[U-Boot-Users] [PATCH v2] MPC85xx, MPC83xx: Add/Fix UPM configuration support
Add support for UPM configuration on the 85xx platform. In addition, on the MPC83xx, remove MPC834x precompiler condition, in order to support all MPC83xx processors.
Signed-off-by: David Saada david.saada@ecitele.com
--- a/include/mpc85xx.h 2008-02-07 12:11:35.873966000 +0200 +++ b/include/mpc85xx.h 2008-02-07 13:13:34.255165000 +0200 @@ -35,7 +35,10 @@ #define BRx_MS_UPMA 0x00000080 /* U.P.M.A Machine Select */ #define BRx_MS_UPMB 0x000000a0 /* U.P.M.B Machine Select */ #define BRx_MS_UPMC 0x000000c0 /* U.P.M.C Machine Select */ +#define BRx_MS_MSK 0x000000e0 /* Machine select mask */ +#define BRx_MS_SHIFT 5 /* Machine select shift */ #define BRx_PS_8 0x00000800 /* 8 bit port size */ +#define BRx_PS_16 0x00001000 /* 16 bit port size */ #define BRx_PS_32 0x00001800 /* 32 bit port size */ #define BRx_BA_MSK 0xffff8000 /* Base Address Mask */
@@ -53,8 +56,10 @@ #define ORxU_AM_MSK 0xffff8000 /* Address Mask Mask */
#define MxMR_OP_NORM 0x00000000 /* Normal Operation */ -#define MxMR_DSx_2_CYCL 0x00400000 /* 2 cycle Disable Period */ #define MxMR_OP_WARR 0x10000000 /* Write to Array */ +#define MxMR_DSx_2_CYCL 0x00400000 /* 2 cycle Disable Period */ +#define MxMR_DSx_3_CYCL 0x00800000 /* 3 cycle Disable Period */ +#define MxMR_GPL4_LPWT 0x00040000 /* LGPL4 Line in LUPWAIT mode */ #define MxMR_BSEL 0x80000000 /* Bus Select */
/* helpers to convert values into an OR address mask (GPCM mode) */ --- a/cpu/mpc83xx/cpu.c 2008-02-07 12:11:27.569737000 +0200 +++ b/cpu/mpc83xx/cpu.c 2008-02-07 13:14:44.975179000 +0200 @@ -227,7 +227,6 @@ int checkcpu(void) */ void upmconfig (uint upm, uint *table, uint size) { -#if defined(CONFIG_MPC834X) volatile immap_t *immap = (immap_t *) CFG_IMMR; volatile lbus83xx_t *lbus = &immap->lbus; volatile uchar *dummy = NULL; @@ -260,10 +259,6 @@ void upmconfig (uint upm, uint *table, u
/* Set the OP field in the MxMR to "normal" and the MAD field to 000000 */ *mxmr &= 0xCFFFFFC0; -#else - printf("Error: %s() not defined for this configuration.\n", __FUNCTION__); - hang(); -#endif }
--- a/cpu/mpc85xx/cpu.c 2008-02-07 12:11:27.616748000 +0200 +++ b/cpu/mpc85xx/cpu.c 2008-02-07 13:30:38.359554000 +0200 @@ -29,6 +29,7 @@ #include <watchdog.h> #include <command.h> #include <asm/cache.h> +#include "asm/immap_85xx.h"
int checkcpu (void) { @@ -263,3 +264,68 @@ int dma_xfer(void *dest, uint count, voi return dma_check(); } #endif + +/* + * Program a UPM with the code supplied in the table. + * + * The 'dummy' variable is used to increment the MAD. 'dummy' is + * supposed to be a pointer to the memory of the device being + * programmed by the UPM. The data in the MDR is written into + * memory and the MAD is incremented every time there's a read + * from 'dummy'. Unfortunately, the current prototype for this + * function doesn't allow for passing the address of this + * device, and changing the prototype will break a number lots + * of other code, so we need to use a round-about way of finding + * the value for 'dummy'. + * + * The value can be extracted from the base address bits of the + * Base Register (BR) associated with the specific UPM. To find + * that BR, we need to scan all BRs until we find the one that + * has its MSEL bits matching the UPM we want. Once we know the + * right BR, we can extract the base address bits from it. + * + * The MxMR and the BR and OR of the chosen bank should all be + * configured before calling this function. + * + * Parameters: + * upm: 0=UPMA, 1=UPMB, 2=UPMC + * table: Pointer to an array of values to program + * size: Number of elements in the array. Must be 64 or less. + */ +void upmconfig (uint upm, uint *table, uint size) +{ + volatile ccsr_lbc_t *lbc = (void *)(CFG_MPC85xx_LBC_ADDR); + volatile uchar *dummy = NULL; + const u32 msel = (upm + 4) << BRx_MS_SHIFT; /* What the MSEL field in BRn