openwrt/package/boot/uboot-lantiq/patches/0042-arx100-cgu-fixes.patch

From patchwork Tue Jan 20 11:28:45 2015
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Subject: [OpenWrt-Devel] uboot-lantiq cgu settings for ramboot image
From: Ben Mulvihill <ben.mulvihill@gmail.com>
X-Patchwork-Id: 431024
Message-Id: <1421753325.25187.58.camel@merveille.lan>
To: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Cc: OpenWrt Development List <openwrt-devel@lists.openwrt.org>
Date: Tue, 20 Jan 2015 12:28:45 +0100

On Tue, 2015-01-20 at 00:39 +0100, Ben Mulvihill wrote:
> On Mon, 2015-01-19 at 19:21 +0100, Ben Mulvihill wrote:
> > On Mon, 2015-01-19 at 16:47 +0100, Daniel Schwierzeck wrote:
> > > 2015-01-19 15:44 GMT+01:00 Ben Mulvihill <ben.mulvihill@gmail.com>:
> > > > On Mon, 2015-01-19 at 11:51 +0000, Conor O'Gorman wrote:
> > > >> On 19/01/15 10:46, Ben Mulvihill wrote:
> > > >> > Hello,
> > > >> >
> > > >> > I am trying to build uboot-lantiq for the BT Home Hub 3A (lantiq
> > > >> > ar9), and am wondering where to initialise the cgu, in the case
> > > >> > of a ramboot image for uart booting. Normally the cgu is initialised
> > > >> > in lowlevel_init, but that code is bypassed in ramboot images. The
> > > >> > result is that the board boots with the wrong cgu settings, which
> > > >> > sends the console haywire. So far I have tried two solutions:
> > > >>
> > > >> Another option is to try and not change anything. The console is already
> > > >> configured and running. The ram does need config.
> > > >>
> > > >> I was used to seeing the ramboot version running at half clock speed, at
> > > >> least on danube, previous to ar9.
> > > >>
> > > >> Conor
> > > >
> > > > Hi Conor,
> > > >
> > > > Thanks for the reply. But with the latest uboot-lantiq, not changing
> > > > anything means that I don't get a usable console. With an older
> > > > version I do at least get a uboot console, but no linux console when
> > > > I boot openwrt. Correcting the cgu settings solves both problems.
> > > >
> > > 
> > > could you try this?
> > > 
> > > diff --git a/arch/mips/cpu/mips32/arx100/cgu.c
> > > b/arch/mips/cpu/mips32/arx100/cgu.c
> > > index 6e71ee7..e0afbda 100644
> > > --- a/arch/mips/cpu/mips32/arx100/cgu.c
> > > +++ b/arch/mips/cpu/mips32/arx100/cgu.c
> > > @@ -95,15 +95,5 @@ unsigned long ltq_get_cpu_clock(void)
> > > 
> > >  unsigned long ltq_get_bus_clock(void)
> > >  {
> > > -       u32 fpi_sel;
> > > -       unsigned long clk;
> > > -
> > > -       fpi_sel = ltq_cgu_sys_readl(1, CGU_SYS_FPI_SEL);
> > > -
> > > -       if (fpi_sel)
> > > -               clk = ltq_get_io_region_clock() / 2;
> > > -       else
> > > -               clk = ltq_get_io_region_clock();
> > > -
> > > -       return clk;
> > > +       return ltq_get_io_region_clock();
> > >  }
> > > 
> > > the UART driver calculates the baudrate from the FPI bus clock, but
> > > FPI_SEL is not available on AR9. FPI bus clock is always the same as
> > > DDR clock, Obviously a copy&paste error from VR9 code ;)
> > > 
> > 
> > No, even with this patch, I still don't get a working console I'm
> > afraid. If I don't set anything explicitly, the board comes up with
> > CGU_SYS set to 0x05, ie CGU_SYS_SYSSEL_PLL0_333_MHZ |
> > CGU_SYS_CPUSEL_EQUAL_DDRCLK | CGU_SYS_DDRSEL_THIRD_SYSCLK.
> > Is this a valid combination without CGU_SYS_PPESEL_250_MHZ ?
> > I don't understand what CGU_SYS_PPESEL_250_MHZ does?
> > The "right setting", as set by the stock uboot, is 0x80.
> 
> P.S. There also seems to be a discrepancy between the uboot and
> linux code. I take it from what you say above that fpi clock, ddr
> clock and io region clock are all the same. Now if the least 
> significant bit of CGU_SYS is set, then according to the uboot
> code - function ltq_get_bus_clock() - their value is one
> third of the system clock. But according to the linux code
> - function ltq_ar9_fpi_hz() in arch/mips/lantiq/xway/clk.c -
> their value in this case is equal to the system clock.
> 
> Or am I getting muddled? It's past my bedtime!
> 
> 

