aboutsummaryrefslogtreecommitdiffstats
path: root/arch/mips/cavium-octeon
diff options
context:
space:
mode:
authorDavid Daney <ddaney@caviumnetworks.com>2009-01-08 19:46:40 -0500
committerRalf Baechle <ralf@linux-mips.org>2009-01-11 04:57:21 -0500
commit5b3b16880f404ca54126210ca86141cceeafc0cf (patch)
treef69d30450a923782534d4ae257f20aace0a0be74 /arch/mips/cavium-octeon
parent58f07778ce9d32c22cecb1d8ef348001f0e705c9 (diff)
MIPS: Add Cavium OCTEON processor support files to arch/mips/cavium-octeon.
These are the rest of the new files needed to add OCTEON processor support to the Linux kernel. Other than Makefile and Kconfig which should be obvious, we have: csrc-octeon.c -- Clock source driver for OCTEON. dma-octeon.c -- Helper functions for mapping DMA memory. flash_setup.c -- Register on-board flash with the MTD subsystem. octeon-irq.c -- OCTEON interrupt controller managment. octeon-memcpy.S -- Optimized memcpy() implementation. serial.c -- Register 8250 platform driver and early console. setup.c -- Early architecture initialization. smp.c -- OCTEON SMP support. octeon_switch.S -- Scheduler context switch for OCTEON. c-octeon.c -- OCTEON cache controller support. cex-oct.S -- OCTEON cache exception handler. asm/mach-cavium-octeon/*.h -- Architecture include files. Signed-off-by: Tomaso Paoletti <tpaoletti@caviumnetworks.com> Signed-off-by: David Daney <ddaney@caviumnetworks.com> Signed-off-by: Ralf Baechle <ralf@linux-mips.org> create mode 100644 arch/mips/cavium-octeon/Kconfig create mode 100644 arch/mips/cavium-octeon/Makefile create mode 100644 arch/mips/cavium-octeon/csrc-octeon.c create mode 100644 arch/mips/cavium-octeon/dma-octeon.c create mode 100644 arch/mips/cavium-octeon/flash_setup.c create mode 100644 arch/mips/cavium-octeon/octeon-irq.c create mode 100644 arch/mips/cavium-octeon/octeon-memcpy.S create mode 100644 arch/mips/cavium-octeon/serial.c create mode 100644 arch/mips/cavium-octeon/setup.c create mode 100644 arch/mips/cavium-octeon/smp.c create mode 100644 arch/mips/include/asm/mach-cavium-octeon/cpu-feature-overrides.h create mode 100644 arch/mips/include/asm/mach-cavium-octeon/dma-coherence.h create mode 100644 arch/mips/include/asm/mach-cavium-octeon/irq.h create mode 100644 arch/mips/include/asm/mach-cavium-octeon/kernel-entry-init.h create mode 100644 arch/mips/include/asm/mach-cavium-octeon/war.h create mode 100644 arch/mips/include/asm/octeon/octeon.h create mode 100644 arch/mips/kernel/octeon_switch.S create mode 100644 arch/mips/mm/c-octeon.c create mode 100644 arch/mips/mm/cex-oct.S
Diffstat (limited to 'arch/mips/cavium-octeon')
-rw-r--r--arch/mips/cavium-octeon/Kconfig85
-rw-r--r--arch/mips/cavium-octeon/Makefile16
-rw-r--r--arch/mips/cavium-octeon/csrc-octeon.c58
-rw-r--r--arch/mips/cavium-octeon/dma-octeon.c32
-rw-r--r--arch/mips/cavium-octeon/flash_setup.c84
-rw-r--r--arch/mips/cavium-octeon/octeon-irq.c497
-rw-r--r--arch/mips/cavium-octeon/octeon-memcpy.S521
-rw-r--r--arch/mips/cavium-octeon/serial.c136
-rw-r--r--arch/mips/cavium-octeon/setup.c929
-rw-r--r--arch/mips/cavium-octeon/smp.c211
10 files changed, 2569 insertions, 0 deletions
diff --git a/arch/mips/cavium-octeon/Kconfig b/arch/mips/cavium-octeon/Kconfig
new file mode 100644
index 000000000000..094c17e38e16
--- /dev/null
+++ b/arch/mips/cavium-octeon/Kconfig
@@ -0,0 +1,85 @@
1config CAVIUM_OCTEON_SPECIFIC_OPTIONS
2 bool "Enable Octeon specific options"
3 depends on CPU_CAVIUM_OCTEON
4 default "y"
5
6config CAVIUM_OCTEON_2ND_KERNEL
7 bool "Build the kernel to be used as a 2nd kernel on the same chip"
8 depends on CAVIUM_OCTEON_SPECIFIC_OPTIONS
9 default "n"
10 help
11 This option configures this kernel to be linked at a different
12 address and use the 2nd uart for output. This allows a kernel built
13 with this option to be run at the same time as one built without this
14 option.
15
16config CAVIUM_OCTEON_HW_FIX_UNALIGNED
17 bool "Enable hardware fixups of unaligned loads and stores"
18 depends on CAVIUM_OCTEON_SPECIFIC_OPTIONS
19 default "y"
20 help
21 Configure the Octeon hardware to automatically fix unaligned loads
22 and stores. Normally unaligned accesses are fixed using a kernel
23 exception handler. This option enables the hardware automatic fixups,
24 which requires only an extra 3 cycles. Disable this option if you
25 are running code that relies on address exceptions on unaligned
26 accesses.
27
28config CAVIUM_OCTEON_CVMSEG_SIZE
29 int "Number of L1 cache lines reserved for CVMSEG memory"
30 depends on CAVIUM_OCTEON_SPECIFIC_OPTIONS
31 range 0 54
32 default 1
33 help
34 CVMSEG LM is a segment that accesses portions of the dcache as a
35 local memory; the larger CVMSEG is, the smaller the cache is.
36 This selects the size of CVMSEG LM, which is in cache blocks. The
37 legally range is from zero to 54 cache blocks (i.e. CVMSEG LM is
38 between zero and 6192 bytes).
39
40config CAVIUM_OCTEON_LOCK_L2
41 bool "Lock often used kernel code in the L2"
42 depends on CAVIUM_OCTEON_SPECIFIC_OPTIONS
43 default "y"
44 help
45 Enable locking parts of the kernel into the L2 cache.
46
47config CAVIUM_OCTEON_LOCK_L2_TLB
48 bool "Lock the TLB handler in L2"
49 depends on CAVIUM_OCTEON_LOCK_L2
50 default "y"
51 help
52 Lock the low level TLB fast path into L2.
53
54config CAVIUM_OCTEON_LOCK_L2_EXCEPTION
55 bool "Lock the exception handler in L2"
56 depends on CAVIUM_OCTEON_LOCK_L2
57 default "y"
58 help
59 Lock the low level exception handler into L2.
60
61config CAVIUM_OCTEON_LOCK_L2_LOW_LEVEL_INTERRUPT
62 bool "Lock the interrupt handler in L2"
63 depends on CAVIUM_OCTEON_LOCK_L2
64 default "y"
65 help
66 Lock the low level interrupt handler into L2.
67
68config CAVIUM_OCTEON_LOCK_L2_INTERRUPT
69 bool "Lock the 2nd level interrupt handler in L2"
70 depends on CAVIUM_OCTEON_LOCK_L2
71 default "y"
72 help
73 Lock the 2nd level interrupt handler in L2.
74
75config CAVIUM_OCTEON_LOCK_L2_MEMCPY
76 bool "Lock memcpy() in L2"
77 depends on CAVIUM_OCTEON_LOCK_L2
78 default "y"
79 help
80 Lock the kernel's implementation of memcpy() into L2.
81
82config ARCH_SPARSEMEM_ENABLE
83 def_bool y
84 select SPARSEMEM_STATIC
85 depends on CPU_CAVIUM_OCTEON
diff --git a/arch/mips/cavium-octeon/Makefile b/arch/mips/cavium-octeon/Makefile
new file mode 100644
index 000000000000..1c2a7faf5881
--- /dev/null
+++ b/arch/mips/cavium-octeon/Makefile
@@ -0,0 +1,16 @@
1#
2# Makefile for the Cavium Octeon specific kernel interface routines
3# under Linux.
4#
5# This file is subject to the terms and conditions of the GNU General Public
6# License. See the file "COPYING" in the main directory of this archive
7# for more details.
8#
9# Copyright (C) 2005-2008 Cavium Networks
10#
11
12obj-y := setup.o serial.o octeon-irq.o csrc-octeon.o
13obj-y += dma-octeon.o flash_setup.o
14obj-y += octeon-memcpy.o
15
16obj-$(CONFIG_SMP) += smp.o
diff --git a/arch/mips/cavium-octeon/csrc-octeon.c b/arch/mips/cavium-octeon/csrc-octeon.c
new file mode 100644
index 000000000000..70fd92c31657
--- /dev/null
+++ b/arch/mips/cavium-octeon/csrc-octeon.c
@@ -0,0 +1,58 @@
1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Copyright (C) 2007 by Ralf Baechle
7 */
8#include <linux/clocksource.h>
9#include <linux/init.h>
10
11#include <asm/time.h>
12
13#include <asm/octeon/octeon.h>
14#include <asm/octeon/cvmx-ipd-defs.h>
15
16/*
17 * Set the current core's cvmcount counter to the value of the
18 * IPD_CLK_COUNT. We do this on all cores as they are brought
19 * on-line. This allows for a read from a local cpu register to
20 * access a synchronized counter.
21 *
22 */
23void octeon_init_cvmcount(void)
24{
25 unsigned long flags;
26 unsigned loops = 2;
27
28 /* Clobber loops so GCC will not unroll the following while loop. */
29 asm("" : "+r" (loops));
30
31 local_irq_save(flags);
32 /*
33 * Loop several times so we are executing from the cache,
34 * which should give more deterministic timing.
35 */
36 while (loops--)
37 write_c0_cvmcount(cvmx_read_csr(CVMX_IPD_CLK_COUNT));
38 local_irq_restore(flags);
39}
40
41static cycle_t octeon_cvmcount_read(void)
42{
43 return read_c0_cvmcount();
44}
45
46static struct clocksource clocksource_mips = {
47 .name = "OCTEON_CVMCOUNT",
48 .read = octeon_cvmcount_read,
49 .mask = CLOCKSOURCE_MASK(64),
50 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
51};
52
53void __init plat_time_init(void)
54{
55 clocksource_mips.rating = 300;
56 clocksource_set_clock(&clocksource_mips, mips_hpt_frequency);
57 clocksource_register(&clocksource_mips);
58}
diff --git a/arch/mips/cavium-octeon/dma-octeon.c b/arch/mips/cavium-octeon/dma-octeon.c
new file mode 100644
index 000000000000..01b1ef94b361
--- /dev/null
+++ b/arch/mips/cavium-octeon/dma-octeon.c
@@ -0,0 +1,32 @@
1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Copyright (C) 2000 Ani Joshi <ajoshi@unixbox.com>
7 * Copyright (C) 2000, 2001 Ralf Baechle <ralf@gnu.org>
8 * Copyright (C) 2005 Ilya A. Volynets-Evenbakh <ilya@total-knowledge.com>
9 * swiped from i386, and cloned for MIPS by Geert, polished by Ralf.
10 * IP32 changes by Ilya.
11 * Cavium Networks: Create new dma setup for Cavium Networks Octeon based on
12 * the kernels original.
13 */
14#include <linux/types.h>
15#include <linux/mm.h>
16
17#include <dma-coherence.h>
18
19dma_addr_t octeon_map_dma_mem(struct device *dev, void *ptr, size_t size)
20{
21 /* Without PCI/PCIe this function can be called for Octeon internal
22 devices such as USB. These devices all support 64bit addressing */
23 mb();
24 return virt_to_phys(ptr);
25}
26
27void octeon_unmap_dma_mem(struct device *dev, dma_addr_t dma_addr)
28{
29 /* Without PCI/PCIe this function can be called for Octeon internal
30 * devices such as USB. These devices all support 64bit addressing */
31 return;
32}
diff --git a/arch/mips/cavium-octeon/flash_setup.c b/arch/mips/cavium-octeon/flash_setup.c
new file mode 100644
index 000000000000..553d36cbcc42
--- /dev/null
+++ b/arch/mips/cavium-octeon/flash_setup.c
@@ -0,0 +1,84 @@
1/*
2 * Octeon Bootbus flash setup
3 *
4 * This file is subject to the terms and conditions of the GNU General Public
5 * License. See the file "COPYING" in the main directory of this archive
6 * for more details.
