action *action) ++{ ++ splay_tree_node node = splay_tree_successor (action_list->tree, ++ (splay_tree_key)action); ++ return node ? (text_action *)node->value : NULL; ++} ++ + /* Add a modification action to the text. For the case of adding or + removing space, modify any current fill and assume that + "unreachable_space" bytes can be freely contracted. Note that a +@@ -5521,8 +5562,8 @@ text_action_add (text_action_list *l, + bfd_vma offset, + int removed) + { +- text_action **m_p; + text_action *ta; ++ text_action a; + + /* It is not necessary to fill at the end of a section. */ + if (action == ta_fill && sec->size == offset) +@@ -5532,34 +5573,30 @@ text_action_add (text_action_list *l, + if (action == ta_fill && removed == 0) + return; + +- for (m_p = &l->head; *m_p && (*m_p)->offset <= offset; m_p = &(*m_p)->next) ++ a.action = action; ++ a.offset = offset; ++ ++ if (action == ta_fill) + { +- text_action *t = *m_p; ++ splay_tree_node node = splay_tree_lookup (l->tree, (splay_tree_key)&a); + +- if (action == ta_fill) ++ if (node) + { +- /* When the action is another fill at the same address, +- just increase the size. */ +- if (t->offset == offset && t->action == ta_fill) +- { +- t->removed_bytes += removed; +- return; +- } +- /* Fills need to happen before widens so that we don't +- insert fill bytes into the instruction stream. */ +- if (t->offset == offset && t->action == ta_widen_insn) +- break; ++ ta = (text_action *)node->value; ++ ta->removed_bytes += removed; ++ return; + } + } ++ else ++ BFD_ASSERT (splay_tree_lookup (l->tree, (splay_tree_key)&a) == NULL); + +- /* Create a new record and fill it up. */ + ta = (text_action *) bfd_zmalloc (sizeof (text_action)); + ta->action = action; + ta->sec = sec; + ta->offset = offset; + ta->removed_bytes = removed; +- ta->next = (*m_p); +- *m_p = ta; ++ splay_tree_insert (l->tree, (splay_tree_key)ta, (splay_tree_value)ta); ++ ++l->count; + } + + +@@ -5570,7 +5607,6 @@ text_action_add_literal (text_action_list *l, + const literal_value *value, + int removed) + { +- text_action **m_p; + text_action *ta; + asection *sec = r_reloc_get_section (loc); + bfd_vma offset = loc->target_offset; +@@ -5578,14 +5614,6 @@ text_action_add_literal (text_action_list *l, + + BFD_ASSERT (action == ta_add_literal); + +- for (m_p = &l->head; *m_p != NULL; m_p = &(*m_p)->next) +- { +- if ((*m_p)->offset > offset +- && ((*m_p)->offset != offset +- || (*m_p)->virtual_offset > virtual_offset)) +- break; +- } +- + /* Create a new record and fill it up. */ + ta = (text_action *) bfd_zmalloc (sizeof (text_action)); + ta->action = action; +@@ -5594,8 +5622,10 @@ text_action_add_literal (text_action_list *l, + ta->virtual_offset = virtual_offset; + ta->value = *value; + ta->removed_bytes = removed; +- ta->next = (*m_p); +- *m_p = ta; ++ ++ BFD_ASSERT (splay_tree_lookup (l->tree, (splay_tree_key)ta) == NULL); ++ splay_tree_insert (l->tree, (splay_tree_key)ta, (splay_tree_value)ta); ++ ++l->count; + } + + +@@ -5606,7 +5636,8 @@ text_action_add_literal (text_action_list *l, + so that each search may begin where the previous one left off. */ + + static int +-removed_by_actions (text_action **p_start_action, ++removed_by_actions (text_action_list *action_list, ++ text_action **p_start_action, + bfd_vma offset, + bfd_boolean before_fill) + { +@@ -5614,6 +5645,13 @@ removed_by_actions (text_action **p_start_action, + int removed = 0; + + r = *p_start_action; ++ if (r) ++ { ++ splay_tree_node node = splay_tree_lookup (action_list->tree, ++ (splay_tree_key)r); ++ BFD_ASSERT (node != NULL && r == (text_action *)node->value); ++ } ++ + while (r) + { + if (r->offset > offset) +@@ -5625,7 +5663,7 @@ removed_by_actions (text_action **p_start_action, + + removed += r->removed_bytes; + +- r = r->next; ++ r = action_next (action_list, r); + } + + *p_start_action = r; +@@ -5636,68 +5674,74 @@ removed_by_actions (text_action **p_start_action, + static bfd_vma + offset_with_removed_text (text_action_list *action_list, bfd_vma offset) + { +- text_action *r = action_list->head; +- return offset - removed_by_actions (&r, offset, FALSE); ++ text_action *r = action_first (action_list); ++ ++ return offset - removed_by_actions (action_list, &r, offset, FALSE); + } + + + static unsigned + action_list_count (text_action_list *action_list) + { +- text_action *r = action_list->head; +- unsigned count = 0; +- for (r = action_list->head; r != NULL; r = r->next) +- { +- count++; +- } +- return count; ++ return action_list->count; + } + +-static void +-map_removal_by_action (text_action_list *action_list) ++typedef struct map_action_fn_context_struct map_action_fn_context; ++struct map_action_fn_context_struct + { +- text_action *r; +- int removed = 0; ++ int removed; + removal_by_action_map map; + bfd_boolean eq_complete; ++}; + +- map.n_entries = 0; +- map.entry = bfd_malloc (action_list_count (action_list) * +- sizeof (removal_by_action_entry)); +- eq_complete = FALSE; ++static int ++map_action_fn (splay_tree_node node, void *p) ++{ ++ map_action_fn_context *ctx = p; ++ text_action *r = (text_action *)node->value; ++ removal_by_action_entry *ientry = ctx->map.entry + ctx->map.n_entries; + +- for (r = action_list->head; r;) ++ if (ctx->map.n_entries && (ientry - 1)->offset == r->offset) + { +- removal_by_action_entry *ientry = map.entry + map.n_entries; ++ --ientry; ++ } ++ else ++ { ++ ++ctx->map.n_entries; ++ ctx->eq_complete = FALSE; ++ ientry->offset = r->offset; ++ ientry->eq_removed_before_fill = ctx->removed; ++ } + +- if (map.n_entries && (ientry - 1)->offset == r->offset) ++ if (!ctx->eq_complete) ++ { ++ if (r->action != ta_fill || r->removed_bytes >= 0) + { +- --ientry; ++ ientry->eq_removed = ctx->removed; ++ ctx->eq_complete = TRUE; + } + else +- { +- ++map.n_entries; +- eq_complete = FALSE; +- ientry->offset = r->offset; +- ientry->eq_removed_before_fill = removed; +- } ++ ientry->eq_removed = ctx->removed + r->removed_bytes; ++ } + +- if (!eq_complete) +- { +- if (r->action != ta_fill || r->removed_bytes >= 0) +- { +- ientry->eq_removed = removed; +- eq_complete = TRUE; +- } +- else +- ientry->eq_removed = removed + r->removed_bytes; +- } ++ ctx->removed += r->removed_bytes; ++ ientry->removed = ctx->removed; ++ return 0; ++} + +- removed += r->removed_bytes; +- ientry->removed = removed; +- r = r->next; +- } +- action_list->map = map; ++static void ++map_removal_by_action (text_action_list *action_list) ++{ ++ map_action_fn_context ctx; ++ ++ ctx.removed = 0; ++ ctx.map.n_entries = 0; ++ ctx.map.entry = bfd_malloc (action_list_count (action_list) * ++ sizeof (removal_by_action_entry)); ++ ctx.eq_complete = FALSE; ++ ++ splay_tree_foreach (action_list->tree, map_action_fn, &ctx); ++ action_list->map = ctx.map; + } + + static int +@@ -5754,28 +5798,26 @@ offset_with_removed_text_map (text_action_list *action_list, bfd_vma offset) + static text_action * + find_insn_action (text_action_list *action_list, bfd_vma offset) + { +- text_action *t; +- for (t = action_list->head; t; t = t->next) ++ static const text_action_t action[] = + { +- if (t->offset == offset) +- { +- switch (t->action) +- { +- case ta_none: +- case ta_fill: +- break; +- case ta_remove_insn: +- case ta_remove_longcall: +- case ta_convert_longcall: +- case ta_narrow_insn: +- case ta_widen_insn: +- return t; +- case ta_remove_literal: +- case ta_add_literal: +- BFD_ASSERT (0); +- break; +- } +- } ++ ta_convert_longcall, ++ ta_remove_longcall, ++ ta_widen_insn, ++ ta_narrow_insn, ++ ta_remove_insn, ++ }; ++ text_action a; ++ unsigned i; ++ ++ a.offset = offset; ++ for (i = 0; i < sizeof (action) / sizeof (*action); ++i) ++ { ++ splay_tree_node node; ++ ++ a.action = action[i]; ++ node = splay_tree_lookup (action_list->tree, (splay_tree_key)&a); ++ if (node) ++ return (text_action *)node->value; + } + return NULL; + } +@@ -5784,40 +5826,50 @@ find_insn_action (text_action_list *action_list, bfd_vma offset) + #if DEBUG + + static void +-print_action_list (FILE *fp, text_action_list *action_list) ++print_action (FILE *fp, text_action *r) ++{ ++ const char *t = "unknown"; ++ switch (r->action) ++ { ++ case ta_remove_insn: ++ t = "remove_insn"; break; ++ case ta_remove_longcall: ++ t = "remove_longcall"; break; ++ case ta_convert_longcall: ++ t = "convert_longcall"; break; ++ case ta_narrow_insn: ++ t = "narrow_insn"; break; ++ case ta_widen_insn: ++ t = "widen_insn"; break; ++ case ta_fill: ++ t = "fill"; break; ++ case ta_none: ++ t = "none"; break; ++ case ta_remove_literal: ++ t = "remove_literal"; break; ++ case ta_add_literal: ++ t = "add_literal"; break; ++ } ++ ++ fprintf (fp, "%s: %s[0x%lx] \"%s\" %d\n", ++ r->sec->owner->filename, ++ r->sec->name, (unsigned long) r->offset, t, r->removed_bytes); ++} ++ ++static int ++print_action_list_fn (splay_tree_node node, void *p) + { +- text_action *r; ++ text_action *r = (text_action *)node->value; + +- fprintf (fp, "Text Action\n"); +- for (r = action_list->head; r != NULL; r = r->next) +- { +- const char *t = "unknown"; +- switch (r->action) +- { +- case ta_remove_insn: +- t = "remove_insn"; break; +- case ta_remove_longcall: +- t = "remove_longcall"; break; +- case ta_convert_longcall: +- t = "convert_longcall"; break; +- case ta_narrow_insn: +- t = "narrow_insn"; break; +- case ta_widen_insn: +- t = "widen_insn"; break; +- case ta_fill: +- t = "fill"; break; +- case ta_none: +- t = "none"; break; +- case ta_remove_literal: +- t = "remove_literal"; break; +- case ta_add_literal: +- t = "add_literal"; break; +- } ++ print_action (p, r); ++ return 0; ++} + +- fprintf (fp, "%s: %s[0x%lx] \"%s\" %d\n", +- r->sec->owner->filename, +- r->sec->name, (unsigned long) r->offset, t, r->removed_bytes); +- } ++static void ++print_action_list (FILE *fp, text_action_list *action_list) ++{ ++ fprintf (fp, "Text Action\n"); ++ splay_tree_foreach (action_list->tree, print_action_list_fn, fp); + } + + #endif /* DEBUG */ +@@ -6071,8 +6123,8 @@ init_xtensa_relax_info (asection *sec) + relax_info->removed_list.head = NULL; + relax_info->removed_list.tail = NULL; + +- relax_info->action_list.head = NULL; +- ++ relax_info->action_list.tree = splay_tree_new (text_action_compare, ++ NULL, NULL); + relax_info->action_list.map.n_entries = 0; + relax_info->action_list.map.entry = NULL; + +@@ -7762,7 +7814,7 @@ compute_text_actions (bfd *abfd, + free_reloc_range_list (&relevant_relocs); + + #if DEBUG +- if (relax_info->action_list.head) ++ if (action_list_count (&relax_info->action_list)) + print_action_list (stderr, &relax_info->action_list); + #endif + +@@ -8263,6 +8315,54 @@ xlate_offset_with_removed_text (const xlate_map_t *map, + return e->new_address - e->orig_address + offset; + } + ++typedef struct xlate_map_context_struct xlate_map_context; ++struct xlate_map_context_struct ++{ ++ xlate_map_t *map; ++ xlate_map_entry_t *current_entry; ++ int removed; ++}; ++ ++static int ++xlate_map_fn (splay_tree_node node, void *p) ++{ ++ text_action *r = (text_action *)node->value; ++ xlate_map_context *ctx = p; ++ unsigned orig_size = 0; ++ ++ switch (r->action) ++ { ++ case ta_none: ++ case ta_remove_insn: ++ case ta_convert_longcall: ++ case ta_remove_literal: ++ case ta_add_literal: ++ break; ++ case ta_remove_longcall: ++ orig_size = 6; ++ break; ++ case ta_narrow_insn: ++ orig_size = 3; ++ break; ++ case ta_widen_insn: ++ orig_size = 2; ++ break; ++ case ta_fill: ++ break; ++ } ++ ctx->current_entry->size = ++ r->offset + orig_size - ctx->current_entry->orig_address; ++ if (ctx->current_entry->size != 0) ++ { ++ ctx->current_entry++; ++ ctx->map->entry_count++; ++ } ++ ctx->current_entry->orig_address = r->offset + orig_size; ++ ctx->removed += r->removed_bytes; ++ ctx->current_entry->new_address = r->offset + orig_size - ctx->removed; ++ ctx->current_entry->size = 0; ++ return 0; ++} + + /* Build a binary searchable offset translation map from a section's + action list. */ +@@ -8270,75 +8370,40 @@ xlate_offset_with_removed_text (const xlate_map_t *map, + static xlate_map_t * + build_xlate_map (asection *sec, xtensa_relax_info *relax_info) + { +- xlate_map_t *map = (xlate_map_t *) bfd_malloc (sizeof (xlate_map_t)); + text_action_list *action_list = &relax_info->action_list; + unsigned num_actions = 0; +- text_action *r; +- int removed; +- xlate_map_entry_t *current_entry; ++ xlate_map_context ctx; + +- if (map == NULL) ++ ctx.map = (xlate_map_t *) bfd_malloc (sizeof (xlate_map_t)); ++ ++ if (ctx.map == NULL) + return NULL; + + num_actions = action_list_count (action_list); +- map->entry = (xlate_map_entry_t *) ++ ctx.map->entry = (xlate_map_entry_t *) + bfd_malloc (sizeof (xlate_map_entry_t) * (num_actions + 1)); +- if (map->entry == NULL) ++ if (ctx.map->entry == NULL) + { +- free (map); ++ free (ctx.map); + return NULL; + } +- map->entry_count = 0; ++ ctx.map->entry_count = 0; + +- removed = 0; +- current_entry = &map->entry[0]; ++ ctx.removed = 0; ++ ctx.current_entry = &ctx.map->entry[0]; + +- current_entry->orig_address = 0; +- current_entry->new_address = 0; +- current_entry->size = 0; ++ ctx.current_entry->orig_address = 0; ++ ctx.current_entry->new_address = 0; ++ ctx.current_entry->size = 0; + +- for (r = action_list->head; r != NULL; r = r->next) +- { +- unsigned orig_size = 0; +- switch (r->action) +- { +- case ta_none: +- case ta_remove_insn: +- case ta_convert_longcall: +- case ta_remove_literal: +- case ta_add_literal: +- break; +- case ta_remove_longcall: +- orig_size = 6; +- break; +- case ta_narrow_insn: +- orig_size = 3; +- break; +- case ta_widen_insn: +- orig_size = 2; +- break; +- case ta_fill: +- break; +- } +- current_entry->size = +- r->offset + orig_size - current_entry->orig_address; +- if (current_entry->size != 0) +- { +- current_entry++; +- map->entry_count++; +- } +- current_entry->orig_address = r->offset + orig_size; +- removed += r->removed_bytes; +- current_entry->new_address = r->offset + orig_size - removed; +- current_entry->size = 0; +- } ++ splay_tree_foreach (action_list->tree, xlate_map_fn, &ctx); + +- current_entry->size = (bfd_get_section_limit (sec->owner, sec) +- - current_entry->orig_address); +- if (current_entry->size != 0) +- map->entry_count++; ++ ctx.current_entry->size = (bfd_get_section_limit (sec->owner, sec) ++ - ctx.current_entry->orig_address); ++ if (ctx.current_entry->size != 0) ++ ctx.map->entry_count++; + +- return map; ++ return ctx.map; + } + + +@@ -9302,6 +9367,16 @@ move_shared_literal (asection *sec, + + /* Second relaxation pass. */ + ++static int ++action_remove_bytes_fn (splay_tree_node node, void *p) ++{ ++ bfd_size_type *final_size = p; ++ text_action *action = (text_action *)node->value; ++ ++ *final_size -= action->removed_bytes; ++ return 0; ++} ++ + /* Modify all of the relocations to point to the right spot, and if this + is a relaxable section, delete the unwanted literals and fix the + section size. */ +@@ -9334,7 +9409,7 @@ relax_section (bfd *abfd, asection *sec, struct bfd_link_info *link_info) + + internal_relocs = retrieve_internal_relocs (abfd, sec, + link_info->keep_memory); +- if (!internal_relocs && !relax_info->action_list.head) ++ if (!internal_relocs && !action_list_count (&relax_info->action_list)) + return TRUE; + + contents = retrieve_contents (abfd, sec, link_info->keep_memory); +@@ -9412,6 +9487,12 @@ relax_section (bfd *abfd, asection *sec, struct bfd_link_info *link_info) + } + /* Update the action so that the code that moves + the contents will do the right thing. */ ++ /* ta_remove_longcall and ta_remove_insn actions are ++ grouped together in the tree as well as ++ ta_convert_longcall and ta_none, so that changes below ++ can be done w/o removing and reinserting action into ++ the tree. */ ++ + if (action->action == ta_remove_longcall) + action->action = ta_remove_insn; + else +@@ -9584,13 +9665,12 @@ relax_section (bfd *abfd, asection *sec, struct bfd_link_info *link_info) + + if ((relax_info->is_relaxable_literal_section + || relax_info->is_relaxable_asm_section) +- && relax_info->action_list.head) ++ && action_list_count (&relax_info->action_list)) + { + /* Walk through the planned actions and build up a table + of move, copy and fill records. Use the move, copy and + fill records to perform the actions once. */ + +- int removed = 0; + bfd_size_type final_size, copy_size, orig_insn_size; + bfd_byte *scratch = NULL; + bfd_byte *dup_contents = NULL; +@@ -9601,15 +9681,12 @@ relax_section (bfd *abfd, asection *sec, struct bfd_link_info *link_info) + bfd_vma orig_dot_vo = 0; /* Virtual offset from orig_dot. */ + bfd_vma dup_dot = 0; + +- text_action *action = relax_info->action_list.head; ++ text_action *action; + + final_size = sec->size; +- for (action = relax_info->action_list.head; action; +- action = action->next) +- { +- final_size -= action->removed_bytes; +- } + ++ splay_tree_foreach (relax_info->action_list.tree, ++ action_remove_bytes_fn, &final_size); + scratch = (bfd_byte *) bfd_zmalloc (final_size); + dup_contents = (bfd_byte *) bfd_zmalloc (final_size); + +@@ -9618,8 +9695,8 @@ relax_section (bfd *abfd, asection *sec, struct bfd_link_info *link_info) + print_action_list (stderr, &relax_info->action_list); + #endif + +- for (action = relax_info->action_list.head; action; +- action = action->next) ++ for (action = action_first (&relax_info->action_list); action; ++ action = action_next (&relax_info->action_list, action)) + { + virtual_action = FALSE; + if (action->offset > orig_dot) +@@ -9748,7 +9825,6 @@ relax_section (bfd *abfd, asection *sec, struct bfd_link_info *link_info) + break; + } + +- removed += action->removed_bytes; + BFD_ASSERT (dup_dot <= final_size); + BFD_ASSERT (orig_dot <= orig_size); + } +-- +1.8.1.4 + diff --git a/patches/binutils/2.25.1/910-xtensa-optimize-trampolines-relaxation.patch b/patches/binutils/2.25.1/910-xtensa-optimize-trampolines-relaxation.patch new file mode 100644 index 0000000..043ff4d --- /dev/null +++ b/patches/binutils/2.25.1/910-xtensa-optimize-trampolines-relaxation.patch @@ -0,0 +1,345 @@ +From cbe53e134d4c3a656880a906738ce19fdcd38e8b Mon Sep 17 00:00:00 2001 +From: Max Filippov +Date: Fri, 1 May 2015 11:39:12 +0300 +Subject: [PATCH] xtensa: optimize trampolines relaxation + +Currently every fixup in the current segment is checked when relaxing +trampoline frag. This is very expensive. Make a searchable array of +fixups pointing at potentially oversized jumps at the beginning of every +relaxation pass and only check subset of this cache in the reach of +single jump from the trampoline frag currently being relaxed. + +Original profile: + +% time self children called name +----------------------------------------- + 370.16 593.38 12283048/12283048 relax_segment + 98.4 370.16 593.38 12283048 xtensa_relax_frag + 58.91 269.26 2691463834/2699602236 xtensa_insnbuf_from_chars + 68.35 68.17 811266668/813338977 S_GET_VALUE + 36.85 29.51 2684369246/2685538060 xtensa_opcode_decode + 28.34 8.84 2684369246/2685538060 xtensa_format_get_slot + 12.39 5.94 2691463834/2699775044 xtensa_format_decode + 0.03 4.60 4101109/4101109 relax_frag_for_align + 0.18 1.76 994617/994617 relax_frag_immed + 0.07 0.09 24556277/24851220 new_logical_line + 0.06 0.00 12283048/14067410 as_where + 0.04 0.00 7094588/15460506 xtensa_format_num_slots + 0.00 0.00 1/712477 xtensa_insnbuf_alloc +----------------------------------------- + +Same data, after optimization: + +% time self children called name +----------------------------------------- + 0.51 7.47 12283048/12283048 relax_segment + 58.0 0.51 7.47 12283048 xtensa_relax_frag + 0.02 4.08 4101109/4101109 relax_frag_for_align + 0.18 1.39 994617/994617 relax_frag_immed + 0.01 0.98 555/555 xtensa_cache_relaxable_fixups + 0.21 0.25 7094588/16693271 xtensa_insnbuf_from_chars + 0.06 0.12 24556277/24851220 new_logical_line + 0.06 0.00 7094588/15460506 xtensa_format_num_slots + 0.02 0.04 7094588/16866079 xtensa_format_decode + 0.05 0.00 12283048/14067410 as_where + 0.00 0.00 1/712477 xtensa_insnbuf_alloc + 0.00 0.00 93808/93808 xtensa_find_first_cached_fixup +----------------------------------------- + +2015-05-02 Max Filippov +gas/ + * config/tc-xtensa.c (cached_fixupS, fixup_cacheS): New typedefs. + (struct cached_fixup, struct fixup_cache): New structures. + (fixup_order, xtensa_make_cached_fixup), + (xtensa_realloc_fixup_cache, xtensa_cache_relaxable_fixups), + (xtensa_find_first_cached_fixup, xtensa_delete_cached_fixup), + (xtensa_add_cached_fixup): New functions. + (xtensa_relax_frag): Cache fixups