resolve IFUNCs correctly

.gitignore

@@ -1,3 +1,5 @@
 /bin
 /obj
 __pycache__
+smol-*-*-*/
+*.tar.xz

Makefile

@@ -4,6 +4,7 @@ SRCDIR := rt
 TESTDIR:= test
 
 NASM ?= nasm
+OBJCOPY ?= objcopy
 
 BITS ?= $(shell getconf LONG_BIT)
 
@@ -35,12 +36,12 @@ else
 CFLAGS += -march=nocona
 endif
 
-LIBS = $(filter-out -pthread,$(shell pkg-config --libs sdl2)) -lX11 -lc #-lGL
+LIBS = $(filter-out -pthread,$(shell pkg-config --libs sdl2)) -lX11 -lm -lc #-lGL
 
 PWD ?= .
 
 SMOLFLAGS = --smolrt "$(PWD)/rt" --smolld "$(PWD)/ld" \
-     -falign-stack -fuse-interp \
+     -falign-stack -fuse-interp -fifunc-support \
      --verbose #--keeptmp
 # -fuse-dnload-loader -fskip-zero-value -fuse-nx -fskip-entries -fuse-dt-debug
 # -fuse-dl-fini -fno-start-arg -funsafe-dynamic
@@ -67,11 +68,13 @@ $(OBJDIR)/%.o: $(SRCDIR)/%.c $(OBJDIR)/
 $(OBJDIR)/%.o: $(TESTDIR)/%.c $(OBJDIR)/
 	$(CC) $(CFLAGS) -c "$<" -o "$@"
 
-$(BINDIR)/%: $(OBJDIR)/%.o $(BINDIR)/
-	$(PYTHON3) ./smold.py $(SMOLFLAGS) $(LIBS) "$<" "$@"
+$(BINDIR)/%.dbg $(BINDIR)/%: $(OBJDIR)/%.o $(BINDIR)/
+	$(PYTHON3) ./smold.py --debugout $(SMOLFLAGS) --ldflags=-Wl,-Map=$(BINDIR)/$*.map $(LIBS) "$<" "$@.dbg"
+	$(PYTHON3) ./smold.py $(SMOLFLAGS) --ldflags=-Wl,-Map=$(BINDIR)/$*.map $(LIBS) "$<" "$@"
 	$(PYTHON3) ./smoltrunc.py "$@" "$(OBJDIR)/$(notdir $@)" && mv "$(OBJDIR)/$(notdir $@)" "$@" && chmod +x "$@"
 
-$(BINDIR)/%-crt: $(OBJDIR)/%.lto.o $(OBJDIR)/crt1.lto.o $(BINDIR)/
+$(BINDIR)/%-crt.dbg $(BINDIR)/%-crt: $(OBJDIR)/%.lto.o $(OBJDIR)/crt1.lto.o $(BINDIR)/
+	$(PYTHON3) ./smold.py --debugout $(SMOLFLAGS) --ldflags=-Wl,-Map=$(BINDIR)/$*-crt.map $(LIBS) "$<" $(OBJDIR)/crt1.lto.o "$@.dbg"
 	$(PYTHON3) ./smold.py $(SMOLFLAGS) --ldflags=-Wl,-Map=$(BINDIR)/$*-crt.map $(LIBS) "$<" $(OBJDIR)/crt1.lto.o "$@"
 	$(PYTHON3) ./smoltrunc.py "$@" "$(OBJDIR)/$(notdir $@)" && mv "$(OBJDIR)/$(notdir $@)" "$@" && chmod +x "$@"
 

ld/link.ld

@@ -1,10 +1,19 @@
-OUTPUT_FORMAT(binary)
+OUTPUT_FORMAT("elf64-x86-64", "elf64-x86-64", "elf64-x86-64")
+OUTPUT_ARCH(i386:x86-64)
+ENTRY(_smol_start)
+SEARCH_DIR("/usr/x86_64-pc-linux-gnu/lib64"); SEARCH_DIR("/usr/lib"); SEARCH_DIR("/usr/local/lib"); SEARCH_DIR("/usr/x86_64-pc-linux-gnu/lib");
+/*OUTPUT_FORMAT(binary)*/
+
+PHDRS {
+	rodttxt PT_LOAD FLAGS(5); /* r-x */
+	rwdt    PT_LOAD FLAGS(6); /* rw- */
+}
 
 SECTIONS {
   . = 0x10000;
   _smol_origin = .;
 
-  .header : { KEEP(*(.header)) }
+  .header : { KEEP(*(.header)) } :rodttxt
 
   _smol_text_start = .;
   _smol_text_off = _smol_text_start - _smol_origin;
@@ -15,7 +24,7 @@ SECTIONS {
      KEEP(*(.text.startup.smol))
      KEEP(*(.text.startup._start))
      *(.text .text.*)
-  }
+  } :rodttxt
   _smol_text_end = .;
   _smol_text_size = _smol_text_end - _smol_text_start;
 
