upstream u-boot with additional patches for our devices/boards: https://lists.denx.de/pipermail/u-boot/2017-March/282789.html (AXP crashes) ; Gbit ethernet patch for some LIME2 revisions ; with SPI flash support
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
u-boot/drivers/pci/fsl_pci_init.c

922 lines
25 KiB

/*
* Copyright 2007-2012 Freescale Semiconductor, Inc.
*
* SPDX-License-Identifier: GPL-2.0+
*/
#include <common.h>
#include <malloc.h>
#include <asm/fsl_serdes.h>
DECLARE_GLOBAL_DATA_PTR;
/*
* PCI/PCIE Controller initialization for mpc85xx/mpc86xx soc's
*
* Initialize controller and call the common driver/pci pci_hose_scan to
* scan for bridges and devices.
*
* Hose fields which need to be pre-initialized by board specific code:
* regions[]
* first_busno
*
* Fields updated:
* last_busno
*/
#include <pci.h>
#include <asm/io.h>
#include <asm/fsl_pci.h>
#ifndef CONFIG_SYS_PCI_MEMORY_BUS
#define CONFIG_SYS_PCI_MEMORY_BUS 0
#endif
#ifndef CONFIG_SYS_PCI_MEMORY_PHYS
#define CONFIG_SYS_PCI_MEMORY_PHYS 0
#endif
#if defined(CONFIG_SYS_PCI_64BIT) && !defined(CONFIG_SYS_PCI64_MEMORY_BUS)
#define CONFIG_SYS_PCI64_MEMORY_BUS (64ull*1024*1024*1024)
#endif
/* Setup one inbound ATMU window.
*
* We let the caller decide what the window size should be
*/
static void set_inbound_window(volatile pit_t *pi,
struct pci_region *r,
u64 size)
{
u32 sz = (__ilog2_u64(size) - 1);
#ifdef CONFIG_SYS_FSL_ERRATUM_A005434
u32 flag = 0;
#else
u32 flag = PIWAR_LOCAL;
#endif
flag |= PIWAR_EN | PIWAR_READ_SNOOP | PIWAR_WRITE_SNOOP;
out_be32(&pi->pitar, r->phys_start >> 12);
out_be32(&pi->piwbar, r->bus_start >> 12);
#ifdef CONFIG_SYS_PCI_64BIT
out_be32(&pi->piwbear, r->bus_start >> 44);
#else
out_be32(&pi->piwbear, 0);
#endif
if (r->flags & PCI_REGION_PREFETCH)
flag |= PIWAR_PF;
out_be32(&pi->piwar, flag | sz);
}
int fsl_setup_hose(struct pci_controller *hose, unsigned long addr)
{
volatile ccsr_fsl_pci_t *pci = (ccsr_fsl_pci_t *) addr;
/* Reset hose to make sure its in a clean state */
memset(hose, 0, sizeof(struct pci_controller));
pci_setup_indirect(hose, (u32)&pci->cfg_addr, (u32)&pci->cfg_data);
return fsl_is_pci_agent(hose);
}
static int fsl_pci_setup_inbound_windows(struct pci_controller *hose,
u64 out_lo, u8 pcie_cap,
volatile pit_t *pi)
{
struct pci_region *r = hose->regions + hose->region_count;
u64 sz = min((u64)gd->ram_size, (1ull << 32));
phys_addr_t phys_start = CONFIG_SYS_PCI_MEMORY_PHYS;
pci_addr_t bus_start = CONFIG_SYS_PCI_MEMORY_BUS;
pci_size_t pci_sz;
/* we have no space available for inbound memory mapping */
if (bus_start > out_lo) {
printf ("no space for inbound mapping of memory\n");
return 0;
}
/* limit size */
if ((bus_start + sz) > out_lo) {
sz = out_lo - bus_start;
debug ("limiting size to %llx\n", sz);
}
pci_sz = 1ull << __ilog2_u64(sz);
/*
* we can overlap inbound/outbound windows on PCI-E since RX & TX
* links a separate
*/
if ((pcie_cap == PCI_CAP_ID_EXP) && (pci_sz < sz)) {
debug ("R0 bus_start: %llx phys_start: %llx size: %llx\n",
(u64)bus_start, (u64)phys_start, (u64)sz);
pci_set_region(r, bus_start, phys_start, sz,
PCI_REGION_MEM | PCI_REGION_SYS_MEMORY |
PCI_REGION_PREFETCH);
/* if we aren't an exact power of two match, pci_sz is smaller
* round it up to the next power of two. We report the actual
* size to pci region tracking.
*/
if (pci_sz != sz)
sz = 2ull << __ilog2_u64(sz);
set_inbound_window(pi--, r++, sz);
sz = 0; /* make sure we dont set the R2 window */
} else {
debug ("R0 bus_start: %llx phys_start: %llx size: %llx\n",
(u64)bus_start, (u64)phys_start, (u64)pci_sz);
pci_set_region(r, bus_start, phys_start, pci_sz,
PCI_REGION_MEM | PCI_REGION_SYS_MEMORY |
PCI_REGION_PREFETCH);
set_inbound_window(pi--, r++, pci_sz);
sz -= pci_sz;
bus_start += pci_sz;
phys_start += pci_sz;
pci_sz = 1ull << __ilog2_u64(sz);
if (sz) {
debug ("R1 bus_start: %llx phys_start: %llx size: %llx\n",
(u64)bus_start, (u64)phys_start, (u64)pci_sz);
pci_set_region(r, bus_start, phys_start, pci_sz,
PCI_REGION_MEM | PCI_REGION_SYS_MEMORY |
PCI_REGION_PREFETCH);
set_inbound_window(pi--, r++, pci_sz);
sz -= pci_sz;
bus_start += pci_sz;
phys_start += pci_sz;
}
}
#if defined(CONFIG_PHYS_64BIT) && defined(CONFIG_SYS_PCI_64BIT)
/*
* On 64-bit capable systems, set up a mapping for all of DRAM
* in high pci address space.
