s390: compile relocatable kernel without -fPIE

On s390, currently kernel uses the '-fPIE' compiler flag for compiling
vmlinux.  This has a few problems:

  - It uses dynamic symbols (.dynsym), for which the linker refuses to
    allow more than 64k sections.  This can break features which use
    '-ffunction-sections' and '-fdata-sections', including kpatch-build
    [1] and Function Granular KASLR.

  - It unnecessarily uses GOT relocations, adding an extra layer of
    indirection for many memory accesses.

Instead of using '-fPIE', resolve all the relocations at link time and
then manually adjust any absolute relocations (R_390_64) during boot.

This is done by first telling the linker to preserve all relocations
during the vmlinux link.  (Note this is harmless: they are later
stripped in the vmlinux.bin link.)

Then use the 'relocs' tool to find all absolute relocations (R_390_64)
which apply to allocatable sections.  The offsets of those relocations
are saved in a special section which is then used to adjust the
relocations during boot.

(Note: For some reason, Clang occasionally creates a GOT reference, even
without '-fPIE'.  So Clang-compiled kernels have a GOT, which needs to
be adjusted.)

On my mostly-defconfig kernel, this reduces kernel text size by ~1.3%.

[1] dynup/kpatch#1284
[2] https://gcc.gnu.org/pipermail/gcc-patches/2023-June/622872.html
[3] https://gcc.gnu.org/pipermail/gcc-patches/2023-August/625986.html

Compiler consideration:

Gcc recently implemented an optimization [2] for loading symbols without
explicit alignment, aligning with the IBM Z ELF ABI. This ABI mandates
symbols to reside on a 2-byte boundary, enabling the use of the larl
instruction. However, kernel linker scripts may still generate unaligned
symbols. To address this, a new -munaligned-symbols option has been
introduced [3] in recent gcc versions. This option has to be used with
future gcc versions.

Older Clang lacks support for handling unaligned symbols generated
by kernel linker scripts when the kernel is built without -fPIE. However,
future versions of Clang will include support for the -munaligned-symbols
option. When the support is unavailable, compile the kernel with -fPIE
to maintain the existing behavior.

In addition to it:
move vmlinux.relocs to safe relocation

When the kernel is built with CONFIG_KERNEL_UNCOMPRESSED, the entire
uncompressed vmlinux.bin is positioned in the bzImage decompressor
image at the default kernel LMA of 0x100000, enabling it to be executed
in-place. However, the size of .vmlinux.relocs could be large enough to
cause an overlap with the uncompressed kernel at the address 0x100000.
To address this issue, .vmlinux.relocs is positioned after the
.rodata.compressed in the bzImage. Nevertheless, in this configuration,
vmlinux.relocs will overlap with the .bss section of vmlinux.bin. To
overcome that, move vmlinux.relocs to a safe location before clearing
.bss and handling relocs.

Compile warning fix from Sumanth Korikkar:

When kernel is built with CONFIG_LD_ORPHAN_WARN and -fno-PIE, there are
several warnings:

ld: warning: orphan section `.rela.iplt' from
`arch/s390/kernel/head64.o' being placed in section `.rela.dyn'
ld: warning: orphan section `.rela.head.text' from
`arch/s390/kernel/head64.o' being placed in section `.rela.dyn'
ld: warning: orphan section `.rela.init.text' from
`arch/s390/kernel/head64.o' being placed in section `.rela.dyn'
ld: warning: orphan section `.rela.rodata.cst8' from
`arch/s390/kernel/head64.o' being placed in section `.rela.dyn'

Orphan sections are sections that exist in an object file but don't have
a corresponding output section in the final executable. ld raises a
warning when it identifies such sections.

Eliminate the warning by placing all .rela orphan sections in .rela.dyn
and raise an error when size of .rela.dyn is greater than zero. i.e.
Dont just neglect orphan sections.

