u-boot-brain/arch/arm/mach-uniphier/dram/umc-proxstream2.c
Masahiro Yamada 94b756f331 ARM: uniphier: refactor UMC init code for ProXstream2
Currently, a dummy value is defined for the UMC_SPCCTLA register
when the DRAM size is zero.  This seems weird because the controller
does not need setting in the first place if the size is zero.

Also, redefine enum dram_size to represent the DRAM size per 16-bit
unit.  This makes things simpler because the channel 0 and 1 are
connected with 32-bit width DRAM, while the channel 2 is connected
with 16-bit width one.

I am renaming SIZE_* into DRAM_SZ_* (and also FREQ_* to DRAM_FREQ_*
for consistency) while I am here because SIZE_* might be easily
mixed-up with the macros in include/linux/sizes.h.

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
2016-02-29 03:50:16 +09:00

639 lines
16 KiB
C

/*
* Copyright (C) 2015 Masahiro Yamada <yamada.masahiro@socionext.com>
*
* based on commit 21b6e480f92ccc38fe0502e3116411d6509d3bf2 of Diag by:
* Copyright (C) 2015 Socionext Inc.
*
* SPDX-License-Identifier: GPL-2.0+
*/
#include <common.h>
#include <linux/err.h>
#include <linux/io.h>
#include <linux/sizes.h>
#include <asm/processor.h>
#include "../init.h"
#include "../soc-info.h"
#include "ddrmphy-regs.h"
#include "umc-regs.h"
#define DRAM_CH_NR 3
enum dram_freq {
DRAM_FREQ_1866M,
DRAM_FREQ_2133M,
DRAM_FREQ_NR,
};
enum dram_size {
DRAM_SZ_256M,
DRAM_SZ_512M,
DRAM_SZ_NR,
};
static u32 ddrphy_pgcr2[DRAM_FREQ_NR] = {0x00FC7E5D, 0x00FC90AB};
static u32 ddrphy_ptr0[DRAM_FREQ_NR] = {0x0EA09205, 0x10C0A6C6};
static u32 ddrphy_ptr1[DRAM_FREQ_NR] = {0x0DAC041B, 0x0FA104B1};
static u32 ddrphy_ptr3[DRAM_FREQ_NR] = {0x15171e45, 0x18182357};
static u32 ddrphy_ptr4[DRAM_FREQ_NR] = {0x0e9ad8e9, 0x10b34157};
static u32 ddrphy_dtpr0[DRAM_FREQ_NR] = {0x35a00d88, 0x39e40e88};
static u32 ddrphy_dtpr1[DRAM_FREQ_NR] = {0x2288cc2c, 0x228a04d0};
static u32 ddrphy_dtpr2[DRAM_FREQ_NR] = {0x50005e00, 0x50006a00};
static u32 ddrphy_dtpr3[DRAM_FREQ_NR] = {0x0010cb49, 0x0010ec89};
static u32 ddrphy_mr0[DRAM_FREQ_NR] = {0x00000115, 0x00000125};
static u32 ddrphy_mr2[DRAM_FREQ_NR] = {0x000002a0, 0x000002a8};
/* dependent on package and board design */
static u32 ddrphy_acbdlr0[DRAM_CH_NR] = {0x0000000c, 0x0000000c, 0x00000009};
static u32 umc_cmdctla[DRAM_FREQ_NR] = {0x66DD131D, 0x77EE1722};
/*
* The ch2 is a different generation UMC core.
* The register spec is different, unfortunately.
