x86: make percpu symbols zerobased on SMP

[ Based on original patch from Christoph Lameter and Mike Travis. ]

This patch makes percpu symbols zerobased on x86_64 SMP by adding
PERCPU_VADDR() to vmlinux.lds.h which helps setting explicit vaddr on
the percpu output section and using it in vmlinux_64.lds.S.  A new
PHDR is added as existing ones cannot contain sections near address
zero.  PERCPU_VADDR() also adds a new symbol __per_cpu_load which
always points to the vaddr of the loaded percpu data.init region.

The following adjustments have been made to accomodate the address
change.

* code to locate percpu gdt_page in head_64.S is updated to add the
  load address to the gdt_page offset.

* __per_cpu_load is used in places where access to the init data area
  is necessary.

* pda->data_offset is initialized soon after C code is entered as zero
  value doesn't work anymore.

This patch is mostly taken from Mike Travis' "x86_64: Base percpu
variables at zero" patch.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

6 files changed, 88 insertions, 10 deletions

diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
index b9a4d8c4b935..bc2900ca82c7 100644
--- a/arch/x86/kernel/head64.c
+++ b/arch/x86/kernel/head64.c
@@ -44,6 +44,8 @@ void __init x86_64_init_pda(void)
 {
         _cpu_pda = __cpu_pda;
         cpu_pda(0) = &_boot_cpu_pda;
+        cpu_pda(0)->data_offset =
+                (unsigned long)(__per_cpu_load - __per_cpu_start);
         pda_init(0);
 }
 
diff --git a/arch/x86/kernel/head_64.S b/arch/x86/kernel/head_64.S
index 0e275d495563..7ee0363871e8 100644
--- a/arch/x86/kernel/head_64.S
+++ b/arch/x86/kernel/head_64.S
@@ -204,6 +204,23 @@ ENTRY(secondary_startup_64)
         pushq $0
         popfq
 
+#ifdef CONFIG_SMP
+        /*
+         * early_gdt_base should point to the gdt_page in static percpu init
+         * data area.  Computing this requires two symbols - __per_cpu_load
+         * and per_cpu__gdt_page.  As linker can't do no such relocation, do
+         * it by hand.  As early_gdt_descr is manipulated by C code for
+         * secondary CPUs, this should be done only once for the boot CPU
+         * when early_gdt_descr_base contains zero.
+         */
+        movq    early_gdt_descr_base(%rip), %rax
+        testq   %rax, %rax
+        jnz     1f
+        movq    $__per_cpu_load, %rax
+        addq    $per_cpu__gdt_page, %rax
+        movq    %rax, early_gdt_descr_base(%rip)
+1:
+#endif
         /*
          * We must switch to a new descriptor in kernel space for the GDT
          * because soon the kernel won't have access anymore to the userspace
@@ -401,7 +418,12 @@ NEXT_PAGE(level2_spare_pgt)
         .globl early_gdt_descr
 early_gdt_descr:
         .word   GDT_ENTRIES*8-1
-        .quad   per_cpu__gdt_page
+#ifdef CONFIG_SMP
+early_gdt_descr_base:
+        .quad   0x0000000000000000
+#else
+        .quad   per_cpu__gdt_page
+#endif
 
 ENTRY(phys_base)
         /* This must match the first entry in level2_kernel_pgt */
diff --git a/arch/x86/kernel/setup_percpu.c b/arch/x86/kernel/setup_percpu.c
index 56c63ac62b10..44845842e722 100644
--- a/arch/x86/kernel/setup_percpu.c
+++ b/arch/x86/kernel/setup_percpu.c
@@ -213,7 +213,7 @@ void __init setup_per_cpu_areas(void)
                 }
 #endif
                 per_cpu_offset(cpu) = ptr - __per_cpu_start;
-                memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
+                memcpy(ptr, __per_cpu_load, __per_cpu_end - __per_cpu_start);
 
                 DBG("PERCPU: cpu %4d %p\n", cpu, ptr);
         }
diff --git a/arch/x86/kernel/vmlinux_64.lds.S b/arch/x86/kernel/vmlinux_64.lds.S
index 1a614c0e6bef..f50280db0dfe 100644
--- a/arch/x86/kernel/vmlinux_64.lds.S
+++ b/arch/x86/kernel/vmlinux_64.lds.S
@@ -19,6 +19,9 @@ PHDRS {
         data PT_LOAD FLAGS(7);  /* RWE */
         user PT_LOAD FLAGS(7);  /* RWE */
         data.init PT_LOAD FLAGS(7);     /* RWE */
+#ifdef CONFIG_SMP
+        percpu PT_LOAD FLAGS(7);        /* RWE */
+#endif
         note PT_NOTE FLAGS(0);  /* ___ */
 }
 SECTIONS
@@ -208,14 +211,26 @@ SECTIONS
   __initramfs_end = .;
 #endif
 
