Re: [PATCH 3/4] LoongArch: Add support for kernel relocation

From: Jinyang He
Date: Sat Jan 14 2023 - 03:31:34 EST



On 2023-01-14 15:50, Youling Tang wrote:
Hi, Jinyang

On 01/14/2023 02:54 PM, Jinyang He wrote:

On 2023-01-09 17:07, Youling Tang wrote:
arch/loongarch/kernel/relocate.c contains the functions necessary to
relocate the kernel elsewhere in memory.

The kernel makes a copy of itself at the new address. It uses the
relocation table inserted by the relocs tool to fix symbol references
within the new image.

If copy/relocation is successful then the entry point of the new kernel
is returned, otherwise fall back to starting the kernel in place.

Signed-off-by: Youling Tang <tangyouling@xxxxxxxxxxx>
---
  arch/loongarch/Kconfig              | 15 +++++
  arch/loongarch/Makefile             |  5 ++
  arch/loongarch/kernel/Makefile      |  2 +
  arch/loongarch/kernel/head.S        | 18 ++++++
  arch/loongarch/kernel/relocate.c    | 96 +++++++++++++++++++++++++++++
  arch/loongarch/kernel/vmlinux.lds.S | 11 +++-
  6 files changed, 145 insertions(+), 2 deletions(-)
  create mode 100644 arch/loongarch/kernel/relocate.c

