cmd/link: separate virtual address layout from file layout

Currently these two forms of layout are done in a single pass. This
makes it difficult to compress DWARF sections because that must be
done after relocations are applied, which must happen after virtual
address layout, but we can't layout the file until we've compressed
the DWARF sections.

Fix this by separating the two layout steps. In the process, we can
also unify the copy-pasted code in Link.address to compute file
offsets, which currently has some unnecessary variation.

Unlike the current file offset computation, which depends on virtual
addresses, the new computation only uses file offsets and sizes. This
will let us compress the file representation of a segment and create
the file layout based on its on-disk size rather than its original
in-memory size.

Tested by comparing the test binary for the "strings" package on all
supported GOOS/GOARCH combinations. All binaries are identical
(except, of course, their build IDs).

This is a second attempt at CL 111682.

For #11799.
Fixes #25863.

Change-Id: If09f28771bb4d78dd392fd58b8d7c9d5f22b0b9e
Reviewed-on: https://go-review.googlesource.com/118716
Run-TryBot: Austin Clements <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>
Reviewed-by: Heschi Kreinick <[email protected]>
Reviewed-by: Ian Lance Taylor <[email protected]>

Author: aclements

src/cmd/link/internal/ld/data.go

@@ -1903,20 +1903,22 @@ func assignAddress(ctxt *Link, sect *sym.Section, n int, s *sym.Symbol, va uint6
 	return sect, n, va
 }
 
-// assign addresses
-func (ctxt *Link) address() {
+// address assigns virtual addresses to all segments and sections and
+// returns all segments in file order.
+func (ctxt *Link) address() []*sym.Segment {
+	var order []*sym.Segment // Layout order
+
 	va := uint64(*FlagTextAddr)
+	order = append(order, &Segtext)
 	Segtext.Rwx = 05
 	Segtext.Vaddr = va
-	Segtext.Fileoff = uint64(HEADR)
 	for _, s := range Segtext.Sections {
 		va = uint64(Rnd(int64(va), int64(s.Align)))
 		s.Vaddr = va
 		va += s.Length
 	}
 
 	Segtext.Length = va - uint64(*FlagTextAddr)
-	Segtext.Filelen = Segtext.Length
 	if ctxt.HeadType == objabi.Hnacl {
 		va += 32 // room for the "halt sled"
 	}
@@ -1937,52 +1939,38 @@ func (ctxt *Link) address() {
 		// writable even for this short period.
 		va = uint64(Rnd(int64(va), int64(*FlagRound)))
 
+		order = append(order, &Segrodata)
 		Segrodata.Rwx = 04
 		Segrodata.Vaddr = va
-		Segrodata.Fileoff = va - Segtext.Vaddr + Segtext.Fileoff
-		Segrodata.Filelen = 0
-		if ctxt.HeadType == objabi.Hwindows {
-			Segrodata.Fileoff = Segtext.Fileoff + uint64(Rnd(int64(Segtext.Length), PEFILEALIGN))
-		}
 		for _, s := range Segrodata.Sections {
 			va = uint64(Rnd(int64(va), int64(s.Align)))
 			s.Vaddr = va
 			va += s.Length
 		}
 
 		Segrodata.Length = va - Segrodata.Vaddr
-		Segrodata.Filelen = Segrodata.Length
 	}
 	if len(Segrelrodata.Sections) > 0 {
 		// align to page boundary so as not to mix
 		// rodata, rel-ro data, and executable text.
 		va = uint64(Rnd(int64(va), int64(*FlagRound)))
 
+		order = append(order, &Segrelrodata)
 		Segrelrodata.Rwx = 06
 		Segrelrodata.Vaddr = va
-		Segrelrodata.Fileoff = va - Segrodata.Vaddr + Segrodata.Fileoff
-		Segrelrodata.Filelen = 0
 		for _, s := range Segrelrodata.Sections {
 			va = uint64(Rnd(int64(va), int64(s.Align)))
 			s.Vaddr = va
 			va += s.Length
 		}
 