*/
pci_sz = 1ull << __ilog2_u64(gd->ram_size);
/* round up to the next largest power of two */
if (gd->ram_size > pci_sz)
pci_sz = 1ull << (__ilog2_u64(gd->ram_size) + 1);
debug ("R64 bus_start: %llx phys_start: %llx size: %llx\n",
(u64)CONFIG_SYS_PCI64_MEMORY_BUS,
(u64)CONFIG_SYS_PCI_MEMORY_PHYS,
(u64)pci_sz);
pci_set_region(r,
CONFIG_SYS_PCI64_MEMORY_BUS,
CONFIG_SYS_PCI_MEMORY_PHYS,
pci_sz,
PCI_REGION_MEM | PCI_REGION_SYS_MEMORY |
PCI_REGION_PREFETCH);
set_inbound_window(pi--, r++, pci_sz);
#else
pci_sz = 1ull << __ilog2_u64(sz);
if (sz) {
debug ("R2 bus_start: %llx phys_start: %llx size: %llx\n",
(u64)bus_start, (u64)phys_start, (u64)pci_sz);
pci_set_region(r, bus_start, phys_start, pci_sz,
PCI_REGION_MEM | PCI_REGION_SYS_MEMORY |
PCI_REGION_PREFETCH);
sz -= pci_sz;
bus_start += pci_sz;
phys_start += pci_sz;
set_inbound_window(pi--, r++, pci_sz);
}
#endif
#ifdef CONFIG_PHYS_64BIT
if (sz && (((u64)gd->ram_size) < (1ull << 32)))
printf("Was not able to map all of memory via "
"inbound windows -- %lld remaining\n", sz);
#endif
hose->region_count = r - hose->regions;
return 1;
}
#ifdef CONFIG_SRIO_PCIE_BOOT_MASTER
powerpc/corenet_ds: Master module for boot from PCIE For the powerpc processors with PCIE interface, boot location can be configured from one PCIE interface by RCW. The processor booting from PCIE can do without flash for u-boot image. The image can be fetched from another processor's memory space by PCIE link connected between them. The processor booting from PCIE is slave, the processor booting from normal flash memory space is master, and it can help slave to boot from master's memory space. When boot from PCIE, slave's core should be in holdoff after powered on for some specific requirements. Master will release the slave's core at the right time by PCIE interface. Environment and requirement: master: 1. NOR flash for its own u-boot image, ucode and ENV space. 2. Slave's u-boot image is in master NOR flash. 3. Normally boot from local NOR flash. 4. Configure PCIE system if needed. slave: 1. Just has EEPROM for RCW. No flash for u-boot image, ucode and ENV. 2. Boot location should be set to one PCIE interface by RCW. 3. RCW should configure the SerDes, PCIE interfaces correctly. 4. Must set all the cores in holdoff by RCW. 5. Must be powered on before master's boot. For the master module, need to finish these processes: 1. Initialize the PCIE port and address space. 2. Set inbound PCIE windows covered slave's u-boot image stored in master's NOR flash. 3. Set outbound windows in order to configure slave's registers for the core's releasing. 4. Should set the environment variable "bootmaster" to "PCIE1", "PCIE2" or "PCIE3" using the following command: setenv bootmaster PCIE1 saveenv Signed-off-by: Liu Gang <Gang.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
13 years ago
static void fsl_pcie_boot_master(pit_t *pi)
{
/* configure inbound window for slave's u-boot image */
debug("PCIEBOOT - MASTER: Inbound window for slave's image; "
"Local = 0x%llx, Bus = 0x%llx, Size = 0x%x\n",
(u64)CONFIG_SRIO_PCIE_BOOT_IMAGE_MEM_PHYS,
(u64)CONFIG_SRIO_PCIE_BOOT_IMAGE_MEM_BUS1,
CONFIG_SRIO_PCIE_BOOT_IMAGE_SIZE);
struct pci_region r_inbound;
u32 sz_inbound = __ilog2_u64(CONFIG_SRIO_PCIE_BOOT_IMAGE_SIZE)
- 1;
pci_set_region(&r_inbound,
CONFIG_SRIO_PCIE_BOOT_IMAGE_MEM_BUS1,
CONFIG_SRIO_PCIE_BOOT_IMAGE_MEM_PHYS,
sz_inbound,
PCI_REGION_MEM | PCI_REGION_SYS_MEMORY);
set_inbound_window(pi--, &r_inbound,
CONFIG_SRIO_PCIE_BOOT_IMAGE_SIZE);
/* configure inbound window for slave's u-boot image */
debug("PCIEBOOT - MASTER: Inbound window for slave's image; "
"Local = 0x%llx, Bus = 0x%llx, Size = 0x%x\n",
(u64)CONFIG_SRIO_PCIE_BOOT_IMAGE_MEM_PHYS,
(u64)CONFIG_SRIO_PCIE_BOOT_IMAGE_MEM_BUS2,
CONFIG_SRIO_PCIE_BOOT_IMAGE_SIZE);
pci_set_region(&r_inbound,
CONFIG_SRIO_PCIE_BOOT_IMAGE_MEM_BUS2,
CONFIG_SRIO_PCIE_BOOT_IMAGE_MEM_PHYS,
sz_inbound,
PCI_REGION_MEM | PCI_REGION_SYS_MEMORY);
set_inbound_window(pi--, &r_inbound,
CONFIG_SRIO_PCIE_BOOT_IMAGE_SIZE);
/* configure inbound window for slave's ucode and ENV */
debug("PCIEBOOT - MASTER: Inbound window for slave's "
"ucode and ENV; "
"Local = 0x%llx, Bus = 0x%llx, Size = 0x%x\n",
(u64)CONFIG_SRIO_PCIE_BOOT_UCODE_ENV_MEM_PHYS,
(u64)CONFIG_SRIO_PCIE_BOOT_UCODE_ENV_MEM_BUS,
CONFIG_SRIO_PCIE_BOOT_UCODE_ENV_SIZE);
sz_inbound = __ilog2_u64(CONFIG_SRIO_PCIE_BOOT_UCODE_ENV_SIZE)
- 1;
pci_set_region(&r_inbound,
CONFIG_SRIO_PCIE_BOOT_UCODE_ENV_MEM_BUS,
CONFIG_SRIO_PCIE_BOOT_UCODE_ENV_MEM_PHYS,
sz_inbound,
PCI_REGION_MEM | PCI_REGION_SYS_MEMORY);
set_inbound_window(pi--, &r_inbound,
CONFIG_SRIO_PCIE_BOOT_UCODE_ENV_SIZE);
}
static void fsl_pcie_boot_master_release_slave(int port)
{
unsigned long release_addr;
/* now release slave's core 0 */
switch (port) {
case 1:
release_addr = CONFIG_SYS_PCIE1_MEM_VIRT
+ CONFIG_SRIO_PCIE_BOOT_BRR_OFFSET;
break;
#ifdef CONFIG_SYS_PCIE2_MEM_VIRT
powerpc/corenet_ds: Master module for boot from PCIE For the powerpc processors with PCIE interface, boot location can be configured from one PCIE interface by RCW. The processor booting from PCIE can do without flash for u-boot image. The image can be fetched from another processor's memory space by PCIE link connected between them. The processor booting from PCIE is slave, the processor booting from normal flash memory space is master, and it can help slave to boot from master's memory space. When boot from PCIE, slave's core should be in holdoff after powered on for some specific requirements. Master will release the slave's core at the right time by PCIE interface. Environment and requirement: master: 1. NOR flash for its own u-boot image, ucode and ENV space. 2. Slave's u-boot image is in master NOR flash. 3. Normally boot from local NOR flash. 4. Configure PCIE system if needed. slave: 1. Just has EEPROM for RCW. No flash for u-boot image, ucode and ENV. 2. Boot location should be set to one PCIE interface by RCW. 3. RCW should configure the SerDes, PCIE interfaces correctly. 4. Must set all the cores in holdoff by RCW. 5. Must be powered on before master's boot. For the master module, need to finish these processes: 1. Initialize the PCIE port and address space. 2. Set inbound PCIE windows covered slave's u-boot image stored in master's NOR flash. 3. Set outbound windows in order to configure slave's registers for the core's releasing. 4. Should set the environment variable "bootmaster" to "PCIE1", "PCIE2" or "PCIE3" using the following command: setenv bootmaster PCIE1 saveenv Signed-off-by: Liu Gang <Gang.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
13 years ago
case 2:
release_addr = CONFIG_SYS_PCIE2_MEM_VIRT
+ CONFIG_SRIO_PCIE_BOOT_BRR_OFFSET;
break;
#endif
#ifdef CONFIG_SYS_PCIE3_MEM_VIRT