This is similar to adjustment performed in x86, where kernel is built
with -fno-PIE.
commit 5354e84 ("x86/build: Add asserts for unwanted sections")

[[email protected]: rebased Josh Poimboeuf patches and move
 vmlinux.relocs to safe location]
[[email protected]: merged compile warning fix from Sumanth]
Tested-by: Sumanth Korikkar <[email protected]>
Acked-by: Vasily Gorbik <[email protected]>
Signed-off-by: Josh Poimboeuf <[email protected]>
Signed-off-by: Sumanth Korikkar <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Heiko Carstens <[email protected]>

arch/s390/Kconfig

@@ -583,14 +583,23 @@ config RELOCATABLE
 	help
 	  This builds a kernel image that retains relocation information
 	  so it can be loaded at an arbitrary address.
-	  The kernel is linked as a position-independent executable (PIE)
-	  and contains dynamic relocations which are processed early in the
-	  bootup process.
 	  The relocations make the kernel image about 15% larger (compressed
 	  10%), but are discarded at runtime.
 	  Note: this option exists only for documentation purposes, please do
 	  not remove it.
 
+config PIE_BUILD
+	def_bool CC_IS_CLANG && !$(cc-option,-munaligned-symbols)
+	help
+	  If the compiler is unable to generate code that can manage unaligned
+	  symbols, the kernel is linked as a position-independent executable
+	  (PIE) and includes dynamic relocations that are processed early
+	  during bootup.
+
+	  For kpatch functionality, it is recommended to build the kernel
+	  without the PIE_BUILD option. PIE_BUILD is only enabled when the
+	  compiler lacks proper support for handling unaligned symbols.
+
 config RANDOMIZE_BASE
 	bool "Randomize the address of the kernel image (KASLR)"
 	default y

arch/s390/Makefile

@@ -14,8 +14,14 @@ KBUILD_AFLAGS_MODULE += -fPIC
 KBUILD_CFLAGS_MODULE += -fPIC
 KBUILD_AFLAGS	+= -m64
 KBUILD_CFLAGS	+= -m64
+ifdef CONFIG_PIE_BUILD
 KBUILD_CFLAGS	+= -fPIE
 LDFLAGS_vmlinux	:= -pie -z notext
+else
+KBUILD_CFLAGS	+= $(call cc-option,-munaligned-symbols,)
+LDFLAGS_vmlinux	:= --emit-relocs --discard-none
+extra_tools	:= relocs
+endif
 aflags_dwarf	:= -Wa,-gdwarf-2
 KBUILD_AFLAGS_DECOMPRESSOR := $(CLANG_FLAGS) -m64 -D__ASSEMBLY__
 ifndef CONFIG_AS_IS_LLVM
@@ -143,7 +149,7 @@ archheaders:
 
 archprepare:
 	$(Q)$(MAKE) $(build)=$(syscalls) kapi
-	$(Q)$(MAKE) $(build)=$(tools) kapi
+	$(Q)$(MAKE) $(build)=$(tools) kapi $(extra_tools)
 ifeq ($(KBUILD_EXTMOD),)
 # We need to generate vdso-offsets.h before compiling certain files in kernel/.
 # In order to do that, we should use the archprepare target, but we can't since

arch/s390/boot/.gitignore

@@ -1,6 +1,7 @@
 # SPDX-License-Identifier: GPL-2.0-only
 image
 bzImage
+relocs.S
 section_cmp.*
 vmlinux
 vmlinux.lds

arch/s390/boot/Makefile

@@ -37,7 +37,8 @@ CFLAGS_sclp_early_core.o += -I$(srctree)/drivers/s390/char
 
 obj-y	:= head.o als.o startup.o physmem_info.o ipl_parm.o ipl_report.o vmem.o
 obj-y	+= string.o ebcdic.o sclp_early_core.o mem.o ipl_vmparm.o cmdline.o
-obj-y	+= version.o pgm_check_info.o ctype.o ipl_data.o machine_kexec_reloc.o
+obj-y	+= version.o pgm_check_info.o ctype.o ipl_data.o
+obj-y	+= $(if $(CONFIG_PIE_BUILD),machine_kexec_reloc.o,relocs.o)
 obj-$(findstring y, $(CONFIG_PROTECTED_VIRTUALIZATION_GUEST) $(CONFIG_PGSTE))	+= uv.o
 obj-$(CONFIG_RANDOMIZE_BASE)	+= kaslr.o
 obj-y	+= $(if $(CONFIG_KERNEL_UNCOMPRESSED),,decompressor.o) info.o
@@ -48,6 +49,9 @@ targets	:= bzImage section_cmp.boot.data section_cmp.boot.preserved.data $(obj-y
 targets	+= vmlinux.lds vmlinux vmlinux.bin vmlinux.bin.gz vmlinux.bin.bz2
 targets += vmlinux.bin.xz vmlinux.bin.lzma vmlinux.bin.lzo vmlinux.bin.lz4
 targets += vmlinux.bin.zst info.bin syms.bin vmlinux.syms $(obj-all)
+ifndef CONFIG_PIE_BUILD
+targets += relocs.S
+endif
 