pointing at potentially + oversized jumps at the beginning of every relaxation pass. Only + check subset of this cache in the reach of single jump from the + trampoline frag currently being relaxed. + +Signed-off-by: Max Filippov +--- +Backported from: b76f99d702c3501ac320396ea06bc7f9237173c3 +Changes to ChangeLog are dropped. + + gas/config/tc-xtensa.c | 220 +++++++++++++++++++++++++++++++++++++++++++------ + 1 file changed, 194 insertions(+), 26 deletions(-) + +diff --git a/gas/config/tc-xtensa.c b/gas/config/tc-xtensa.c +index 3e85b69..31c0b6b 100644 +--- a/gas/config/tc-xtensa.c ++++ b/gas/config/tc-xtensa.c +@@ -8785,6 +8785,154 @@ static long relax_frag_for_align (fragS *, long); + static long relax_frag_immed + (segT, fragS *, long, int, xtensa_format, int, int *, bfd_boolean); + ++typedef struct cached_fixup cached_fixupS; ++struct cached_fixup ++{ ++ int addr; ++ int target; ++ int delta; ++ fixS *fixP; ++}; ++ ++typedef struct fixup_cache fixup_cacheS; ++struct fixup_cache ++{ ++ cached_fixupS *fixups; ++ unsigned n_fixups; ++ unsigned n_max; ++ ++ segT seg; ++ fragS *first_frag; ++}; ++ ++static int fixup_order (const void *a, const void *b) ++{ ++ const cached_fixupS *pa = a; ++ const cached_fixupS *pb = b; ++ ++ if (pa->addr == pb->addr) ++ { ++ if (pa->target == pb->target) ++ { ++ if (pa->fixP->fx_r_type == pb->fixP->fx_r_type) ++ return 0; ++ return pa->fixP->fx_r_type < pb->fixP->fx_r_type ? -1 : 1; ++ } ++ return pa->target - pb->target; ++ } ++ return pa->addr - pb->addr; ++} ++ ++static bfd_boolean xtensa_make_cached_fixup (cached_fixupS *o, fixS *fixP) ++{ ++ xtensa_isa isa = xtensa_default_isa; ++ int addr = fixP->fx_frag->fr_address; ++ int target; ++ int delta; ++ symbolS *s = fixP->fx_addsy; ++ int slot; ++ xtensa_format fmt; ++ xtensa_opcode opcode; ++ ++ if (fixP->fx_r_type < BFD_RELOC_XTENSA_SLOT0_OP || ++ fixP->fx_r_type > BFD_RELOC_XTENSA_SLOT14_OP) ++ return FALSE; ++ target = S_GET_VALUE (s); ++ delta = target - addr; ++ ++ if (abs(delta) < J_RANGE / 2) ++ return FALSE; ++ ++ xtensa_insnbuf_from_chars (isa, trampoline_buf, ++ (unsigned char *) fixP->fx_frag->fr_literal + ++ fixP->fx_where, 0); ++ fmt = xtensa_format_decode (isa, trampoline_buf); ++ gas_assert (fmt != XTENSA_UNDEFINED); ++ slot = fixP->tc_fix_data.slot; ++ xtensa_format_get_slot (isa, fmt, slot, trampoline_buf, trampoline_slotbuf); ++ opcode = xtensa_opcode_decode (isa, fmt, slot, trampoline_slotbuf); ++ if (opcode != xtensa_j_opcode) ++ return FALSE; ++ ++ o->addr = addr; ++ o->target = target; ++ o->delta = delta; ++ o->fixP = fixP; ++ ++ return TRUE; ++} ++ ++static void xtensa_realloc_fixup_cache (fixup_cacheS *cache, unsigned add) ++{ ++ if (cache->n_fixups + add > cache->n_max) ++ { ++ cache->n_max = (cache->n_fixups + add) * 2; ++ cache->fixups = xrealloc (cache->fixups, ++ sizeof (*cache->fixups) * cache->n_max); ++ } ++} ++ ++static void xtensa_cache_relaxable_fixups (fixup_cacheS *cache, ++ segment_info_type *seginfo) ++{ ++ fixS *fixP; ++ ++ cache->n_fixups = 0; ++ ++ for (fixP = seginfo->fix_root; fixP ; fixP = fixP->fx_next) ++ { ++ xtensa_realloc_fixup_cache (cache, 1); ++ ++ if (xtensa_make_cached_fixup (cache->fixups + cache->n_fixups, fixP)) ++ ++cache->n_fixups; ++ } ++ qsort (cache->fixups, cache->n_fixups, sizeof (*cache->fixups), fixup_order); ++} ++ ++static unsigned xtensa_find_first_cached_fixup (const fixup_cacheS *cache, ++ int addr) ++{ ++ unsigned a = 0; ++ unsigned b = cache->n_fixups; ++ ++ while (b - a > 1) ++ { ++ unsigned c = (a + b) / 2; ++ ++ if (cache->fixups[c].addr < addr) ++ a = c; ++ else ++ b = c; ++ } ++ return a; ++} ++ ++static void xtensa_delete_cached_fixup (fixup_cacheS *cache, unsigned i) ++{ ++ memmove (cache->fixups + i, cache->fixups + i + 1, ++ (cache->n_fixups - i - 1) * sizeof (*cache->fixups)); ++ --cache->n_fixups; ++} ++ ++static bfd_boolean xtensa_add_cached_fixup (fixup_cacheS *cache, fixS *fixP) ++{ ++ cached_fixupS o; ++ unsigned i; ++ ++ if (!xtensa_make_cached_fixup (&o, fixP)) ++ return FALSE; ++ xtensa_realloc_fixup_cache (cache, 1); ++ i = xtensa_find_first_cached_fixup (cache, o.addr); ++ if (i < cache->n_fixups) ++ { ++ ++i; ++ memmove (cache->fixups + i + 1, cache->fixups + i, ++ (cache->n_fixups - i) * sizeof (*cache->fixups)); ++ } ++ cache->fixups[i] = o; ++ ++cache->n_fixups; ++ return TRUE; ++} + + /* Return the number of bytes added to this fragment, given that the + input has been stretched already by "stretch". */ +@@ -8896,35 +9044,42 @@ xtensa_relax_frag (fragS *fragP, long stretch, int *stretched_p) + case RELAX_TRAMPOLINE: + if (fragP->tc_frag_data.relax_seen) + { +- segment_info_type *seginfo = seg_info (now_seg); +- fragS *fP; /* The out-of-range jump. */ +- fixS *fixP; ++ static fixup_cacheS fixup_cache; ++ segment_info_type *seginfo = seg_info (now_seg); ++ int trampaddr = fragP->fr_address + fragP->fr_fix; ++ int searchaddr = trampaddr < J_RANGE ? 0 : trampaddr - J_RANGE; ++ unsigned i; ++ ++ if (now_seg != fixup_cache.seg || ++ fragP == fixup_cache.first_frag || ++ fixup_cache.first_frag == NULL) ++ { ++ xtensa_cache_relaxable_fixups (&fixup_cache, seginfo); ++ fixup_cache.seg = now_seg; ++ fixup_cache.first_frag = fragP; ++ } + + /* Scan for jumps that will not reach. */ +- for (fixP = seginfo->fix_root; fixP ; fixP = fixP->fx_next) ++ for (i = xtensa_find_first_cached_fixup (&fixup_cache, searchaddr); ++ i < fixup_cache.n_fixups; ++i) ++ + { +- symbolS *s = fixP->fx_addsy; +- xtensa_opcode opcode; +- int target; +- int addr; +- int delta; +- +- if (fixP->fx_r_type < BFD_RELOC_XTENSA_SLOT0_OP || +- fixP->fx_r_type > BFD_RELOC_XTENSA_SLOT14_OP) +- continue; +- xtensa_insnbuf_from_chars (isa, trampoline_buf, +- (unsigned char *) fixP->fx_frag->fr_literal + fixP->fx_where, +- 0); +- fmt = xtensa_format_decode (isa, trampoline_buf); +- gas_assert (fmt != XTENSA_UNDEFINED); +- slot = fixP->tc_fix_data.slot; +- xtensa_format_get_slot (isa, fmt, slot, trampoline_buf, trampoline_slotbuf); +- opcode = xtensa_opcode_decode (isa, fmt, slot, trampoline_slotbuf); +- if (opcode != xtensa_j_opcode) ++ fixS *fixP = fixup_cache.fixups[i].fixP; ++ int target = fixup_cache.fixups[i].target; ++ int addr = fixup_cache.fixups[i].addr; ++ int delta = fixup_cache.fixups[i].delta + stretch; ++ ++ trampaddr = fragP->fr_address + fragP->fr_fix; ++ ++ if (addr + J_RANGE < trampaddr) + continue; +- target = S_GET_VALUE (s); +- addr = fixP->fx_frag->fr_address; +- delta = target - addr + stretch; ++ if (addr > trampaddr + J_RANGE) ++ break; ++ if (abs (delta) < J_RANGE) ++ continue; ++ ++ slot = fixP->tc_fix_data.slot; ++ + if (delta > J_RANGE || delta < -1 * J_RANGE) + { /* Found an out-of-range jump; scan the list of trampolines for the best match. */ + struct trampoline_seg *ts = find_trampoline_seg (now_seg); +@@ -8978,14 +9133,13 @@ xtensa_relax_frag (fragS *fragP, long stretch, int *stretched_p) + } + if (tf->fragP == fragP) + { +- int trampaddr = fragP->fr_address + fragP->fr_fix; +- + if (abs (addr - trampaddr) < J_RANGE) + { /* The trampoline is in range of original; fix it! */ + fixS *newfixP; + int offset; + TInsn insn; + symbolS *lsym; ++ fragS *fP; /* The out-of-range jump. */ + + new_stretch += init_trampoline_frag (tf); + offset = fragP->fr_fix; /* Where to assemble the j insn. */ +@@ -9009,10 +9163,20 @@ xtensa_relax_frag (fragS *fragP, long stretch, int *stretched_p) + newfixP->tc_fix_data.X_add_symbol = lsym; + newfixP->tc_fix_data.X_add_number = offset; + newfixP->tc_fix_data.slot = slot; ++ ++ xtensa_delete_cached_fixup (&fixup_cache, i); ++ xtensa_add_cached_fixup (&fixup_cache, newfixP); ++ + /* Move the fix-up from the original j insn to this one. */ + fixP->fx_frag = fragP; + fixP->fx_where = fragP->fr_fix - 3; + fixP->tc_fix_data.slot = 0; ++ ++ xtensa_add_cached_fixup (&fixup_cache, fixP); ++ ++ /* re-do current fixup */ ++ --i; ++ + /* Adjust the jump around this trampoline (if present). */ + if (tf->fixP != NULL) + { +@@ -9027,6 +9191,8 @@ xtensa_relax_frag (fragS *fragP, long stretch, int *stretched_p) + fragP->fr_subtype = 0; + /* Remove from the trampoline_list. */ + prev->next = tf->next; ++ if (fragP == fixup_cache.first_frag) ++ fixup_cache.first_frag = NULL; + break; + } + } +-- +1.8.1.4 + diff --git a/patches/binutils/2.25.1/911-xtensa-fix-localized-symbol-refcounting-with-gc-sect.patch b/patches/binutils/2.25.1/911-xtensa-fix-localized-symbol-refcounting-with-gc-sect.patch new file mode 100644 index 0000000..9ad6b3b --- /dev/null +++ b/patches/binutils/2.25.1/911-xtensa-fix-localized-symbol-refcounting-with-gc-sect.patch @@ -0,0 +1,57 @@ +From 8ec76b16f62d1bf386fb2c39af5f66c3afddc5cb Mon Sep 17 00:00:00 2001 +From: Max Filippov +Date: Thu, 14 May 2015 05:22:55 +0300 +Subject: [PATCH] xtensa: fix localized symbol refcounting with --gc-sections + +elf_xtensa_gc_sweep_hook doesn't correctly unreference symbols that were +made local, that results in link failure with the following message: + + BFD (GNU Binutils) 2.24 internal error, aborting at elf32-xtensa.c line + 3372 in elf_xtensa_finish_dynamic_sections + +elf_xtensa_gc_sweep_hook determines symbol reference type (PLT or GOT) by +relocation type. Relocation types are not changed when symbol becomes +local, but its PLT references are added to GOT references and +plt.refcount is set to 0. Such symbol cannot be unreferences in the +elf_xtensa_gc_sweep_hook and its extra references make calculated GOT +relocations section size not match number of GOT relocations. + +Fix it by treating PLT reference as GOT reference when plt.refcount is +not positive. + +2015-05-14 Max Filippov +bfd/ + * elf32-xtensa.c (elf_xtensa_gc_sweep_hook): Treat PLT reference + as GOT reference when plt.refcount is not positive. + +Signed-off-by: Max Filippov +--- +Backported from: e6c9a083ec5ae7a45bd71682b26aae1939849388 +Changes to ChangeLog are dropped. + + bfd/elf32-xtensa.c | 6 +++++- + 1 file changed, 5 insertions(+), 1 deletion(-) + +diff --git a/bfd/elf32-xtensa.c b/bfd/elf32-xtensa.c +index 53af1c6..2523670 100644 +--- a/bfd/elf32-xtensa.c ++++ b/bfd/elf32-xtensa.c +@@ -1360,10 +1360,14 @@ elf_xtensa_gc_sweep_hook (bfd *abfd, + { + if (is_plt) + { ++ /* If the symbol has been localized its plt.refcount got moved ++ to got.refcount. Handle it as GOT. */ + if (h->plt.refcount > 0) + h->plt.refcount--; ++ else ++ is_got = TRUE; + } +- else if (is_got) ++ if (is_got) + { + if (h->got.refcount > 0) + h->got.refcount--; +-- +1.8.1.4 + diff --git a/patches/binutils/2.25.1/912-xtensa-fix-gas-segfault-with-text-section-literals.patch b/patches/binutils/2.25.1/912-xtensa-fix-gas-segfault-with-text-section-literals.patch new file mode 100644 index 0000000..4a3de2c --- /dev/null +++ b/patches/binutils/2.25.1/912-xtensa-fix-gas-segfault-with-text-section-literals.patch @@ -0,0 +1,56 @@ +From 2d0522e76e4afeeb2e104e0a4332d94fa0d2fbf6 Mon Sep 17 00:00:00 2001 +From: Max Filippov +Date: Sun, 17 May 2015 06:46:15 +0300 +Subject: [PATCH] xtensa: fix gas segfault with --text-section-literals + +When --text-section-literals is used and code in the .init or .fini +emits literal in the absence of .literal_position, xtensa_move_literals +segfaults. + +Check that search_frag is non-NULL in the xtensa_move_literals and +report error otherwise. + +2015-05-26 Max Filippov +gas/ + * config/tc-xtensa.c (xtensa_move_literals): Check that + search_frag is non-NULL. Report error if literal frag is not + found. + +Signed-off-by: Max Filippov +--- +Backported from: 4de0562a4c69fef4952aa7e19d7bda359f02e8b4 +Changes to ChangeLog are dropped. + + gas/config/tc-xtensa.c | 10 +++++++++- + 1 file changed, 9 insertions(+), 1 deletion(-) + +diff --git a/gas/config/tc-xtensa.c b/gas/config/tc-xtensa.c +index 31c0b6b..18307c1 100644 +--- a/gas/config/tc-xtensa.c ++++ b/gas/config/tc-xtensa.c +@@ -10808,13 +10808,21 @@ xtensa_move_literals (void) + frchain_to = NULL; + frag_splice = &(frchain_from->frch_root); + +- while (!search_frag->tc_frag_data.literal_frag) ++ while (search_frag && !search_frag->tc_frag_data.literal_frag) + { + gas_assert (search_frag->fr_fix == 0 + || search_frag->fr_type == rs_align); + search_frag = search_frag->fr_next; + } + ++ if (!search_frag) ++ { ++ search_frag = frchain_from->frch_root; ++ as_bad_where (search_frag->fr_file, search_frag->fr_line, ++ _("literal pool location required for text-section-literals; specify with .literal_position")); ++ continue; ++ } ++ + gas_assert (search_frag->tc_frag_data.literal_frag->fr_subtype + == RELAX_LITERAL_POOL_BEGIN); + xtensa_switch_section_emit_state (&state, segment->seg, 0); +-- +1.8.1.4 + diff --git a/patches/binutils/2.25.1/913-xtensa-add-auto-litpools-option.patch b/patches/binutils/2.25.1/913-xtensa-add-auto-litpools-option.patch new file mode 100644 index 0000000..3ed9af1 --- /dev/null +++ b/patches/binutils/2.25.1/913-xtensa-add-auto-litpools-option.patch @@ -0,0 +1,699 @@ +From 978adaaa4cd3921842e2be8a31c05f081fb17fcf Mon Sep 17 00:00:00 2001 +From: Max Filippov +Date: Wed, 29 Jul 2015 17:42:54 +0300 +Subject: [PATCH] xtensa: add --auto-litpools option + +Auto-litpools is the automated version of text-section-literals: literal +pool candidate frags are planted every N frags and during relaxation +they are turned into actual literal pools where literals are moved to +become reachable for their first reference by L32R instruction. + +2015-08-12 David Weatherford +gas/ + * config/tc-xtensa.c (struct litpool_frag, struct litpool_seg): + New structures. + (xtensa_maybe_create_literal_pool_frag): New function. + (litpool_seg_list, auto_litpools, auto_litpool_limit) + (litpool_buf, litpool_slotbuf): New static