powerpc/corenet_ds: Master module for boot from PCIE For the powerpc processors with PCIE interface, boot location can be configured from one PCIE interface by RCW. The processor booting from PCIE can do without flash for u-boot image. The image can be fetched from another processor's memory space by PCIE link connected between them. The processor booting from PCIE is slave, the processor booting from normal flash memory space is master, and it can help slave to boot from master's memory space. When boot from PCIE, slave's core should be in holdoff after powered on for some specific requirements. Master will release the slave's core at the right time by PCIE interface. Environment and requirement: master: 1. NOR flash for its own u-boot image, ucode and ENV space. 2. Slave's u-boot image is in master NOR flash. 3. Normally boot from local NOR flash. 4. Configure PCIE system if needed. slave: 1. Just has EEPROM for RCW. No flash for u-boot image, ucode and ENV. 2. Boot location should be set to one PCIE interface by RCW. 3. RCW should configure the SerDes, PCIE interfaces correctly. 4. Must set all the cores in holdoff by RCW. 5. Must be powered on before master's boot. For the master module, need to finish these processes: 1. Initialize the PCIE port and address space. 2. Set inbound PCIE windows covered slave's u-boot image stored in master's NOR flash. 3. Set outbound windows in order to configure slave's registers for the core's releasing. 4. Should set the environment variable "bootmaster" to "PCIE1", "PCIE2" or "PCIE3" using the following command: setenv bootmaster PCIE1 saveenv Signed-off-by: Liu Gang <Gang.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
13 years ago
case 3:
release_addr = CONFIG_SYS_PCIE3_MEM_VIRT
+ CONFIG_SRIO_PCIE_BOOT_BRR_OFFSET;
break;
#endif
powerpc/corenet_ds: Master module for boot from PCIE For the powerpc processors with PCIE interface, boot location can be configured from one PCIE interface by RCW. The processor booting from PCIE can do without flash for u-boot image. The image can be fetched from another processor's memory space by PCIE link connected between them. The processor booting from PCIE is slave, the processor booting from normal flash memory space is master, and it can help slave to boot from master's memory space. When boot from PCIE, slave's core should be in holdoff after powered on for some specific requirements. Master will release the slave's core at the right time by PCIE interface. Environment and requirement: master: 1. NOR flash for its own u-boot image, ucode and ENV space. 2. Slave's u-boot image is in master NOR flash. 3. Normally boot from local NOR flash. 4. Configure PCIE system if needed. slave: 1. Just has EEPROM for RCW. No flash for u-boot image, ucode and ENV. 2. Boot location should be set to one PCIE interface by RCW. 3. RCW should configure the SerDes, PCIE interfaces correctly. 4. Must set all the cores in holdoff by RCW. 5. Must be powered on before master's boot. For the master module, need to finish these processes: 1. Initialize the PCIE port and address space. 2. Set inbound PCIE windows covered slave's u-boot image stored in master's NOR flash. 3. Set outbound windows in order to configure slave's registers for the core's releasing. 4. Should set the environment variable "bootmaster" to "PCIE1", "PCIE2" or "PCIE3" using the following command: setenv bootmaster PCIE1 saveenv Signed-off-by: Liu Gang <Gang.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
13 years ago
default:
release_addr = 0;
break;
}
if (release_addr != 0) {
out_be32((void *)release_addr,
CONFIG_SRIO_PCIE_BOOT_RELEASE_MASK);
debug("PCIEBOOT - MASTER: "
"Release slave successfully! Now the slave should start up!\n");
} else {
debug("PCIEBOOT - MASTER: "
"Release slave failed!\n");
}
}
#endif
void fsl_pci_init(struct pci_controller *hose, struct fsl_pci_info *pci_info)
{
u32 cfg_addr = (u32)&((ccsr_fsl_pci_t *)pci_info->regs)->cfg_addr;
u32 cfg_data = (u32)&((ccsr_fsl_pci_t *)pci_info->regs)->cfg_data;
u16 temp16;
u32 temp32;
u32 block_rev;
int enabled, r, inbound = 0;
u16 ltssm;
u8 temp8, pcie_cap;
int pcie_cap_pos;
int pci_dcr;
int pci_dsr;
int pci_lsr;
#if defined(CONFIG_FSL_PCIE_DISABLE_ASPM)
int pci_lcr;
#endif
volatile ccsr_fsl_pci_t *pci = (ccsr_fsl_pci_t *)cfg_addr;
struct pci_region *reg = hose->regions + hose->region_count;
pci_dev_t dev = PCI_BDF(hose->first_busno, 0, 0);
/* Initialize ATMU registers based on hose regions and flags */
volatile pot_t *po = &pci->pot[1]; /* skip 0 */
volatile pit_t *pi;
u64 out_hi = 0, out_lo = -1ULL;
u32 pcicsrbar, pcicsrbar_sz;
pci_setup_indirect(hose, cfg_addr, cfg_data);
block_rev = in_be32(&pci->block_rev1);
if (PEX_IP_BLK_REV_2_2 <= block_rev) {
pi = &pci->pit[2]; /* 0xDC0 */
} else {
pi = &pci->pit[3]; /* 0xDE0 */
}
/* Handle setup of outbound windows first */
for (r = 0; r < hose->region_count; r++) {
unsigned long flags = hose->regions[r].flags;
u32 sz = (__ilog2_u64((u64)hose->regions[r].size) - 1);
flags &= PCI_REGION_SYS_MEMORY|PCI_REGION_TYPE;
if (flags != PCI_REGION_SYS_MEMORY) {
u64 start = hose->regions[r].bus_start;
u64 end = start + hose->regions[r].size;
out_be32(&po->powbar, hose->regions[r].phys_start >> 12);
out_be32(&po->potar, start >> 12);
#ifdef CONFIG_SYS_PCI_64BIT
out_be32(&po->potear, start >> 44);
#else
out_be32(&po->potear, 0);
#endif
if (hose->regions[r].flags & PCI_REGION_IO) {
out_be32(&po->powar, POWAR_EN | sz |
POWAR_IO_READ | POWAR_IO_WRITE);
} else {
out_be32(&po->powar, POWAR_EN | sz |
POWAR_MEM_READ | POWAR_MEM_WRITE);
out_lo = min(start, out_lo);
out_hi = max(end, out_hi);
}
po++;
}
}
debug("Outbound memory range: %llx:%llx\n", out_lo, out_hi);
/* setup PCSRBAR/PEXCSRBAR */
pci_hose_write_config_dword(hose, dev, PCI_BASE_ADDRESS_0, 0xffffffff);
pci_hose_read_config_dword (hose, dev, PCI_BASE_ADDRESS_0, &pcicsrbar_sz);
pcicsrbar_sz = ~pcicsrbar_sz + 1;
if (out_hi < (0x100000000ull - pcicsrbar_sz) ||
(out_lo > 0x100000000ull))
pcicsrbar = 0x100000000ull - pcicsrbar_sz;
else
pcicsrbar = (out_lo - pcicsrbar_sz) & -pcicsrbar_sz;
pci_hose_write_config_dword(hose, dev, PCI_BASE_ADDRESS_0, pcicsrbar);
out_lo = min(out_lo, (u64)pcicsrbar);
debug("PCICSRBAR @ 0x%x\n", pcicsrbar);
pci_set_region(reg++, pcicsrbar, CONFIG_SYS_CCSRBAR_PHYS,
pcicsrbar_sz, PCI_REGION_SYS_MEMORY);
hose->region_count++;
/* see if we are a PCIe or PCI controller */
pcie_cap_pos = pci_hose_find_capability(hose, dev, PCI_CAP_ID_EXP);
pci_dcr = pcie_cap_pos + 0x08;
pci_dsr = pcie_cap_pos + 0x0a;
pci_lsr = pcie_cap_pos + 0x12;
pci_hose_read_config_byte(hose, dev, pcie_cap_pos, &pcie_cap);
#ifdef CONFIG_SRIO_PCIE_BOOT_MASTER
powerpc/corenet_ds: Master module for boot from PCIE For the powerpc processors with PCIE interface, boot location can be configured from one PCIE interface by RCW. The processor booting from PCIE can do without flash for u-boot image. The image can be fetched from another processor's memory space by PCIE link connected between them. The processor booting from PCIE is slave, the processor booting from normal flash memory space is master, and it can help slave to boot from master's memory space. When boot from PCIE, slave's core should be in holdoff after powered on for some specific requirements. Master will release the slave's core at the right time by PCIE interface. Environment and requirement: master: 1. NOR flash for its own u-boot image, ucode and ENV space. 2. Slave's u-boot image is in master NOR flash. 3. Normally boot from local NOR flash. 4. Configure PCIE system if needed. slave: 1. Just has EEPROM for RCW. No flash for u-boot image, ucode and ENV. 2. Boot location should be set to one PCIE interface by RCW. 3. RCW should configure the SerDes, PCIE interfaces correctly. 4. Must set all the cores in holdoff by RCW. 5. Must be powered on before master's boot. For the master module, need to finish these processes: 1. Initialize the PCIE port and address space. 2. Set inbound PCIE windows covered slave's u-boot image stored in master's NOR flash. 3. Set outbound windows in order to configure slave's registers for the core's releasing. 4. Should set the environment variable "bootmaster" to "PCIE1", "PCIE2" or "PCIE3" using the following command: setenv bootmaster PCIE1 saveenv Signed-off-by: Liu Gang <Gang.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