 OBJECTS := $(addprefix $(obj)/,$(obj-y))
 OBJECTS_ALL := $(addprefix $(obj)/,$(obj-all))
@@ -106,6 +110,14 @@ OBJCOPYFLAGS_vmlinux.bin := -O binary --remove-section=.comment --remove-section
 $(obj)/vmlinux.bin: vmlinux FORCE
 	$(call if_changed,objcopy)
 
+ifndef CONFIG_PIE_BUILD
+CMD_RELOCS=arch/s390/tools/relocs
+quiet_cmd_relocs = RELOCS $@
+	cmd_relocs = $(CMD_RELOCS) $< > $@
+$(obj)/relocs.S: vmlinux FORCE
+	$(call if_changed,relocs)
+endif
+
 suffix-$(CONFIG_KERNEL_GZIP)  := .gz
 suffix-$(CONFIG_KERNEL_BZIP2) := .bz2
 suffix-$(CONFIG_KERNEL_LZ4)  := .lz4

arch/s390/boot/boot.h

@@ -25,9 +25,14 @@ struct vmlinux_info {
 	unsigned long bootdata_size;
 	unsigned long bootdata_preserved_off;
 	unsigned long bootdata_preserved_size;
+#ifdef CONFIG_PIE_BUILD
 	unsigned long dynsym_start;
 	unsigned long rela_dyn_start;
 	unsigned long rela_dyn_end;
+#else
+	unsigned long got_off;
+	unsigned long got_size;
+#endif
 	unsigned long amode31_size;
 	unsigned long init_mm_off;
 	unsigned long swapper_pg_dir_off;
@@ -83,6 +88,7 @@ extern unsigned long vmalloc_size;
 extern int vmalloc_size_set;
 extern char __boot_data_start[], __boot_data_end[];
 extern char __boot_data_preserved_start[], __boot_data_preserved_end[];
+extern int __vmlinux_relocs_64_start[], __vmlinux_relocs_64_end[];
 extern char _decompressor_syms_start[], _decompressor_syms_end[];
 extern char _stack_start[], _stack_end[];
 extern char _end[], _decompressor_end[];

arch/s390/boot/startup.c

@@ -141,7 +141,8 @@ static void copy_bootdata(void)
 	memcpy((void *)vmlinux.bootdata_preserved_off, __boot_data_preserved_start, vmlinux.bootdata_preserved_size);
 }
 
-static void handle_relocs(unsigned long offset)
+#ifdef CONFIG_PIE_BUILD
+static void kaslr_adjust_relocs(unsigned long min_addr, unsigned long offset)
 {
 	Elf64_Rela *rela_start, *rela_end, *rela;
 	int r_type, r_sym, rc;
@@ -172,6 +173,62 @@ static void handle_relocs(unsigned long offset)
 	}
 }
 