+#ifdef CONFIG_SMP
+  /*
+   * percpu offsets are zero-based on SMP.  PERCPU_VADDR() changes the
+   * output PHDR, so the next output section - __data_nosave - should
+   * switch it back to data.init.
+   */
+  . = ALIGN(PAGE_SIZE);
+  PERCPU_VADDR(0, :percpu)
+#else
   PERCPU(PAGE_SIZE)
+#endif
 
   . = ALIGN(PAGE_SIZE);
   __init_end = .;
 
   . = ALIGN(PAGE_SIZE);
   __nosave_begin = .;
-  .data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) { *(.data.nosave) }
+  .data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) {
+      *(.data.nosave)
+  } :data.init  /* switch back to data.init, see PERCPU_VADDR() above */
   . = ALIGN(PAGE_SIZE);
   __nosave_end = .;
 
diff --git a/include/asm-generic/sections.h b/include/asm-generic/sections.h
index 79a7ff925bf8..4ce48e878530 100644
--- a/include/asm-generic/sections.h
+++ b/include/asm-generic/sections.h
@@ -9,7 +9,7 @@ extern char __bss_start[], __bss_stop[];
 extern char __init_begin[], __init_end[];
 extern char _sinittext[], _einittext[];
 extern char _end[];
-extern char __per_cpu_start[], __per_cpu_end[];
+extern char __per_cpu_load[], __per_cpu_start[], __per_cpu_end[];
 extern char __kprobes_text_start[], __kprobes_text_end[];
 extern char __initdata_begin[], __initdata_end[];
 extern char __start_rodata[], __end_rodata[];
diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index c61fab1dd2f8..fc2f55f2dcd6 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -430,12 +430,51 @@
         *(.initcall7.init)                                              \
         *(.initcall7s.init)
 
-#define PERCPU(align)                                                   \
-        . = ALIGN(align);                                               \
-        VMLINUX_SYMBOL(__per_cpu_start) = .;                            \
-        .data.percpu  : AT(ADDR(.data.percpu) - LOAD_OFFSET) {          \
+#define PERCPU_PROLOG(vaddr)                                            \
+        VMLINUX_SYMBOL(__per_cpu_load) = .;                             \
+        .data.percpu vaddr : AT(__per_cpu_load - LOAD_OFFSET) {         \
+                VMLINUX_SYMBOL(__per_cpu_start) = .;
+
+#define PERCPU_EPILOG(phdr)                                             \
+                VMLINUX_SYMBOL(__per_cpu_end) = .;                      \
+        } phdr                                                          \
+        . = __per_cpu_load + SIZEOF(.data.percpu);
+
+/**
+ * PERCPU_VADDR - define output section for percpu area
+ * @vaddr: explicit base address (optional)
+ * @phdr: destination PHDR (optional)
+ *
+ * Macro which expands to output section for percpu area.  If @vaddr
+ * is not blank, it specifies explicit base address and all percpu
+ * symbols will be offset from the given address.  If blank, @vaddr
+ * always equals @laddr + LOAD_OFFSET.
+ *
+ * @phdr defines the output PHDR to use if not blank.  Be warned that
+ * output PHDR is sticky.  If @phdr is specified, the next output
+ * section in the linker script will go there too.  @phdr should have
+ * a leading colon.
+ *
+ * This macro defines three symbols, __per_cpu_load, __per_cpu_start
+ * and __per_cpu_end.  The first one is the vaddr of loaded percpu
+ * init data.  __per_cpu_start equals @vaddr and __per_cpu_end is the
+ * end offset.
+ */
+#define PERCPU_VADDR(vaddr, phdr)                                       \
+        PERCPU_PROLOG(vaddr)                                            \
                 *(.data.percpu.page_aligned)                            \
                 *(.data.percpu)                                         \
                 *(.data.percpu.shared_aligned)                          \
-        }                                                               \
-        VMLINUX_SYMBOL(__per_cpu_end) = .;
+        PERCPU_EPILOG(phdr)
+
+/**
+ * PERCPU - define output section for percpu area, simple version
+ * @align: required alignment
+ *
+ * Align to @align and outputs output section for percpu area.  This
+ * macro doesn't maniuplate @vaddr or @phdr and __per_cpu_load and
+ * __per_cpu_start will be identical.
+ */
+#define PERCPU(align)                                                   \
+        . = ALIGN(align);                                               \
+        PERCPU_VADDR( , )