13 years ago
/* boot from PCIE --master */
char *s = getenv("bootmaster");
char pcie[6];
sprintf(pcie, "PCIE%d", pci_info->pci_num);
if (s && (strcmp(s, pcie) == 0)) {
debug("PCIEBOOT - MASTER: Master port [ %d ] for pcie boot.\n",
pci_info->pci_num);
fsl_pcie_boot_master((pit_t *)pi);
} else {
/* inbound */
inbound = fsl_pci_setup_inbound_windows(hose,
out_lo, pcie_cap, pi);
}
#else
/* inbound */
inbound = fsl_pci_setup_inbound_windows(hose, out_lo, pcie_cap, pi);
powerpc/corenet_ds: Master module for boot from PCIE For the powerpc processors with PCIE interface, boot location can be configured from one PCIE interface by RCW. The processor booting from PCIE can do without flash for u-boot image. The image can be fetched from another processor's memory space by PCIE link connected between them. The processor booting from PCIE is slave, the processor booting from normal flash memory space is master, and it can help slave to boot from master's memory space. When boot from PCIE, slave's core should be in holdoff after powered on for some specific requirements. Master will release the slave's core at the right time by PCIE interface. Environment and requirement: master: 1. NOR flash for its own u-boot image, ucode and ENV space. 2. Slave's u-boot image is in master NOR flash. 3. Normally boot from local NOR flash. 4. Configure PCIE system if needed. slave: 1. Just has EEPROM for RCW. No flash for u-boot image, ucode and ENV. 2. Boot location should be set to one PCIE interface by RCW. 3. RCW should configure the SerDes, PCIE interfaces correctly. 4. Must set all the cores in holdoff by RCW. 5. Must be powered on before master's boot. For the master module, need to finish these processes: 1. Initialize the PCIE port and address space. 2. Set inbound PCIE windows covered slave's u-boot image stored in master's NOR flash. 3. Set outbound windows in order to configure slave's registers for the core's releasing. 4. Should set the environment variable "bootmaster" to "PCIE1", "PCIE2" or "PCIE3" using the following command: setenv bootmaster PCIE1 saveenv Signed-off-by: Liu Gang <Gang.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
13 years ago
#endif
for (r = 0; r < hose->region_count; r++)
debug("PCI reg:%d %016llx:%016llx %016llx %08lx\n", r,
(u64)hose->regions[r].phys_start,
(u64)hose->regions[r].bus_start,
(u64)hose->regions[r].size,
hose->regions[r].flags);
pci_register_hose(hose);
pciauto_config_init(hose); /* grab pci_{mem,prefetch,io} */
hose->current_busno = hose->first_busno;
out_be32(&pci->pedr, 0xffffffff); /* Clear any errors */
out_be32(&pci->peer, ~0x20140); /* Enable All Error Interrupts except
* - Master abort (pci)
* - Master PERR (pci)
* - ICCA (PCIe)
*/
pci_hose_read_config_dword(hose, dev, pci_dcr, &temp32);
temp32 |= 0xf000e; /* set URR, FER, NFER (but not CER) */
pci_hose_write_config_dword(hose, dev, pci_dcr, temp32);
#if defined(CONFIG_FSL_PCIE_DISABLE_ASPM)
pci_lcr = pcie_cap_pos + 0x10;
temp32 = 0;
pci_hose_read_config_dword(hose, dev, pci_lcr, &temp32);
temp32 &= ~0x03; /* Disable ASPM */
pci_hose_write_config_dword(hose, dev, pci_lcr, temp32);
udelay(1);
#endif
if (pcie_cap == PCI_CAP_ID_EXP) {
if (block_rev >= PEX_IP_BLK_REV_3_0) {
#define PEX_CSR0_LTSSM_MASK 0xFC
#define PEX_CSR0_LTSSM_SHIFT 2
ltssm = (in_be32(&pci->pex_csr0)
& PEX_CSR0_LTSSM_MASK) >> PEX_CSR0_LTSSM_SHIFT;
enabled = (ltssm == 0x11) ? 1 : 0;
#ifdef CONFIG_FSL_PCIE_RESET
int i;
/* assert PCIe reset */
setbits_be32(&pci->pdb_stat, 0x08000000);
(void) in_be32(&pci->pdb_stat);
udelay(1000);
/* clear PCIe reset */
clrbits_be32(&pci->pdb_stat, 0x08000000);
asm("sync;isync");
for (i = 0; i < 100 && ltssm < PCI_LTSSM_L0; i++) {
pci_hose_read_config_word(hose, dev, PCI_LTSSM,
&ltssm);
udelay(1000);
}
#endif
} else {
/* pci_hose_read_config_word(hose, dev, PCI_LTSSM, &ltssm); */
/* enabled = ltssm >= PCI_LTSSM_L0; */
pci_hose_read_config_word(hose, dev, PCI_LTSSM, &ltssm);
enabled = ltssm >= PCI_LTSSM_L0;
#ifdef CONFIG_FSL_PCIE_RESET
if (ltssm == 1) {
int i;
debug("....PCIe link error. " "LTSSM=0x%02x.", ltssm);
/* assert PCIe reset */
setbits_be32(&pci->pdb_stat, 0x08000000);
(void) in_be32(&pci->pdb_stat);
udelay(100);
debug(" Asserting PCIe reset @%p = %x\n",
&pci->pdb_stat, in_be32(&pci->pdb_stat));
/* clear PCIe reset */
clrbits_be32(&pci->pdb_stat, 0x08000000);
asm("sync;isync");
for (i=0; i<100 && ltssm < PCI_LTSSM_L0; i++) {
pci_hose_read_config_word(hose, dev, PCI_LTSSM,
&ltssm);
udelay(1000);
debug("....PCIe link error. "
"LTSSM=0x%02x.\n", ltssm);
}
enabled = ltssm >= PCI_LTSSM_L0;
/* we need to re-write the bar0 since a reset will
* clear it
*/
pci_hose_write_config_dword(hose, dev,
PCI_BASE_ADDRESS_0, pcicsrbar);
}
#endif
}
#ifdef CONFIG_SYS_P4080_ERRATUM_PCIE_A003
if (enabled == 0) {
serdes_corenet_t *srds_regs = (void *)CONFIG_SYS_FSL_CORENET_SERDES_ADDR;
temp32 = in_be32(&srds_regs->srdspccr0);
if ((temp32 >> 28) == 3) {
int i;
out_be32(&srds_regs->srdspccr0, 2 << 28);
setbits_be32(&pci->pdb_stat, 0x08000000);
in_be32(&pci->pdb_stat);
udelay(100);
clrbits_be32(&pci->pdb_stat, 0x08000000);
asm("sync;isync");
for (i=0; i < 100 && ltssm < PCI_LTSSM_L0; i++) {
pci_hose_read_config_word(hose, dev, PCI_LTSSM, &ltssm);
udelay(1000);
}
enabled = ltssm >= PCI_LTSSM_L0;
}
}
#endif
if (!enabled) {
/* Let the user know there's no PCIe link for root
* complex. for endpoint, the link may not setup, so
* print undetermined.
*/
if (fsl_is_pci_agent(hose))
printf("undetermined, regs @ 0x%lx\n", pci_info->regs);
else
printf("no link, regs @ 0x%lx\n", pci_info->regs);
hose->last_busno = hose->first_busno;
return;
}
out_be32(&pci->pme_msg_det, 0xffffffff);
out_be32(&pci->pme_msg_int_en, 0xffffffff);
/* Print the negotiated PCIe link width */
pci_hose_read_config_word(hose, dev, pci_lsr, &temp16);
printf("x%d gen%d, regs @ 0x%lx\n", (temp16 & 0x3f0) >> 4,
(temp16 & 0xf), pci_info->regs);
hose->current_busno++; /* Start scan with secondary */
pciauto_prescan_setup_bridge(hose, dev, hose->current_busno);
}
#ifdef CONFIG_SYS_FSL_ERRATUM_A007815
/* The Read-Only Write Enable bit defaults to 1 instead of 0.
* Set to 0 to protect the read-only registers.
*/
clrbits_be32(&pci->dbi_ro_wr_en, 0x01);
#endif
/* Use generic setup_device to initialize standard pci regs,
* but do not allocate any windows since any BAR found (such
* as PCSRBAR) is not in this cpu's memory space.