7 *
8 * Copyright (C) 2007, 2008 Cavium Networks
9 */
10#include <linux/kernel.h>
11#include <linux/mtd/mtd.h>
12#include <linux/mtd/map.h>
13#include <linux/mtd/partitions.h>
14
15#include <asm/octeon/octeon.h>
16
17static struct map_info flash_map;
18static struct mtd_info *mymtd;
19#ifdef CONFIG_MTD_PARTITIONS
20static int nr_parts;
21static struct mtd_partition *parts;
22static const char *part_probe_types[] = {
23 "cmdlinepart",
24#ifdef CONFIG_MTD_REDBOOT_PARTS
25 "RedBoot",
26#endif
27 NULL
28};
29#endif
30
31/**
32 * Module/ driver initialization.
33 *
34 * Returns Zero on success
35 */
36static int __init flash_init(void)
37{
38 /*
39 * Read the bootbus region 0 setup to determine the base
40 * address of the flash.
41 */
42 union cvmx_mio_boot_reg_cfgx region_cfg;
43 region_cfg.u64 = cvmx_read_csr(CVMX_MIO_BOOT_REG_CFGX(0));
44 if (region_cfg.s.en) {
45 /*
46 * The bootloader always takes the flash and sets its
47 * address so the entire flash fits below
48 * 0x1fc00000. This way the flash aliases to
49 * 0x1fc00000 for booting. Software can access the
50 * full flash at the true address, while core boot can
51 * access 4MB.
52 */
53 /* Use this name so old part lines work */
54 flash_map.name = "phys_mapped_flash";
55 flash_map.phys = region_cfg.s.base << 16;
56 flash_map.size = 0x1fc00000 - flash_map.phys;
57 flash_map.bankwidth = 1;
58 flash_map.virt = ioremap(flash_map.phys, flash_map.size);
59 pr_notice("Bootbus flash: Setting flash for %luMB flash at "
60 "0x%08lx\n", flash_map.size >> 20, flash_map.phys);
61 simple_map_init(&flash_map);
62 mymtd = do_map_probe("cfi_probe", &flash_map);
63 if (mymtd) {
64 mymtd->owner = THIS_MODULE;
65
66#ifdef CONFIG_MTD_PARTITIONS
67 nr_parts = parse_mtd_partitions(mymtd,
68 part_probe_types,
69 &parts, 0);
70 if (nr_parts > 0)
71 add_mtd_partitions(mymtd, parts, nr_parts);
72 else
73 add_mtd_device(mymtd);
74#else
75 add_mtd_device(mymtd);
76#endif
77 } else {
78 pr_err("Failed to register MTD device for flash\n");
79 }
80 }
81 return 0;
82}
83
84late_initcall(flash_init);
diff --git a/arch/mips/cavium-octeon/octeon-irq.c b/arch/mips/cavium-octeon/octeon-irq.c
new file mode 100644
index 000000000000..fc72984a5dae
--- /dev/null
+++ b/arch/mips/cavium-octeon/octeon-irq.c
@@ -0,0 +1,497 @@
1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Copyright (C) 2004-2008 Cavium Networks
7 */
8#include <linux/irq.h>
9#include <linux/interrupt.h>
10#include <linux/hardirq.h>
11
12#include <asm/octeon/octeon.h>
13
14DEFINE_RWLOCK(octeon_irq_ciu0_rwlock);
15DEFINE_RWLOCK(octeon_irq_ciu1_rwlock);
16DEFINE_SPINLOCK(octeon_irq_msi_lock);
17
18static void octeon_irq_core_ack(unsigned int irq)
19{
20 unsigned int bit = irq - OCTEON_IRQ_SW0;
21 /*
22 * We don't need to disable IRQs to make these atomic since
23 * they are already disabled earlier in the low level
24 * interrupt code.
25 */
26 clear_c0_status(0x100 << bit);
27 /* The two user interrupts must be cleared manually. */
28 if (bit < 2)
29 clear_c0_cause(0x100 << bit);
30}
31
32static void octeon_irq_core_eoi(unsigned int irq)
33{
34 irq_desc_t *desc = irq_desc + irq;
35 unsigned int bit = irq - OCTEON_IRQ_SW0;
36 /*
37 * If an IRQ is being processed while we are disabling it the
38 * handler will attempt to unmask the interrupt after it has
39 * been disabled.
40 */
41 if (desc->status & IRQ_DISABLED)
42 return;
43
44 /* There is a race here. We should fix it. */
45
46 /*
47 * We don't need to disable IRQs to make these atomic since
48 * they are already disabled earlier in the low level
49 * interrupt code.
50 */
51 set_c0_status(0x100 << bit);
52}
53
54static void octeon_irq_core_enable(unsigned int irq)
55{
56 unsigned long flags;
57 unsigned int bit = irq - OCTEON_IRQ_SW0;
58
59 /*
60 * We need to disable interrupts to make sure our updates are
61 * atomic.
62 */
63 local_irq_save(flags);
64 set_c0_status(0x100 << bit);
65 local_irq_restore(flags);
66}
67
68static void octeon_irq_core_disable_local(unsigned int irq)
69{
70 unsigned long flags;
71 unsigned int bit = irq - OCTEON_IRQ_SW0;
72 /*
73 * We need to disable interrupts to make sure our updates are
74 * atomic.
75 */
76 local_irq_save(flags);
77 clear_c0_status(0x100 << bit);
78 local_irq_restore(flags);
79}
80
81static void octeon_irq_core_disable(unsigned int irq)
82{
83#ifdef CONFIG_SMP
84 on_each_cpu((void (*)(void *)) octeon_irq_core_disable_local,
85 (void *) (long) irq, 1);
86#else
87 octeon_irq_core_disable_local(irq);
88#endif
89}
90
91static struct irq_chip octeon_irq_chip_core = {
92 .name = "Core",
93 .enable = octeon_irq_core_enable,
94 .disable = octeon_irq_core_disable,
95 .ack = octeon_irq_core_ack,
96 .eoi = octeon_irq_core_eoi,
97};
98
99
100static void octeon_irq_ciu0_ack(unsigned int irq)
101{
102 /*
103 * In order to avoid any locking accessing the CIU, we
104 * acknowledge CIU interrupts by disabling all of them. This
105 * way we can use a per core register and avoid any out of
106 * core locking requirements. This has the side affect that
107 * CIU interrupts can't be processed recursively.
108 *
109 * We don't need to disable IRQs to make these atomic since
110 * they are already disabled earlier in the low level
111 * interrupt code.
112 */
113 clear_c0_status(0x100 << 2);
114}
115
116static void octeon_irq_ciu0_eoi(unsigned int irq)
117{
118 /*
119 * Enable all CIU interrupts again. We don't need to disable
120 * IRQs to make these atomic since they are already disabled
121 * earlier in the low level interrupt code.
122 */
123 set_c0_status(0x100 << 2);
124}
125
126static void octeon_irq_ciu0_enable(unsigned int irq)
127{
128 int coreid = cvmx_get_core_num();
129 unsigned long flags;
130 uint64_t en0;
131 int bit = irq - OCTEON_IRQ_WORKQ0; /* Bit 0-63 of EN0 */
132
133 /*
134 * A read lock is used here to make sure only one core is ever
135 * updating the CIU enable bits at a time. During an enable
136 * the cores don't interfere with each other. During a disable
137 * the write lock stops any enables that might cause a
138 * problem.
139 */
140 read_lock_irqsave(&octeon_irq_ciu0_rwlock, flags);
141 en0 = cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
142 en0 |= 1ull << bit;
143 cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), en0);
144 cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
145 read_unlock_irqrestore(&octeon_irq_ciu0_rwlock, flags);
146}
147
148static void octeon_irq_ciu0_disable(unsigned int irq)
149{
150 int bit = irq - OCTEON_IRQ_WORKQ0; /* Bit 0-63 of EN0 */
151 unsigned long flags;
152 uint64_t en0;
153#ifdef CONFIG_SMP
154 int cpu;
155 write_lock_irqsave(&octeon_irq_ciu0_rwlock, flags);
156 for_each_online_cpu(cpu) {
157 int coreid = cpu_logical_map(cpu);
158 en0 = cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
159 en0 &= ~(1ull << bit);
160 cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), en0);
161 }
162 /*
163 * We need to do a read after the last update to make sure all
164 * of them are done.
165 */
166 cvmx_read_csr(CVMX_CIU_INTX_EN0(cvmx_get_core_num() * 2));
167 write_unlock_irqrestore(&octeon_irq_ciu0_rwlock, flags);
168#else
169 int coreid = cvmx_get_core_num();
170 local_irq_save(flags);
171 en0 = cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
172 en0 &= ~(1ull << bit);
173 cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), en0);
174 cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
175 local_irq_restore(flags);
176#endif
177}
178
179#ifdef CONFIG_SMP
180static void octeon_irq_ciu0_set_affinity(unsigned int irq, const struct cpumask *dest)
181{
182 int cpu;
183 int bit = irq - OCTEON_IRQ_WORKQ0; /* Bit 0-63 of EN0 */
184
185 write_lock(&octeon_irq_ciu0_rwlock);
186 for_each_online_cpu(cpu) {
187 int coreid = cpu_logical_map(cpu);
188 uint64_t en0 =
189 cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2));
190 if (cpumask_test_cpu(cpu, dest))
191 en0 |= 1ull << bit;
192 else
193 en0 &= ~(1ull << bit);
194 cvmx_write_csr(CVMX_CIU_INTX_EN0(coreid * 2), en0);
195 }
196 /*
197 * We need to do a read after the last update to make sure all
198 * of them are done.
199 */
200 cvmx_read_csr(CVMX_CIU_INTX_EN0(cvmx_get_core_num() * 2));
201 write_unlock(&octeon_irq_ciu0_rwlock);
202}
203#endif
204
205static struct irq_chip octeon_irq_chip_ciu0 = {
206 .name = "CIU0",
207 .enable = octeon_irq_ciu0_enable,
208 .disable = octeon_irq_ciu0_disable,
209 .ack = octeon_irq_ciu0_ack,
210 .eoi = octeon_irq_ciu0_eoi,
211#ifdef CONFIG_SMP
212 .set_affinity = octeon_irq_ciu0_set_affinity,
213#endif
214};
215
216
217static void octeon_irq_ciu1_ack(unsigned int irq)
218{
219 /*
220 * In order to avoid any locking accessing the CIU, we
221 * acknowledge CIU interrupts by disabling all of them. This
222 * way we can use a per core register and avoid any out of
223 * core locking requirements. This has the side affect that
224 * CIU interrupts can't be processed recursively. We don't
225 * need to disable IRQs to make these atomic since they are
226 * already disabled earlier in the low level interrupt code.
227 */
228 clear_c0_status(0x100 << 3);
229}
230
231static void octeon_irq_ciu1_eoi(unsigned int irq)
232{
233 /*
234 * Enable all CIU interrupts again. We don't need to disable
235 * IRQs to make these atomic since they are already disabled
236 * earlier in the low level interrupt code.
237 */
238 set_c0_status(0x100 << 3);
239}
240
241static void octeon_irq_ciu1_enable(unsigned int irq)
242{
243 int coreid = cvmx_get_core_num();
244 unsigned long flags;
245 uint64_t en1;
246 int bit = irq - OCTEON_IRQ_WDOG0; /* Bit 0-63 of EN1 */
247
248 /*
249 * A read lock is used here to make sure only one core is ever
250 * updating the CIU enable bits at a time. During an enable
251 * the cores don't interfere with each other. During a disable
252 * the write lock stops any enables that might cause a
253 * problem.
254 */
255 read_lock_irqsave(&octeon_irq_ciu1_rwlock, flags);
256 en1 = cvmx_read_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1));
257 en1 |= 1ull << bit;
258 cvmx_write_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1), en1);
259 cvmx_read_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1));
260 read_unlock_irqrestore(&octeon_irq_ciu1_rwlock, flags);
261}
262
263static void octeon_irq_ciu1_disable(unsigned int irq)
264{
265 int bit = irq - OCTEON_IRQ_WDOG0; /* Bit 0-63 of EN1 */
266 unsigned long flags;
267 uint64_t en1;
268#ifdef CONFIG_SMP
269 int cpu;
270 write_lock_irqsave(&octeon_irq_ciu1_rwlock, flags);
271 for_each_online_cpu(cpu) {
272 int coreid = cpu_logical_map(cpu);
273 en1 = cvmx_read_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1));
274 en1 &= ~(1ull << bit);
275 cvmx_write_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1), en1);
276 }
277 /*
278 * We need to do a read after the last update to make sure all
279 * of them are done.