Some of the bitshifting in arch/mips/cpu/mips32/arx100/cgu.c is 1
out. A patch along these lines should fix it:

--- a/arch/mips/cpu/mips32/arx100/cgu.c	2015-01-20 11:57:22.000000000 +0100
+++ b/arch/mips/cpu/mips32/arx100/cgu.c	2015-01-20 12:00:15.000000000 +0100
@@ -10,12 +10,17 @@
 #include <asm/lantiq/clk.h>
 #include <asm/lantiq/io.h>
 
-#define CGU_SYS_DDR_SEL		(1 << 0)
-#define CGU_SYS_CPU_SEL		(1 << 2)
+#define CGU_SYS_DDR_SHIFT	0
+#define CGU_SYS_CPU_SHIFT	2
 #define CGU_SYS_SYS_SHIFT	3
+#define CGU_SYS_FPI_SHIFT	6
+#define CGU_SYS_PPE_SHIFT	7
+
+#define CGU_SYS_DDR_MASK	(1 << CGU_SYS_DDR_SHIFT)
+#define CGU_SYS_CPU_MASK	(1 << CGU_SYS_CPU_SHIFT)
 #define CGU_SYS_SYS_MASK	(0x3 << CGU_SYS_SYS_SHIFT)
-#define CGU_SYS_FPI_SEL		(1 << 6)
-#define CGU_SYS_PPE_SEL		(1 << 7)
+#define CGU_SYS_FPI_MASK	(1 << CGU_SYS_FPI_SHIFT)
+#define CGU_SYS_PPE_MASK	(1 << CGU_SYS_PPE_SHIFT)
 
 struct ltq_cgu_regs {
 	u32	rsvd0;
@@ -68,7 +73,7 @@ unsigned long ltq_get_io_region_clock(vo
 	u32 ddr_sel;
 	unsigned long clk;
 
-	ddr_sel = ltq_cgu_sys_readl(1, CGU_SYS_DDR_SEL);
+	ddr_sel = ltq_cgu_sys_readl(CGU_SYS_DDR_MASK, CGU_SYS_DDR_SHIFT);
 
 	if (ddr_sel)
 		clk = ltq_get_system_clock() / 3;
@@ -83,7 +88,7 @@ unsigned long ltq_get_cpu_clock(void)
 	u32 cpu_sel;
 	unsigned long clk;
 
-	cpu_sel = ltq_cgu_sys_readl(1, CGU_SYS_CPU_SEL);
+	cpu_sel = ltq_cgu_sys_readl(CGU_SYS_CPU_MASK, CGU_SYS_CPU_SHIFT);
 
 	if (cpu_sel)
 		clk = ltq_get_io_region_clock();
@@ -98,7 +103,7 @@ unsigned long ltq_get_bus_clock(void)
 	u32 fpi_sel;
 	unsigned long clk;
 
-	fpi_sel = ltq_cgu_sys_readl(1, CGU_SYS_FPI_SEL);
+	fpi_sel = ltq_cgu_sys_readl(CGU_SYS_FPI_MASK, CGU_SYS_FPI_SHIFT);
 
 	if (fpi_sel)
 		clk = ltq_get_io_region_clock() / 2;