*/
static u32 umc_cmdctlb_ch01[DRAM_FREQ_NR] = {0x13E87C44, 0x18F88C44};
static u32 umc_cmdctlb_ch2[DRAM_FREQ_NR] = {0x19E8DC44, 0x1EF8EC44};
static u32 umc_spcctla[DRAM_FREQ_NR][DRAM_SZ_NR] = {
{0x004A071D, 0x0078071D},
{0x0055081E, 0x0089081E},
};
static u32 umc_spcctlb[] = {0x00FF000A, 0x00FF000B};
/* The ch2 is different for some reason only hardware guys know... */
static u32 umc_flowctla_ch01[] = {0x0800001E, 0x08000022};
static u32 umc_flowctla_ch2[] = {0x0800001E, 0x0800001E};
/* DDR multiPHY */
static inline int ddrphy_get_rank(int dx)
{
return dx / 2;
}
static void ddrphy_fifo_reset(void __iomem *phy_base)
{
u32 tmp;
tmp = readl(phy_base + DMPHY_PGCR0);
tmp &= ~DMPHY_PGCR0_PHYFRST;
writel(tmp, phy_base + DMPHY_PGCR0);
udelay(1);
tmp |= DMPHY_PGCR0_PHYFRST;
writel(tmp, phy_base + DMPHY_PGCR0);
udelay(1);
}
static void ddrphy_vt_ctrl(void __iomem *phy_base, int enable)
{
u32 tmp;
tmp = readl(phy_base + DMPHY_PGCR1);
if (enable)
tmp &= ~DMPHY_PGCR1_INHVT;
else
tmp |= DMPHY_PGCR1_INHVT;
writel(tmp, phy_base + DMPHY_PGCR1);
if (!enable) {
while (!(readl(phy_base + DMPHY_PGSR1) & DMPHY_PGSR1_VTSTOP))
cpu_relax();
}
}
static void ddrphy_dqs_delay_fixup(void __iomem *phy_base, int nr_dx, int step)
{
int dx;
u32 lcdlr1, rdqsd;
void __iomem *dx_base = phy_base + DMPHY_DX_BASE;
ddrphy_vt_ctrl(phy_base, 0);
for (dx = 0; dx < nr_dx; dx++) {
lcdlr1 = readl(dx_base + DMPHY_DX_LCDLR1);
rdqsd = (lcdlr1 >> 8) & 0xff;
rdqsd = clamp(rdqsd + step, 0U, 0xffU);
lcdlr1 = (lcdlr1 & ~(0xff << 8)) | (rdqsd << 8);
writel(lcdlr1, dx_base + DMPHY_DX_LCDLR1);
readl(dx_base + DMPHY_DX_LCDLR1); /* relax */
dx_base += DMPHY_DX_STRIDE;
}
ddrphy_vt_ctrl(phy_base, 1);
}
static int ddrphy_get_system_latency(void __iomem *phy_base, int width)
{
void __iomem *dx_base = phy_base + DMPHY_DX_BASE;
const int nr_dx = width / 8;
int dx, rank;
u32 gtr;
int dgsl, dgsl_min = INT_MAX, dgsl_max = 0;
for (dx = 0; dx < nr_dx; dx++) {
gtr = readl(dx_base + DMPHY_DX_GTR);
for (rank = 0; rank < 4; rank++) {
dgsl = gtr & 0x7;
/* if dgsl is zero, this rank was not trained. skip. */
if (dgsl) {
dgsl_min = min(dgsl_min, dgsl);
dgsl_max = max(dgsl_max, dgsl);
}
gtr >>= 3;
}
dx_base += DMPHY_DX_STRIDE;
}
if (dgsl_min != dgsl_max)
printf("DQS Gateing System Latencies are not all leveled.\n");
return dgsl_max;
}
static void ddrphy_init(void __iomem *phy_base, enum dram_freq freq, int width,
int ch)
{
u32 tmp;
void __iomem *zq_base, *dx_base;
int zq, dx;
int nr_dx;
nr_dx = width / 8;
writel(DMPHY_PIR_ZCALBYP, phy_base + DMPHY_PIR);
/*
* Disable RGLVT bit (Read DQS Gating LCDL Delay VT Compensation)
* to avoid read error issue.