 		Segrelrodata.Length = va - Segrelrodata.Vaddr
-		Segrelrodata.Filelen = Segrelrodata.Length
 	}
 
 	va = uint64(Rnd(int64(va), int64(*FlagRound)))
+	order = append(order, &Segdata)
 	Segdata.Rwx = 06
 	Segdata.Vaddr = va
-	Segdata.Fileoff = va - Segtext.Vaddr + Segtext.Fileoff
-	Segdata.Filelen = 0
-	if ctxt.HeadType == objabi.Hwindows {
-		Segdata.Fileoff = Segrodata.Fileoff + uint64(Rnd(int64(Segrodata.Length), PEFILEALIGN))
-	}
-	if ctxt.HeadType == objabi.Hplan9 {
-		Segdata.Fileoff = Segtext.Fileoff + Segtext.Filelen
-	}
 	var data *sym.Section
 	var noptr *sym.Section
 	var bss *sym.Section
@@ -2012,16 +2000,14 @@ func (ctxt *Link) address() {
 		}
 	}
 
+	// Assign Segdata's Filelen omitting the BSS. We do this here
+	// simply because right now we know where the BSS starts.
 	Segdata.Filelen = bss.Vaddr - Segdata.Vaddr
 
 	va = uint64(Rnd(int64(va), int64(*FlagRound)))
+	order = append(order, &Segdwarf)
 	Segdwarf.Rwx = 06
 	Segdwarf.Vaddr = va
-	Segdwarf.Fileoff = Segdata.Fileoff + uint64(Rnd(int64(Segdata.Filelen), int64(*FlagRound)))
-	Segdwarf.Filelen = 0
-	if ctxt.HeadType == objabi.Hwindows {
-		Segdwarf.Fileoff = Segdata.Fileoff + uint64(Rnd(int64(Segdata.Filelen), PEFILEALIGN))
-	}
 	for i, s := range Segdwarf.Sections {
 		vlen := int64(s.Length)
 		if i+1 < len(Segdwarf.Sections) {
@@ -2035,8 +2021,6 @@ func (ctxt *Link) address() {
 		Segdwarf.Length = va - Segdwarf.Vaddr
 	}
 
-	Segdwarf.Filelen = va - Segdwarf.Vaddr
-
 	var (
 		text     = Segtext.Sections[0]
 		rodata   = ctxt.Syms.Lookup("runtime.rodata", 0).Sect
@@ -2123,6 +2107,41 @@ func (ctxt *Link) address() {
 	ctxt.xdefine("runtime.noptrbss", sym.SNOPTRBSS, int64(noptrbss.Vaddr))
 	ctxt.xdefine("runtime.enoptrbss", sym.SNOPTRBSS, int64(noptrbss.Vaddr+noptrbss.Length))
 	ctxt.xdefine("runtime.end", sym.SBSS, int64(Segdata.Vaddr+Segdata.Length))
+
+	return order
+}
+
+// layout assigns file offsets and lengths to the segments in order.
+func (ctxt *Link) layout(order []*sym.Segment) {
+	var prev *sym.Segment
+	for _, seg := range order {
+		if prev == nil {
+			seg.Fileoff = uint64(HEADR)
+		} else {
+			switch ctxt.HeadType {
+			default:
+				// Assuming the previous segment was
+				// aligned, the following rounding
+				// should ensure that this segment's
+				// VA ≡ Fileoff mod FlagRound.
+				seg.Fileoff = uint64(Rnd(int64(prev.Fileoff+prev.Filelen), int64(*FlagRound)))
+				if seg.Vaddr%uint64(*FlagRound) != seg.Fileoff%uint64(*FlagRound) {
+					Exitf("bad segment rounding (Vaddr=%#x Fileoff=%#x FlagRound=%#x)", seg.Vaddr, seg.Fileoff, *FlagRound)
+				}
+			case objabi.Hwindows:
+				seg.Fileoff = prev.Fileoff + uint64(Rnd(int64(prev.Filelen), PEFILEALIGN))
+			case objabi.Hplan9:
+				seg.Fileoff = prev.Fileoff + prev.Filelen
+			}
+		}
+		if seg != &Segdata {
+			// Link.address already set Segdata.Filelen to
+			// account for BSS.
+			seg.Filelen = seg.Length
+		}
+		prev = seg
+	}
+
 }
 
 // add a trampoline with symbol s (to be laid down after the current function)

src/cmd/link/internal/ld/main.go

@@ -224,8 +224,9 @@ func Main(arch *sys.Arch, theArch Arch) {
 	ctxt.typelink()
 	ctxt.symtab()
 	ctxt.dodata()
-	ctxt.address()
+	order := ctxt.address()
 	ctxt.reloc()
+	ctxt.layout(order)
 	thearch.Asmb(ctxt)
 	ctxt.undef()
 	ctxt.hostlink()