ia64: Still more gate DSO tuning. Turns out a linker bug prevented

	us from building the gate DSO in a way that makes it fit in
	<= 1 page. If a fixed linker is available, we do it in this
	space-saving way now. Otherwise, we'll do it the old way
	(the gate DSO will then take up about 18KB instead of just
	~3KB).  Thanks to Roland McGrath for making this all work.

arch/ia64/Makefile

@@ -27,6 +27,10 @@ GCC_MINOR_VERSION=$(shell $(CC) -v 2>&1 | fgrep 'gcc version' | cut -f3 -d' ' |
 
 GAS_STATUS=$(shell arch/ia64/scripts/check-gas $(CC) $(OBJDUMP))
 
+arch-cppflags	:= $(shell arch/ia64/scripts/toolchain-flags $(CC) $(LD) $(OBJDUMP))
+cflags-y	+= $(arch-cppflags)
+AFLAGS		+= $(arch-cppflags)
+
 ifeq ($(GAS_STATUS),buggy)
 $(error Sorry, you need a newer version of the assember, one that is built from	\
 	a source-tree that post-dates 18-Dec-2002.  You can find a pre-compiled	\

arch/ia64/kernel/gate.lds.S

@@ -12,21 +12,46 @@ SECTIONS
 {
   . = GATE_ADDR + SIZEOF_HEADERS;
 
-  .hash				: { *(.hash) }					:rodata
+  .hash				: { *(.hash) }				:readable
   .dynsym			: { *(.dynsym) }
   .dynstr			: { *(.dynstr) }
   .gnu.version			: { *(.gnu.version) }
   .gnu.version_d		: { *(.gnu.version_d) }
   .gnu.version_r		: { *(.gnu.version_r) }
-  .data.patch.mckinley_e9	: { *(.data.patch.mckinley_e9) }
-  .data.patch.fsyscall_table	: { *(.data.patch.fsyscall_table) }
-  .data.patch.vtop		: { *(.data.patch.vtop) }
-  .data.patch.brl_fsys_bubble_down : { *(.data.patch.brl_fsys_bubble_down) }
-  .IA_64.unwind_info		: { *(.IA_64.unwind_info*) }
-  .IA_64.unwind			: { *(.IA_64.unwind*) }				:rodata :unwind
-  .dynamic			: { *(.dynamic) }				:rodata :dynamic
+  .dynamic			: { *(.dynamic) }			:readable :dynamic
+
+  /*
+   * This linker script is used both with -r and with -shared.  For the layouts to match,
+   * we need to skip more than enough space for the dynamic symbol table et al.  If this
+   * amount is insufficient, ld -shared will barf.  Just increase it here.
+   */
+  . = GATE_ADDR + 0x500;
+
+  .data.patch			: {
+				    __start_gate_mckinley_e9_patchlist = .;
+				    *(.data.patch.mckinley_e9)
+				    __end_gate_mckinley_e9_patchlist = .;
 
-  .text	(GATE_ADDR + PAGE_SIZE)	: { *(.text) }					:rodata
+				    __start_gate_vtop_patchlist = .;
+				    *(.data.patch.vtop)
+				    __end_gate_vtop_patchlist = .;
+
+				    __start_gate_fsyscall_patchlist = .;
+				    *(.data.patch.fsyscall_table)
+				    __end_gate_fsyscall_patchlist = .;
+
+				    __start_gate_brl_fsys_bubble_down_patchlist = .;
+				    *(.data.patch.brl_fsys_bubble_down)
+				    __end_gate_brl_fsys_bubble_down_patchlist = .;
+  }									:readable
+  .IA_64.unwind_info		: { *(.IA_64.unwind_info*) }
+  .IA_64.unwind			: { *(.IA_64.unwind*) }			:readable :unwind
+#ifdef HAVE_BUGGY_SEGREL
+  .text (GATE_ADDR + PAGE_SIZE)	: { *(.text) *(.text.*) }		:readable
+#else
+  . = ALIGN (PERCPU_PAGE_SIZE) + (. & (PERCPU_PAGE_SIZE - 1));
+  .text				: { *(.text) *(.text.*) }		:epc
+#endif
 