*/
writel(0x07d81e37, phy_base + DMPHY_PGCR0);
writel(0x0200c4e0, phy_base + DMPHY_PGCR1);
tmp = ddrphy_pgcr2[freq];
if (width >= 32)
tmp |= DMPHY_PGCR2_DUALCHN | DMPHY_PGCR2_ACPDDC;
writel(tmp, phy_base + DMPHY_PGCR2);
writel(ddrphy_ptr0[freq], phy_base + DMPHY_PTR0);
writel(ddrphy_ptr1[freq], phy_base + DMPHY_PTR1);
writel(0x00083def, phy_base + DMPHY_PTR2);
writel(ddrphy_ptr3[freq], phy_base + DMPHY_PTR3);
writel(ddrphy_ptr4[freq], phy_base + DMPHY_PTR4);
writel(ddrphy_acbdlr0[ch], phy_base + DMPHY_ACBDLR0);
writel(0x55555555, phy_base + DMPHY_ACIOCR1);
writel(0x00000000, phy_base + DMPHY_ACIOCR2);
writel(0x55555555, phy_base + DMPHY_ACIOCR3);
writel(0x00000000, phy_base + DMPHY_ACIOCR4);
writel(0x00000055, phy_base + DMPHY_ACIOCR5);
writel(0x00181aa4, phy_base + DMPHY_DXCCR);
writel(0x0024641e, phy_base + DMPHY_DSGCR);
writel(0x0000040b, phy_base + DMPHY_DCR);
writel(ddrphy_dtpr0[freq], phy_base + DMPHY_DTPR0);
writel(ddrphy_dtpr1[freq], phy_base + DMPHY_DTPR1);
writel(ddrphy_dtpr2[freq], phy_base + DMPHY_DTPR2);
writel(ddrphy_dtpr3[freq], phy_base + DMPHY_DTPR3);
writel(ddrphy_mr0[freq], phy_base + DMPHY_MR0);
writel(0x00000006, phy_base + DMPHY_MR1);
writel(ddrphy_mr2[freq], phy_base + DMPHY_MR2);
writel(0x00000000, phy_base + DMPHY_MR3);
tmp = 0;
for (dx = 0; dx < nr_dx; dx++)
tmp |= BIT(DMPHY_DTCR_RANKEN_SHIFT + ddrphy_get_rank(dx));
writel(0x90003087 | tmp, phy_base + DMPHY_DTCR);
writel(0x00000000, phy_base + DMPHY_DTAR0);
writel(0x00000008, phy_base + DMPHY_DTAR1);
writel(0x00000010, phy_base + DMPHY_DTAR2);
writel(0x00000018, phy_base + DMPHY_DTAR3);
writel(0xdd22ee11, phy_base + DMPHY_DTDR0);
writel(0x7788bb44, phy_base + DMPHY_DTDR1);
/* impedance control settings */
writel(0x04048900, phy_base + DMPHY_ZQCR);
zq_base = phy_base + DMPHY_ZQ_BASE;
for (zq = 0; zq < 4; zq++) {
/*
* board-dependent
* PXS2: CH0ZQ0=0x5B, CH1ZQ0=0x5B, CH2ZQ0=0x59, others=0x5D
*/
writel(0x0007BB5D, zq_base + DMPHY_ZQ_PR);
zq_base += DMPHY_ZQ_STRIDE;
}
/* DATX8 settings */
dx_base = phy_base + DMPHY_DX_BASE;
for (dx = 0; dx < 4; dx++) {
tmp = readl(dx_base + DMPHY_DX_GCR0);
tmp &= ~DMPHY_DX_GCR0_WLRKEN_MASK;
tmp |= BIT(DMPHY_DX_GCR0_WLRKEN_SHIFT + ddrphy_get_rank(dx)) &
DMPHY_DX_GCR0_WLRKEN_MASK;
writel(tmp, dx_base + DMPHY_DX_GCR0);
writel(0x00000000, dx_base + DMPHY_DX_GCR1);
writel(0x00000000, dx_base + DMPHY_DX_GCR2);
writel(0x00000000, dx_base + DMPHY_DX_GCR3);
dx_base += DMPHY_DX_STRIDE;
}
while (!(readl(phy_base + DMPHY_PGSR0) & DMPHY_PGSR0_IDONE))
cpu_relax();
ddrphy_dqs_delay_fixup(phy_base, nr_dx, -4);
}
struct ddrphy_init_sequence {
char *description;
u32 init_flag;
u32 done_flag;
u32 err_flag;
};
static const struct ddrphy_init_sequence impedance_calibration_sequence[] = {
{
"Impedance Calibration",
DMPHY_PIR_ZCAL,
DMPHY_PGSR0_ZCDONE,
DMPHY_PGSR0_ZCERR,
},
{ /* sentinel */ }
};
static const struct ddrphy_init_sequence dram_init_sequence[] = {
{
"DRAM Initialization",