@@ -26,10 +35,10 @@ SECTIONS {
   .data : {
      KEEP(*(.data.smolgot))
      *(.data .data.* .tdata .tdata.*)
-  }
+  } :rwdt
 
-  .dynamic : { *(.dynamic) } :all :dyn
-  .dynstuff : { *(.symtab .strtab .shstrtab .rel.text .got.plt .gnu.linkonce.* .plt .plt.got .interp) } :all
+  .dynamic : { *(.dynamic) } :rodttxt :dyn
+  .dynstuff : { *(.symtab .strtab .shstrtab .rel.text .got.plt .gnu.linkonce.* .plt .plt.got .interp) } :rodttxt
   _smol_data_end = .;
   _smol_data_size = _smol_data_end - _smol_data_start;
 
@@ -39,7 +48,7 @@ SECTIONS {
   _smol_bss_off = _smol_bss_start - _smol_origin;
   .bss : {
       *(.bss .bss.* .tbss .tbss.* .sbss .sbss.*)
-  }
+  } :rwdt
   _smol_bss_end = .;
   _smol_bss_size = _smol_bss_end - _smol_bss_start;
 

rt/elf.inc

@@ -34,17 +34,22 @@
 %define DT_SYMTAB ( 6)
 %define DT_DEBUG  (21)
 
-%define ST_NAME_OFF  ( 0)
+%define ST_NAME_OFF       ( 0)
+%define ST_INFO__STT_MASK (15)
+%define STT_GNU_IFUNC     (10)
+
 ;  ,---- not 16? ; what's this comment??!
 ; v
 %if __BITS__ == 32
 %define D_UN_PTR_OFF         ( 4)
 %define ST_VALUE_OFF         ( 4)
+%define ST_INFO_OFF          (12)
 %define SYMTAB_SIZE          (16)
 %define ELF_DYN_SZ           ( 8)
 %else
 %define D_UN_PTR_OFF         ( 8)
 %define ST_VALUE_OFF         ( 8)
+%define ST_INFO_OFF          ( 4)
 %define SYMTAB_SIZE          (24)
 %define ELF_DYN_SZ           (16)
 %endif

rt/loader64.asm

@@ -127,16 +127,51 @@ _smol_start:
             jmp short .next_link
 
         .hasheq:
+%ifdef IFUNC_SUPPORT
+            mov cl , [rdx + ST_INFO_OFF]
+%endif
             mov rax, [rdx + ST_VALUE_OFF]
 %ifdef SKIP_ZERO_VALUE
              or rax, rax
              jz short .next_link
 %endif
             add rax, [r12 + L_ADDR_OFF]
+%ifdef IFUNC_SUPPORT
+           and cl, ST_INFO__STT_MASK
+           cmp cl, STT_GNU_IFUNC
+            je .ifunc
+        .no_ifunc:
+%endif
           stosq
             cmp word [rdi], 0
             jne short .next_hash
 
+%ifdef IFUNC_SUPPORT
+            jmp short .break_loop
+        .ifunc:
+         ;;int3 ; in this call, we lose rax rcx rdx rsi rdi r8 r9 r10 r11
+                ; we only need persistence for rdi and r11 tho
+          ;push rcx
+          ;push rdx
+          ;push rsi
+           push rdi
+          ;push r8
+          ;push r9
+          ;push r10
+           push r11
+           call rax
+            pop r11
+           ;pop r10
+           ;pop r9
+           ;pop r8
+            pop rdi
+           ;pop rsi
+           ;pop rdx
+           ;pop rcx
+            jmp short .no_ifunc
+        .break_loop:
+%endif
+
 ; if USE_DNLOAD_LOADER
 %else
        push _smol_start
@@ -219,6 +254,12 @@ repne scasd ; technically, scasq should be used, but meh. this is 1 byte smaller
             mov rax, [rax + D_UN_PTR_OFF]
                 ; ElfW(Addr) symoff(rax) = symtab[bucket].st_value
             lea rdx, [rcx + rcx * 2]
+
+%ifdef IFUNC_SUPPORT
+                ; large opcode, but, ~almost the same as the next one, so,
+                ; should compress well
+            mov cl, [rax + rdx * 8 + ST_INFO_OFF] ; TODO: actually mov cl, ...
+%endif
             mov rax, [rax + rdx * 8 + ST_VALUE_OFF]
 %ifdef SKIP_ZERO_VALUE
              or rax, rax ; zero value => weak symbol or sth
@@ -227,6 +268,20 @@ repne scasd ; technically, scasq should be used, but meh. this is 1 byte smaller
                 ; void* finaladdr(rax) = symoff + entry->l_addr
             add rax, [r12 + L_ADDR_OFF]
 