*/
pciauto_setup_device(hose, dev, 0, hose->pci_mem,
hose->pci_prefetch, hose->pci_io);
if (inbound) {
pci_hose_read_config_word(hose, dev, PCI_COMMAND, &temp16);
pci_hose_write_config_word(hose, dev, PCI_COMMAND,
temp16 | PCI_COMMAND_MEMORY);
}
#ifndef CONFIG_PCI_NOSCAN
if (!fsl_is_pci_agent(hose)) {
debug(" Scanning PCI bus %02x\n",
hose->current_busno);
hose->last_busno = pci_hose_scan_bus(hose, hose->current_busno);
} else {
debug(" Not scanning PCI bus %02x. PI=%x\n",
hose->current_busno, temp8);
hose->last_busno = hose->current_busno;
}
/* if we are PCIe - update limit regs and subordinate busno
* for the virtual P2P bridge
*/
if (pcie_cap == PCI_CAP_ID_EXP) {
pciauto_postscan_setup_bridge(hose, dev, hose->last_busno);
}
#else
hose->last_busno = hose->current_busno;
#endif
/* Clear all error indications */
if (pcie_cap == PCI_CAP_ID_EXP)
out_be32(&pci->pme_msg_det, 0xffffffff);
out_be32(&pci->pedr, 0xffffffff);
pci_hose_read_config_word(hose, dev, pci_dsr, &temp16);
if (temp16) {
pci_hose_write_config_word(hose, dev, pci_dsr, 0xffff);
}
pci_hose_read_config_word (hose, dev, PCI_SEC_STATUS, &temp16);
if (temp16) {
pci_hose_write_config_word(hose, dev, PCI_SEC_STATUS, 0xffff);
}
}
int fsl_is_pci_agent(struct pci_controller *hose)
{
int pcie_cap_pos;
u8 pcie_cap;
pci_dev_t dev = PCI_BDF(hose->first_busno, 0, 0);
pcie_cap_pos = pci_hose_find_capability(hose, dev, PCI_CAP_ID_EXP);
pci_hose_read_config_byte(hose, dev, pcie_cap_pos, &pcie_cap);
if (pcie_cap == PCI_CAP_ID_EXP) {
u8 header_type;
pci_hose_read_config_byte(hose, dev, PCI_HEADER_TYPE,
&header_type);
return (header_type & 0x7f) == PCI_HEADER_TYPE_NORMAL;
} else {
u8 prog_if;
pci_hose_read_config_byte(hose, dev, PCI_CLASS_PROG, &prog_if);
/* Programming Interface (PCI_CLASS_PROG)
* 0 == pci host or pcie root-complex,
* 1 == pci agent or pcie end-point
*/
return (prog_if == FSL_PROG_IF_AGENT);
}
}
int fsl_pci_init_port(struct fsl_pci_info *pci_info,
struct pci_controller *hose, int busno)
{
volatile ccsr_fsl_pci_t *pci;
struct pci_region *r;
pci_dev_t dev = PCI_BDF(busno,0,0);
int pcie_cap_pos;
u8 pcie_cap;
pci = (ccsr_fsl_pci_t *) pci_info->regs;
/* on non-PCIe controllers we don't have pme_msg_det so this code
* should do nothing since the read will return 0
*/
if (in_be32(&pci->pme_msg_det)) {
out_be32(&pci->pme_msg_det, 0xffffffff);
debug (" with errors. Clearing. Now 0x%08x",
pci->pme_msg_det);
}
r = hose->regions + hose->region_count;
/* outbound memory */
pci_set_region(r++,
pci_info->mem_bus,
pci_info->mem_phys,
pci_info->mem_size,
PCI_REGION_MEM);
/* outbound io */
pci_set_region(r++,
pci_info->io_bus,
pci_info->io_phys,
pci_info->io_size,
PCI_REGION_IO);
hose->region_count = r - hose->regions;
hose->first_busno = busno;
fsl_pci_init(hose, pci_info);
if (fsl_is_pci_agent(hose)) {
fsl_pci_config_unlock(hose);
hose->last_busno = hose->first_busno;
#ifdef CONFIG_SRIO_PCIE_BOOT_MASTER
powerpc/corenet_ds: Master module for boot from PCIE For the powerpc processors with PCIE interface, boot location can be configured from one PCIE interface by RCW. The processor booting from PCIE can do without flash for u-boot image. The image can be fetched from another processor's memory space by PCIE link connected between them. The processor booting from PCIE is slave, the processor booting from normal flash memory space is master, and it can help slave to boot from master's memory space. When boot from PCIE, slave's core should be in holdoff after powered on for some specific requirements. Master will release the slave's core at the right time by PCIE interface. Environment and requirement: master: 1. NOR flash for its own u-boot image, ucode and ENV space. 2. Slave's u-boot image is in master NOR flash. 3. Normally boot from local NOR flash. 4. Configure PCIE system if needed. slave: 1. Just has EEPROM for RCW. No flash for u-boot image, ucode and ENV. 2. Boot location should be set to one PCIE interface by RCW. 3. RCW should configure the SerDes, PCIE interfaces correctly. 4. Must set all the cores in holdoff by RCW. 5. Must be powered on before master's boot. For the master module, need to finish these processes: 1. Initialize the PCIE port and address space. 2. Set inbound PCIE windows covered slave's u-boot image stored in master's NOR flash. 3. Set outbound windows in order to configure slave's registers for the core's releasing. 4. Should set the environment variable "bootmaster" to "PCIE1", "PCIE2" or "PCIE3" using the following command: setenv bootmaster PCIE1 saveenv Signed-off-by: Liu Gang <Gang.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
13 years ago
} else {
/* boot from PCIE --master releases slave's core 0 */
char *s = getenv("bootmaster");
char pcie[6];
sprintf(pcie, "PCIE%d", pci_info->pci_num);
if (s && (strcmp(s, pcie) == 0))
fsl_pcie_boot_master_release_slave(pci_info->pci_num);
#endif
}
pcie_cap_pos = pci_hose_find_capability(hose, dev, PCI_CAP_ID_EXP);
pci_hose_read_config_byte(hose, dev, pcie_cap_pos, &pcie_cap);
printf("PCI%s%x: Bus %02x - %02x\n", pcie_cap == PCI_CAP_ID_EXP ?
"e" : "", pci_info->pci_num,
hose->first_busno, hose->last_busno);
return(hose->last_busno + 1);
}
/* Enable inbound PCI config cycles for agent/endpoint interface */
void fsl_pci_config_unlock(struct pci_controller *hose)
{
pci_dev_t dev = PCI_BDF(hose->first_busno,0,0);
int pcie_cap_pos;
u8 pcie_cap;
u16 pbfr;
if (!fsl_is_pci_agent(hose))
return;
pcie_cap_pos = pci_hose_find_capability(hose, dev, PCI_CAP_ID_EXP);
pci_hose_read_config_byte(hose, dev, pcie_cap_pos, &pcie_cap);
if (pcie_cap != 0x0) {
ccsr_fsl_pci_t *pci = (ccsr_fsl_pci_t *)hose->cfg_addr;
u32 block_rev = in_be32(&pci->block_rev1);
/* PCIe - set CFG_READY bit of Configuration Ready Register */
if (block_rev >= PEX_IP_BLK_REV_3_0)
setbits_be32(&pci->config, FSL_PCIE_V3_CFG_RDY);
else
pci_hose_write_config_byte(hose, dev,
FSL_PCIE_CFG_RDY, 0x1);
} else {
/* PCI - clear ACL bit of PBFR */
pci_hose_read_config_word(hose, dev, FSL_PCI_PBFR, &pbfr);
pbfr &= ~0x20;
pci_hose_write_config_word(hose, dev, FSL_PCI_PBFR, pbfr);
}
}
#if defined(CONFIG_PCIE1) || defined(CONFIG_PCIE2) || \
defined(CONFIG_PCIE3) || defined(CONFIG_PCIE4)
int fsl_configure_pcie(struct fsl_pci_info *info,
struct pci_controller *hose,
const char *connected, int busno)
{
int is_endpoint;
set_next_law(info->mem_phys, law_size_bits(info->mem_size), info->law);
set_next_law(info->io_phys, law_size_bits(info->io_size), info->law);
is_endpoint = fsl_setup_hose(hose, info->regs);
printf("PCIe%u: %s", info->pci_num,
is_endpoint ? "Endpoint" : "Root Complex");
if (connected)
printf(" of %s", connected);
puts(", ");
return fsl_pci_init_port(info, hose, busno);
}
#if defined(CONFIG_FSL_CORENET)