DMPHY_PIR_DRAMRST | DMPHY_PIR_DRAMINIT,
DMPHY_PGSR0_DIDONE,
0,
},
{ /* sentinel */ }
};
static const struct ddrphy_init_sequence training_sequence[] = {
{
"Write Leveling",
DMPHY_PIR_WL,
DMPHY_PGSR0_WLDONE,
DMPHY_PGSR0_WLERR,
},
{
"Read DQS Gate Training",
DMPHY_PIR_QSGATE,
DMPHY_PGSR0_QSGDONE,
DMPHY_PGSR0_QSGERR,
},
{
"Write Leveling Adjustment",
DMPHY_PIR_WLADJ,
DMPHY_PGSR0_WLADONE,
DMPHY_PGSR0_WLAERR,
},
{
"Read Bit Deskew",
DMPHY_PIR_RDDSKW,
DMPHY_PGSR0_RDDONE,
DMPHY_PGSR0_RDERR,
},
{
"Write Bit Deskew",
DMPHY_PIR_WRDSKW,
DMPHY_PGSR0_WDDONE,
DMPHY_PGSR0_WDERR,
},
{
"Read Eye Training",
DMPHY_PIR_RDEYE,
DMPHY_PGSR0_REDONE,
DMPHY_PGSR0_REERR,
},
{
"Write Eye Training",
DMPHY_PIR_WREYE,
DMPHY_PGSR0_WEDONE,
DMPHY_PGSR0_WEERR,
},
{ /* sentinel */ }
};
static int __ddrphy_training(void __iomem *phy_base,
const struct ddrphy_init_sequence *seq)
{
const struct ddrphy_init_sequence *s;
u32 pgsr0;
u32 init_flag = DMPHY_PIR_INIT;
u32 done_flag = DMPHY_PGSR0_IDONE;
int timeout = 50000; /* 50 msec is long enough */
#ifdef DISPLAY_ELAPSED_TIME
ulong start = get_timer(0);
#endif
for (s = seq; s->description; s++) {
init_flag |= s->init_flag;
done_flag |= s->done_flag;
}
writel(init_flag, phy_base + DMPHY_PIR);
do {
if (--timeout < 0) {
pr_err("%s: error: timeout during DDR training\n",
__func__);
return -ETIMEDOUT;
}
udelay(1);
pgsr0 = readl(phy_base + DMPHY_PGSR0);
} while ((pgsr0 & done_flag) != done_flag);
for (s = seq; s->description; s++) {
if (pgsr0 & s->err_flag) {
pr_err("%s: error: %s failed\n", __func__,
s->description);
return -EIO;
}
}
#ifdef DISPLAY_ELAPSED_TIME
printf("%s: info: elapsed time %ld msec\n", get_timer(start));
#endif
return 0;
}
static int ddrphy_impedance_calibration(void __iomem *phy_base)
{
int ret;
u32 tmp;
ret = __ddrphy_training(phy_base, impedance_calibration_sequence);
if (ret)
return ret;
/*
* Because of a hardware bug, IDONE flag is set when the first ZQ block
* is calibrated. The flag does not guarantee the completion for all
* the ZQ blocks. Wait a little more just in case.
*/
udelay(1);
/* reflect ZQ settings and enable average algorithm*/
tmp = readl(phy_base + DMPHY_ZQCR);
tmp |= DMPHY_ZQCR_FORCE_ZCAL_VT_UPDATE;
writel(tmp, phy_base + DMPHY_ZQCR);
tmp &= ~DMPHY_ZQCR_FORCE_ZCAL_VT_UPDATE;
tmp |= DMPHY_ZQCR_AVGEN;
writel(tmp, phy_base + DMPHY_ZQCR);
return 0;
}
static int ddrphy_dram_init(void __iomem *phy_base)
{
return __ddrphy_training(phy_base, dram_init_sequence);
}
static int ddrphy_training(void __iomem *phy_base)
{
return __ddrphy_training(phy_base, training_sequence);
}
/* UMC */
static void umc_set_system_latency(void __iomem *umc_dc_base, int phy_latency)
{
u32 val;
int latency;
val = readl(umc_dc_base + UMC_RDATACTL_D0);
latency = (val & UMC_RDATACTL_RADLTY_MASK) >> UMC_RDATACTL_RADLTY_SHIFT;
latency += (val & UMC_RDATACTL_RAD2LTY_MASK) >>
UMC_RDATACTL_RAD2LTY_SHIFT;
/*
* UMC works at the half clock rate of the PHY.