diff --git a/arch/loongarch/Kconfig b/arch/loongarch/Kconfig
index 9cc8b84f7eb0..089a4695b1b3 100644
--- a/arch/loongarch/Kconfig
+++ b/arch/loongarch/Kconfig
@@ -48,6 +48,7 @@ config LOONGARCH
      select ARCH_SUPPORTS_ATOMIC_RMW
      select ARCH_SUPPORTS_HUGETLBFS
      select ARCH_SUPPORTS_NUMA_BALANCING
+    select SYS_SUPPORTS_RELOCATABLE
      select ARCH_USE_BUILTIN_BSWAP
      select ARCH_USE_CMPXCHG_LOCKREF
      select ARCH_USE_QUEUED_RWLOCKS
@@ -229,6 +230,11 @@ config SCHED_OMIT_FRAME_POINTER
  config AS_HAS_EXPLICIT_RELOCS
      def_bool $(as-instr,x:pcalau12i \$t0$(comma)%pc_hi20(x))
  +config SYS_SUPPORTS_RELOCATABLE
+    bool
+    help
+      Selected if the platform supports relocating the kernel.
+
  menu "Kernel type and options"
    source "kernel/Kconfig.hz"
@@ -474,6 +480,15 @@ config PHYSICAL_START
        specified in the "crashkernel=YM@XM" command line boot parameter
        passed to the panic-ed kernel).
  +config RELOCATABLE
+    bool "Relocatable kernel"
+    depends on SYS_SUPPORTS_RELOCATABLE
+    help
+      This builds the kernel as a Position Independent Executable (PIE),
+      which retains all relocation metadata required to relocate the
+      kernel binary at runtime to a different virtual address than the
+      address it was linked at.
+
  config SECCOMP
      bool "Enable seccomp to safely compute untrusted bytecode"
      depends on PROC_FS
diff --git a/arch/loongarch/Makefile b/arch/loongarch/Makefile
index 4402387d2755..27b5a70ff31c 100644
--- a/arch/loongarch/Makefile
+++ b/arch/loongarch/Makefile
@@ -71,6 +71,11 @@ KBUILD_AFLAGS_MODULE        +=
-Wa,-mla-global-with-abs
  KBUILD_CFLAGS_MODULE        += -fplt
-Wa,-mla-global-with-abs,-mla-local-with-abs
  endif
  +ifeq ($(CONFIG_RELOCATABLE),y)
+LDFLAGS_vmlinux            += -static -pie --no-dynamic-linker -z notext
+KBUILD_CFLAGS_KERNEL        += -fPIE
+endif
+
  cflags-y += -ffreestanding
  cflags-y += $(call cc-option, -mno-check-zero-division)
  diff --git a/arch/loongarch/kernel/Makefile
b/arch/loongarch/kernel/Makefile
index fcaa024a685e..33787d22e6f4 100644
--- a/arch/loongarch/kernel/Makefile
+++ b/arch/loongarch/kernel/Makefile
@@ -31,6 +31,8 @@ endif
  obj-$(CONFIG_MODULES)        += module.o module-sections.o
  obj-$(CONFIG_STACKTRACE)    += stacktrace.o
  +obj-$(CONFIG_RELOCATABLE)    += relocate.o
+
  obj-$(CONFIG_PROC_FS)        += proc.o
    obj-$(CONFIG_SMP)        += smp.o
diff --git a/arch/loongarch/kernel/head.S b/arch/loongarch/kernel/head.S
index e8a4bf9d7599..6db1549177ad 100644
--- a/arch/loongarch/kernel/head.S
+++ b/arch/loongarch/kernel/head.S
@@ -88,7 +88,25 @@ SYM_CODE_START(kernel_entry)            # kernel
entry point
      PTR_ADD        sp, sp, tp
      set_saved_sp    sp, t0, t1
  +#ifdef CONFIG_RELOCATABLE
+    /* Copy kernel and apply the relocations */
+    bl        relocate_kernel
+
+    /* Repoint the sp into the new kernel image */
+    PTR_LI        sp, (_THREAD_SIZE - 32 - PT_SIZE)
+    PTR_ADD        sp, sp, tp
+    set_saved_sp    sp, t0, t1
+    PTR_ADDI    sp, sp, -4 * SZREG      # init stack pointer
+
+    /*
+     * relocate_kernel returns the entry point either
+     * in the relocated kernel or the original if for
+     * some reason relocation failed.
+    */
+    jr        a0
+#else
      bl        start_kernel
+#endif
      ASM_BUG()
    SYM_CODE_END(kernel_entry)
diff --git a/arch/loongarch/kernel/relocate.c
b/arch/loongarch/kernel/relocate.c
new file mode 100644
index 000000000000..a58551c0698d
--- /dev/null
+++ b/arch/loongarch/kernel/relocate.c
@@ -0,0 +1,96 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Support for Kernel relocation at boot time
+ *
+ * Copyright (C) 2023 Loongson Technology Corporation Limited
+ */
+
+#include <linux/elf.h>
+#include <linux/kernel.h>
+#include <linux/start_kernel.h>
+#include <linux/printk.h>
+#include <linux/panic_notifier.h>
+#include <asm/bootinfo.h>
+#include <asm/inst.h>
+#include <asm/sections.h>
+
+#define RELOCATED(x) ((void *)((long)x + offset))
+
+extern long __rela_dyn_start;
+extern long __rela_dyn_end;
+
+/*
+ * Choose a new address for the kernel, for now we'll hard
+ * code the destination.
+ */
+static inline void __init *determine_relocation_address(void)
+{
+    return (void *)(CACHE_BASE + 0x02000000);
+}
+
+static inline int __init relocation_addr_valid(void *loc_new)
+{
+    if ((unsigned long)loc_new & 0x0000ffff) {

Hi, Youling,


12bits-aligned is OK as actually R_LARCH_PCALA_LO12 is absolute.

Yes, 12bits-aligned is enough.

But for the convenience of debugging, is __kaslr_offset more friendly
to align with 1M? (For example, use objdump -d vmlinux --start-address=
(real_addr - __kaslr_offset)).

This will reduce the random range, but it's up to you.





BTW, I think the relocation of kernel looks like the '.so' solve its
Relocs at user space so that it can be placed any legal address. The
address is determined first, and relocation make it works on that address.
Thus, this patch always relocate kernel to 'CACHE_BASE + 0x02000000'
makes me puzzled. I think the relocation only works when the kernel
runtime address is different from the link-time address.

IMHO, an immutable offset (eg: 0x1e00000) for the relocation kernel may
not make much sense, it may only be used during debugging to confirm
whether the relocation kernel offset is as expected. (We can remove the
immutable offset code if you think it is unnecessary) A random offset
will be generated when KASLR is enabled.

