[PATCH] ppc64: kexec support for ppc64

This patch implements the kexec support for ppc64 platforms.

A couple of notes:

1)  We copy the pages in virtual mode, using the full base kernel
    and a statically allocated stack.   At kexec_prepare time we
    scan the pages and if any overlap our (0, _end[]) range we
    return -ETXTBSY.

    On PowerPC 64 systems running in LPAR (logical partitioning)
    mode, only a small region of memory, referred to as the RMO,
    can be accessed in real mode.  Since Linux runs with only one
    zone of memory in the memory allocator, and it can be orders of
    magnitude more memory than the RMO, looping until we allocate
    pages in the source region is not feasible.  Copying in virtual
    means we don't have to write a hash table generation and call
    hypervisor to insert translations, instead we rely on the pinned
    kernel linear mapping.  The kernel already has move to linked
    location built in, so there is no requirement to load it at 0.

    If we want to load something other than a kernel, then a stub
    can be written to copy a linear chunk in real mode.

2)  The start entry point gets passed parameters from the kernel.
    Slaves are started at a fixed address after copying code from
    the entry point.

    All CPUs get passed their firmware assigned physical id in r3
    (most calling conventions use this register for the first
    argument).

    This is used to distinguish each CPU from all other CPUs.
    Since firmware is not around, there is no other way to obtain
    this information other than to pass it somewhere.

    A single CPU, referred to here as the master and the one executing
    the kexec call, branches to start with the address of start in r4.
    While this can be calculated, we have to load it through a gpr to
    branch to this point so defining the register this is contained
    in is free.  A stack of unspecified size is available at r1
    (also common calling convention).

    All remaining running CPUs are sent to start at absolute address
    0x60 after copying the first 0x100 bytes from start to address 0.
    This convention was chosen because it matches what the kernel
    has been doing itself.  (only gpr3 is defined).

    Note: This is not quite the convention of the kexec bootblock v2
    in the kernel.  A stub has been written to convert between them,
    and we may adjust the kernel in the future to allow this directly
    without any stub.

3)  Destination pages can be placed anywhere, even where they
    would not be accessible in real mode.  This will allow us to
    place ram disks above the RMO if we choose.

Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: R Sharada <sharada@in.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

1 files changed, 173 insertions, 2 deletions

diff --git a/arch/ppc64/kernel/misc.S b/arch/ppc64/kernel/misc.S
index e3c73b3425dc..f3dea0c5a88c 100644
--- a/arch/ppc64/kernel/misc.S
+++ b/arch/ppc64/kernel/misc.S
@@ -680,6 +680,177 @@ _GLOBAL(kernel_thread)
         ld      r30,-16(r1)
         blr
 
