Message ID | 1351391173-32379-1-git-send-email-horms@verge.net.au |
---|---|
State | New |
Headers | show |
Dear Simon Horman, On Sun, 28 Oct 2012 11:26:13 +0900, Simon Horman wrote: > -MACHINE_START(MACKEREL, "mackerel") > +static const char *mackerel_boards_compat_dt[] __initdata = { > + "renesas,mackerel", > + NULL, > +}; > + > +DT_MACHINE_START(MACKEREL_DT, "mackerel") > .map_io = sh7372_map_io, > .init_early = sh7372_add_early_devices, > .init_irq = sh7372_init_irq, > @@ -1659,4 +1664,5 @@ MACHINE_START(MACKEREL, "mackerel") > .init_machine = mackerel_init, > .init_late = sh7372_pm_init_late, > .timer = &shmobile_timer, > + .dt_compat = mackerel_boards_compat_dt, > MACHINE_END Not sure what is the status of mach-shmobile with regard to the device tree, but one of the idea of the device tree is to avoid having one .c file per board, and therefore one DT_MACHINE_START definition per board. There should be only only DT_MACHINE_START definition per SoC or SoC family, with the DT giving the detailed hardware description. Of course, during a transition period, not all your drivers may have DT bindings, and therefore some devices may need to be probed in the "old" way (manual registration of platform_device). In that case, your ->init_machine() hook can call initialization function on a per-board basis by testing if the machine is such or such board. See arch/arm/mach-kirkwood/board-dt.c for an example. The advantage is that with this solution, the board specific C files are progressively reducing in size as more and more devices get registered through the Device Tree, until the point where they become empty and can be removed. Best regards, Thomas
On Sun, Oct 28, 2012 at 10:37:11AM +0100, Thomas Petazzoni wrote: > Dear Simon Horman, > > On Sun, 28 Oct 2012 11:26:13 +0900, Simon Horman wrote: > > > -MACHINE_START(MACKEREL, "mackerel") > > +static const char *mackerel_boards_compat_dt[] __initdata = { > > + "renesas,mackerel", > > + NULL, > > +}; > > + > > +DT_MACHINE_START(MACKEREL_DT, "mackerel") > > .map_io = sh7372_map_io, > > .init_early = sh7372_add_early_devices, > > .init_irq = sh7372_init_irq, > > @@ -1659,4 +1664,5 @@ MACHINE_START(MACKEREL, "mackerel") > > .init_machine = mackerel_init, > > .init_late = sh7372_pm_init_late, > > .timer = &shmobile_timer, > > + .dt_compat = mackerel_boards_compat_dt, > > MACHINE_END > > Not sure what is the status of mach-shmobile with regard to the device > tree, but one of the idea of the device tree is to avoid having one .c > file per board, and therefore one DT_MACHINE_START definition per > board. There should be only only DT_MACHINE_START definition per SoC or > SoC family, with the DT giving the detailed hardware description. > > Of course, during a transition period, not all your drivers may have DT > bindings, and therefore some devices may need to be probed in the "old" > way (manual registration of platform_device). In that case, your > ->init_machine() hook can call initialization function on a per-board > basis by testing if the machine is such or such board. > > See arch/arm/mach-kirkwood/board-dt.c for an example. The advantage is > that with this solution, the board specific C files are progressively > reducing in size as more and more devices get registered through the > Device Tree, until the point where they become empty and can be removed. Hi Thomas, shmobile is very much in the transition phase and we are, patch by patch, working towards full(er) DT support for boards, SoCs and drivers. I am not sure that now is the right time to jump to a approach similar to that taken by mach-kirkwood/board-dt.c.