280 */
281 cvmx_read_csr(CVMX_CIU_INTX_EN1(cvmx_get_core_num() * 2 + 1));
282 write_unlock_irqrestore(&octeon_irq_ciu1_rwlock, flags);
283#else
284 int coreid = cvmx_get_core_num();
285 local_irq_save(flags);
286 en1 = cvmx_read_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1));
287 en1 &= ~(1ull << bit);
288 cvmx_write_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1), en1);
289 cvmx_read_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1));
290 local_irq_restore(flags);
291#endif
292}
293
294#ifdef CONFIG_SMP
295static void octeon_irq_ciu1_set_affinity(unsigned int irq, const struct cpumask *dest)
296{
297 int cpu;
298 int bit = irq - OCTEON_IRQ_WDOG0; /* Bit 0-63 of EN1 */
299
300 write_lock(&octeon_irq_ciu1_rwlock);
301 for_each_online_cpu(cpu) {
302 int coreid = cpu_logical_map(cpu);
303 uint64_t en1 =
304 cvmx_read_csr(CVMX_CIU_INTX_EN1
305 (coreid * 2 + 1));
306 if (cpumask_test_cpu(cpu, dest))
307 en1 |= 1ull << bit;
308 else
309 en1 &= ~(1ull << bit);
310 cvmx_write_csr(CVMX_CIU_INTX_EN1(coreid * 2 + 1), en1);
311 }
312 /*
313 * We need to do a read after the last update to make sure all
314 * of them are done.
315 */
316 cvmx_read_csr(CVMX_CIU_INTX_EN1(cvmx_get_core_num() * 2 + 1));
317 write_unlock(&octeon_irq_ciu1_rwlock);
318}
319#endif
320
321static struct irq_chip octeon_irq_chip_ciu1 = {
322 .name = "CIU1",
323 .enable = octeon_irq_ciu1_enable,
324 .disable = octeon_irq_ciu1_disable,
325 .ack = octeon_irq_ciu1_ack,
326 .eoi = octeon_irq_ciu1_eoi,
327#ifdef CONFIG_SMP
328 .set_affinity = octeon_irq_ciu1_set_affinity,
329#endif
330};
331
332#ifdef CONFIG_PCI_MSI
333
334static void octeon_irq_msi_ack(unsigned int irq)
335{
336 if (!octeon_has_feature(OCTEON_FEATURE_PCIE)) {
337 /* These chips have PCI */
338 cvmx_write_csr(CVMX_NPI_NPI_MSI_RCV,
339 1ull << (irq - OCTEON_IRQ_MSI_BIT0));
340 } else {
341 /*
342 * These chips have PCIe. Thankfully the ACK doesn't
343 * need any locking.
344 */
345 cvmx_write_csr(CVMX_PEXP_NPEI_MSI_RCV0,
346 1ull << (irq - OCTEON_IRQ_MSI_BIT0));
347 }
348}
349
350static void octeon_irq_msi_eoi(unsigned int irq)
351{
352 /* Nothing needed */
353}
354
355static void octeon_irq_msi_enable(unsigned int irq)
356{
357 if (!octeon_has_feature(OCTEON_FEATURE_PCIE)) {
358 /*
359 * Octeon PCI doesn't have the ability to mask/unmask
360 * MSI interrupts individually. Instead of
361 * masking/unmasking them in groups of 16, we simple
362 * assume MSI devices are well behaved. MSI
363 * interrupts are always enable and the ACK is assumed
364 * to be enough.
365 */
366 } else {
367 /* These chips have PCIe. Note that we only support
368 * the first 64 MSI interrupts. Unfortunately all the
369 * MSI enables are in the same register. We use
370 * MSI0's lock to control access to them all.
371 */
372 uint64_t en;
373 unsigned long flags;
374 spin_lock_irqsave(&octeon_irq_msi_lock, flags);
375 en = cvmx_read_csr(CVMX_PEXP_NPEI_MSI_ENB0);
376 en |= 1ull << (irq - OCTEON_IRQ_MSI_BIT0);
377 cvmx_write_csr(CVMX_PEXP_NPEI_MSI_ENB0, en);
378 cvmx_read_csr(CVMX_PEXP_NPEI_MSI_ENB0);
379 spin_unlock_irqrestore(&octeon_irq_msi_lock, flags);
380 }
381}
382
383static void octeon_irq_msi_disable(unsigned int irq)
384{
385 if (!octeon_has_feature(OCTEON_FEATURE_PCIE)) {
386 /* See comment in enable */
387 } else {
388 /*
389 * These chips have PCIe. Note that we only support
390 * the first 64 MSI interrupts. Unfortunately all the
391 * MSI enables are in the same register. We use
392 * MSI0's lock to control access to them all.
393 */
394 uint64_t en;
395 unsigned long flags;
396 spin_lock_irqsave(&octeon_irq_msi_lock, flags);
397 en = cvmx_read_csr(CVMX_PEXP_NPEI_MSI_ENB0);
398 en &= ~(1ull << (irq - OCTEON_IRQ_MSI_BIT0));
399 cvmx_write_csr(CVMX_PEXP_NPEI_MSI_ENB0, en);
400 cvmx_read_csr(CVMX_PEXP_NPEI_MSI_ENB0);
401 spin_unlock_irqrestore(&octeon_irq_msi_lock, flags);
402 }
403}
404
405static struct irq_chip octeon_irq_chip_msi = {
406 .name = "MSI",
407 .enable = octeon_irq_msi_enable,
408 .disable = octeon_irq_msi_disable,
409 .ack = octeon_irq_msi_ack,
410 .eoi = octeon_irq_msi_eoi,
411};
412#endif
413
414void __init arch_init_irq(void)
415{
416 int irq;
417
418#ifdef CONFIG_SMP
419 /* Set the default affinity to the boot cpu. */
420 cpumask_clear(irq_default_affinity);
421 cpumask_set_cpu(smp_processor_id(), irq_default_affinity);
422#endif
423
424 if (NR_IRQS < OCTEON_IRQ_LAST)
425 pr_err("octeon_irq_init: NR_IRQS is set too low\n");
426
427 /* 0 - 15 reserved for i8259 master and slave controller. */
428
429 /* 17 - 23 Mips internal */
430 for (irq = OCTEON_IRQ_SW0; irq <= OCTEON_IRQ_TIMER; irq++) {
431 set_irq_chip_and_handler(irq, &octeon_irq_chip_core,
432 handle_percpu_irq);
433 }
434
435 /* 24 - 87 CIU_INT_SUM0 */
436 for (irq = OCTEON_IRQ_WORKQ0; irq <= OCTEON_IRQ_BOOTDMA; irq++) {
437 set_irq_chip_and_handler(irq, &octeon_irq_chip_ciu0,
438 handle_percpu_irq);
439 }
440
441 /* 88 - 151 CIU_INT_SUM1 */
442 for (irq = OCTEON_IRQ_WDOG0; irq <= OCTEON_IRQ_RESERVED151; irq++) {
443 set_irq_chip_and_handler(irq, &octeon_irq_chip_ciu1,
444 handle_percpu_irq);
445 }
446
447#ifdef CONFIG_PCI_MSI
448 /* 152 - 215 PCI/PCIe MSI interrupts */
449 for (irq = OCTEON_IRQ_MSI_BIT0; irq <= OCTEON_IRQ_MSI_BIT63; irq++) {
450 set_irq_chip_and_handler(irq, &octeon_irq_chip_msi,
451 handle_percpu_irq);
452 }
453#endif
454 set_c0_status(0x300 << 2);
455}
456
457asmlinkage void plat_irq_dispatch(void)
458{
459 const unsigned long core_id = cvmx_get_core_num();
460 const uint64_t ciu_sum0_address = CVMX_CIU_INTX_SUM0(core_id * 2);
461 const uint64_t ciu_en0_address = CVMX_CIU_INTX_EN0(core_id * 2);
462 const uint64_t ciu_sum1_address = CVMX_CIU_INT_SUM1;
463 const uint64_t ciu_en1_address = CVMX_CIU_INTX_EN1(core_id * 2 + 1);
464 unsigned long cop0_cause;
465 unsigned long cop0_status;
466 uint64_t ciu_en;
467 uint64_t ciu_sum;
468
469 while (1) {
470 cop0_cause = read_c0_cause();
471 cop0_status = read_c0_status();
472 cop0_cause &= cop0_status;
473 cop0_cause &= ST0_IM;
474
475 if (unlikely(cop0_cause & STATUSF_IP2)) {
476 ciu_sum = cvmx_read_csr(ciu_sum0_address);
477 ciu_en = cvmx_read_csr(ciu_en0_address);
478 ciu_sum &= ciu_en;
479 if (likely(ciu_sum))
480 do_IRQ(fls64(ciu_sum) + OCTEON_IRQ_WORKQ0 - 1);
481 else
482 spurious_interrupt();
483 } else if (unlikely(cop0_cause & STATUSF_IP3)) {
484 ciu_sum = cvmx_read_csr(ciu_sum1_address);
485 ciu_en = cvmx_read_csr(ciu_en1_address);
486 ciu_sum &= ciu_en;
487 if (likely(ciu_sum))
488 do_IRQ(fls64(ciu_sum) + OCTEON_IRQ_WDOG0 - 1);
489 else
490 spurious_interrupt();
491 } else if (likely(cop0_cause)) {
492 do_IRQ(fls(cop0_cause) - 9 + MIPS_CPU_IRQ_BASE);
493 } else {
494 break;
495 }
496 }
497}
diff --git a/arch/mips/cavium-octeon/octeon-memcpy.S b/arch/mips/cavium-octeon/octeon-memcpy.S
new file mode 100644
index 000000000000..88e0cddca205
--- /dev/null
+++ b/arch/mips/cavium-octeon/octeon-memcpy.S
@@ -0,0 +1,521 @@
1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Unified implementation of memcpy, memmove and the __copy_user backend.
7 *
8 * Copyright (C) 1998, 99, 2000, 01, 2002 Ralf Baechle (ralf@gnu.org)
9 * Copyright (C) 1999, 2000, 01, 2002 Silicon Graphics, Inc.
10 * Copyright (C) 2002 Broadcom, Inc.
11 * memcpy/copy_user author: Mark Vandevoorde
12 *
13 * Mnemonic names for arguments to memcpy/__copy_user
14 */
15
16#include <asm/asm.h>
17#include <asm/asm-offsets.h>
18#include <asm/regdef.h>
19
20#define dst a0
21#define src a1
22#define len a2
23
24/*
25 * Spec
26 *
27 * memcpy copies len bytes from src to dst and sets v0 to dst.
28 * It assumes that
29 * - src and dst don't overlap
30 * - src is readable
31 * - dst is writable
32 * memcpy uses the standard calling convention
33 *
34 * __copy_user copies up to len bytes from src to dst and sets a2 (len) to
35 * the number of uncopied bytes due to an exception caused by a read or write.
36 * __copy_user assumes that src and dst don't overlap, and that the call is
37 * implementing one of the following:
38 * copy_to_user
39 * - src is readable (no exceptions when reading src)
40 * copy_from_user
41 * - dst is writable (no exceptions when writing dst)
42 * __copy_user uses a non-standard calling convention; see
43 * arch/mips/include/asm/uaccess.h
44 *
45 * When an exception happens on a load, the handler must
46 # ensure that all of the destination buffer is overwritten to prevent
47 * leaking information to user mode programs.
48 */
49
50/*
51 * Implementation
52 */
53
54/*
55 * The exception handler for loads requires that:
56 * 1- AT contain the address of the byte just past the end of the source
57 * of the copy,
58 * 2- src_entry <= src < AT, and
59 * 3- (dst - src) == (dst_entry - src_entry),
60 * The _entry suffix denotes values when __copy_user was called.
61 *
62 * (1) is set up up by uaccess.h and maintained by not writing AT in copy_user
63 * (2) is met by incrementing src by the number of bytes copied
64 * (3) is met by not doing loads between a pair of increments of dst and src
65 *
66 * The exception handlers for stores adjust len (if necessary) and return.
67 * These handlers do not need to overwrite any data.
68 *
69 * For __rmemcpy and memmove an exception is always a kernel bug, therefore
70 * they're not protected.
71 */
72
73#define EXC(inst_reg,addr,handler) \
749: inst_reg, addr; \
75 .section __ex_table,"a"; \
76 PTR 9b, handler; \
77 .previous
78
79/*
80 * Only on the 64-bit kernel we can made use of 64-bit registers.