+/* kexec_wait(phys_cpu)
+ *
+ * wait for the flag to change, indicating this kernel is going away but
+ * the slave code for the next one is at addresses 0 to 100.
+ *
+ * This is used by all slaves.
+ *
+ * Physical (hardware) cpu id should be in r3.
+ */
+_GLOBAL(kexec_wait)
+        bl      1f
+1:      mflr    r5
+        addi    r5,r5,kexec_flag-1b
+
+99:     HMT_LOW
+#ifdef CONFIG_KEXEC             /* use no memory without kexec */
+        lwz     r4,0(r5)
+        cmpwi   0,r4,0
+        bnea    0x60
+#endif
+        b       99b
+
+/* this can be in text because we won't change it until we are
+ * running in real anyways
+ */
+kexec_flag:
+        .long   0
+
+
+#ifdef CONFIG_KEXEC
+
+/* kexec_smp_wait(void)
+ *
+ * call with interrupts off
+ * note: this is a terminal routine, it does not save lr
+ *
+ * get phys id from paca
+ * set paca id to -1 to say we got here
+ * switch to real mode
+ * join other cpus in kexec_wait(phys_id)
+ */
+_GLOBAL(kexec_smp_wait)
+        lhz     r3,PACAHWCPUID(r13)
+        li      r4,-1
+        sth     r4,PACAHWCPUID(r13)     /* let others know we left */
+        bl      real_mode
+        b       .kexec_wait
+
+/*
+ * switch to real mode (turn mmu off)
+ * we use the early kernel trick that the hardware ignores bits
+ * 0 and 1 (big endian) of the effective address in real mode
+ *
+ * don't overwrite r3 here, it is live for kexec_wait above.
+ */
+real_mode:      /* assume normal blr return */
+1:      li      r9,MSR_RI
+        li      r10,MSR_DR|MSR_IR
+        mflr    r11             /* return address to SRR0 */
+        mfmsr   r12
+        andc    r9,r12,r9
+        andc    r10,r12,r10
+
+        mtmsrd  r9,1
+        mtspr   SPRN_SRR1,r10
+        mtspr   SPRN_SRR0,r11
+        rfid
+
+
+/*
+ * kexec_sequence(newstack, start, image, control, clear_all())
+ *
+ * does the grungy work with stack switching and real mode switches
+ * also does simple calls to other code
+ */
+
+_GLOBAL(kexec_sequence)
+        mflr    r0
+        std     r0,16(r1)
+
+        /* switch stacks to newstack -- &kexec_stack.stack */
+        stdu    r1,THREAD_SIZE-112(r3)
+        mr      r1,r3
+
+        li      r0,0
+        std     r0,16(r1)
+
+        /* save regs for local vars on new stack.
+         * yes, we won't go back, but ...
+         */
+        std     r31,-8(r1)
+        std     r30,-16(r1)
+        std     r29,-24(r1)
+        std     r28,-32(r1)
+        std     r27,-40(r1)
+        std     r26,-48(r1)
+        std     r25,-56(r1)
+
+        stdu    r1,-112-64(r1)
+
+        /* save args into preserved regs */
+        mr      r31,r3                  /* newstack (both) */
+        mr      r30,r4                  /* start (real) */
+        mr      r29,r5                  /* image (virt) */
+        mr      r28,r6                  /* control, unused */
+        mr      r27,r7                  /* clear_all() fn desc */
+        mr      r26,r8                  /* spare */
+        lhz     r25,PACAHWCPUID(r13)    /* get our phys cpu from paca */
+
+        /* disable interrupts, we are overwriting kernel data next */
+        mfmsr   r3
+        rlwinm  r3,r3,0,17,15
+        mtmsrd  r3,1
+
+        /* copy dest pages, flush whole dest image */
+        mr      r3,r29
+        bl      .kexec_copy_flush       /* (image) */
+
+        /* turn off mmu */
+        bl      real_mode
+
+        /* clear out hardware hash page table and tlb */
+        ld      r5,0(r27)               /* deref function descriptor */
+        mtctr   r5
+        bctrl                           /* ppc_md.hash_clear_all(void); */
+
+/*
+ *   kexec image calling is:
+ *      the first 0x100 bytes of the entry point are copied to 0
+ *
+ *      all slaves branch to slave = 0x60 (absolute)
+ *              slave(phys_cpu_id);
+ *
+ *      master goes to start = entry point
+ *              start(phys_cpu_id, start, 0);
+ *
+ *
+ *   a wrapper is needed to call existing kernels, here is an approximate
+ *   description of one method:
+ *
+ * v2: (2.6.10)
+ *   start will be near the boot_block (maybe 0x100 bytes before it?)
+ *   it will have a 0x60, which will b to boot_block, where it will wait
+ *   and 0 will store phys into struct boot-block and load r3 from there,
+ *   copy kernel 0-0x100 and tell slaves to back down to 0x60 again
+ *
+ * v1: (2.6.9)
+ *    boot block will have all cpus scanning device tree to see if they
+ *    are the boot cpu ?????
+ *    other device tree differences (prop sizes, va vs pa, etc)...
+ */
+
+        /* copy  0x100 bytes starting at start to 0 */
+        li      r3,0
+        mr      r4,r30
+        li      r5,0x100
+        li      r6,0
+        bl      .copy_and_flush /* (dest, src, copy limit, start offset) */
+1:      /* assume normal blr return */
+
+        /* release other cpus to the new kernel secondary start at 0x60 */
+        mflr    r5
+        li      r6,1
+        stw     r6,kexec_flag-1b(5)
+        mr      r3,r25  # my phys cpu
+        mr      r4,r30  # start, aka phys mem offset
+        mtlr    4
+        li      r5,0
+        blr     /* image->start(physid, image->start, 0); */
+#endif /* CONFIG_KEXEC */
+
 /* Why isn't this a) automatic, b) written in 'C'? */   
         .balign 8
 _GLOBAL(sys_call_table32)
@@ -951,7 +1122,7 @@ _GLOBAL(sys_call_table32)
         .llong .compat_sys_mq_timedreceive /* 265 */
         .llong .compat_sys_mq_notify
         .llong .compat_sys_mq_getsetattr
-        .llong .sys_ni_syscall          /* 268 reserved for sys_kexec_load */
+        .llong .compat_sys_kexec_load
         .llong .sys32_add_key
         .llong .sys32_request_key
         .llong .compat_sys_keyctl
@@ -1227,7 +1398,7 @@ _GLOBAL(sys_call_table)
         .llong .sys_mq_timedreceive     /* 265 */
         .llong .sys_mq_notify
         .llong .sys_mq_getsetattr
-        .llong .sys_ni_syscall          /* 268 reserved for sys_kexec_load */
+        .llong .sys_kexec_load
         .llong .sys_add_key
         .llong .sys_request_key         /* 270 */
         .llong .sys_keyctl