+static void kaslr_adjust_got(unsigned long offset) {}
+static void rescue_relocs(void) {}
+static void free_relocs(void) {}
+#else
+int *vmlinux_relocs_64_start;
+int *vmlinux_relocs_64_end;
+
+static void rescue_relocs(void)
+{
+	unsigned long size, nrelocs;
+
+	nrelocs = __vmlinux_relocs_64_end - __vmlinux_relocs_64_start;
+	size = nrelocs * sizeof(uint32_t);
+	vmlinux_relocs_64_start = (void *)physmem_alloc_top_down(RR_RELOC, size, 0);
+	memmove(vmlinux_relocs_64_start, (void *)__vmlinux_relocs_64_start, size);
+	vmlinux_relocs_64_end = vmlinux_relocs_64_start + nrelocs;
+}
+
+static void free_relocs(void)
+{
+	physmem_free(RR_RELOC);
+}
+
+static void kaslr_adjust_relocs(unsigned long min_addr, unsigned long offset)
+{
+	int *reloc;
+	unsigned long max_addr = min_addr + vmlinux.image_size;
+	long loc;
+
+	/* Adjust R_390_64 relocations */
+	for (reloc = vmlinux_relocs_64_start;
+		reloc < vmlinux_relocs_64_end && *reloc;
+		reloc++) {
+		loc = (long)*reloc + offset;
+		if (loc < min_addr || loc > max_addr)
+			error("64-bit relocation outside of kernel!\n");
+		*(u64 *)loc += offset;
+	}
+}
+
+static void kaslr_adjust_got(unsigned long offset)
+{
+	u64 *entry;
+
+	/*
+	 * Even without -fPIE, Clang still uses a global offset table for some
+	 * reason.  Adjust the GOT entries.
+	 */
+	for (entry = (u64 *)vmlinux.got_off;
+	     entry < (u64 *)(vmlinux.got_off + vmlinux.got_size);
+	     entry++) {
+		*entry += offset;
+	}
+}
+#endif
+
 /*
  * Merge information from several sources into a single ident_map_size value.
  * "ident_map_size" represents the upper limit of physical memory we may ever
@@ -299,14 +356,18 @@ static void setup_vmalloc_size(void)
 	vmalloc_size = max(size, vmalloc_size);
 }
 
-static void offset_vmlinux_info(unsigned long offset)
+static void kaslr_adjust_vmlinux_info(unsigned long offset)
 {
 	*(unsigned long *)(&vmlinux.entry) += offset;
 	vmlinux.bootdata_off += offset;
 	vmlinux.bootdata_preserved_off += offset;
+#ifdef CONFIG_PIE_BUILD
 	vmlinux.rela_dyn_start += offset;
 	vmlinux.rela_dyn_end += offset;
 	vmlinux.dynsym_start += offset;
+#else
+	vmlinux.got_off += offset;
+#endif
 	vmlinux.init_mm_off += offset;
 	vmlinux.swapper_pg_dir_off += offset;
 	vmlinux.invalid_pg_dir_off += offset;
@@ -361,14 +422,15 @@ void startup_kernel(void)
 	detect_physmem_online_ranges(max_physmem_end);
 	save_ipl_cert_comp_list();
 	rescue_initrd(safe_addr, ident_map_size);
+	rescue_relocs();
 
 	if (kaslr_enabled()) {
 		vmlinux_lma = randomize_within_range(vmlinux.image_size + vmlinux.bss_size,
 						     THREAD_SIZE, vmlinux.default_lma,
 						     ident_map_size);
 		if (vmlinux_lma) {
 			__kaslr_offset = vmlinux_lma - vmlinux.default_lma;
-			offset_vmlinux_info(__kaslr_offset);
+			kaslr_adjust_vmlinux_info(__kaslr_offset);
 		}
 	}
 	vmlinux_lma = vmlinux_lma ?: vmlinux.default_lma;
@@ -393,18 +455,20 @@ void startup_kernel(void)
 	/*
 	 * The order of the following operations is important:
 	 *
-	 * - handle_relocs() must follow clear_bss_section() to establish static
+	 * - kaslr_adjust_relocs() must follow clear_bss_section() to establish static
 	 *   memory references to data in .bss to be used by setup_vmem()
 	 *   (i.e init_mm.pgd)
 	 *
-	 * - setup_vmem() must follow handle_relocs() to be able using
+	 * - setup_vmem() must follow kaslr_adjust_relocs() to be able using
 	 *   static memory references to data in .bss (i.e init_mm.pgd)
 	 *
-	 * - copy_bootdata() must follow setup_vmem() to propagate changes to
-	 *   bootdata made by setup_vmem()
+	 * - copy_bootdata() must follow setup_vmem() to propagate changes
+	 *   to bootdata made by setup_vmem()
 	 */
 	clear_bss_section(vmlinux_lma);
-	handle_relocs(__kaslr_offset);
+	kaslr_adjust_relocs(vmlinux_lma, __kaslr_offset);
+	kaslr_adjust_got(__kaslr_offset);
+	free_relocs();
 	setup_vmem(asce_limit);
 	copy_bootdata();
 