81 */
82#ifdef CONFIG_64BIT
83#define USE_DOUBLE
84#endif
85
86#ifdef USE_DOUBLE
87
88#define LOAD ld
89#define LOADL ldl
90#define LOADR ldr
91#define STOREL sdl
92#define STORER sdr
93#define STORE sd
94#define ADD daddu
95#define SUB dsubu
96#define SRL dsrl
97#define SRA dsra
98#define SLL dsll
99#define SLLV dsllv
100#define SRLV dsrlv
101#define NBYTES 8
102#define LOG_NBYTES 3
103
104/*
105 * As we are sharing code base with the mips32 tree (which use the o32 ABI
106 * register definitions). We need to redefine the register definitions from
107 * the n64 ABI register naming to the o32 ABI register naming.
108 */
109#undef t0
110#undef t1
111#undef t2
112#undef t3
113#define t0 $8
114#define t1 $9
115#define t2 $10
116#define t3 $11
117#define t4 $12
118#define t5 $13
119#define t6 $14
120#define t7 $15
121
122#else
123
124#define LOAD lw
125#define LOADL lwl
126#define LOADR lwr
127#define STOREL swl
128#define STORER swr
129#define STORE sw
130#define ADD addu
131#define SUB subu
132#define SRL srl
133#define SLL sll
134#define SRA sra
135#define SLLV sllv
136#define SRLV srlv
137#define NBYTES 4
138#define LOG_NBYTES 2
139
140#endif /* USE_DOUBLE */
141
142#ifdef CONFIG_CPU_LITTLE_ENDIAN
143#define LDFIRST LOADR
144#define LDREST LOADL
145#define STFIRST STORER
146#define STREST STOREL
147#define SHIFT_DISCARD SLLV
148#else
149#define LDFIRST LOADL
150#define LDREST LOADR
151#define STFIRST STOREL
152#define STREST STORER
153#define SHIFT_DISCARD SRLV
154#endif
155
156#define FIRST(unit) ((unit)*NBYTES)
157#define REST(unit) (FIRST(unit)+NBYTES-1)
158#define UNIT(unit) FIRST(unit)
159
160#define ADDRMASK (NBYTES-1)
161
162 .text
163 .set noreorder
164 .set noat
165
166/*
167 * A combined memcpy/__copy_user
168 * __copy_user sets len to 0 for success; else to an upper bound of
169 * the number of uncopied bytes.
170 * memcpy sets v0 to dst.
171 */
172 .align 5
173LEAF(memcpy) /* a0=dst a1=src a2=len */
174 move v0, dst /* return value */
175__memcpy:
176FEXPORT(__copy_user)
177 /*
178 * Note: dst & src may be unaligned, len may be 0
179 * Temps
180 */
181 #
182 # Octeon doesn't care if the destination is unaligned. The hardware
183 # can fix it faster than we can special case the assembly.
184 #
185 pref 0, 0(src)
186 sltu t0, len, NBYTES # Check if < 1 word
187 bnez t0, copy_bytes_checklen
188 and t0, src, ADDRMASK # Check if src unaligned
189 bnez t0, src_unaligned
190 sltu t0, len, 4*NBYTES # Check if < 4 words
191 bnez t0, less_than_4units
192 sltu t0, len, 8*NBYTES # Check if < 8 words
193 bnez t0, less_than_8units
194 sltu t0, len, 16*NBYTES # Check if < 16 words
195 bnez t0, cleanup_both_aligned
196 sltu t0, len, 128+1 # Check if len < 129
197 bnez t0, 1f # Skip prefetch if len is too short
198 sltu t0, len, 256+1 # Check if len < 257
199 bnez t0, 1f # Skip prefetch if len is too short
200 pref 0, 128(src) # We must not prefetch invalid addresses
201 #
202 # This is where we loop if there is more than 128 bytes left
2032: pref 0, 256(src) # We must not prefetch invalid addresses
204 #
205 # This is where we loop if we can't prefetch anymore
2061:
207EXC( LOAD t0, UNIT(0)(src), l_exc)
208EXC( LOAD t1, UNIT(1)(src), l_exc_copy)
209EXC( LOAD t2, UNIT(2)(src), l_exc_copy)
210EXC( LOAD t3, UNIT(3)(src), l_exc_copy)
211 SUB len, len, 16*NBYTES
212EXC( STORE t0, UNIT(0)(dst), s_exc_p16u)
213EXC( STORE t1, UNIT(1)(dst), s_exc_p15u)
214EXC( STORE t2, UNIT(2)(dst), s_exc_p14u)
215EXC( STORE t3, UNIT(3)(dst), s_exc_p13u)
216EXC( LOAD t0, UNIT(4)(src), l_exc_copy)
217EXC( LOAD t1, UNIT(5)(src), l_exc_copy)
218EXC( LOAD t2, UNIT(6)(src), l_exc_copy)
219EXC( LOAD t3, UNIT(7)(src), l_exc_copy)
220EXC( STORE t0, UNIT(4)(dst), s_exc_p12u)
221EXC( STORE t1, UNIT(5)(dst), s_exc_p11u)
222EXC( STORE t2, UNIT(6)(dst), s_exc_p10u)
223 ADD src, src, 16*NBYTES
224EXC( STORE t3, UNIT(7)(dst), s_exc_p9u)
225 ADD dst, dst, 16*NBYTES
226EXC( LOAD t0, UNIT(-8)(src), l_exc_copy)
227EXC( LOAD t1, UNIT(-7)(src), l_exc_copy)
228EXC( LOAD t2, UNIT(-6)(src), l_exc_copy)
229EXC( LOAD t3, UNIT(-5)(src), l_exc_copy)
230EXC( STORE t0, UNIT(-8)(dst), s_exc_p8u)
231EXC( STORE t1, UNIT(-7)(dst), s_exc_p7u)
232EXC( STORE t2, UNIT(-6)(dst), s_exc_p6u)
233EXC( STORE t3, UNIT(-5)(dst), s_exc_p5u)
234EXC( LOAD t0, UNIT(-4)(src), l_exc_copy)
235EXC( LOAD t1, UNIT(-3)(src), l_exc_copy)
236EXC( LOAD t2, UNIT(-2)(src), l_exc_copy)
237EXC( LOAD t3, UNIT(-1)(src), l_exc_copy)
238EXC( STORE t0, UNIT(-4)(dst), s_exc_p4u)
239EXC( STORE t1, UNIT(-3)(dst), s_exc_p3u)
240EXC( STORE t2, UNIT(-2)(dst), s_exc_p2u)
241EXC( STORE t3, UNIT(-1)(dst), s_exc_p1u)
242 sltu t0, len, 256+1 # See if we can prefetch more
243 beqz t0, 2b
244 sltu t0, len, 128 # See if we can loop more time
245 beqz t0, 1b
246 nop
247 #
248 # Jump here if there are less than 16*NBYTES left.
249 #
250cleanup_both_aligned:
251 beqz len, done
252 sltu t0, len, 8*NBYTES
253 bnez t0, less_than_8units
254 nop
255EXC( LOAD t0, UNIT(0)(src), l_exc)
256EXC( LOAD t1, UNIT(1)(src), l_exc_copy)
257EXC( LOAD t2, UNIT(2)(src), l_exc_copy)
258EXC( LOAD t3, UNIT(3)(src), l_exc_copy)
259 SUB len, len, 8*NBYTES
260EXC( STORE t0, UNIT(0)(dst), s_exc_p8u)
261EXC( STORE t1, UNIT(1)(dst), s_exc_p7u)
262EXC( STORE t2, UNIT(2)(dst), s_exc_p6u)
263EXC( STORE t3, UNIT(3)(dst), s_exc_p5u)
264EXC( LOAD t0, UNIT(4)(src), l_exc_copy)
265EXC( LOAD t1, UNIT(5)(src), l_exc_copy)
266EXC( LOAD t2, UNIT(6)(src), l_exc_copy)
267EXC( LOAD t3, UNIT(7)(src), l_exc_copy)
268EXC( STORE t0, UNIT(4)(dst), s_exc_p4u)
269EXC( STORE t1, UNIT(5)(dst), s_exc_p3u)
270EXC( STORE t2, UNIT(6)(dst), s_exc_p2u)
271EXC( STORE t3, UNIT(7)(dst), s_exc_p1u)
272 ADD src, src, 8*NBYTES
273 beqz len, done
274 ADD dst, dst, 8*NBYTES
275 #
276 # Jump here if there are less than 8*NBYTES left.
277 #
278less_than_8units:
279 sltu t0, len, 4*NBYTES
280 bnez t0, less_than_4units
281 nop
282EXC( LOAD t0, UNIT(0)(src), l_exc)
283EXC( LOAD t1, UNIT(1)(src), l_exc_copy)
284EXC( LOAD t2, UNIT(2)(src), l_exc_copy)
285EXC( LOAD t3, UNIT(3)(src), l_exc_copy)
286 SUB len, len, 4*NBYTES
287EXC( STORE t0, UNIT(0)(dst), s_exc_p4u)
288EXC( STORE t1, UNIT(1)(dst), s_exc_p3u)
289EXC( STORE t2, UNIT(2)(dst), s_exc_p2u)
290EXC( STORE t3, UNIT(3)(dst), s_exc_p1u)
291 ADD src, src, 4*NBYTES
292 beqz len, done
293 ADD dst, dst, 4*NBYTES
294 #
295 # Jump here if there are less than 4*NBYTES left. This means
296 # we may need to copy up to 3 NBYTES words.
297 #
298less_than_4units:
299 sltu t0, len, 1*NBYTES
300 bnez t0, copy_bytes_checklen
301 nop
302 #
303 # 1) Copy NBYTES, then check length again
304 #
305EXC( LOAD t0, 0(src), l_exc)
306 SUB len, len, NBYTES
307 sltu t1, len, 8
308EXC( STORE t0, 0(dst), s_exc_p1u)
309 ADD src, src, NBYTES
310 bnez t1, copy_bytes_checklen
311 ADD dst, dst, NBYTES
312 #
313 # 2) Copy NBYTES, then check length again
314 #
315EXC( LOAD t0, 0(src), l_exc)
316 SUB len, len, NBYTES
317 sltu t1, len, 8
318EXC( STORE t0, 0(dst), s_exc_p1u)
319 ADD src, src, NBYTES
320 bnez t1, copy_bytes_checklen
321 ADD dst, dst, NBYTES
322 #
323 # 3) Copy NBYTES, then check length again
324 #
325EXC( LOAD t0, 0(src), l_exc)
326 SUB len, len, NBYTES
327 ADD src, src, NBYTES
328 ADD dst, dst, NBYTES
329 b copy_bytes_checklen
330EXC( STORE t0, -8(dst), s_exc_p1u)
331
332src_unaligned:
333#define rem t8
334 SRL t0, len, LOG_NBYTES+2 # +2 for 4 units/iter
335 beqz t0, cleanup_src_unaligned
336 and rem, len, (4*NBYTES-1) # rem = len % 4*NBYTES
3371:
338/*
339 * Avoid consecutive LD*'s to the same register since some mips
340 * implementations can't issue them in the same cycle.
341 * It's OK to load FIRST(N+1) before REST(N) because the two addresses
342 * are to the same unit (unless src is aligned, but it's not).
343 */
344EXC( LDFIRST t0, FIRST(0)(src), l_exc)
345EXC( LDFIRST t1, FIRST(1)(src), l_exc_copy)
346 SUB len, len, 4*NBYTES
347EXC( LDREST t0, REST(0)(src), l_exc_copy)
348EXC( LDREST t1, REST(1)(src), l_exc_copy)
349EXC( LDFIRST t2, FIRST(2)(src), l_exc_copy)
350EXC( LDFIRST t3, FIRST(3)(src), l_exc_copy)
351EXC( LDREST t2, REST(2)(src), l_exc_copy)
352EXC( LDREST t3, REST(3)(src), l_exc_copy)
353 ADD src, src, 4*NBYTES
354EXC( STORE t0, UNIT(0)(dst), s_exc_p4u)
355EXC( STORE t1, UNIT(1)(dst), s_exc_p3u)
356EXC( STORE t2, UNIT(2)(dst), s_exc_p2u)
357EXC( STORE t3, UNIT(3)(dst), s_exc_p1u)
358 bne len, rem, 1b
359 ADD dst, dst, 4*NBYTES
360
361cleanup_src_unaligned:
362 beqz len, done
363 and rem, len, NBYTES-1 # rem = len % NBYTES
364 beq rem, len, copy_bytes
365 nop
3661:
367EXC( LDFIRST t0, FIRST(0)(src), l_exc)
368EXC( LDREST t0, REST(0)(src), l_exc_copy)
369 SUB len, len, NBYTES
370EXC( STORE t0, 0(dst), s_exc_p1u)
371 ADD src, src, NBYTES
372 bne len, rem, 1b
373 ADD dst, dst, NBYTES
374
375copy_bytes_checklen:
376 beqz len, done
377 nop
378copy_bytes:
379 /* 0 < len < NBYTES */
380#define COPY_BYTE(N) \
381EXC( lb t0, N(src), l_exc); \
382 SUB len, len, 1; \
383 beqz len, done; \
384EXC( sb t0, N(dst), s_exc_p1)
385
386 COPY_BYTE(0)
387 COPY_BYTE(1)
388#ifdef USE_DOUBLE
389 COPY_BYTE(2)
390 COPY_BYTE(3)
391 COPY_BYTE(4)
392 COPY_BYTE(5)
393#endif
394EXC( lb t0, NBYTES-2(src), l_exc)
395 SUB len, len, 1
396 jr ra
397EXC( sb t0, NBYTES-2(dst), s_exc_p1)
398done:
399 jr ra
400 nop
401 END(memcpy)
402
403l_exc_copy:
404 /*
405 * Copy bytes from src until faulting load address (or until a
406 * lb faults)
407 *
408 * When reached by a faulting LDFIRST/LDREST, THREAD_BUADDR($28)
409 * may be more than a byte beyond the last address.