powerpc/mpc85xx: Add T4240 SoC Add support for Freescale T4240 SoC. Feature of T4240 are (incomplete list): 12 dual-threaded e6500 cores built on Power Architecture® technology Arranged as clusters of four cores sharing a 2 MB L2 cache. Up to 1.8 GHz at 1.0 V with 64-bit ISA support (Power Architecture v2.06-compliant) Three levels of instruction: user, supervisor, and hypervisor 1.5 MB CoreNet Platform Cache (CPC) Hierarchical interconnect fabric CoreNet fabric supporting coherent and non-coherent transactions with prioritization and bandwidth allocation amongst CoreNet end-points 1.6 Tbps coherent read bandwidth Queue Manager (QMan) fabric supporting packet-level queue management and quality of service scheduling Three 64-bit DDR3/3L SDRAM memory controllers with ECC and interleaving support Memory prefetch engine (PMan) Data Path Acceleration Architecture (DPAA) incorporating acceleration for the following functions: Packet parsing, classification, and distribution (Frame Manager 1.1) Queue management for scheduling, packet sequencing, and congestion management (Queue Manager 1.1) Hardware buffer management for buffer allocation and de-allocation (BMan 1.1) Cryptography acceleration (SEC 5.0) at up to 40 Gbps RegEx Pattern Matching Acceleration (PME 2.1) at up to 10 Gbps Decompression/Compression Acceleration (DCE 1.0) at up to 20 Gbps DPAA chip-to-chip interconnect via RapidIO Message Manager (RMAN 1.0) 32 SerDes lanes at up to 10.3125 GHz Ethernet interfaces Up to four 10 Gbps Ethernet MACs Up to sixteen 1 Gbps Ethernet MACs Maximum configuration of 4 x 10 GE + 8 x 1 GE High-speed peripheral interfaces Four PCI Express 2.0/3.0 controllers Two Serial RapidIO 2.0 controllers/ports running at up to 5 GHz with Type 11 messaging and Type 9 data streaming support Interlaken look-aside interface for serial TCAM connection Additional peripheral interfaces Two serial ATA (SATA 2.0) controllers Two high-speed USB 2.0 controllers with integrated PHY Enhanced secure digital host controller (SD/MMC/eMMC) Enhanced serial peripheral interface (eSPI) Four I2C controllers Four 2-pin or two 4-pin UARTs Integrated Flash controller supporting NAND and NOR flash Two eight-channel DMA engines Support for hardware virtualization and partitioning enforcement QorIQ Platform's Trust Architecture 1.1 Signed-off-by: York Sun <yorksun@freescale.com> Signed-off-by: Kumar Gala <galak@kernel.crashing.org> Signed-off-by: Andy Fleming <afleming@freescale.com> Signed-off-by: Roy Zang <tie-fei.zang@freescale.com> Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com> Signed-off-by: Shengzhou Liu <Shengzhou.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
12 years ago
#ifdef CONFIG_SYS_FSL_QORIQ_CHASSIS2
#define _DEVDISR_PCIE1 FSL_CORENET_DEVDISR3_PCIE1
#define _DEVDISR_PCIE2 FSL_CORENET_DEVDISR3_PCIE2
#define _DEVDISR_PCIE3 FSL_CORENET_DEVDISR3_PCIE3
#define _DEVDISR_PCIE4 FSL_CORENET_DEVDISR3_PCIE4
#else
#define _DEVDISR_PCIE1 FSL_CORENET_DEVDISR_PCIE1
#define _DEVDISR_PCIE2 FSL_CORENET_DEVDISR_PCIE2
#define _DEVDISR_PCIE3 FSL_CORENET_DEVDISR_PCIE3
#define _DEVDISR_PCIE4 FSL_CORENET_DEVDISR_PCIE4
powerpc/mpc85xx: Add T4240 SoC Add support for Freescale T4240 SoC. Feature of T4240 are (incomplete list): 12 dual-threaded e6500 cores built on Power Architecture® technology Arranged as clusters of four cores sharing a 2 MB L2 cache. Up to 1.8 GHz at 1.0 V with 64-bit ISA support (Power Architecture v2.06-compliant) Three levels of instruction: user, supervisor, and hypervisor 1.5 MB CoreNet Platform Cache (CPC) Hierarchical interconnect fabric CoreNet fabric supporting coherent and non-coherent transactions with prioritization and bandwidth allocation amongst CoreNet end-points 1.6 Tbps coherent read bandwidth Queue Manager (QMan) fabric supporting packet-level queue management and quality of service scheduling Three 64-bit DDR3/3L SDRAM memory controllers with ECC and interleaving support Memory prefetch engine (PMan) Data Path Acceleration Architecture (DPAA) incorporating acceleration for the following functions: Packet parsing, classification, and distribution (Frame Manager 1.1) Queue management for scheduling, packet sequencing, and congestion management (Queue Manager 1.1) Hardware buffer management for buffer allocation and de-allocation (BMan 1.1) Cryptography acceleration (SEC 5.0) at up to 40 Gbps RegEx Pattern Matching Acceleration (PME 2.1) at up to 10 Gbps Decompression/Compression Acceleration (DCE 1.0) at up to 20 Gbps DPAA chip-to-chip interconnect via RapidIO Message Manager (RMAN 1.0) 32 SerDes lanes at up to 10.3125 GHz Ethernet interfaces Up to four 10 Gbps Ethernet MACs Up to sixteen 1 Gbps Ethernet MACs Maximum configuration of 4 x 10 GE + 8 x 1 GE High-speed peripheral interfaces Four PCI Express 2.0/3.0 controllers Two Serial RapidIO 2.0 controllers/ports running at up to 5 GHz with Type 11 messaging and Type 9 data streaming support Interlaken look-aside interface for serial TCAM connection Additional peripheral interfaces Two serial ATA (SATA 2.0) controllers Two high-speed USB 2.0 controllers with integrated PHY Enhanced secure digital host controller (SD/MMC/eMMC) Enhanced serial peripheral interface (eSPI) Four I2C controllers Four 2-pin or two 4-pin UARTs Integrated Flash controller supporting NAND and NOR flash Two eight-channel DMA engines Support for hardware virtualization and partitioning enforcement QorIQ Platform's Trust Architecture 1.1 Signed-off-by: York Sun <yorksun@freescale.com> Signed-off-by: Kumar Gala <galak@kernel.crashing.org> Signed-off-by: Andy Fleming <afleming@freescale.com> Signed-off-by: Roy Zang <tie-fei.zang@freescale.com> Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com> Signed-off-by: Shengzhou Liu <Shengzhou.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
12 years ago
#endif
#define CONFIG_SYS_MPC8xxx_GUTS_ADDR CONFIG_SYS_MPC85xx_GUTS_ADDR
#elif defined(CONFIG_MPC85xx)
#define _DEVDISR_PCIE1 MPC85xx_DEVDISR_PCIE
#define _DEVDISR_PCIE2 MPC85xx_DEVDISR_PCIE2
#define _DEVDISR_PCIE3 MPC85xx_DEVDISR_PCIE3
#define _DEVDISR_PCIE4 0
#define CONFIG_SYS_MPC8xxx_GUTS_ADDR CONFIG_SYS_MPC85xx_GUTS_ADDR
#elif defined(CONFIG_MPC86xx)
#define _DEVDISR_PCIE1 MPC86xx_DEVDISR_PCIE1
#define _DEVDISR_PCIE2 MPC86xx_DEVDISR_PCIE2
#define _DEVDISR_PCIE3 0
#define _DEVDISR_PCIE4 0
#define CONFIG_SYS_MPC8xxx_GUTS_ADDR \
(&((immap_t *)CONFIG_SYS_IMMR)->im_gur)
#else
#error "No defines for DEVDISR_PCIE"
#endif
/* Implement a dummy function for those platforms w/o SERDES */
static const char *__board_serdes_name(enum srds_prtcl device)
{
switch (device) {
#ifdef CONFIG_SYS_PCIE1_NAME
case PCIE1:
return CONFIG_SYS_PCIE1_NAME;
#endif
#ifdef CONFIG_SYS_PCIE2_NAME
case PCIE2:
return CONFIG_SYS_PCIE2_NAME;
#endif
#ifdef CONFIG_SYS_PCIE3_NAME
case PCIE3:
return CONFIG_SYS_PCIE3_NAME;
#endif
#ifdef CONFIG_SYS_PCIE4_NAME
case PCIE4:
return CONFIG_SYS_PCIE4_NAME;
#endif
default:
return NULL;
}
return NULL;
}
__attribute__((weak, alias("__board_serdes_name"))) const char *
board_serdes_name(enum srds_prtcl device);
static u32 devdisr_mask[] = {
_DEVDISR_PCIE1,
_DEVDISR_PCIE2,
_DEVDISR_PCIE3,
_DEVDISR_PCIE4,
};
int fsl_pcie_init_ctrl(int busno, u32 devdisr, enum srds_prtcl dev,
struct fsl_pci_info *pci_info)
{
struct pci_controller *hose;
int num = dev - PCIE1;
hose = calloc(1, sizeof(struct pci_controller));
if (!hose)
return busno;
if (is_serdes_configured(dev) && !(devdisr & devdisr_mask[num])) {
busno = fsl_configure_pcie(pci_info, hose,
board_serdes_name(dev), busno);
} else {
printf("PCIe%d: disabled\n", num + 1);
}
return busno;
}
int fsl_pcie_init_board(int busno)
{
struct fsl_pci_info pci_info;
ccsr_gur_t *gur = (void *)CONFIG_SYS_MPC8xxx_GUTS_ADDR;