   /DISCARD/			: {
   	*(.got.plt) *(.got)
@@ -43,9 +68,12 @@ SECTIONS
  */
 PHDRS
 {
-  rodata  PT_LOAD	FILEHDR	PHDRS	FLAGS(4);	/* PF_R */
-  dynamic PT_DYNAMIC			FLAGS(4);	/* PF_R */
-  unwind  0x70000001;					/* PT_IA_64_UNWIND */
+  readable  PT_LOAD	FILEHDR	PHDRS	FLAGS(4);	/* PF_R */
+#ifndef HAVE_BUGGY_SEGREL
+  epc	    PT_LOAD	FILEHDR PHDRS	FLAGS(1);	/* PF_X */
+#endif
+  dynamic   PT_DYNAMIC			FLAGS(4);	/* PF_R */
+  unwind    0x70000001; /* PT_IA_64_UNWIND, but ld doesn't match the name */
 }
 
 /*
@@ -64,4 +92,4 @@ VERSION
 }
 
 /* The ELF entry point can be used to set the AT_SYSINFO value.  */
-ENTRY(syscall_via_epc)
+ENTRY(__kernel_syscall_via_epc)

arch/ia64/kernel/patch.c

@@ -99,7 +99,7 @@ ia64_patch_vtop (unsigned long start, unsigned long end)
 
 		/* replace virtual address with corresponding physical address: */
 		ia64_patch_imm64(ip, ia64_tpa(get_imm64(ip)));
-		ia64_fc(ip);
+		ia64_fc((void *) ip);
 		++offp;
 	}
 	ia64_sync_i();
@@ -143,11 +143,16 @@ patch_fsyscall_table (unsigned long start, unsigned long end)
 {
 	extern unsigned long fsyscall_table[NR_syscalls];
 	s32 *offp = (s32 *) start;
+	u64 ip;
 
 	while (offp < (s32 *) end) {
-		ia64_patch_imm64((u64) ia64_imva((char *) offp + *offp), (u64) fsyscall_table);
+		ip = (u64) ia64_imva((char *) offp + *offp);
+		ia64_patch_imm64(ip, (u64) fsyscall_table);
+		ia64_fc((void *) ip);
 		++offp;
 	}
+	ia64_sync_i();
+	ia64_srlz_i();
 }
 
 static void
@@ -161,52 +166,29 @@ patch_brl_fsys_bubble_down (unsigned long start, unsigned long end)
 		ip = (u64) offp + *offp;
 		ia64_patch_imm60((u64) ia64_imva((void *) ip),
 				 (u64) (fsys_bubble_down - (ip & -16)) / 16);
+		ia64_fc((void *) ip);
 		++offp;
 	}
+	ia64_sync_i();
+	ia64_srlz_i();
 }
 
 void
-ia64_patch_gate (Elf64_Ehdr *ehdr)
+ia64_patch_gate (void)
 {
-	Elf64_Shdr *shdr, *strsec, *sec;
-	Elf64_Half i;
-	unsigned long start, end;
-	char *strtab, *name;
-
-	if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) != 0 || ehdr->e_type != ET_DYN
-	    || !elf_check_arch(ehdr))
-		/*
-		 * Without the gate shared library, we can't do signal return or fast
-		 * system calls.  In other words, the kernel will crash quickly anyhow, so
-		 * we might just as well do it now...
-		 */
-		panic("%s: gate shared library at %p not a valid ELF object!\n",
-		      __FUNCTION__, ehdr);
-
-	shdr = (void *) ((char *) ehdr + ehdr->e_shoff);
-	strsec = (void *) ((char *) shdr + ehdr->e_shstrndx * ehdr->e_shentsize);
-	strtab = (void *) ((char *) ehdr + strsec->sh_offset);
-	for (i = 0; i < ehdr->e_shnum; ++i) {
-		sec = (void *) ((char *) shdr + i * ehdr->e_shentsize);
-
-		if (strncmp(strtab + sec->sh_name, ".data.patch.", 12) != 0)
-			continue;
-
-		if (sec->sh_size == 0)
-			continue;
-
-		name = strtab + sec->sh_name + 12;
-		start = sec->sh_addr;
-		end = start + sec->sh_size;
-		if (strcmp(name, "fsyscall_table") == 0)
-			patch_fsyscall_table(start, end);
-		else if (strcmp(name, "brl_fsys_bubble_down") == 0)
-			patch_brl_fsys_bubble_down(start, end);
-		else if (strcmp(name, "top") == 0)
-			ia64_patch_vtop(start, end);
-		else if (strcmp(name, "mckinley_e9") == 0)
-			ia64_patch_mckinley_e9(start, end);
-		else
-			panic("%s: found unknown patch-list `%s'\n", __FUNCTION__, name);
-	}
+	extern char __start_gate_mckinley_e9_patchlist;
+	extern char __end_gate_mckinley_e9_patchlist;
+	extern char __start_gate_vtop_patchlist;
+	extern char __end_gate_vtop_patchlist;
+	extern char __start_gate_fsyscall_patchlist;
+	extern char __end_gate_fsyscall_patchlist;
+	extern char __start_gate_brl_fsys_bubble_down_patchlist;
+	extern char __end_gate_brl_fsys_bubble_down_patchlist;
+#	define START(name)	((unsigned long) &__start_gate_##name##_patchlist)
+#	define END(name)	((unsigned long)&__end_gate_##name##_patchlist)
+
+	patch_fsyscall_table(START(fsyscall), END(fsyscall));
+	patch_brl_fsys_bubble_down(START(brl_fsys_bubble_down), END(brl_fsys_bubble_down));
+	ia64_patch_vtop(START(vtop), END(vtop));
+	ia64_patch_mckinley_e9(START(mckinley_e9), END(mckinley_e9));
 }