* The LSB of latency is ignored
*/
latency += phy_latency & ~1;
val &= ~(UMC_RDATACTL_RADLTY_MASK | UMC_RDATACTL_RAD2LTY_MASK);
if (latency > 0xf) {
val |= 0xf << UMC_RDATACTL_RADLTY_SHIFT;
val |= (latency - 0xf) << UMC_RDATACTL_RAD2LTY_SHIFT;
} else {
val |= latency << UMC_RDATACTL_RADLTY_SHIFT;
}
writel(val, umc_dc_base + UMC_RDATACTL_D0);
writel(val, umc_dc_base + UMC_RDATACTL_D1);
readl(umc_dc_base + UMC_RDATACTL_D1); /* relax */
}
/* enable/disable auto refresh */
void umc_refresh_ctrl(void __iomem *umc_dc_base, int enable)
{
u32 tmp;
tmp = readl(umc_dc_base + UMC_SPCSETB);
tmp &= ~UMC_SPCSETB_AREFMD_MASK;
if (enable)
tmp |= UMC_SPCSETB_AREFMD_ARB;
else
tmp |= UMC_SPCSETB_AREFMD_REG;
writel(tmp, umc_dc_base + UMC_SPCSETB);
udelay(1);
}
static void umc_ud_init(void __iomem *umc_base, int ch)
{
writel(0x00000003, umc_base + UMC_BITPERPIXELMODE_D0);
if (ch == 2)
writel(0x00000033, umc_base + UMC_PAIR1DOFF_D0);
}
static int umc_dc_init(void __iomem *umc_dc_base, enum dram_freq freq,
unsigned long size, int width, int ch)
{
enum dram_size size_e;
int latency;
u32 val;
switch (size) {
case 0:
return 0;
case SZ_256M:
size_e = DRAM_SZ_256M;
break;
case SZ_512M:
size_e = DRAM_SZ_512M;
break;
default:
pr_err("unsupported DRAM size 0x%08lx (per 16bit) for ch%d\n",
size, ch);
return -EINVAL;
}
writel(umc_cmdctla[freq], umc_dc_base + UMC_CMDCTLA);
writel(ch == 2 ? umc_cmdctlb_ch2[freq] : umc_cmdctlb_ch01[freq],
umc_dc_base + UMC_CMDCTLB);
writel(umc_spcctla[freq][size_e],
umc_dc_base + UMC_SPCCTLA);
writel(umc_spcctlb[freq], umc_dc_base + UMC_SPCCTLB);
val = 0x000e000e;
latency = 12;
/* ES2 inserted one more FF to the logic. */
if (uniphier_get_soc_model() >= 2)
latency += 2;
if (latency > 0xf) {
val |= 0xf << UMC_RDATACTL_RADLTY_SHIFT;
val |= (latency - 0xf) << UMC_RDATACTL_RAD2LTY_SHIFT;
} else {
val |= latency << UMC_RDATACTL_RADLTY_SHIFT;
}
writel(val, umc_dc_base + UMC_RDATACTL_D0);
if (width >= 32)
writel(val, umc_dc_base + UMC_RDATACTL_D1);
writel(0x04060A02, umc_dc_base + UMC_WDATACTL_D0);
if (width >= 32)
writel(0x04060A02, umc_dc_base + UMC_WDATACTL_D1);
writel(0x04000000, umc_dc_base + UMC_DATASET);
writel(0x00400020, umc_dc_base + UMC_DCCGCTL);
writel(0x00000084, umc_dc_base + UMC_FLOWCTLG);
writel(0x00000000, umc_dc_base + UMC_ACSSETA);
writel(ch == 2 ? umc_flowctla_ch2[freq] : umc_flowctla_ch01[freq],
umc_dc_base + UMC_FLOWCTLA);
writel(0x00004400, umc_dc_base + UMC_FLOWCTLC);
writel(0x200A0A00, umc_dc_base + UMC_SPCSETB);
writel(0x00000520, umc_dc_base + UMC_DFICUPDCTLA);