410 * Hence, the lb below may get an exception.
411 *
412 * Assumes src < THREAD_BUADDR($28)
413 */
414 LOAD t0, TI_TASK($28)
415 nop
416 LOAD t0, THREAD_BUADDR(t0)
4171:
418EXC( lb t1, 0(src), l_exc)
419 ADD src, src, 1
420 sb t1, 0(dst) # can't fault -- we're copy_from_user
421 bne src, t0, 1b
422 ADD dst, dst, 1
423l_exc:
424 LOAD t0, TI_TASK($28)
425 nop
426 LOAD t0, THREAD_BUADDR(t0) # t0 is just past last good address
427 nop
428 SUB len, AT, t0 # len number of uncopied bytes
429 /*
430 * Here's where we rely on src and dst being incremented in tandem,
431 * See (3) above.
432 * dst += (fault addr - src) to put dst at first byte to clear
433 */
434 ADD dst, t0 # compute start address in a1
435 SUB dst, src
436 /*
437 * Clear len bytes starting at dst. Can't call __bzero because it
438 * might modify len. An inefficient loop for these rare times...
439 */
440 beqz len, done
441 SUB src, len, 1
4421: sb zero, 0(dst)
443 ADD dst, dst, 1
444 bnez src, 1b
445 SUB src, src, 1
446 jr ra
447 nop
448
449
450#define SEXC(n) \
451s_exc_p ## n ## u: \
452 jr ra; \
453 ADD len, len, n*NBYTES
454
455SEXC(16)
456SEXC(15)
457SEXC(14)
458SEXC(13)
459SEXC(12)
460SEXC(11)
461SEXC(10)
462SEXC(9)
463SEXC(8)
464SEXC(7)
465SEXC(6)
466SEXC(5)
467SEXC(4)
468SEXC(3)
469SEXC(2)
470SEXC(1)
471
472s_exc_p1:
473 jr ra
474 ADD len, len, 1
475s_exc:
476 jr ra
477 nop
478
479 .align 5
480LEAF(memmove)
481 ADD t0, a0, a2
482 ADD t1, a1, a2
483 sltu t0, a1, t0 # dst + len <= src -> memcpy
484 sltu t1, a0, t1 # dst >= src + len -> memcpy
485 and t0, t1
486 beqz t0, __memcpy
487 move v0, a0 /* return value */
488 beqz a2, r_out
489 END(memmove)
490
491 /* fall through to __rmemcpy */
492LEAF(__rmemcpy) /* a0=dst a1=src a2=len */
493 sltu t0, a1, a0
494 beqz t0, r_end_bytes_up # src >= dst
495 nop
496 ADD a0, a2 # dst = dst + len
497 ADD a1, a2 # src = src + len
498
499r_end_bytes:
500 lb t0, -1(a1)
501 SUB a2, a2, 0x1
502 sb t0, -1(a0)
503 SUB a1, a1, 0x1
504 bnez a2, r_end_bytes
505 SUB a0, a0, 0x1
506
507r_out:
508 jr ra
509 move a2, zero
510
511r_end_bytes_up:
512 lb t0, (a1)
513 SUB a2, a2, 0x1
514 sb t0, (a0)
515 ADD a1, a1, 0x1
516 bnez a2, r_end_bytes_up
517 ADD a0, a0, 0x1
518
519 jr ra
520 move a2, zero
521 END(__rmemcpy)
diff --git a/arch/mips/cavium-octeon/serial.c b/arch/mips/cavium-octeon/serial.c
new file mode 100644
index 000000000000..8240728d485a
--- /dev/null
+++ b/arch/mips/cavium-octeon/serial.c
@@ -0,0 +1,136 @@
1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Copyright (C) 2004-2007 Cavium Networks
7 */
8#include <linux/console.h>
9#include <linux/module.h>
10#include <linux/init.h>
11#include <linux/platform_device.h>
12#include <linux/serial.h>
13#include <linux/serial_8250.h>
14#include <linux/serial_reg.h>
15#include <linux/tty.h>
16
17#include <asm/time.h>
18
19#include <asm/octeon/octeon.h>
20
21#ifdef CONFIG_GDB_CONSOLE
22#define DEBUG_UART 0
23#else
24#define DEBUG_UART 1
25#endif
26
27unsigned int octeon_serial_in(struct uart_port *up, int offset)
28{
29 int rv = cvmx_read_csr((uint64_t)(up->membase + (offset << 3)));
30 if (offset == UART_IIR && (rv & 0xf) == 7) {
31 /* Busy interrupt, read the USR (39) and try again. */
32 cvmx_read_csr((uint64_t)(up->membase + (39 << 3)));
33 rv = cvmx_read_csr((uint64_t)(up->membase + (offset << 3)));
34 }
35 return rv;
36}
37
38void octeon_serial_out(struct uart_port *up, int offset, int value)
39{
40 /*
41 * If bits 6 or 7 of the OCTEON UART's LCR are set, it quits
42 * working.
43 */
44 if (offset == UART_LCR)
45 value &= 0x9f;
46 cvmx_write_csr((uint64_t)(up->membase + (offset << 3)), (u8)value);
47}
48
49/*
50 * Allocated in .bss, so it is all zeroed.
51 */
52#define OCTEON_MAX_UARTS 3
53static struct plat_serial8250_port octeon_uart8250_data[OCTEON_MAX_UARTS + 1];
54static struct platform_device octeon_uart8250_device = {
55 .name = "serial8250",
56 .id = PLAT8250_DEV_PLATFORM,
57 .dev = {
58 .platform_data = octeon_uart8250_data,
59 },
60};
61
62static void __init octeon_uart_set_common(struct plat_serial8250_port *p)
63{
64 p->flags = ASYNC_SKIP_TEST | UPF_SHARE_IRQ | UPF_FIXED_TYPE;
65 p->type = PORT_OCTEON;
66 p->iotype = UPIO_MEM;
67 p->regshift = 3; /* I/O addresses are every 8 bytes */
68 p->uartclk = mips_hpt_frequency;
69 p->serial_in = octeon_serial_in;
70 p->serial_out = octeon_serial_out;
71}
72
73static int __init octeon_serial_init(void)
74{
75 int enable_uart0;
76 int enable_uart1;
77 int enable_uart2;
78 struct plat_serial8250_port *p;
79
80#ifdef CONFIG_CAVIUM_OCTEON_2ND_KERNEL
81 /*
82 * If we are configured to run as the second of two kernels,
83 * disable uart0 and enable uart1. Uart0 is owned by the first
84 * kernel
85 */
86 enable_uart0 = 0;
87 enable_uart1 = 1;
88#else
89 /*
90 * We are configured for the first kernel. We'll enable uart0
91 * if the bootloader told us to use 0, otherwise will enable
92 * uart 1.
93 */
94 enable_uart0 = (octeon_get_boot_uart() == 0);
95 enable_uart1 = (octeon_get_boot_uart() == 1);
96#ifdef CONFIG_KGDB
97 enable_uart1 = 1;
98#endif
99#endif
100
101 /* Right now CN52XX is the only chip with a third uart */
102 enable_uart2 = OCTEON_IS_MODEL(OCTEON_CN52XX);
103
104 p = octeon_uart8250_data;
105 if (enable_uart0) {
106 /* Add a ttyS device for hardware uart 0 */
107 octeon_uart_set_common(p);
108 p->membase = (void *) CVMX_MIO_UARTX_RBR(0);
109 p->mapbase = CVMX_MIO_UARTX_RBR(0) & ((1ull << 49) - 1);
110 p->irq = OCTEON_IRQ_UART0;
111 p++;
112 }
113
114 if (enable_uart1) {
115 /* Add a ttyS device for hardware uart 1 */
116 octeon_uart_set_common(p);
117 p->membase = (void *) CVMX_MIO_UARTX_RBR(1);
118 p->mapbase = CVMX_MIO_UARTX_RBR(1) & ((1ull << 49) - 1);
119 p->irq = OCTEON_IRQ_UART1;
120 p++;
121 }
122 if (enable_uart2) {
123 /* Add a ttyS device for hardware uart 2 */
124 octeon_uart_set_common(p);
125 p->membase = (void *) CVMX_MIO_UART2_RBR;
126 p->mapbase = CVMX_MIO_UART2_RBR & ((1ull << 49) - 1);
127 p->irq = OCTEON_IRQ_UART2;
128 p++;
129 }
130
131 BUG_ON(p > &octeon_uart8250_data[OCTEON_MAX_UARTS]);
132
133 return platform_device_register(&octeon_uart8250_device);
134}
135
136device_initcall(octeon_serial_init);
diff --git a/arch/mips/cavium-octeon/setup.c b/arch/mips/cavium-octeon/setup.c
new file mode 100644
index 000000000000..e085feddb4a4
--- /dev/null
+++ b/arch/mips/cavium-octeon/setup.c
@@ -0,0 +1,929 @@
1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Copyright (C) 2004-2007 Cavium Networks
7 * Copyright (C) 2008 Wind River Systems
8 */
9#include <linux/init.h>
10#include <linux/console.h>
11#include <linux/delay.h>
12#include <linux/interrupt.h>
13#include <linux/io.h>
14#include <linux/irq.h>
15#include <linux/serial.h>
16#include <linux/types.h>
17#include <linux/string.h> /* for memset */
18#include <linux/serial.h>
19#include <linux/tty.h>
20#include <linux/time.h>
21#include <linux/platform_device.h>
22#include <linux/serial_core.h>
23#include <linux/serial_8250.h>
24#include <linux/string.h>
25
26#include <asm/processor.h>
27#include <asm/reboot.h>
28#include <asm/smp-ops.h>
29#include <asm/system.h>
30#include <asm/irq_cpu.h>
31#include <asm/mipsregs.h>
32#include <asm/bootinfo.h>
33#include <asm/sections.h>
34#include <asm/time.h>
35
36#include <asm/octeon/octeon.h>
37
38#ifdef CONFIG_CAVIUM_DECODE_RSL
39extern void cvmx_interrupt_rsl_decode(void);
40extern int __cvmx_interrupt_ecc_report_single_bit_errors;
41extern void cvmx_interrupt_rsl_enable(void);
42#endif
43
44extern struct plat_smp_ops octeon_smp_ops;
45
46#ifdef CONFIG_PCI
47extern void pci_console_init(const char *arg);
48#endif
49
50#ifdef CONFIG_CAVIUM_RESERVE32
51extern uint64_t octeon_reserve32_memory;
52#endif
53static unsigned long long MAX_MEMORY = 512ull << 20;
54
55struct octeon_boot_descriptor *octeon_boot_desc_ptr;
56
57struct cvmx_bootinfo *octeon_bootinfo;
58EXPORT_SYMBOL(octeon_bootinfo);
59
60#ifdef CONFIG_CAVIUM_RESERVE32
61uint64_t octeon_reserve32_memory;
62EXPORT_SYMBOL(octeon_reserve32_memory);
63#endif
64
65static int octeon_uart;
66
67extern asmlinkage void handle_int(void);
68extern asmlinkage void plat_irq_dispatch(void);
69
70/**
71 * Return non zero if we are currently running in the Octeon simulator
72 *
73 * Returns
74 */
75int octeon_is_simulation(void)
76{
77 return octeon_bootinfo->board_type == CVMX_BOARD_TYPE_SIM;
78}
79EXPORT_SYMBOL(octeon_is_simulation);
80
81/**
82 * Return true if Octeon is in PCI Host mode. This means
83 * Linux can control the PCI bus.