arch/ia64/mm/init.c

@@ -287,19 +287,22 @@ put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
 static void
 setup_gate (void)
 {
-	struct page *page0, *page1;
+	struct page *page;
 	extern char __start_gate_section[];
 
 	/*
-	 * Install the gate pages: one read-only page containing the ELF headers etc. and
-	 * one execute-only page, which enables privilege-promotion via "epc":
+	 * Map the gate page twice: once read-only to export the ELF headers etc. and once
+	 * execute-only page to enable privilege-promotion via "epc":
 	 */
-	page0 = virt_to_page(ia64_imva(__start_gate_section));
-	page1 = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE));
-	put_kernel_page(page0, GATE_ADDR, PAGE_READONLY);
-	put_kernel_page(page1, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
-
-	ia64_patch_gate((Elf64_Ehdr *) __start_gate_section);
+	page = virt_to_page(ia64_imva(__start_gate_section));
+	put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
+#ifdef HAVE_BUGGY_SEGREL
+	page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE));
+	put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
+#else
+	put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
+#endif
+	ia64_patch_gate();
 }
 
 void __init

arch/ia64/scripts/check-gas

@@ -2,8 +2,11 @@
 dir=$(dirname $0)
 CC=$1
 OBJDUMP=$2
-$CC -c $dir/check-gas-asm.S
-res=$($OBJDUMP -r --section .data check-gas-asm.o | fgrep 00004 | tr -s ' ' |cut -f3 -d' ')
+tmp=${TMPDIR:-/tmp}
+out=$tmp/out$$.o
+$CC -c $dir/check-gas-asm.S -o $out
+res=$($OBJDUMP -r --section .data $out | fgrep 00004 | tr -s ' ' |cut -f3 -d' ')
+rm -f $out
 if [ $res != ".text" ]; then
 	echo buggy
 else

arch/ia64/scripts/check-segrel.S

+	.rodata
+	data4 @segrel(start)
+	.data
+start:

arch/ia64/scripts/check-segrel.lds

+SECTIONS {
+	. = SIZEOF_HEADERS;
+	.rodata : { *(.rodata) } :ro
+	. = 0xa0000;
+	.data : { *(.data) } :dat
+	/DISCARD/ : { *(*) }
+}
+PHDRS {
+  ro PT_LOAD FILEHDR PHDRS;
+  dat PT_LOAD;
+}

arch/ia64/scripts/toolchain-flags

+#!/bin/sh
+#
+# Check whether linker can handle cross-segment @segrel():
+#
+CC=$1
+LD=$2
+OBJDUMP=$3
+dir=$(dirname $0)
+tmp=${TMPDIR:-/tmp}
+out=$tmp/out$$
+$CC -c $dir/check-segrel.S -o $out.o
+$LD -static -T$dir/check-segrel.lds $out.o -o $out
+res=$($OBJDUMP --full --section .rodata $out | fgrep 000 | cut -f3 -d' ')
+rm -f $out $out.o
+if [ $res != 00000a00 ]; then
+    echo " -DHAVE_BUGGY_SEGREL"
+    cat >&2 <<EOF
+warning: your linker cannot handle cross-segment segment-relative relocations.
+         please upgrade to a newer version (it is safe to use this linker, but
+         the kernel will be bigger than strictly necessary).
+EOF
+fi

include/asm-ia64/patch.h

@@ -20,6 +20,6 @@ extern void ia64_patch_imm60 (u64 insn_addr, u64 val);		/* patch "brl" w/ip-rel
 
 extern void ia64_patch_mckinley_e9 (unsigned long start, unsigned long end);
 extern void ia64_patch_vtop (unsigned long start, unsigned long end);
-extern void ia64_patch_gate (Elf64_Ehdr *ehdr);
+extern void ia64_patch_gate (void);
 
 #endif /* _ASM_IA64_PATCH_H */