[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, 25 insertions, 0 deletions

diff --git a/arch/ppc64/kernel/xics.c b/arch/ppc64/kernel/xics.c
index 879f39b90a3..677c4450984 100644
--- a/arch/ppc64/kernel/xics.c
+++ b/arch/ppc64/kernel/xics.c
@@ -647,6 +647,31 @@ static void xics_set_affinity(unsigned int virq, cpumask_t cpumask)
 	}
 }
 
+void xics_teardown_cpu(void)
+{
+	int cpu = smp_processor_id();
+	int status;
+
+	ops->cppr_info(cpu, 0x00);
+	iosync();
+
+	/*
+	 * we need to EOI the IPI if we got here from kexec down IPI
+	 *
+	 * xics doesn't care if we duplicate an EOI as long as we
+	 * don't EOI and raise priority.
+	 *
+	 * probably need to check all the other interrupts too
+	 * should we be flagging idle loop instead?
+	 * or creating some task to be scheduled?
+	 */
+	ops->xirr_info_set(cpu, XICS_IPI);
+
+	status = rtas_set_indicator(GLOBAL_INTERRUPT_QUEUE,
+		(1UL << interrupt_server_size) - 1 - default_distrib_server, 0);
+	WARN_ON(status != 0);
+}
+
 #ifdef CONFIG_HOTPLUG_CPU
 
 /* Interrupts are disabled. */