84 *
85 * Returns Non zero if Octeon in host mode.
86 */
87int octeon_is_pci_host(void)
88{
89#ifdef CONFIG_PCI
90 return octeon_bootinfo->config_flags & CVMX_BOOTINFO_CFG_FLAG_PCI_HOST;
91#else
92 return 0;
93#endif
94}
95
96/**
97 * Get the clock rate of Octeon
98 *
99 * Returns Clock rate in HZ
100 */
101uint64_t octeon_get_clock_rate(void)
102{
103 if (octeon_is_simulation())
104 octeon_bootinfo->eclock_hz = 6000000;
105 return octeon_bootinfo->eclock_hz;
106}
107EXPORT_SYMBOL(octeon_get_clock_rate);
108
109/**
110 * Write to the LCD display connected to the bootbus. This display
111 * exists on most Cavium evaluation boards. If it doesn't exist, then
112 * this function doesn't do anything.
113 *
114 * @s: String to write
115 */
116void octeon_write_lcd(const char *s)
117{
118 if (octeon_bootinfo->led_display_base_addr) {
119 void __iomem *lcd_address =
120 ioremap_nocache(octeon_bootinfo->led_display_base_addr,
121 8);
122 int i;
123 for (i = 0; i < 8; i++, s++) {
124 if (*s)
125 iowrite8(*s, lcd_address + i);
126 else
127 iowrite8(' ', lcd_address + i);
128 }
129 iounmap(lcd_address);
130 }
131}
132
133/**
134 * Return the console uart passed by the bootloader
135 *
136 * Returns uart (0 or 1)
137 */
138int octeon_get_boot_uart(void)
139{
140 int uart;
141#ifdef CONFIG_CAVIUM_OCTEON_2ND_KERNEL
142 uart = 1;
143#else
144 uart = (octeon_boot_desc_ptr->flags & OCTEON_BL_FLAG_CONSOLE_UART1) ?
145 1 : 0;
146#endif
147 return uart;
148}
149
150/**
151 * Get the coremask Linux was booted on.
152 *
153 * Returns Core mask
154 */
155int octeon_get_boot_coremask(void)
156{
157 return octeon_boot_desc_ptr->core_mask;
158}
159
160/**
161 * Check the hardware BIST results for a CPU
162 */
163void octeon_check_cpu_bist(void)
164{
165 const int coreid = cvmx_get_core_num();
166 unsigned long long mask;
167 unsigned long long bist_val;
168
169 /* Check BIST results for COP0 registers */
170 mask = 0x1f00000000ull;
171 bist_val = read_octeon_c0_icacheerr();
172 if (bist_val & mask)
173 pr_err("Core%d BIST Failure: CacheErr(icache) = 0x%llx\n",
174 coreid, bist_val);
175
176 bist_val = read_octeon_c0_dcacheerr();
177 if (bist_val & 1)
178 pr_err("Core%d L1 Dcache parity error: "
179 "CacheErr(dcache) = 0x%llx\n",
180 coreid, bist_val);
181
182 mask = 0xfc00000000000000ull;
183 bist_val = read_c0_cvmmemctl();
184 if (bist_val & mask)
185 pr_err("Core%d BIST Failure: COP0_CVM_MEM_CTL = 0x%llx\n",
186 coreid, bist_val);
187
188 write_octeon_c0_dcacheerr(0);
189}
190
191#ifdef CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB
192/**
193 * Called on every core to setup the wired tlb entry needed
194 * if CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB is set.
195 *
196 */
197static void octeon_hal_setup_per_cpu_reserved32(void *unused)
198{
199 /*
200 * The config has selected to wire the reserve32 memory for all
201 * userspace applications. We need to put a wired TLB entry in for each
202 * 512MB of reserve32 memory. We only handle double 256MB pages here,
203 * so reserve32 must be multiple of 512MB.
204 */
205 uint32_t size = CONFIG_CAVIUM_RESERVE32;
206 uint32_t entrylo0 =
207 0x7 | ((octeon_reserve32_memory & ((1ul << 40) - 1)) >> 6);
208 uint32_t entrylo1 = entrylo0 + (256 << 14);
209 uint32_t entryhi = (0x80000000UL - (CONFIG_CAVIUM_RESERVE32 << 20));
210 while (size >= 512) {
211#if 0
212 pr_info("CPU%d: Adding double wired TLB entry for 0x%lx\n",
213 smp_processor_id(), entryhi);
214#endif
215 add_wired_entry(entrylo0, entrylo1, entryhi, PM_256M);
216 entrylo0 += 512 << 14;
217 entrylo1 += 512 << 14;
218 entryhi += 512 << 20;
219 size -= 512;
220 }
221}
222#endif /* CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB */
223
224/**
225 * Called to release the named block which was used to made sure
226 * that nobody used the memory for something else during
227 * init. Now we'll free it so userspace apps can use this
228 * memory region with bootmem_alloc.
229 *
230 * This function is called only once from prom_free_prom_memory().
231 */
232void octeon_hal_setup_reserved32(void)
233{
234#ifdef CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB
235 on_each_cpu(octeon_hal_setup_per_cpu_reserved32, NULL, 0, 1);
236#endif
237}
238
239/**
240 * Reboot Octeon
241 *
242 * @command: Command to pass to the bootloader. Currently ignored.
243 */
244static void octeon_restart(char *command)
245{
246 /* Disable all watchdogs before soft reset. They don't get cleared */
247#ifdef CONFIG_SMP
248 int cpu;
249 for_each_online_cpu(cpu)
250 cvmx_write_csr(CVMX_CIU_WDOGX(cpu_logical_map(cpu)), 0);
251#else
252 cvmx_write_csr(CVMX_CIU_WDOGX(cvmx_get_core_num()), 0);
253#endif
254
255 mb();
256 while (1)
257 cvmx_write_csr(CVMX_CIU_SOFT_RST, 1);
258}
259
260
261/**
262 * Permanently stop a core.
263 *
264 * @arg: Ignored.
265 */
266static void octeon_kill_core(void *arg)
267{
268 mb();
269 if (octeon_is_simulation()) {
270 /* The simulator needs the watchdog to stop for dead cores */
271 cvmx_write_csr(CVMX_CIU_WDOGX(cvmx_get_core_num()), 0);
272 /* A break instruction causes the simulator stop a core */
273 asm volatile ("sync\nbreak");
274 }
275}
276
277
278/**
279 * Halt the system
280 */
281static void octeon_halt(void)
282{
283 smp_call_function(octeon_kill_core, NULL, 0);
284
285 switch (octeon_bootinfo->board_type) {
286 case CVMX_BOARD_TYPE_NAO38:
287 /* Driving a 1 to GPIO 12 shuts off this board */
288 cvmx_write_csr(CVMX_GPIO_BIT_CFGX(12), 1);
289 cvmx_write_csr(CVMX_GPIO_TX_SET, 0x1000);
290 break;
291 default:
292 octeon_write_lcd("PowerOff");
293 break;
294 }
295
296 octeon_kill_core(NULL);
297}
298
299#if 0
300/**
301 * Platform time init specifics.
302 * Returns
303 */
304void __init plat_time_init(void)
305{
306 /* Nothing special here, but we are required to have one */
307}
308
309#endif
310
311/**
312 * Handle all the error condition interrupts that might occur.
313 *
314 */
315#ifdef CONFIG_CAVIUM_DECODE_RSL
316static irqreturn_t octeon_rlm_interrupt(int cpl, void *dev_id)
317{
318 cvmx_interrupt_rsl_decode();
319 return IRQ_HANDLED;
320}
321#endif
322
323/**
324 * Return a string representing the system type
325 *
326 * Returns
327 */
328const char *octeon_board_type_string(void)
329{
330 static char name[80];
331 sprintf(name, "%s (%s)",
332 cvmx_board_type_to_string(octeon_bootinfo->board_type),
333 octeon_model_get_string(read_c0_prid()));
334 return name;
335}
336
337const char *get_system_type(void)
338 __attribute__ ((alias("octeon_board_type_string")));
339
340void octeon_user_io_init(void)
341{
342 union octeon_cvmemctl cvmmemctl;
343 union cvmx_iob_fau_timeout fau_timeout;
344 union cvmx_pow_nw_tim nm_tim;
345 uint64_t cvmctl;
346
347 /* Get the current settings for CP0_CVMMEMCTL_REG */
348 cvmmemctl.u64 = read_c0_cvmmemctl();
349 /* R/W If set, marked write-buffer entries time out the same
350 * as as other entries; if clear, marked write-buffer entries
351 * use the maximum timeout. */
352 cvmmemctl.s.dismarkwblongto = 1;
353 /* R/W If set, a merged store does not clear the write-buffer
354 * entry timeout state. */
355 cvmmemctl.s.dismrgclrwbto = 0;
356 /* R/W Two bits that are the MSBs of the resultant CVMSEG LM
357 * word location for an IOBDMA. The other 8 bits come from the
358 * SCRADDR field of the IOBDMA. */
359 cvmmemctl.s.iobdmascrmsb = 0;
360 /* R/W If set, SYNCWS and SYNCS only order marked stores; if
361 * clear, SYNCWS and SYNCS only order unmarked
362 * stores. SYNCWSMARKED has no effect when DISSYNCWS is
363 * set. */
364 cvmmemctl.s.syncwsmarked = 0;
365 /* R/W If set, SYNCWS acts as SYNCW and SYNCS acts as SYNC. */
366 cvmmemctl.s.dissyncws = 0;
367 /* R/W If set, no stall happens on write buffer full. */
368 if (OCTEON_IS_MODEL(OCTEON_CN38XX_PASS2))
369 cvmmemctl.s.diswbfst = 1;
370 else
371 cvmmemctl.s.diswbfst = 0;
372 /* R/W If set (and SX set), supervisor-level loads/stores can
373 * use XKPHYS addresses with <48>==0 */
374 cvmmemctl.s.xkmemenas = 0;
375
376 /* R/W If set (and UX set), user-level loads/stores can use
377 * XKPHYS addresses with VA<48>==0 */
378 cvmmemctl.s.xkmemenau = 0;
379
380 /* R/W If set (and SX set), supervisor-level loads/stores can
381 * use XKPHYS addresses with VA<48>==1 */
382 cvmmemctl.s.xkioenas = 0;
383
384 /* R/W If set (and UX set), user-level loads/stores can use
385 * XKPHYS addresses with VA<48>==1 */
386 cvmmemctl.s.xkioenau = 0;
387
388 /* R/W If set, all stores act as SYNCW (NOMERGE must be set
389 * when this is set) RW, reset to 0. */
390 cvmmemctl.s.allsyncw = 0;
391
392 /* R/W If set, no stores merge, and all stores reach the
393 * coherent bus in order. */
394 cvmmemctl.s.nomerge = 0;
395 /* R/W Selects the bit in the counter used for DID time-outs 0
396 * = 231, 1 = 230, 2 = 229, 3 = 214. Actual time-out is
397 * between 1x and 2x this interval. For example, with
398 * DIDTTO=3, expiration interval is between 16K and 32K. */
399 cvmmemctl.s.didtto = 0;
400 /* R/W If set, the (mem) CSR clock never turns off. */
401 cvmmemctl.s.csrckalwys = 0;
402 /* R/W If set, mclk never turns off. */
403 cvmmemctl.s.mclkalwys = 0;
404 /* R/W Selects the bit in the counter used for write buffer
405 * flush time-outs (WBFLT+11) is the bit position in an
406 * internal counter used to determine expiration. The write
407 * buffer expires between 1x and 2x this interval. For
408 * example, with WBFLT = 0, a write buffer expires between 2K
409 * and 4K cycles after the write buffer entry is allocated. */
410 cvmmemctl.s.wbfltime = 0;
411 /* R/W If set, do not put Istream in the L2 cache. */
412 cvmmemctl.s.istrnol2 = 0;
413 /* R/W The write buffer threshold. */
414 cvmmemctl.s.wbthresh = 10;
415 /* R/W If set, CVMSEG is available for loads/stores in
416 * kernel/debug mode. */
417#if CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE > 0
418 cvmmemctl.s.cvmsegenak = 1;
419#else
420 cvmmemctl.s.cvmsegenak = 0;
421#endif
422 /* R/W If set, CVMSEG is available for loads/stores in
423 * supervisor mode. */
424 cvmmemctl.s.cvmsegenas = 0;
425 /* R/W If set, CVMSEG is available for loads/stores in user
426 * mode. */
427 cvmmemctl.s.cvmsegenau = 0;
428 /* R/W Size of local memory in cache blocks, 54 (6912 bytes)
429 * is max legal value. */
430 cvmmemctl.s.lmemsz = CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE;
431
432
433 if (smp_processor_id() == 0)
434 pr_notice("CVMSEG size: %d cache lines (%d bytes)\n",
435 CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE,
436 CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE * 128);
437
438 write_c0_cvmmemctl(cvmmemctl.u64);
439
440 /* Move the performance counter interrupts to IRQ 6 */
441 cvmctl = read_c0_cvmctl();
442 cvmctl &= ~(7 << 7);
443 cvmctl |= 6 << 7;
444 write_c0_cvmctl(cvmctl);
445
446 /* Set a default for the hardware timeouts */
447 fau_timeout.u64 = 0;
448 fau_timeout.s.tout_val = 0xfff;
449 /* Disable tagwait FAU timeout */
450 fau_timeout.s.tout_enb = 0;
451 cvmx_write_csr(CVMX_IOB_FAU_TIMEOUT, fau_timeout.u64);
452
453 nm_tim.u64 = 0;
454 /* 4096 cycles */
455 nm_tim.s.nw_tim = 3;
456 cvmx_write_csr(CVMX_POW_NW_TIM, nm_tim.u64);
457
458 write_octeon_c0_icacheerr(0);
459 write_c0_derraddr1(0);
460}
461
462/**
463 * Early entry point for arch setup
464 */
465void __init prom_init(void)
466{
467 struct cvmx_sysinfo *sysinfo;
468 const int coreid = cvmx_get_core_num();
469 int i;
470 int argc;
471 struct uart_port octeon_port;
472#ifdef CONFIG_CAVIUM_RESERVE32
473 int64_t addr = -1;
474#endif
475 /*
476 * The bootloader passes a pointer to the boot descriptor in
477 * $a3, this is available as fw_arg3.