should be */ + volatile uint *mxmr = &lbc->mamr + upm; /* Pointer to mamr, mbmr, or mcmr */ + volatile uint *brx = &lbc->br0; /* Pointer to BRx */ + uint i; + + /* Scan all the banks to determine the base address of the device */ + for (i = 0; i < 8; i++) { + if ((*brx & BRx_MS_MSK) == msel) { + dummy = (uchar *) (*brx & BRx_BA_MSK); + break; + } + brx += 2; /* Skip to next BRx */ + } + + if (!dummy) { + printf("Error: %s() could not find matching BR\n", __FUNCTION__); + hang(); + } + + /* Set the OP field in the MxMR to "write" and the MAD field to 000000 */ + *mxmr = (*mxmr & 0xCFFFFFC0) | MxMR_OP_WARR; + + for (i = 0; i < size; i++) { + lbc->mdr = table[i]; + __asm__ __volatile__ ("sync"); + *dummy; /* Write the value to memory and increment MAD */ + __asm__ __volatile__ ("sync"); + } + + /* Set the OP field in the MxMR to "normal" and the MAD field to 000000 */ + *mxmr &= 0xCFFFFFC0; +} +
On Thu, 7 Feb 2008 08:53:46 -0800 (PST) David Saada David.Saada@ecitele.com wrote:
Add support for UPM configuration on the 85xx platform. In addition, on the MPC83xx, remove MPC834x precompiler condition, in order to support all MPC83xx processors.
Signed-off-by: David Saada david.saada@ecitele.com
ack on the 83xx change, but the 85xx code looks identical to the 83xx version, and 86xx bits should be identical, too, so we would be better off refraining from triplicating code again and find a more common place to put this. Jon? Andy?
btw, David, can you start including diffstats with your patch submissions? It makes it easier to see what's being touched at a glance, especially on the larger patches. You can simply use diffstat if you're not ready to use git-format-patch (which does this automatically for you).
Kim
ack on the 83xx change, but the 85xx code looks identical to the 83xx version, and 86xx bits should be identical, too, so we would be better off refraining from triplicating code again and find a more common place to put this. Jon? Andy?
Definitely true, but that's a delicate change. For instance, this UPM patch has the same logics for the 83xx & 85xx platforms, but the registers are different. Unifying this code should result in lots of ugly ifdefs.
btw, David, can you start including diffstats with your patch submissions? It makes it easier to see what's being touched at a glance, especially on the larger patches. You can simply use diffstat if you're not ready to use git-format-patch (which does this automatically for you).
Sure thing - will do.
David.
On Sun, 10 Feb 2008 15:14:45 +0200 David Saada David.Saada@ecitele.com wrote:
ack on the 83xx change, but the 85xx code looks identical to the 83xx version, and 86xx bits should be identical, too, so we would be better off refraining from triplicating code again and find a more common place to put this. Jon? Andy?
Definitely true, but that's a delicate change. For instance, this UPM patch has the same logics for the 83xx & 85xx platforms, but the registers are different. Unifying this code should result in lots of ugly ifdefs.
which registers are different?
Kim
Definitely true, but that's a delicate change. For instance, this
UPM
patch has the same logics for the 83xx & 85xx platforms, but the registers are different. Unifying this code should result in lots of ugly ifdefs.
which registers are different?
Well, not quite the register structures, but more the way they are accessed. For instance, the BRx registers are accessed differently in the 83xx (via the lbus array) comparing to the 85xx (via lbc). In the ioports code, for instance, the code has the same logics for both platforms, but the registers are accessed via different structures, and even have different names (dir1 vs. cpdir1 for instance). I think this kind of change would also require changes in the register definition files, if we want the unified code to be clear.
David.
participants (2)
-
David Saada -
Kim Phillips