[PATCH 4/4] ARM: tegra: Set SCU base address dynamically from DT

[Hiroshi Doyu]

Rob Herring <robherring2 at gmail.com> wrote @ Tue, 18 Dec 2012 14:46:36 +0100:

> On 12/18/2012 03:21 AM, Hiroshi Doyu wrote:
> > Hi Rob,
> > 
> > Rob Herring <robherring2 at gmail.com> wrote @ Mon, 17 Dec 2012 15:00:46 +0100:
> > 
> >> On 12/17/2012 12:18 AM, Hiroshi Doyu wrote:
> >>> Set Snoop Control Unit(SCU) register base address dynamically from DT.
> >>>
> >>> Signed-off-by: Hiroshi Doyu <hdoyu at nvidia.com>
> >>> ---
> >>>  arch/arm/mach-tegra/platsmp.c |   23 ++++++++++++++++++++---
> >>>  1 file changed, 20 insertions(+), 3 deletions(-)
> >>>
> >>> diff --git a/arch/arm/mach-tegra/platsmp.c b/arch/arm/mach-tegra/platsmp.c
> >>> index 1b926df..45c0b79 100644
> >>> --- a/arch/arm/mach-tegra/platsmp.c
> >>> +++ b/arch/arm/mach-tegra/platsmp.c
> >>> @@ -18,6 +18,8 @@
> >>>  #include <linux/jiffies.h>
> >>>  #include <linux/smp.h>
> >>>  #include <linux/io.h>
> >>> +#include <linux/of.h>
> >>> +#include <linux/of_address.h>
> >>>  
> >>>  #include <asm/cacheflush.h>
> >>>  #include <asm/hardware/gic.h>
> >>> @@ -36,7 +38,7 @@
> >>>  
> >>>  extern void tegra_secondary_startup(void);
> >>>  
> >>> -static void __iomem *scu_base = IO_ADDRESS(TEGRA_ARM_PERIF_BASE);
> >>> +static void __iomem *scu_base;
> >>>  
> >>>  #define EVP_CPU_RESET_VECTOR \
> >>>  	(IO_ADDRESS(TEGRA_EXCEPTION_VECTORS_BASE) + 0x100)
> >>> @@ -143,14 +145,28 @@ done:
> >>>  	return status;
> >>>  }
> >>>  
> >>> +static const struct of_device_id cortex_a9_scu_match[] __initconst = {
> >>> +	{ .compatible = "arm,cortex-a9-scu", },
> >>> +	{}
> >>> +};
> >>> +
> >>>  /*
> >>>   * Initialise the CPU possible map early - this describes the CPUs
> >>>   * which may be present or become present in the system.
> >>>   */
> >>>  static void __init tegra_smp_init_cpus(void)
> >>>  {
> >>> -	unsigned int i, ncores = scu_get_core_count(scu_base);
> >>> +	struct device_node *np;
> >>> +	unsigned int i, ncores = 1;
> >>> +
> >>> +	np = of_find_matching_node(NULL, cortex_a9_scu_match);
> >>> +	if (!np)
> >>> +		return;
> >>> +	scu_base = of_iomap(np, 0);
> >>
> >> Did you actually test this? Unless something changed, ioremap does not
> >> work this early. The only reason to have it mapped this early is to get
> >> the core count, but that doesn't work on A15 or A7. So we really need to
> >> get core count/mask in a standard way. At least some work to get core
> >> count from DT went into 3.8.
> >>
> >> BTW, you can get the scu address on the A9 by reading cp15 register:
> >>
> >> 	/* Get SCU base */
> >> 	asm("mrc p15, 4, %0, c15, c0, 0" : "=r" (base));
> >>
> >> It's still probably good to have the DT node, but the reg property can
> >> be optional in this case.
> > 
> > I'm simply wondering, if the above cp15 works with Cortex-A9, do we
> > still need SCU DT node? At least from Cortex-A15 TRM, it seems that
> > SCU is tighly integrated into CPU core and it doesn't have any user
> > control. So Cortex-A15 doesn't seem to need to configure SCU. For
> > Cortex-A7, I haven't yet found S/W configurable register definitions
> > in TRM. So if neither of A15/A7 need SCU base, would the above cp15
> > intructions be enough?
> 
> The A15/A7 still have the register for the other peripherals like the
> GIC, but there are no SCU registers or other way to get a core count
> from the h/w.
> 
> The DT node could be used to determine if you have an SCU or not. I just
> used the cpu node compatible value to determine that.

Taking a look at A15 TRM again, it seems that A15 can get number of
processors via(*1):

  asm("mrc p15, 1, %0, c9, c0, 2\n" : "=r" (l2ctlr));

At least, A15 is ok without SCU/DT to get the number of processors. I
haven't found the above instruction in A7, but is there any way to get
the number of core in A7?

If both A15/A7 can get # of CPU cores via coprocessor instruction,
what others should we take care of with DT node?

*1: http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0438e/BABBACEE.html

> >> We need to move away from having the DT matching code within the
> >> platforms. This should all be moved to the scu code in a scu_of_init
> >> function that could be called from common code.
> > 
> > True if SCU DT node is still necessary.
> 
> Well, reading the cp15 register and mapping the registers could be
> common code independent of DT. I'm not sure if there are non-A9
> implementations of the SCU which don't have the cp15 register.