writel(0x0000000D, umc_dc_base + UMC_RESPCTL);
if (ch != 2) {
writel(0x00202000, umc_dc_base + UMC_FLOWCTLB);
writel(0xFDBFFFFF, umc_dc_base + UMC_FLOWCTLOB0);
writel(0xFFFFFFFF, umc_dc_base + UMC_FLOWCTLOB1);
writel(0x00080700, umc_dc_base + UMC_BSICMAPSET);
} else {
writel(0x00200000, umc_dc_base + UMC_FLOWCTLB);
writel(0x00000000, umc_dc_base + UMC_BSICMAPSET);
}
writel(0x00000000, umc_dc_base + UMC_ERRMASKA);
writel(0x00000000, umc_dc_base + UMC_ERRMASKB);
return 0;
}
static int umc_ch_init(void __iomem *umc_ch_base, enum dram_freq freq,
unsigned long size, unsigned int width, int ch)
{
void __iomem *umc_dc_base = umc_ch_base + 0x00011000;
void __iomem *phy_base = umc_ch_base + 0x00030000;
int ret;
writel(0x00000002, umc_dc_base + UMC_INITSET);
while (readl(umc_dc_base + UMC_INITSTAT) & BIT(2))
cpu_relax();
/* deassert PHY reset signals */
writel(UMC_DIOCTLA_CTL_NRST | UMC_DIOCTLA_CFG_NRST,
umc_dc_base + UMC_DIOCTLA);
ddrphy_init(phy_base, freq, width, ch);
ret = ddrphy_impedance_calibration(phy_base);
if (ret)
return ret;
ddrphy_dram_init(phy_base);
if (ret)
return ret;
ret = umc_dc_init(umc_dc_base, freq, size, width, ch);
if (ret)
return ret;
umc_ud_init(umc_ch_base, ch);
ret = ddrphy_training(phy_base);
if (ret)
return ret;
udelay(1);
/* match the system latency between UMC and PHY */
umc_set_system_latency(umc_dc_base,
ddrphy_get_system_latency(phy_base, width));
udelay(1);
/* stop auto refresh before clearing FIFO in PHY */
umc_refresh_ctrl(umc_dc_base, 0);
ddrphy_fifo_reset(phy_base);
umc_refresh_ctrl(umc_dc_base, 1);
udelay(10);
return 0;
}
static void um_init(void __iomem *um_base)
{
writel(0x000000ff, um_base + UMC_MBUS0);
writel(0x000000ff, um_base + UMC_MBUS1);
writel(0x000000ff, um_base + UMC_MBUS2);
writel(0x000000ff, um_base + UMC_MBUS3);
}
int proxstream2_umc_init(const struct uniphier_board_data *bd)
{
void __iomem *um_base = (void __iomem *)0x5b600000;
void __iomem *umc_ch_base = (void __iomem *)0x5b800000;
enum dram_freq freq;
int ch, ret;
switch (bd->dram_freq) {
case 1866:
freq = DRAM_FREQ_1866M;
break;
case 2133:
freq = DRAM_FREQ_2133M;
break;
default:
pr_err("unsupported DRAM frequency %d MHz\n", bd->dram_freq);
return -EINVAL;
}
for (ch = 0; ch < bd->dram_nr_ch; ch++) {
unsigned long size = bd->dram_ch[ch].size;
unsigned int width = bd->dram_ch[ch].width;
ret = umc_ch_init(umc_ch_base, freq, size / (width / 16),
width, ch);
if (ret) {
pr_err("failed to initialize UMC ch%d\n", ch);
return ret;
}
umc_ch_base += 0x00200000;
}
um_init(um_base);
return 0;
}