478 */
479 octeon_boot_desc_ptr = (struct octeon_boot_descriptor *)fw_arg3;
480 octeon_bootinfo =
481 cvmx_phys_to_ptr(octeon_boot_desc_ptr->cvmx_desc_vaddr);
482 cvmx_bootmem_init(cvmx_phys_to_ptr(octeon_bootinfo->phy_mem_desc_addr));
483
484 /*
485 * Only enable the LED controller if we're running on a CN38XX, CN58XX,
486 * or CN56XX. The CN30XX and CN31XX don't have an LED controller.
487 */
488 if (!octeon_is_simulation() &&
489 octeon_has_feature(OCTEON_FEATURE_LED_CONTROLLER)) {
490 cvmx_write_csr(CVMX_LED_EN, 0);
491 cvmx_write_csr(CVMX_LED_PRT, 0);
492 cvmx_write_csr(CVMX_LED_DBG, 0);
493 cvmx_write_csr(CVMX_LED_PRT_FMT, 0);
494 cvmx_write_csr(CVMX_LED_UDD_CNTX(0), 32);
495 cvmx_write_csr(CVMX_LED_UDD_CNTX(1), 32);
496 cvmx_write_csr(CVMX_LED_UDD_DATX(0), 0);
497 cvmx_write_csr(CVMX_LED_UDD_DATX(1), 0);
498 cvmx_write_csr(CVMX_LED_EN, 1);
499 }
500#ifdef CONFIG_CAVIUM_RESERVE32
501 /*
502 * We need to temporarily allocate all memory in the reserve32
503 * region. This makes sure the kernel doesn't allocate this
504 * memory when it is getting memory from the
505 * bootloader. Later, after the memory allocations are
506 * complete, the reserve32 will be freed.
507 */
508#ifdef CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB
509 if (CONFIG_CAVIUM_RESERVE32 & 0x1ff)
510 pr_err("CAVIUM_RESERVE32 isn't a multiple of 512MB. "
511 "This is required if CAVIUM_RESERVE32_USE_WIRED_TLB "
512 "is set\n");
513 else
514 addr = cvmx_bootmem_phy_named_block_alloc(CONFIG_CAVIUM_RESERVE32 << 20,
515 0, 0, 512 << 20,
516 "CAVIUM_RESERVE32", 0);
517#else
518 /*
519 * Allocate memory for RESERVED32 aligned on 2MB boundary. This
520 * is in case we later use hugetlb entries with it.
521 */
522 addr = cvmx_bootmem_phy_named_block_alloc(CONFIG_CAVIUM_RESERVE32 << 20,
523 0, 0, 2 << 20,
524 "CAVIUM_RESERVE32", 0);
525#endif
526 if (addr < 0)
527 pr_err("Failed to allocate CAVIUM_RESERVE32 memory area\n");
528 else
529 octeon_reserve32_memory = addr;
530#endif
531
532#ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2
533 if (cvmx_read_csr(CVMX_L2D_FUS3) & (3ull << 34)) {
534 pr_info("Skipping L2 locking due to reduced L2 cache size\n");
535 } else {
536 uint32_t ebase = read_c0_ebase() & 0x3ffff000;
537#ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2_TLB
538 /* TLB refill */
539 cvmx_l2c_lock_mem_region(ebase, 0x100);
540#endif
541#ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2_EXCEPTION
542 /* General exception */
543 cvmx_l2c_lock_mem_region(ebase + 0x180, 0x80);
544#endif
545#ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2_LOW_LEVEL_INTERRUPT
546 /* Interrupt handler */
547 cvmx_l2c_lock_mem_region(ebase + 0x200, 0x80);
548#endif
549#ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2_INTERRUPT
550 cvmx_l2c_lock_mem_region(__pa_symbol(handle_int), 0x100);
551 cvmx_l2c_lock_mem_region(__pa_symbol(plat_irq_dispatch), 0x80);
552#endif
553#ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2_MEMCPY
554 cvmx_l2c_lock_mem_region(__pa_symbol(memcpy), 0x480);
555#endif
556 }
557#endif
558
559 sysinfo = cvmx_sysinfo_get();
560 memset(sysinfo, 0, sizeof(*sysinfo));
561 sysinfo->system_dram_size = octeon_bootinfo->dram_size << 20;
562 sysinfo->phy_mem_desc_ptr =
563 cvmx_phys_to_ptr(octeon_bootinfo->phy_mem_desc_addr);
564 sysinfo->core_mask = octeon_bootinfo->core_mask;
565 sysinfo->exception_base_addr = octeon_bootinfo->exception_base_addr;
566 sysinfo->cpu_clock_hz = octeon_bootinfo->eclock_hz;
567 sysinfo->dram_data_rate_hz = octeon_bootinfo->dclock_hz * 2;
568 sysinfo->board_type = octeon_bootinfo->board_type;
569 sysinfo->board_rev_major = octeon_bootinfo->board_rev_major;
570 sysinfo->board_rev_minor = octeon_bootinfo->board_rev_minor;
571 memcpy(sysinfo->mac_addr_base, octeon_bootinfo->mac_addr_base,
572 sizeof(sysinfo->mac_addr_base));
573 sysinfo->mac_addr_count = octeon_bootinfo->mac_addr_count;
574 memcpy(sysinfo->board_serial_number,
575 octeon_bootinfo->board_serial_number,
576 sizeof(sysinfo->board_serial_number));
577 sysinfo->compact_flash_common_base_addr =
578 octeon_bootinfo->compact_flash_common_base_addr;
579 sysinfo->compact_flash_attribute_base_addr =
580 octeon_bootinfo->compact_flash_attribute_base_addr;
581 sysinfo->led_display_base_addr = octeon_bootinfo->led_display_base_addr;
582 sysinfo->dfa_ref_clock_hz = octeon_bootinfo->dfa_ref_clock_hz;
583 sysinfo->bootloader_config_flags = octeon_bootinfo->config_flags;
584
585
586 octeon_check_cpu_bist();
587
588 octeon_uart = octeon_get_boot_uart();
589
590 /*
591 * Disable All CIU Interrupts. The ones we need will be
592 * enabled later. Read the SUM register so we know the write
593 * completed.
594 */
595 cvmx_write_csr(CVMX_CIU_INTX_EN0((coreid * 2)), 0);
596 cvmx_write_csr(CVMX_CIU_INTX_EN0((coreid * 2 + 1)), 0);
597 cvmx_write_csr(CVMX_CIU_INTX_EN1((coreid * 2)), 0);
598 cvmx_write_csr(CVMX_CIU_INTX_EN1((coreid * 2 + 1)), 0);
599 cvmx_read_csr(CVMX_CIU_INTX_SUM0((coreid * 2)));
600
601#ifdef CONFIG_SMP
602 octeon_write_lcd("LinuxSMP");
603#else
604 octeon_write_lcd("Linux");
605#endif
606
607#ifdef CONFIG_CAVIUM_GDB
608 /*
609 * When debugging the linux kernel, force the cores to enter
610 * the debug exception handler to break in.
611 */
612 if (octeon_get_boot_debug_flag()) {
613 cvmx_write_csr(CVMX_CIU_DINT, 1 << cvmx_get_core_num());
614 cvmx_read_csr(CVMX_CIU_DINT);
615 }
616#endif
617
618 /*
619 * BIST should always be enabled when doing a soft reset. L2
620 * Cache locking for instance is not cleared unless BIST is
621 * enabled. Unfortunately due to a chip errata G-200 for
622 * Cn38XX and CN31XX, BIST msut be disabled on these parts.
623 */
624 if (OCTEON_IS_MODEL(OCTEON_CN38XX_PASS2) ||
625 OCTEON_IS_MODEL(OCTEON_CN31XX))
626 cvmx_write_csr(CVMX_CIU_SOFT_BIST, 0);
627 else
628 cvmx_write_csr(CVMX_CIU_SOFT_BIST, 1);
629
630 /* Default to 64MB in the simulator to speed things up */
631 if (octeon_is_simulation())
632 MAX_MEMORY = 64ull << 20;
633
634 arcs_cmdline[0] = 0;
635 argc = octeon_boot_desc_ptr->argc;
636 for (i = 0; i < argc; i++) {
637 const char *arg =
638 cvmx_phys_to_ptr(octeon_boot_desc_ptr->argv[i]);
639 if ((strncmp(arg, "MEM=", 4) == 0) ||
640 (strncmp(arg, "mem=", 4) == 0)) {
641 sscanf(arg + 4, "%llu", &MAX_MEMORY);
642 MAX_MEMORY <<= 20;
643 if (MAX_MEMORY == 0)
644 MAX_MEMORY = 32ull << 30;
645 } else if (strcmp(arg, "ecc_verbose") == 0) {
646#ifdef CONFIG_CAVIUM_REPORT_SINGLE_BIT_ECC
647 __cvmx_interrupt_ecc_report_single_bit_errors = 1;
648 pr_notice("Reporting of single bit ECC errors is "
649 "turned on\n");
650#endif
651 } else if (strlen(arcs_cmdline) + strlen(arg) + 1 <
652 sizeof(arcs_cmdline) - 1) {
653 strcat(arcs_cmdline, " ");
654 strcat(arcs_cmdline, arg);
655 }
656 }
657
658 if (strstr(arcs_cmdline, "console=") == NULL) {
659#ifdef CONFIG_GDB_CONSOLE
660 strcat(arcs_cmdline, " console=gdb");
661#else
662#ifdef CONFIG_CAVIUM_OCTEON_2ND_KERNEL
663 strcat(arcs_cmdline, " console=ttyS0,115200");
664#else
665 if (octeon_uart == 1)
666 strcat(arcs_cmdline, " console=ttyS1,115200");
667 else
668 strcat(arcs_cmdline, " console=ttyS0,115200");
669#endif
670#endif
671 }
672
673 if (octeon_is_simulation()) {
674 /*
675 * The simulator uses a mtdram device pre filled with
676 * the filesystem. Also specify the calibration delay
677 * to avoid calculating it every time.
678 */
679 strcat(arcs_cmdline, " rw root=1f00"
680 " lpj=60176 slram=root,0x40000000,+1073741824");
681 }
682
683 mips_hpt_frequency = octeon_get_clock_rate();
684
685 octeon_init_cvmcount();
686
687 _machine_restart = octeon_restart;
688 _machine_halt = octeon_halt;
689
690 memset(&octeon_port, 0, sizeof(octeon_port));
691 /*
692 * For early_serial_setup we don't set the port type or
693 * UPF_FIXED_TYPE.