arch/s390/boot/vmlinux.lds.S

@@ -110,6 +110,24 @@ SECTIONS
 		_compressed_end = .;
 	}
 
+#ifndef CONFIG_PIE_BUILD
+	/*
+	 * When the kernel is built with CONFIG_KERNEL_UNCOMPRESSED, the entire
+	 * uncompressed vmlinux.bin is positioned in the bzImage decompressor
+	 * image at the default kernel LMA of 0x100000, enabling it to be
+	 * executed in-place. However, the size of .vmlinux.relocs could be
+	 * large enough to cause an overlap with the uncompressed kernel at the
+	 * address 0x100000. To address this issue, .vmlinux.relocs is
+	 * positioned after the .rodata.compressed.
+	 */
+	. = ALIGN(4);
+	.vmlinux.relocs : {
+		__vmlinux_relocs_64_start = .;
+		*(.vmlinux.relocs_64)
+		__vmlinux_relocs_64_end = .;
+	}
+#endif
+
 #define SB_TRAILER_SIZE 32
 	/* Trailer needed for Secure Boot */
 	. += SB_TRAILER_SIZE; /* make sure .sb.trailer does not overwrite the previous section */

arch/s390/include/asm/physmem_info.h

@@ -22,6 +22,7 @@ enum reserved_range_type {
 	RR_DECOMPRESSOR,
 	RR_INITRD,
 	RR_VMLINUX,
+	RR_RELOC,
 	RR_AMODE31,
 	RR_IPLREPORT,
 	RR_CERT_COMP_LIST,

arch/s390/kernel/vmlinux.lds.S

@@ -63,7 +63,9 @@ SECTIONS
 		*(.data.rel.ro .data.rel.ro.*)
 	}
 	.got : {
+		__got_start = .;
 		*(.got)
+		__got_end = .;
 	}
 
 	. = ALIGN(PAGE_SIZE);
@@ -190,6 +192,7 @@ SECTIONS
 
 	PERCPU_SECTION(0x100)
 
+#ifdef CONFIG_PIE_BUILD
 	.dynsym ALIGN(8) : {
 		__dynsym_start = .;
 		*(.dynsym)
@@ -206,6 +209,7 @@ SECTIONS
 	.dynstr ALIGN(8) : {
 		*(.dynstr)
 	}
+#endif
 	.hash ALIGN(8) : {
 		*(.hash)
 	}
@@ -235,9 +239,14 @@ SECTIONS
 		QUAD(__boot_data_preserved_start)		/* bootdata_preserved_off */
 		QUAD(__boot_data_preserved_end -
 		     __boot_data_preserved_start)		/* bootdata_preserved_size */
+#ifdef CONFIG_PIE_BUILD
 		QUAD(__dynsym_start)				/* dynsym_start */
 		QUAD(__rela_dyn_start)				/* rela_dyn_start */
 		QUAD(__rela_dyn_end)				/* rela_dyn_end */
+#else
+		QUAD(__got_start)				/* got_off */
+		QUAD(__got_end - __got_start)			/* got_size */
+#endif
 		QUAD(_eamode31 - _samode31)			/* amode31_size */
 		QUAD(init_mm)
 		QUAD(swapper_pg_dir)
@@ -268,6 +277,12 @@ SECTIONS
 		*(.plt) *(.plt.*) *(.iplt) *(.igot .igot.plt)
 	}
 	ASSERT(SIZEOF(.plt) == 0, "Unexpected run-time procedure linkages detected!")
+#ifndef CONFIG_PIE_BUILD
+	.rela.dyn : {
+		*(.rela.*) *(.rela_*)
+	}
+	ASSERT(SIZEOF(.rela.dyn) == 0, "Unexpected run-time relocations (.rela) detected!")
+#endif
 
 	/* Sections to be discarded */
 	DISCARDS