powerpc/mpc85xx: Add T4240 SoC Add support for Freescale T4240 SoC. Feature of T4240 are (incomplete list): 12 dual-threaded e6500 cores built on Power Architecture® technology Arranged as clusters of four cores sharing a 2 MB L2 cache. Up to 1.8 GHz at 1.0 V with 64-bit ISA support (Power Architecture v2.06-compliant) Three levels of instruction: user, supervisor, and hypervisor 1.5 MB CoreNet Platform Cache (CPC) Hierarchical interconnect fabric CoreNet fabric supporting coherent and non-coherent transactions with prioritization and bandwidth allocation amongst CoreNet end-points 1.6 Tbps coherent read bandwidth Queue Manager (QMan) fabric supporting packet-level queue management and quality of service scheduling Three 64-bit DDR3/3L SDRAM memory controllers with ECC and interleaving support Memory prefetch engine (PMan) Data Path Acceleration Architecture (DPAA) incorporating acceleration for the following functions: Packet parsing, classification, and distribution (Frame Manager 1.1) Queue management for scheduling, packet sequencing, and congestion management (Queue Manager 1.1) Hardware buffer management for buffer allocation and de-allocation (BMan 1.1) Cryptography acceleration (SEC 5.0) at up to 40 Gbps RegEx Pattern Matching Acceleration (PME 2.1) at up to 10 Gbps Decompression/Compression Acceleration (DCE 1.0) at up to 20 Gbps DPAA chip-to-chip interconnect via RapidIO Message Manager (RMAN 1.0) 32 SerDes lanes at up to 10.3125 GHz Ethernet interfaces Up to four 10 Gbps Ethernet MACs Up to sixteen 1 Gbps Ethernet MACs Maximum configuration of 4 x 10 GE + 8 x 1 GE High-speed peripheral interfaces Four PCI Express 2.0/3.0 controllers Two Serial RapidIO 2.0 controllers/ports running at up to 5 GHz with Type 11 messaging and Type 9 data streaming support Interlaken look-aside interface for serial TCAM connection Additional peripheral interfaces Two serial ATA (SATA 2.0) controllers Two high-speed USB 2.0 controllers with integrated PHY Enhanced secure digital host controller (SD/MMC/eMMC) Enhanced serial peripheral interface (eSPI) Four I2C controllers Four 2-pin or two 4-pin UARTs Integrated Flash controller supporting NAND and NOR flash Two eight-channel DMA engines Support for hardware virtualization and partitioning enforcement QorIQ Platform's Trust Architecture 1.1 Signed-off-by: York Sun <yorksun@freescale.com> Signed-off-by: Kumar Gala <galak@kernel.crashing.org> Signed-off-by: Andy Fleming <afleming@freescale.com> Signed-off-by: Roy Zang <tie-fei.zang@freescale.com> Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com> Signed-off-by: Shengzhou Liu <Shengzhou.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
12 years ago
u32 devdisr;
u32 *addr;
#ifdef CONFIG_SYS_FSL_QORIQ_CHASSIS2
addr = &gur->devdisr3;
#else
addr = &gur->devdisr;
#endif
devdisr = in_be32(addr);
#ifdef CONFIG_PCIE1
SET_STD_PCIE_INFO(pci_info, 1);
busno = fsl_pcie_init_ctrl(busno, devdisr, PCIE1, &pci_info);
#else
powerpc/mpc85xx: Add T4240 SoC Add support for Freescale T4240 SoC. Feature of T4240 are (incomplete list): 12 dual-threaded e6500 cores built on Power Architecture® technology Arranged as clusters of four cores sharing a 2 MB L2 cache. Up to 1.8 GHz at 1.0 V with 64-bit ISA support (Power Architecture v2.06-compliant) Three levels of instruction: user, supervisor, and hypervisor 1.5 MB CoreNet Platform Cache (CPC) Hierarchical interconnect fabric CoreNet fabric supporting coherent and non-coherent transactions with prioritization and bandwidth allocation amongst CoreNet end-points 1.6 Tbps coherent read bandwidth Queue Manager (QMan) fabric supporting packet-level queue management and quality of service scheduling Three 64-bit DDR3/3L SDRAM memory controllers with ECC and interleaving support Memory prefetch engine (PMan) Data Path Acceleration Architecture (DPAA) incorporating acceleration for the following functions: Packet parsing, classification, and distribution (Frame Manager 1.1) Queue management for scheduling, packet sequencing, and congestion management (Queue Manager 1.1) Hardware buffer management for buffer allocation and de-allocation (BMan 1.1) Cryptography acceleration (SEC 5.0) at up to 40 Gbps RegEx Pattern Matching Acceleration (PME 2.1) at up to 10 Gbps Decompression/Compression Acceleration (DCE 1.0) at up to 20 Gbps DPAA chip-to-chip interconnect via RapidIO Message Manager (RMAN 1.0) 32 SerDes lanes at up to 10.3125 GHz Ethernet interfaces Up to four 10 Gbps Ethernet MACs Up to sixteen 1 Gbps Ethernet MACs Maximum configuration of 4 x 10 GE + 8 x 1 GE High-speed peripheral interfaces Four PCI Express 2.0/3.0 controllers Two Serial RapidIO 2.0 controllers/ports running at up to 5 GHz with Type 11 messaging and Type 9 data streaming support Interlaken look-aside interface for serial TCAM connection Additional peripheral interfaces Two serial ATA (SATA 2.0) controllers Two high-speed USB 2.0 controllers with integrated PHY Enhanced secure digital host controller (SD/MMC/eMMC) Enhanced serial peripheral interface (eSPI) Four I2C controllers Four 2-pin or two 4-pin UARTs Integrated Flash controller supporting NAND and NOR flash Two eight-channel DMA engines Support for hardware virtualization and partitioning enforcement QorIQ Platform's Trust Architecture 1.1 Signed-off-by: York Sun <yorksun@freescale.com> Signed-off-by: Kumar Gala <galak@kernel.crashing.org> Signed-off-by: Andy Fleming <afleming@freescale.com> Signed-off-by: Roy Zang <tie-fei.zang@freescale.com> Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com> Signed-off-by: Shengzhou Liu <Shengzhou.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
12 years ago
setbits_be32(addr, _DEVDISR_PCIE1); /* disable */
#endif
#ifdef CONFIG_PCIE2
SET_STD_PCIE_INFO(pci_info, 2);
busno = fsl_pcie_init_ctrl(busno, devdisr, PCIE2, &pci_info);
#else
powerpc/mpc85xx: Add T4240 SoC Add support for Freescale T4240 SoC. Feature of T4240 are (incomplete list): 12 dual-threaded e6500 cores built on Power Architecture® technology Arranged as clusters of four cores sharing a 2 MB L2 cache. Up to 1.8 GHz at 1.0 V with 64-bit ISA support (Power Architecture v2.06-compliant) Three levels of instruction: user, supervisor, and hypervisor 1.5 MB CoreNet Platform Cache (CPC) Hierarchical interconnect fabric CoreNet fabric supporting coherent and non-coherent transactions with prioritization and bandwidth allocation amongst CoreNet end-points 1.6 Tbps coherent read bandwidth Queue Manager (QMan) fabric supporting packet-level queue management and quality of service scheduling Three 64-bit DDR3/3L SDRAM memory controllers with ECC and interleaving support Memory prefetch engine (PMan) Data Path Acceleration Architecture (DPAA) incorporating acceleration for the following functions: Packet parsing, classification, and distribution (Frame Manager 1.1) Queue management for scheduling, packet sequencing, and congestion management (Queue Manager 1.1) Hardware buffer management for buffer allocation and de-allocation (BMan 1.1) Cryptography acceleration (SEC 5.0) at up to 40 Gbps RegEx Pattern Matching Acceleration (PME 2.1) at up to 10 Gbps Decompression/Compression Acceleration (DCE 1.0) at up to 20 Gbps DPAA chip-to-chip interconnect via RapidIO Message Manager (RMAN 1.0) 32 SerDes lanes at up to 10.3125 GHz Ethernet interfaces Up to four 10 Gbps Ethernet MACs Up to sixteen 1 Gbps Ethernet MACs Maximum configuration of 4 x 10 GE + 8 x 1 GE High-speed peripheral interfaces Four PCI Express 2.0/3.0 controllers Two Serial RapidIO 2.0 controllers/ports running at up to 5 GHz with Type 11 messaging and Type 9 data streaming support Interlaken look-aside interface for serial TCAM connection Additional peripheral interfaces Two serial ATA (SATA 2.0) controllers Two high-speed USB 2.0 controllers with integrated PHY Enhanced secure digital host controller (SD/MMC/eMMC) Enhanced serial peripheral interface (eSPI) Four I2C controllers Four 2-pin or two 4-pin UARTs Integrated Flash controller supporting NAND and NOR flash Two eight-channel DMA engines Support for hardware virtualization and partitioning enforcement QorIQ Platform's Trust Architecture 1.1 Signed-off-by: York Sun <yorksun@freescale.com> Signed-off-by: Kumar Gala <galak@kernel.crashing.org> Signed-off-by: Andy Fleming <afleming@freescale.com> Signed-off-by: Roy Zang <tie-fei.zang@freescale.com> Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com> Signed-off-by: Shengzhou Liu <Shengzhou.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