variables. + (option_auto_litpools, option_no_auto_litpools) + (option_auto_litpool_limit): New enum identifiers. + (md_longopts): Add entries for auto-litpools, no-auto-litpools + and auto-litpool-limit. + (md_parse_option): Handle option_auto_litpools, + option_no_auto_litpools and option_auto_litpool_limit. + (md_show_usage): Add help for --[no-]auto-litpools and + --auto-litpool-limit. + (xtensa_mark_literal_pool_location): Record a place for literal + pool with a call to xtensa_maybe_create_literal_pool_frag. + (get_literal_pool_location): Find highest priority literal pool + or convert candidate to literal pool when auto-litpools are used. + (xg_assemble_vliw_tokens): Create literal pool after jump + instruction. + (xtensa_check_frag_count): Create candidate literal pool every + auto_litpool_limit frags. + (xtensa_relax_frag): Add jump around literals to non-empty + literal pool. + (xtensa_move_literals): Estimate literal pool addresses and move + unreachable literals closer to their users, converting candidate + to literal pool if needed. + (xtensa_switch_to_non_abs_literal_fragment): Only emit error + about missing .literal_position in case auto-litpools are not + used. + * config/tc-xtensa.h (xtensa_relax_statesE): New relaxation + state: RELAX_LITERAL_POOL_CANDIDATE_BEGIN. + +2015-08-12 Max Filippov +gas/testsuite/ + * gas/xtensa/all.exp: Add auto-litpools to the list of xtensa + tests. + * gas/xtensa/auto-litpools.s: New file: auto-litpools test. + * gas/xtensa/auto-litpools.s: New file: auto-litpools test + result pattern. + +Signed-off-by: Max Filippov +--- +Backported from: b46824bd49648c575372e6d9bc6a6defeabd6ed5 +Changes to ChangeLogs and documentation are dropped. + + gas/config/tc-xtensa.c | 432 ++++++++++++++++++++++++++++++- + gas/config/tc-xtensa.h | 1 + + gas/testsuite/gas/xtensa/all.exp | 1 + + gas/testsuite/gas/xtensa/auto-litpools.d | 12 + + gas/testsuite/gas/xtensa/auto-litpools.s | 13 + + 5 files changed, 454 insertions(+), 5 deletions(-) + create mode 100644 gas/testsuite/gas/xtensa/auto-litpools.d + create mode 100644 gas/testsuite/gas/xtensa/auto-litpools.s + +diff --git a/gas/config/tc-xtensa.c b/gas/config/tc-xtensa.c +index 7311a05..b8b1e7d 100644 +--- a/gas/config/tc-xtensa.c ++++ b/gas/config/tc-xtensa.c +@@ -440,6 +440,29 @@ bfd_boolean directive_state[] = + #endif + }; + ++/* A circular list of all potential and actual literal pool locations ++ in a segment. */ ++struct litpool_frag ++{ ++ struct litpool_frag *next; ++ struct litpool_frag *prev; ++ fragS *fragP; ++ addressT addr; ++ short priority; /* 1, 2, or 3 -- 1 is highest */ ++ short original_priority; ++}; ++ ++/* Map a segment to its litpool_frag list. */ ++struct litpool_seg ++{ ++ struct litpool_seg *next; ++ asection *seg; ++ struct litpool_frag frag_list; ++ int frag_count; /* since last litpool location */ ++}; ++ ++static struct litpool_seg litpool_seg_list; ++ + + /* Directive functions. */ + +@@ -474,6 +497,9 @@ static void xtensa_create_trampoline_frag (bfd_boolean); + static void xtensa_maybe_create_trampoline_frag (void); + struct trampoline_frag; + static int init_trampoline_frag (struct trampoline_frag *); ++static void xtensa_maybe_create_literal_pool_frag (bfd_boolean, bfd_boolean); ++static bfd_boolean auto_litpools = FALSE; ++static int auto_litpool_limit = 10000; + + /* Alignment Functions. */ + +@@ -698,6 +724,10 @@ enum + + option_trampolines, + option_no_trampolines, ++ ++ option_auto_litpools, ++ option_no_auto_litpools, ++ option_auto_litpool_limit, + }; + + const char *md_shortopts = ""; +@@ -773,6 +803,10 @@ struct option md_longopts[] = + { "trampolines", no_argument, NULL, option_trampolines }, + { "no-trampolines", no_argument, NULL, option_no_trampolines }, + ++ { "auto-litpools", no_argument, NULL, option_auto_litpools }, ++ { "no-auto-litpools", no_argument, NULL, option_no_auto_litpools }, ++ { "auto-litpool-limit", required_argument, NULL, option_auto_litpool_limit }, ++ + { NULL, no_argument, NULL, 0 } + }; + +@@ -961,6 +995,34 @@ md_parse_option (int c, char *arg) + use_trampolines = FALSE; + return 1; + ++ case option_auto_litpools: ++ auto_litpools = TRUE; ++ use_literal_section = FALSE; ++ return 1; ++ ++ case option_no_auto_litpools: ++ auto_litpools = FALSE; ++ auto_litpool_limit = -1; ++ return 1; ++ ++ case option_auto_litpool_limit: ++ { ++ int value = 0; ++ if (auto_litpool_limit < 0) ++ as_fatal (_("no-auto-litpools is incompatible with auto-litpool-limit")); ++ if (*arg == 0 || *arg == '-') ++ as_fatal (_("invalid auto-litpool-limit argument")); ++ value = strtol (arg, &arg, 10); ++ if (*arg != 0) ++ as_fatal (_("invalid auto-litpool-limit argument")); ++ if (value < 100 || value > 10000) ++ as_fatal (_("invalid auto-litpool-limit argument (range is 100-10000)")); ++ auto_litpool_limit = value; ++ auto_litpools = TRUE; ++ use_literal_section = FALSE; ++ return 1; ++ } ++ + default: + return 0; + } +@@ -986,7 +1048,12 @@ Xtensa options:\n\ + flix bundles\n\ + --rename-section old=new Rename section 'old' to 'new'\n\ + --[no-]trampolines [Do not] generate trampolines (jumps to jumps)\n\ +- when jumps do not reach their targets\n", stream); ++ when jumps do not reach their targets\n\ ++ --[no-]auto-litpools [Do not] automatically create literal pools\n\ ++ --auto-litpool-limit=\n\ ++ (range 100-10000) Maximum number of blocks of\n\ ++ instructions to emit between literal pool\n\ ++ locations; implies --auto-litpools flag\n", stream); + } + + +@@ -4728,6 +4795,8 @@ xtensa_mark_literal_pool_location (void) + pool_location = frag_now; + frag_now->tc_frag_data.lit_frchain = frchain_now; + frag_now->tc_frag_data.literal_frag = frag_now; ++ /* Just record this frag. */ ++ xtensa_maybe_create_literal_pool_frag (FALSE, FALSE); + frag_variant (rs_machine_dependent, 0, 0, + RELAX_LITERAL_POOL_BEGIN, NULL, 0, NULL); + xtensa_set_frag_assembly_state (frag_now); +@@ -4832,6 +4901,31 @@ get_expanded_loop_offset (xtensa_opcode opcode) + static fragS * + get_literal_pool_location (segT seg) + { ++ struct litpool_seg *lps = litpool_seg_list.next; ++ struct litpool_frag *lpf; ++ for ( ; lps && lps->seg->id != seg->id; lps = lps->next) ++ ; ++ if (lps) ++ { ++ for (lpf = lps->frag_list.prev; lpf->fragP; lpf = lpf->prev) ++ { /* Skip "candidates" for now. */ ++ if (lpf->fragP->fr_subtype == RELAX_LITERAL_POOL_BEGIN && ++ lpf->priority == 1) ++ return lpf->fragP; ++ } ++ /* Must convert a lower-priority pool. */ ++ for (lpf = lps->frag_list.prev; lpf->fragP; lpf = lpf->prev) ++ { ++ if (lpf->fragP->fr_subtype == RELAX_LITERAL_POOL_BEGIN) ++ return lpf->fragP; ++ } ++ /* Still no match -- try for a low priority pool. */ ++ for (lpf = lps->frag_list.prev; lpf->fragP; lpf = lpf->prev) ++ { ++ if (lpf->fragP->fr_subtype == RELAX_LITERAL_POOL_CANDIDATE_BEGIN) ++ return lpf->fragP; ++ } ++ } + return seg_info (seg)->tc_segment_info_data.literal_pool_loc; + } + +@@ -7098,6 +7192,11 @@ xg_assemble_vliw_tokens (vliw_insn *vinsn) + frag_now->tc_frag_data.slot_symbols[slot] = tinsn->symbol; + frag_now->tc_frag_data.slot_offsets[slot] = tinsn->offset; + frag_now->tc_frag_data.literal_frags[slot] = tinsn->literal_frag; ++ if (tinsn->opcode == xtensa_l32r_opcode) ++ { ++ frag_now->tc_frag_data.literal_frags[slot] = ++ tinsn->tok[1].X_add_symbol->sy_frag; ++ } + if (tinsn->literal_space != 0) + xg_assemble_literal_space (tinsn->literal_space, slot); + frag_now->tc_frag_data.free_reg[slot] = tinsn->extra_arg; +@@ -7170,6 +7269,8 @@ xg_assemble_vliw_tokens (vliw_insn *vinsn) + frag_now->fr_symbol, frag_now->fr_offset, NULL); + xtensa_set_frag_assembly_state (frag_now); + xtensa_maybe_create_trampoline_frag (); ++ /* Always create one here. */ ++ xtensa_maybe_create_literal_pool_frag (TRUE, FALSE); + } + else if (is_branch && do_align_targets ()) + { +@@ -7314,11 +7415,18 @@ xtensa_check_frag_count (void) + clear_frag_count (); + unreachable_count = 0; + } ++ ++ /* We create an area for a possible literal pool every N (default 5000) ++ frags or so. */ ++ xtensa_maybe_create_literal_pool_frag (TRUE, TRUE); + } + + static xtensa_insnbuf trampoline_buf = NULL; + static xtensa_insnbuf trampoline_slotbuf = NULL; + ++static xtensa_insnbuf litpool_buf = NULL; ++static xtensa_insnbuf litpool_slotbuf = NULL; ++ + #define TRAMPOLINE_FRAG_SIZE 3000 + + static void +@@ -7410,6 +7518,135 @@ dump_trampolines (void) + } + } + ++static void dump_litpools (void) __attribute__ ((unused)); ++ ++static void ++dump_litpools (void) ++{ ++ struct litpool_seg *lps = litpool_seg_list.next; ++ struct litpool_frag *lpf; ++ ++ for ( ; lps ; lps = lps->next ) ++ { ++ printf("litpool seg %s\n", lps->seg->name); ++ for ( lpf = lps->frag_list.next; lpf->fragP; lpf = lpf->next ) ++ { ++ fragS *litfrag = lpf->fragP->fr_next; ++ int count = 0; ++ while (litfrag && litfrag->fr_subtype != RELAX_LITERAL_POOL_END) ++ { ++ if (litfrag->fr_fix == 4) ++ count++; ++ litfrag = litfrag->fr_next; ++ } ++ printf(" %ld <%d:%d> (%d) [%d]: ", ++ lpf->addr, lpf->priority, lpf->original_priority, ++ lpf->fragP->fr_line, count); ++ //dump_frag(lpf->fragP); ++ } ++ } ++} ++ ++static void ++xtensa_maybe_create_literal_pool_frag (bfd_boolean create, ++ bfd_boolean only_if_needed) ++{ ++ struct litpool_seg *lps = litpool_seg_list.next; ++ fragS *fragP; ++ struct litpool_frag *lpf; ++ bfd_boolean needed = FALSE; ++ ++ if (use_literal_section || !auto_litpools) ++ return; ++ ++ for ( ; lps ; lps = lps->next ) ++ { ++ if (lps->seg == now_seg) ++ break; ++ } ++ ++ if (lps == NULL) ++ { ++ lps = (struct litpool_seg *)xcalloc (sizeof (struct litpool_seg), 1); ++ lps->next = litpool_seg_list.next; ++ litpool_seg_list.next = lps; ++ lps->seg = now_seg; ++ lps->frag_list.next = &lps->frag_list; ++ lps->frag_list.prev = &lps->frag_list; ++ } ++ ++ lps->frag_count++; ++ ++ if (create) ++ { ++ if (only_if_needed) ++ { ++ if (past_xtensa_end || !use_transform() || ++ frag_now->tc_frag_data.is_no_transform) ++ { ++ return; ++ } ++ if (auto_litpool_limit <= 0) ++ { ++ /* Don't create a litpool based only on frag count. */ ++ return; ++ } ++ else if (lps->frag_count > auto_litpool_limit) ++ { ++ needed = TRUE; ++ } ++ else ++ { ++ return; ++ } ++ } ++ else ++ { ++ needed = TRUE; ++ } ++ } ++ ++ if (needed) ++ { ++ int size = (only_if_needed) ? 3 : 0; /* Space for a "j" insn. */ ++ /* Create a potential site for a literal pool. */ ++ frag_wane (frag_now); ++ frag_new (0); ++ xtensa_set_frag_assembly_state (frag_now); ++ fragP = frag_now; ++ fragP->tc_frag_data.lit_frchain = frchain_now; ++ fragP->tc_frag_data.literal_frag = fragP; ++ frag_var (rs_machine_dependent, size, size, ++ (only_if_needed) ? ++ RELAX_LITERAL_POOL_CANDIDATE_BEGIN : ++ RELAX_LITERAL_POOL_BEGIN, ++ NULL, 0, NULL); ++ frag_now->tc_frag_data.lit_seg = now_seg; ++ frag_variant (rs_machine_dependent, 0, 0, ++ RELAX_LITERAL_POOL_END, NULL, 0, NULL); ++ xtensa_set_frag_assembly_state (frag_now); ++ } ++ else ++ { ++ /* RELAX_LITERAL_POOL_BEGIN frag is being created; ++ just record it here. */ ++ fragP = frag_now; ++ } ++ ++ lpf = (struct litpool_frag *)xmalloc(sizeof (struct litpool_frag)); ++ /* Insert at tail of circular list. */ ++ lpf->addr = 0; ++ lps->frag_list.prev->next = lpf; ++ lpf->next = &lps->frag_list; ++ lpf->prev = lps->frag_list.prev; ++ lps->frag_list.prev = lpf; ++ lpf->fragP = fragP; ++ lpf->priority = (needed) ? (only_if_needed) ? 3 : 2 : 1; ++ lpf->original_priority = lpf->priority; ++ ++ lps->frag_count = 0; ++} ++ + static void + xtensa_cleanup_align_frags (void) + { +@@ -9029,7 +9266,41 @@ xtensa_relax_frag (fragS *fragP, long stretch, int *stretched_p) + break; + + case RELAX_LITERAL_POOL_BEGIN: ++ if (fragP->fr_var != 0) ++ { ++ /* We have a converted "candidate" literal pool; ++ assemble a jump around it. */ ++ TInsn insn; ++ if (!litpool_slotbuf) ++ { ++ litpool_buf = xtensa_insnbuf_alloc (isa); ++ litpool_slotbuf = xtensa_insnbuf_alloc (isa); ++ } ++ new_stretch += 3; ++ fragP->tc_frag_data.relax_seen = FALSE; /* Need another pass. */ ++ fragP->tc_frag_data.is_insn = TRUE; ++ tinsn_init (&insn); ++ insn.insn_type = ITYPE_INSN; ++ insn.opcode = xtensa_j_opcode; ++ insn.ntok = 1; ++ set_expr_symbol_offset (&insn.tok[0], fragP->fr_symbol, ++ fragP->fr_fix); ++ fmt = xg_get_single_format (xtensa_j_opcode); ++ tinsn_to_slotbuf (fmt, 0, &insn, litpool_slotbuf); ++ xtensa_format_set_slot (isa, fmt, 0, litpool_buf, litpool_slotbuf); ++ xtensa_insnbuf_to_chars (isa, litpool_buf, ++ (unsigned char *)fragP->fr_literal + ++ fragP->fr_fix, 3); ++ fragP->fr_fix += 3; ++ fragP->fr_var -= 3; ++ /* Add a fix-up. */ ++ fix_new (fragP, 0, 3, fragP->fr_symbol, 0, TRUE, ++ BFD_RELOC_XTENSA_SLOT0_OP); ++ } ++ break; ++ + case RELAX_LITERAL_POOL_END: ++ case RELAX_LITERAL_POOL_CANDIDATE_BEGIN: + case RELAX_MAYBE_UNREACHABLE: + case RELAX_MAYBE_DESIRE_ALIGN: + /* No relaxation required. */ +@@ -10789,12 +11060,115 @@ xtensa_move_literals (void) + segT dest_seg; + fixS *fix, *next_fix, **fix_splice; + sym_list *lit; ++ struct litpool_seg *lps; + + mark_literal_frags (literal_head->next); + + if (use_literal_section) + return; + ++ /* Assign addresses (rough estimates) to the potential literal pool locations ++ and create new ones if the gaps are too large. */ ++ ++ for (lps = litpool_seg_list.next; lps; lps = lps->next) ++ { ++ frchainS *frchP = seg_info (lps->seg)->frchainP; ++ struct litpool_frag *lpf = lps->frag_list.next; ++ addressT addr = 0; ++ ++ for ( ; frchP; frchP = frchP->frch_next) ++ { ++ fragS *fragP; ++ for (fragP = frchP->frch_root; fragP; fragP = fragP->fr_next) ++ { ++ if (lpf && fragP == lpf->fragP) ++ { ++ gas_assert(fragP->fr_type == rs_machine_dependent && ++ (fragP->fr_subtype == RELAX_LITERAL_POOL_BEGIN || ++ fragP->fr_subtype == RELAX_LITERAL_POOL_CANDIDATE_BEGIN)); ++ /* Found a litpool location. */ ++ lpf->addr = addr; ++ lpf = lpf->next; ++ } ++ if (fragP->fr_type == rs_machine_dependent && ++ fragP->fr_subtype == RELAX_SLOTS) ++ { ++ int slot; ++ for (slot = 0; slot < MAX_SLOTS; slot++) ++ { ++ if (fragP->tc_frag_data.literal_frags[slot]) ++ { ++ /* L32R; point its literal to the nearest litpool ++ preferring non-"candidate" positions to avoid ++ the jump-around. */ ++ fragS *litfrag = fragP->tc_frag_data.literal_frags[slot]; ++ struct litpool_frag *lp = lpf->prev; ++ if (!lp->fragP) ++ { ++ break; ++ } ++ while (lp->fragP->fr_subtype == ++ RELAX_LITERAL_POOL_CANDIDATE_BEGIN) ++ { ++ lp = lp->prev; ++ if (lp->fragP == NULL) ++ { ++ /* End of list; have to bite the bullet. ++ Take the nearest. */ ++ lp = lpf->prev; ++ break; ++ } ++ /* Does it (conservatively) reach? */ ++ if (addr - lp->addr <= 128 * 1024) ++ { ++ if (lp->fragP->fr_subtype == RELAX_LITERAL_POOL_BEGIN) ++ { ++ /* Found a good one. */ ++ break; ++ } ++ else if (lp->prev->fragP && ++ addr - lp->prev->addr > 128 * 1024) ++ { ++ /* This is still a "candidate" but the next one ++ will be too far away, so revert to the nearest ++ one, convert it and add the jump around. */ ++ fragS *poolbeg; ++ fragS *poolend; ++ symbolS *lsym; ++ char label[10 + 2 * sizeof (fragS *)]; ++ lp = lpf->prev; ++ poolbeg = lp->fragP; ++ lp->priority = 1; ++ poolbeg->fr_subtype = RELAX_LITERAL_POOL_BEGIN; ++ poolend = poolbeg->fr_next; ++ gas_assert (poolend->fr_type == rs_machine_dependent && ++ poolend->fr_subtype == RELAX_LITERAL_POOL_END); ++ /* Create a local symbol pointing to the ++ end of the pool. */ ++ sprintf (label, ".L0_LT_%p", poolbeg); ++ lsym = (symbolS *)local_symbol_make (label, lps->seg, ++ 0, poolend); ++ poolbeg->fr_symbol = lsym; ++ /* Rest is done in xtensa_relax_frag. */ ++ } ++ } ++ } ++ if (! litfrag->tc_frag_data.literal_frag) ++ { ++ /* Take earliest use of this literal to avoid ++ forward refs. */ ++ litfrag->tc_frag_data.literal_frag = lp->fragP; ++ } ++ } ++ } ++ } ++ addr += fragP->fr_fix; ++ if (fragP->fr_type == rs_fill) ++ addr += fragP->fr_offset; ++ } ++ } ++ } ++ + for (segment = literal_head->next; segment; segment = segment->next) + { + /* Keep the literals for .init and .fini in separate sections. */ +@@ -10839,9 +11213,6 @@ xtensa_move_literals (void) + while (search_frag != frag_now) + { + next_frag = search_frag->fr_next; +- +- /* First, move the frag out of the literal section and +- to the appropriate place. */ + if (search_frag->tc_frag_data.literal_frag) + { + literal_pool = search_frag->tc_frag_data.literal_frag; +@@ -10849,8 +11220,56 @@ xtensa_move_literals (void) + frchain_to = literal_pool->tc_frag_data.lit_frchain; + gas_assert (frchain_to); + } ++ ++ if (search_frag->fr_type == rs_fill && search_frag->fr_fix == 0) ++ { ++ /* Skip empty fill frags. */ ++ *frag_splice = next_frag; ++ search_frag = next_frag; ++ continue; ++ } ++ ++ if (search_frag->fr_type == rs_align) ++ { ++ /* Skip alignment frags, because the pool as a whole will be ++ aligned if used, and we don't want to force alignment if the ++ pool is unused. */ ++ *frag_splice = next_frag; ++ search_frag = next_frag; ++ continue; ++ } ++ ++ /* First, move the frag out of the literal section and ++ to the appropriate place. */ ++ ++ /* Insert an aligmnent frag at start of pool. */ ++ if (literal_pool->fr_next->fr_type == rs_machine_dependent && ++ literal_pool->fr_next->fr_subtype == RELAX_LITERAL_POOL_END) ++ { ++ segT pool_seg = literal_pool->fr_next->tc_frag_data.lit_seg; ++ emit_state prev_state; ++ fragS *prev_frag; ++ fragS *align_frag; ++ xtensa_switch_section_emit_state (&prev_state, pool_seg, 0); ++ prev_frag = frag_now; ++ frag_variant (rs_fill, 0, 0, 0, NULL, 0, NULL); ++ align_frag = frag_now; ++ frag_align (2, 0, 0); ++ /* Splice it into the right place. */ ++ prev_frag->fr_next = align_frag->fr_next; ++ align_frag->fr_next = literal_pool->fr_next; ++ literal_pool->fr_next = align_frag; ++ /* Insert after this one. */ ++ literal_pool->tc_frag_data.literal_frag = align_frag; ++ xtensa_restore_emit_state (&prev_state); ++ } + insert_after = literal_pool->tc_frag_data.literal_frag; + dest_seg = insert_after->fr_next->tc_frag_data.lit_seg; ++ /* Skip align frag. */ ++ if (insert_after->fr_next->fr_type == rs_align) ++ { ++ insert_after = insert_after->fr_next; ++ } + + *frag_splice = next_frag; + search_frag->fr_next = insert_after->fr_next; +@@ -11014,7 +11433,10 @@ xtensa_switch_to_non_abs_literal_fragment (emit_state *result) + && !recursive + && !is_init && ! is_fini) + { +- as_bad (_("literal pool location required for text-section-literals; specify with .literal_position")); ++ if (!auto_litpools) ++ { ++ as_bad (_("literal pool location required for text-section-literals; specify with .literal_position")); ++ } + + /* When we mark a literal pool location, we want to put a frag in + the literal pool that points to it. But to do that, we want to +diff --git a/gas/config/tc-xtensa.h b/gas/config/tc-xtensa.h +index b2e43fa..290d902 100644 +--- a/gas/config/tc-xtensa.h ++++ b/gas/config/tc-xtensa.h +@@ -124,6 +124,7 @@ enum xtensa_relax_statesE + + RELAX_LITERAL_POOL_BEGIN, + RELAX_LITERAL_POOL_END, ++ RELAX_LITERAL_POOL_CANDIDATE_BEGIN, + /* Technically these are not relaxations at all but mark a location + to store literals later. Note that fr_var stores the frchain for + BEGIN frags and fr_var stores now_seg for END frags. */ +diff --git a/gas/testsuite/gas/xtensa/all.exp b/gas/testsuite/gas/xtensa/all.exp +index d197ec8..db39629 100644 +--- a/gas/testsuite/gas/xtensa/all.exp ++++ b/gas/testsuite/gas/xtensa/all.exp +@@ -100,6 +100,7 @@ if [istarget xtensa*-*-*] then { + run_dump_test "jlong" + run_dump_test "trampoline" + run_dump_test "first_frag_align" ++ run_dump_test "auto-litpools" + } + + if [info exists errorInfo] then { +diff --git a/gas/testsuite/gas/xtensa/auto-litpools.d b/gas/testsuite/gas/xtensa/auto-litpools.d +new file mode 100644 +index 0000000..4d1a690 +--- /dev/null ++++ b/gas/testsuite/gas/xtensa/auto-litpools.d +@@ -0,0 +1,12 @@ ++#as: --auto-litpools ++#objdump: -d ++#name: auto literal pool placement ++ ++.*: +file format .*xtensa.* ++#... ++.*4:.*l32r.a2, 0 .* ++#... ++.*3e437:.*j.3e440 .* ++#... ++.*40750:.*l32r.a2, 3e43c .* ++#... +diff --git a/gas/testsuite/gas/xtensa/auto-litpools.s b/gas/testsuite/gas/xtensa/auto-litpools.s +new file mode 100644 +index 0000000..9a5b26b +--- /dev/null ++++ b/gas/testsuite/gas/xtensa/auto-litpools.s +@@ -0,0 +1,13 @@ ++ .text ++ .align 4 ++ .literal .L0, 0x12345 ++ .literal .L1, 0x12345 ++ ++f: ++ l32r a2, .L0 ++ .rep 44000 ++ _nop ++ _nop ++ .endr ++ l32r a2, .L1 ++ ret +-- +1.8.1.4 + -- cgit v0.10.2-6-g49f6