Simon, On Sun, 28 Oct 2012 19:44:38 +0900, Simon Horman wrote: > shmobile is very much in the transition phase and we are, patch by patch, > working towards full(er) DT support for boards, SoCs and drivers. I am not > sure that now is the right time to jump to a approach similar to that taken > by mach-kirkwood/board-dt.c. Ok, I understand that for transition needs, each SoC family may need a different method. My comment was more a general comment to make sure that the right final direction is being taken. For sure the intermediate steps may be different. Best regards, Thomas
Hello. On 28-10-2012 6:26, Simon Horman wrote: > From: Nobuhiro Iwamatsu <nobuhiro.iwamatsu.yj@renesas.com> > This patch enables PMU for r8a7740. And enables performance events on the Armadillo board. Perhaps the defconfig part was worth a separate patch... > Signed-off-by: Nobuhiro Iwamatsu <nobuhiro.iwamatsu.yj@renesas.com> > Signed-off-by: Simon Horman <horms@verge.net.au> [...] > diff --git a/arch/arm/mach-shmobile/setup-r8a7740.c b/arch/arm/mach-shmobile/setup-r8a7740.c > index 11bb1d9..1e032cb 100644 > --- a/arch/arm/mach-shmobile/setup-r8a7740.c > +++ b/arch/arm/mach-shmobile/setup-r8a7740.c > @@ -590,6 +590,21 @@ static struct platform_device i2c1_device = { > .num_resources = ARRAY_SIZE(i2c1_resources), > }; > > +static struct resource pmu_resources[] = { > + [0] = { > + .start = evt2irq(0x19a0), > + .end = evt2irq(0x19a0), > + .flags = IORESOURCE_IRQ, > + }, > +}; > + > +static struct platform_device pmu_device = { > + .name = "arm-pmu", > + .id = -1, I think you should either align '=' with tabs, or not bother doing this, not both. :-) > + .num_resources = ARRAY_SIZE(pmu_resources), > + .resource = pmu_resources, > +}; > + > static struct platform_device *r8a7740_late_devices[] __initdata = { > &i2c0_device, > &i2c1_device, WBR, Sergei
On Sun, Oct 28, 2012 at 03:28:47PM +0400, Sergei Shtylyov wrote: > Hello. > > On 28-10-2012 6:26, Simon Horman wrote: > > >From: Nobuhiro Iwamatsu <nobuhiro.iwamatsu.yj@renesas.com> > > >This patch enables PMU for r8a7740. > > And enables performance events on the Armadillo board. Perhaps > the defconfig part was worth a separate patch... As I understand things, the Armadillo is currently the only board in-tree that uses the r8a7740 SoC and it seems reasonable to enable performance events to activate the code that comprises the rest of the patch. So while I don't feel strongly about this I do have a slight preference for the current single-patch approach. > >Signed-off-by: Nobuhiro Iwamatsu <nobuhiro.iwamatsu.yj@renesas.com> > >Signed-off-by: Simon Horman <horms@verge.net.au> > [...] > > >diff --git a/arch/arm/mach-shmobile/setup-r8a7740.c b/arch/arm/mach-shmobile/setup-r8a7740.c > >index 11bb1d9..1e032cb 100644 > >--- a/arch/arm/mach-shmobile/setup-r8a7740.c > >+++ b/arch/arm/mach-shmobile/setup-r8a7740.c > >@@ -590,6 +590,21 @@ static struct platform_device i2c1_device = { > > .num_resources = ARRAY_SIZE(i2c1_resources), > > }; > > > >+static struct resource pmu_resources[] = { > >+ [0] = { > >+ .start = evt2irq(0x19a0), > >+ .end = evt2irq(0x19a0), > >+ .flags = IORESOURCE_IRQ, > >+ }, > >+}; > >+ > >+static struct platform_device pmu_device = { > >+ .name = "arm-pmu", > >+ .id = -1, > > I think you should either align '=' with tabs, or not bother > doing this, not both. :-) Thanks, there seems to be an extra tab there. I'll roll this into an updated patch and series once we come to a consensus about the defconfig change. > >+ .num_resources = ARRAY_SIZE(pmu_resources), > >+ .resource = pmu_resources, > >+}; > >+ > > static struct platform_device *r8a7740_late_devices[] __initdata = { > > &i2c0_device, > > &i2c1_device, > > WBR, Sergei >