What I expected is that when the kernel runtime address is different
from the link-time address, the relocatable kernel can solve its
Relocs so that it can work well again. And the immutable offset is
relocate kernel to another address, that looks like one of random
offsets. Although relocatable and kaslr are similar, they are not the same.


Thanks,

Jinyang



Thanks,
Youling



Thanks,
Jinyang


+        /* Inappropriately aligned new location */
+        return 0;
+    }
+    if ((unsigned long)loc_new < (unsigned long)_end) {
+        /* New location overlaps original kernel */
+        return 0;
+    }
+    return 1;
+}
+
+void *__init relocate_kernel(void)
+{
+    Elf64_Rela *rela, *rela_end;
+    void *loc_new;
+    unsigned long kernel_length;
+    long offset = 0;
+    int res = 1;
+    /* Default to original kernel entry point */
+    void *kernel_entry = start_kernel;
+
+    kernel_length = (long)(_end) - (long)(_text);
+
+    loc_new = determine_relocation_address();
+
+    /* Sanity check relocation address */
+    if (relocation_addr_valid(loc_new))
+        offset = (unsigned long)loc_new - (unsigned long)(_text);
+
+    if (offset) {
+        /* Copy the kernel to it's new location */
+        memcpy(loc_new, _text, kernel_length);
+
+        /* Sync the caches ready for execution of new kernel */
+        __asm__ __volatile__ (
+            "ibar 0 \t\n"
+            "dbar 0 \t\n");
+
+        rela = (Elf64_Rela *)RELOCATED(&__rela_dyn_start);
+        rela_end = (Elf64_Rela *)RELOCATED(&__rela_dyn_end);
+
+        for ( ; rela < rela_end; rela++) {
+            Elf64_Addr addr = rela->r_offset;
+            Elf64_Addr relocated_addr = rela->r_addend;
+
+            if (rela->r_info != R_LARCH_RELATIVE)
+                continue;
+
+            if (relocated_addr >= VMLINUX_LOAD_ADDRESS)
+                relocated_addr = RELOCATED(relocated_addr);
+
+            *(Elf64_Addr *)RELOCATED(addr) = relocated_addr;
+
+        }
+
+        /* The current thread is now within the relocated image */
+        __current_thread_info = RELOCATED(__current_thread_info);
+
+        /* Return the new kernel's entry point */
+        kernel_entry = RELOCATED(start_kernel);
+    }
+out:
+    return kernel_entry;
+}
diff --git a/arch/loongarch/kernel/vmlinux.lds.S
b/arch/loongarch/kernel/vmlinux.lds.S
index 733b16e8d55d..aec0b6567d24 100644
--- a/arch/loongarch/kernel/vmlinux.lds.S
+++ b/arch/loongarch/kernel/vmlinux.lds.S
@@ -70,6 +70,8 @@ SECTIONS
      .plt : ALIGN(16) { *(.plt) }
      .got.plt : ALIGN(16) { *(.got.plt) }
  +    .data.rel : { *(.data.rel*) }
+
      . = ALIGN(PECOFF_SEGMENT_ALIGN);
      __init_begin = .;
      __inittext_begin = .;
@@ -93,8 +95,6 @@ SECTIONS
      PERCPU_SECTION(1 << CONFIG_L1_CACHE_SHIFT)
  #endif
  -    .rela.dyn : ALIGN(8) { *(.rela.dyn) *(.rela*) }
-
      .init.bss : {
          *(.init.bss)
      }
@@ -107,6 +107,12 @@ SECTIONS
      RO_DATA(4096)
      RW_DATA(1 << CONFIG_L1_CACHE_SHIFT, PAGE_SIZE, THREAD_SIZE)
  +    .rela.dyn : ALIGN(8) {
+        __rela_dyn_start = .;
+         *(.rela.dyn) *(.rela*)
+        __rela_dyn_end = .;
+    }
+
      .sdata : {
          *(.sdata)
      }
@@ -133,6 +139,7 @@ SECTIONS
        DISCARDS
      /DISCARD/ : {
+        *(.dynamic .dynsym .dynstr .hash .gnu.hash)
          *(.gnu.attributes)
          *(.options)
          *(.eh_frame)