| Message ID | 1418179187-18486-2-git-send-email-chris.packham@alliedtelesis.co.nz (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wednesday 10 December 2014 15:39:44 Chris Packham wrote: > > +static void __init armada_370_xp_dt_fixup(void) > +{ > + smp_set_ops(smp_ops(armada_xp_smp_ops)); > +} > + > The dt_fixup callback pointer is meant to fix up a legacy dtb file in memory. I think this would be fairly easy in this case, just add in the missing enable-method property here to make the normal boot path work for old dtbs. Arnd
Hi Arnd, On 12/10/2014 10:44 PM, Arnd Bergmann wrote: > On Wednesday 10 December 2014 15:39:44 Chris Packham wrote: >> >> +static void __init armada_370_xp_dt_fixup(void) >> +{ >> + smp_set_ops(smp_ops(armada_xp_smp_ops)); >> +} >> + >> > > The dt_fixup callback pointer is meant to fix up a legacy dtb file in > memory. I think this would be fairly easy in this case, just add in the > missing enable-method property here to make the normal boot path > work for old dtbs. > > Arnd > I briefly explored that approach here[1]. The tricky part would be handling the fact that the enable method can be attached to either the /cpus node or and individual /cpu entry (or is that something I can ignore?). In the end I thought that the unconditional setting of smp_ops was easier to implement and would achieve the same result. -- [1] - http://lists.infradead.org/pipermail/linux-arm-kernel/2014-November/303465.html
diff --git a/arch/arm/mach-mvebu/board-v7.c b/arch/arm/mach-mvebu/board-v7.c index d0d39f1..ce23c6a 100644 --- a/arch/arm/mach-mvebu/board-v7.c +++ b/arch/arm/mach-mvebu/board-v7.c @@ -198,6 +198,11 @@ static void __init mvebu_dt_init(void) of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL); } +static void __init armada_370_xp_dt_fixup(void) +{ + smp_set_ops(smp_ops(armada_xp_smp_ops)); +} + static const char * const armada_370_xp_dt_compat[] = { "marvell,armada-370-xp", NULL, @@ -206,11 +211,11 @@ static const char * const armada_370_xp_dt_compat[] = { DT_MACHINE_START(ARMADA_370_XP_DT, "Marvell Armada 370/XP (Device Tree)") .l2c_aux_val = 0, .l2c_aux_mask = ~0, - .smp = smp_ops(armada_xp_smp_ops), .init_machine = mvebu_dt_init, .init_irq = mvebu_init_irq, .restart = mvebu_restart, .dt_compat = armada_370_xp_dt_compat, + .dt_fixup = armada_370_xp_dt_fixup, MACHINE_END static const char * const armada_375_dt_compat[] = {
We need to maintain backwards compatibility with device trees that don't define an enable method. At the same time we want the device tree to be able to specify an enable-method and have it stick. Previously by having smp assigned in the DT_MACHINE definition this would be picked up by setup_arch() and override whatever arm_dt_init_cpu_maps() had configured. Now we move the initial assignment of default smp_ops to a dt_fixup and let arm_dt_init_cpu_maps() override that if the device tree defines an enable-method. Signed-off-by: Chris Packham <chris.packham@alliedtelesis.co.nz> --- This didn't get much attention last time and is unchanged from v4. I'm pretty happy with this incarnation. It doesn't touch core code. It provides a fallback for old device trees and it achieves my original goal of allowing the device tree to configure the smp_ops via the enable-method property. v1: http://lists.infradead.org/pipermail/linux-arm-kernel/2014-November/300182.html v2: http://lists.infradead.org/pipermail/linux-arm-kernel/2014-November/300480.html v3: http://lists.infradead.org/pipermail/linux-arm-kernel/2014-November/302945.html v4: http://lists.infradead.org/pipermail/linux-arm-kernel/2014-November/303899.html arch/arm/mach-mvebu/board-v7.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)