+%ifdef IFUNC_SUPPORT
+                ; is this an ifunc?
+            and cl, ST_INFO__STT_MASK
+            cmp cl, STT_GNU_IFUNC
+            jne .no_ifunc
+                ; if so: call the resolver
+           push rdi
+           push r11
+           call rax
+            pop r11
+            pop rdi
+        .no_ifunc:
+%endif
+
                 ; *phash = finaladdr
           stosq
             cmp word [rdi], 0

smol/cnl.py

@@ -28,12 +28,14 @@ def nasm_assemble_elfhdr(verbose, nasm_bin, arch, rtdir, intbl, output, asflags)
     if verbose: eprintf("nasm: %s" % repr(args))
     subprocess.check_call(args, stdout=subprocess.DEVNULL)
 
-def ld_link_final(verbose, cc_bin, arch, lddir, inobjs, output, ldflags):
+def ld_link_final(verbose, cc_bin, arch, lddir, inobjs, output, ldflags, debug):
     archflag = '-m64' if arch == "x86_64" else '-m32'
 
-    args = [cc_bin, archflag, '-T', lddir+'/link.ld', \
-            '-Wl,--oformat=binary', '-nostartfiles', '-nostdlib', \
-            '-o', output] + inobjs + ldflags
+    args = [cc_bin, archflag, '-T', lddir+'/link.ld', '-no-pie']
+    if not debug:
+        args.append('-Wl,--oformat=binary')
+        #args = [*args, '-T', lddir+'/link.ld', '-Wl,--oformat=binary']
+    args += ['-nostartfiles', '-nostdlib', '-o', output, *inobjs, *ldflags]
 
     if verbose: eprintf("ld: %s" % repr(args))
     subprocess.check_call(args, stdout=subprocess.DEVNULL)

smold.py

@@ -74,6 +74,9 @@ def main():
         help="Don't end the ELF Dyn table with a DT_NULL entry. This might "+\
              "cause ld.so to interpret the entire binary as the Dyn table, "+\
              "so only enable this if you're sure this won't break things!")
+    parser.add_argument('-fifunc-support', default=False, action='store_true', \
+        help="Support linking to IFUNCs. Probably needed on x86_64, but costs "+\
+             "~16 bytes. Ignored on platforms other than x86_64 or aarch64.")
 
     parser.add_argument('--nasm', default=os.getenv('NASM') or shutil.which('nasm'), \
         help="which nasm binary to use")
@@ -99,6 +102,10 @@ def main():
         help="Be verbose about what happens and which subcommands are invoked")
     parser.add_argument('--keeptmp', default=False, action='store_true', \
         help="Keep temp files (only useful for debugging)")
+    parser.add_argument('--debugout', default=False, action='store_true', \
+        help="Output an unrunnable debug ELF file with symbol information. "+\
+             "(Useful for debugging with gdb, cannot be ran due to broken "+\
+             "relocations.)")
 
     parser.add_argument('input', nargs='+', help="input object file")
     parser.add_argument('output', type=str, help="output binary")
@@ -118,6 +125,7 @@ def main():
     if args.fuse_dnload_loader: args.asflags.insert(0, "-DUSE_DNLOAD_LOADER")
     if args.fuse_interp: args.asflags.insert(0, "-DUSE_INTERP")
     if args.falign_stack: args.asflags.insert(0, "-DALIGN_STACK")
+    if args.fifunc_support: args.asflags.insert(0, "-DIFUNC_SUPPORT")
 
     for x in ['nasm','cc','scanelf','readelf']:
         val = args.__dict__[x]
@@ -172,7 +180,7 @@ def main():
 
         # link with LD into the final executable, w/ special linker script
         ld_link_final(args.verbose, args.cc, arch, args.smolld, [objinput, tmp_elf_file],
-                      args.output, args.ldflags)
+                      args.output, args.ldflags, args.debugout)
     finally:
         if not args.keeptmp:
             if objinputistemp: os.remove(objinput)

test/hello.c

@@ -1,6 +1,7 @@
 
 #include <stdlib.h>
 #include <stdio.h>
+#include <math.h>
 
 const char *f = "foo";
 
@@ -10,6 +11,7 @@ int main(int argc, char* argv[]) {
     printf("argv=%p\n", (void*)argv);
     for (int i = 0; i < argc; ++i)
         printf("argv[%d](%p)=%s\n", i, (void*)argv[i], argv[i]);
+    printf("sin(%d)=%f\n", argc, sinf(argc));
     exit(42);
 }