12 years ago
setbits_be32(addr, _DEVDISR_PCIE2); /* disable */
#endif
#ifdef CONFIG_PCIE3
SET_STD_PCIE_INFO(pci_info, 3);
busno = fsl_pcie_init_ctrl(busno, devdisr, PCIE3, &pci_info);
#else
powerpc/mpc85xx: Add T4240 SoC Add support for Freescale T4240 SoC. Feature of T4240 are (incomplete list): 12 dual-threaded e6500 cores built on Power Architecture® technology Arranged as clusters of four cores sharing a 2 MB L2 cache. Up to 1.8 GHz at 1.0 V with 64-bit ISA support (Power Architecture v2.06-compliant) Three levels of instruction: user, supervisor, and hypervisor 1.5 MB CoreNet Platform Cache (CPC) Hierarchical interconnect fabric CoreNet fabric supporting coherent and non-coherent transactions with prioritization and bandwidth allocation amongst CoreNet end-points 1.6 Tbps coherent read bandwidth Queue Manager (QMan) fabric supporting packet-level queue management and quality of service scheduling Three 64-bit DDR3/3L SDRAM memory controllers with ECC and interleaving support Memory prefetch engine (PMan) Data Path Acceleration Architecture (DPAA) incorporating acceleration for the following functions: Packet parsing, classification, and distribution (Frame Manager 1.1) Queue management for scheduling, packet sequencing, and congestion management (Queue Manager 1.1) Hardware buffer management for buffer allocation and de-allocation (BMan 1.1) Cryptography acceleration (SEC 5.0) at up to 40 Gbps RegEx Pattern Matching Acceleration (PME 2.1) at up to 10 Gbps Decompression/Compression Acceleration (DCE 1.0) at up to 20 Gbps DPAA chip-to-chip interconnect via RapidIO Message Manager (RMAN 1.0) 32 SerDes lanes at up to 10.3125 GHz Ethernet interfaces Up to four 10 Gbps Ethernet MACs Up to sixteen 1 Gbps Ethernet MACs Maximum configuration of 4 x 10 GE + 8 x 1 GE High-speed peripheral interfaces Four PCI Express 2.0/3.0 controllers Two Serial RapidIO 2.0 controllers/ports running at up to 5 GHz with Type 11 messaging and Type 9 data streaming support Interlaken look-aside interface for serial TCAM connection Additional peripheral interfaces Two serial ATA (SATA 2.0) controllers Two high-speed USB 2.0 controllers with integrated PHY Enhanced secure digital host controller (SD/MMC/eMMC) Enhanced serial peripheral interface (eSPI) Four I2C controllers Four 2-pin or two 4-pin UARTs Integrated Flash controller supporting NAND and NOR flash Two eight-channel DMA engines Support for hardware virtualization and partitioning enforcement QorIQ Platform's Trust Architecture 1.1 Signed-off-by: York Sun <yorksun@freescale.com> Signed-off-by: Kumar Gala <galak@kernel.crashing.org> Signed-off-by: Andy Fleming <afleming@freescale.com> Signed-off-by: Roy Zang <tie-fei.zang@freescale.com> Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com> Signed-off-by: Shengzhou Liu <Shengzhou.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
12 years ago
setbits_be32(addr, _DEVDISR_PCIE3); /* disable */
#endif
#ifdef CONFIG_PCIE4
SET_STD_PCIE_INFO(pci_info, 4);
busno = fsl_pcie_init_ctrl(busno, devdisr, PCIE4, &pci_info);
#else
powerpc/mpc85xx: Add T4240 SoC Add support for Freescale T4240 SoC. Feature of T4240 are (incomplete list): 12 dual-threaded e6500 cores built on Power Architecture® technology Arranged as clusters of four cores sharing a 2 MB L2 cache. Up to 1.8 GHz at 1.0 V with 64-bit ISA support (Power Architecture v2.06-compliant) Three levels of instruction: user, supervisor, and hypervisor 1.5 MB CoreNet Platform Cache (CPC) Hierarchical interconnect fabric CoreNet fabric supporting coherent and non-coherent transactions with prioritization and bandwidth allocation amongst CoreNet end-points 1.6 Tbps coherent read bandwidth Queue Manager (QMan) fabric supporting packet-level queue management and quality of service scheduling Three 64-bit DDR3/3L SDRAM memory controllers with ECC and interleaving support Memory prefetch engine (PMan) Data Path Acceleration Architecture (DPAA) incorporating acceleration for the following functions: Packet parsing, classification, and distribution (Frame Manager 1.1) Queue management for scheduling, packet sequencing, and congestion management (Queue Manager 1.1) Hardware buffer management for buffer allocation and de-allocation (BMan 1.1) Cryptography acceleration (SEC 5.0) at up to 40 Gbps RegEx Pattern Matching Acceleration (PME 2.1) at up to 10 Gbps Decompression/Compression Acceleration (DCE 1.0) at up to 20 Gbps DPAA chip-to-chip interconnect via RapidIO Message Manager (RMAN 1.0) 32 SerDes lanes at up to 10.3125 GHz Ethernet interfaces Up to four 10 Gbps Ethernet MACs Up to sixteen 1 Gbps Ethernet MACs Maximum configuration of 4 x 10 GE + 8 x 1 GE High-speed peripheral interfaces Four PCI Express 2.0/3.0 controllers Two Serial RapidIO 2.0 controllers/ports running at up to 5 GHz with Type 11 messaging and Type 9 data streaming support Interlaken look-aside interface for serial TCAM connection Additional peripheral interfaces Two serial ATA (SATA 2.0) controllers Two high-speed USB 2.0 controllers with integrated PHY Enhanced secure digital host controller (SD/MMC/eMMC) Enhanced serial peripheral interface (eSPI) Four I2C controllers Four 2-pin or two 4-pin UARTs Integrated Flash controller supporting NAND and NOR flash Two eight-channel DMA engines Support for hardware virtualization and partitioning enforcement QorIQ Platform's Trust Architecture 1.1 Signed-off-by: York Sun <yorksun@freescale.com> Signed-off-by: Kumar Gala <galak@kernel.crashing.org> Signed-off-by: Andy Fleming <afleming@freescale.com> Signed-off-by: Roy Zang <tie-fei.zang@freescale.com> Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com> Signed-off-by: Shengzhou Liu <Shengzhou.Liu@freescale.com> Signed-off-by: Andy Fleming <afleming@freescale.com>
12 years ago
setbits_be32(addr, _DEVDISR_PCIE4); /* disable */
#endif
return busno;
}
#else
int fsl_pcie_init_ctrl(int busno, u32 devdisr, enum srds_prtcl dev,
struct fsl_pci_info *pci_info)
{
return busno;
}
int fsl_pcie_init_board(int busno)
{
return busno;
}
#endif
#ifdef CONFIG_OF_BOARD_SETUP
#include <libfdt.h>
#include <fdt_support.h>
void ft_fsl_pci_setup(void *blob, const char *pci_compat,
unsigned long ctrl_addr)
{
int off;
u32 bus_range[2];
phys_addr_t p_ctrl_addr = (phys_addr_t)ctrl_addr;
struct pci_controller *hose;
hose = find_hose_by_cfg_addr((void *)(ctrl_addr));
/* convert ctrl_addr to true physical address */
p_ctrl_addr = (phys_addr_t)ctrl_addr - CONFIG_SYS_CCSRBAR;
p_ctrl_addr += CONFIG_SYS_CCSRBAR_PHYS;
off = fdt_node_offset_by_compat_reg(blob, pci_compat, p_ctrl_addr);
if (off < 0)
return;
/* We assume a cfg_addr not being set means we didn't setup the controller */
if ((hose == NULL) || (hose->cfg_addr == NULL)) {
fdt_del_node(blob, off);
} else {
bus_range[0] = 0;
bus_range[1] = hose->last_busno - hose->first_busno;
fdt_setprop(blob, off, "bus-range", &bus_range[0], 2*4);
fdt_pci_dma_ranges(blob, off, hose);
}
}
#endif