694 */
695 octeon_port.flags = ASYNC_SKIP_TEST | UPF_SHARE_IRQ;
696 octeon_port.iotype = UPIO_MEM;
697 /* I/O addresses are every 8 bytes */
698 octeon_port.regshift = 3;
699 /* Clock rate of the chip */
700 octeon_port.uartclk = mips_hpt_frequency;
701 octeon_port.fifosize = 64;
702 octeon_port.mapbase = 0x0001180000000800ull + (1024 * octeon_uart);
703 octeon_port.membase = cvmx_phys_to_ptr(octeon_port.mapbase);
704 octeon_port.serial_in = octeon_serial_in;
705 octeon_port.serial_out = octeon_serial_out;
706#ifdef CONFIG_CAVIUM_OCTEON_2ND_KERNEL
707 octeon_port.line = 0;
708#else
709 octeon_port.line = octeon_uart;
710#endif
711 octeon_port.irq = 42 + octeon_uart;
712 early_serial_setup(&octeon_port);
713
714 octeon_user_io_init();
715 register_smp_ops(&octeon_smp_ops);
716}
717
718void __init plat_mem_setup(void)
719{
720 uint64_t mem_alloc_size;
721 uint64_t total;
722 int64_t memory;
723
724 total = 0;
725
726 /* First add the init memory we will be returning. */
727 memory = __pa_symbol(&__init_begin) & PAGE_MASK;
728 mem_alloc_size = (__pa_symbol(&__init_end) & PAGE_MASK) - memory;
729 if (mem_alloc_size > 0) {
730 add_memory_region(memory, mem_alloc_size, BOOT_MEM_RAM);
731 total += mem_alloc_size;
732 }
733
734 /*
735 * The Mips memory init uses the first memory location for
736 * some memory vectors. When SPARSEMEM is in use, it doesn't
737 * verify that the size is big enough for the final
738 * vectors. Making the smallest chuck 4MB seems to be enough
739 * to consistantly work.
740 */
741 mem_alloc_size = 4 << 20;
742 if (mem_alloc_size > MAX_MEMORY)
743 mem_alloc_size = MAX_MEMORY;
744
745 /*
746 * When allocating memory, we want incrementing addresses from
747 * bootmem_alloc so the code in add_memory_region can merge
748 * regions next to each other.
749 */
750 cvmx_bootmem_lock();
751 while ((boot_mem_map.nr_map < BOOT_MEM_MAP_MAX)
752 && (total < MAX_MEMORY)) {
753#if defined(CONFIG_64BIT) || defined(CONFIG_64BIT_PHYS_ADDR)
754 memory = cvmx_bootmem_phy_alloc(mem_alloc_size,
755 __pa_symbol(&__init_end), -1,
756 0x100000,
757 CVMX_BOOTMEM_FLAG_NO_LOCKING);
758#elif defined(CONFIG_HIGHMEM)
759 memory = cvmx_bootmem_phy_alloc(mem_alloc_size, 0, 1ull << 31,
760 0x100000,
761 CVMX_BOOTMEM_FLAG_NO_LOCKING);
762#else
763 memory = cvmx_bootmem_phy_alloc(mem_alloc_size, 0, 512 << 20,
764 0x100000,
765 CVMX_BOOTMEM_FLAG_NO_LOCKING);
766#endif
767 if (memory >= 0) {
768 /*
769 * This function automatically merges address
770 * regions next to each other if they are
771 * received in incrementing order.
772 */
773 add_memory_region(memory, mem_alloc_size, BOOT_MEM_RAM);
774 total += mem_alloc_size;
775 } else {
776 break;
777 }
778 }
779 cvmx_bootmem_unlock();
780
781#ifdef CONFIG_CAVIUM_RESERVE32
782 /*
783 * Now that we've allocated the kernel memory it is safe to
784 * free the reserved region. We free it here so that builtin
785 * drivers can use the memory.
786 */
787 if (octeon_reserve32_memory)
788 cvmx_bootmem_free_named("CAVIUM_RESERVE32");
789#endif /* CONFIG_CAVIUM_RESERVE32 */
790
791 if (total == 0)
792 panic("Unable to allocate memory from "
793 "cvmx_bootmem_phy_alloc\n");
794}
795
796
797int prom_putchar(char c)
798{
799 uint64_t lsrval;
800
801 /* Spin until there is room */
802 do {
803 lsrval = cvmx_read_csr(CVMX_MIO_UARTX_LSR(octeon_uart));
804 } while ((lsrval & 0x20) == 0);
805
806 /* Write the byte */
807 cvmx_write_csr(CVMX_MIO_UARTX_THR(octeon_uart), c);
808 return 1;
809}
810
811void prom_free_prom_memory(void)
812{
813#ifdef CONFIG_CAVIUM_DECODE_RSL
814 cvmx_interrupt_rsl_enable();
815
816 /* Add an interrupt handler for general failures. */
817 if (request_irq(OCTEON_IRQ_RML, octeon_rlm_interrupt, IRQF_SHARED,
818 "RML/RSL", octeon_rlm_interrupt)) {
819 panic("Unable to request_irq(OCTEON_IRQ_RML)\n");
820 }
821#endif
822
823 /* This call is here so that it is performed after any TLB
824 initializations. It needs to be after these in case the
825 CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB option is set */
826 octeon_hal_setup_reserved32();
827}
828
829static struct octeon_cf_data octeon_cf_data;
830
831static int __init octeon_cf_device_init(void)
832{
833 union cvmx_mio_boot_reg_cfgx mio_boot_reg_cfg;
834 unsigned long base_ptr, region_base, region_size;
835 struct platform_device *pd;
836 struct resource cf_resources[3];
837 unsigned int num_resources;
838 int i;
839 int ret = 0;
840
841 /* Setup octeon-cf platform device if present. */
842 base_ptr = 0;
843 if (octeon_bootinfo->major_version == 1
844 && octeon_bootinfo->minor_version >= 1) {
845 if (octeon_bootinfo->compact_flash_common_base_addr)
846 base_ptr =
847 octeon_bootinfo->compact_flash_common_base_addr;
848 } else {
849 base_ptr = 0x1d000800;
850 }
851
852 if (!base_ptr)
853 return ret;
854
855 /* Find CS0 region. */
856 for (i = 0; i < 8; i++) {
857 mio_boot_reg_cfg.u64 = cvmx_read_csr(CVMX_MIO_BOOT_REG_CFGX(i));
858 region_base = mio_boot_reg_cfg.s.base << 16;
859 region_size = (mio_boot_reg_cfg.s.size + 1) << 16;
860 if (mio_boot_reg_cfg.s.en && base_ptr >= region_base
861 && base_ptr < region_base + region_size)
862 break;
863 }
864 if (i >= 7) {
865 /* i and i + 1 are CS0 and CS1, both must be less than 8. */
866 goto out;
867 }
868 octeon_cf_data.base_region = i;
869 octeon_cf_data.is16bit = mio_boot_reg_cfg.s.width;
870 octeon_cf_data.base_region_bias = base_ptr - region_base;
871 memset(cf_resources, 0, sizeof(cf_resources));
872 num_resources = 0;
873 cf_resources[num_resources].flags = IORESOURCE_MEM;
874 cf_resources[num_resources].start = region_base;
875 cf_resources[num_resources].end = region_base + region_size - 1;
876 num_resources++;
877
878
879 if (!(base_ptr & 0xfffful)) {
880 /*
881 * Boot loader signals availability of DMA (true_ide
882 * mode) by setting low order bits of base_ptr to
883 * zero.
884 */
885
886 /* Asume that CS1 immediately follows. */
887 mio_boot_reg_cfg.u64 =
888 cvmx_read_csr(CVMX_MIO_BOOT_REG_CFGX(i + 1));
889 region_base = mio_boot_reg_cfg.s.base << 16;
890 region_size = (mio_boot_reg_cfg.s.size + 1) << 16;
891 if (!mio_boot_reg_cfg.s.en)
892 goto out;
893
894 cf_resources[num_resources].flags = IORESOURCE_MEM;
895 cf_resources[num_resources].start = region_base;
896 cf_resources[num_resources].end = region_base + region_size - 1;
897 num_resources++;
898
899 octeon_cf_data.dma_engine = 0;
900 cf_resources[num_resources].flags = IORESOURCE_IRQ;
901 cf_resources[num_resources].start = OCTEON_IRQ_BOOTDMA;
902 cf_resources[num_resources].end = OCTEON_IRQ_BOOTDMA;
903 num_resources++;
904 } else {
905 octeon_cf_data.dma_engine = -1;
906 }
907
908 pd = platform_device_alloc("pata_octeon_cf", -1);
909 if (!pd) {
910 ret = -ENOMEM;
911 goto out;
912 }
913 pd->dev.platform_data = &octeon_cf_data;
914
915 ret = platform_device_add_resources(pd, cf_resources, num_resources);
916 if (ret)
917 goto fail;
918
919 ret = platform_device_add(pd);
920 if (ret)
921 goto fail;
922
923 return ret;
924fail:
925 platform_device_put(pd);
926out:
927 return ret;
928}
929device_initcall(octeon_cf_device_init);
diff --git a/arch/mips/cavium-octeon/smp.c b/arch/mips/cavium-octeon/smp.c
new file mode 100644
index 000000000000..24e0ad63980a
--- /dev/null
+++ b/arch/mips/cavium-octeon/smp.c
@@ -0,0 +1,211 @@
1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Copyright (C) 2004-2008 Cavium Networks
7 */
8#include <linux/init.h>
9#include <linux/delay.h>
10#include <linux/smp.h>
11#include <linux/interrupt.h>
12#include <linux/kernel_stat.h>
13#include <linux/sched.h>
14#include <linux/module.h>
15
16#include <asm/mmu_context.h>
17#include <asm/system.h>
18#include <asm/time.h>
19
20#include <asm/octeon/octeon.h>
21
22volatile unsigned long octeon_processor_boot = 0xff;
23volatile unsigned long octeon_processor_sp;
24volatile unsigned long octeon_processor_gp;
25
26static irqreturn_t mailbox_interrupt(int irq, void *dev_id)
27{
28 const int coreid = cvmx_get_core_num();
29 uint64_t action;
30
31 /* Load the mailbox register to figure out what we're supposed to do */
32 action = cvmx_read_csr(CVMX_CIU_MBOX_CLRX(coreid));
33
34 /* Clear the mailbox to clear the interrupt */
35 cvmx_write_csr(CVMX_CIU_MBOX_CLRX(coreid), action);
36
37 if (action & SMP_CALL_FUNCTION)
38 smp_call_function_interrupt();
39
40 /* Check if we've been told to flush the icache */
41 if (action & SMP_ICACHE_FLUSH)
42 asm volatile ("synci 0($0)\n");
43 return IRQ_HANDLED;
44}
45
46/**
47 * Cause the function described by call_data to be executed on the passed
48 * cpu. When the function has finished, increment the finished field of
49 * call_data.
50 */
51void octeon_send_ipi_single(int cpu, unsigned int action)
52{
53 int coreid = cpu_logical_map(cpu);
54 /*
55 pr_info("SMP: Mailbox send cpu=%d, coreid=%d, action=%u\n", cpu,
56 coreid, action);
57 */
58 cvmx_write_csr(CVMX_CIU_MBOX_SETX(coreid), action);
59}
60
61static inline void octeon_send_ipi_mask(cpumask_t mask, unsigned int action)
62{
63 unsigned int i;
64
65 for_each_cpu_mask(i, mask)
66 octeon_send_ipi_single(i, action);
67}
68
69/**
70 * Detect available CPUs, populate phys_cpu_present_map
71 */
72static void octeon_smp_setup(void)
73{
74 const int coreid = cvmx_get_core_num();
75 int cpus;
76 int id;
77
78 int core_mask = octeon_get_boot_coremask();
79
80 cpus_clear(cpu_possible_map);
81 __cpu_number_map[coreid] = 0;
82 __cpu_logical_map[0] = coreid;
83 cpu_set(0, cpu_possible_map);
84
85 cpus = 1;
86 for (id = 0; id < 16; id++) {
87 if ((id != coreid) && (core_mask & (1 << id))) {
88 cpu_set(cpus, cpu_possible_map);
89 __cpu_number_map[id] = cpus;
90 __cpu_logical_map[cpus] = id;
91 cpus++;
92 }
93 }
94}
95
96/**
97 * Firmware CPU startup hook
98 *
99 */
100static void octeon_boot_secondary(int cpu, struct task_struct *idle)
101{
102 int count;
103
104 pr_info("SMP: Booting CPU%02d (CoreId %2d)...\n", cpu,
105 cpu_logical_map(cpu));
106
107 octeon_processor_sp = __KSTK_TOS(idle);
108 octeon_processor_gp = (unsigned long)(task_thread_info(idle));
109 octeon_processor_boot = cpu_logical_map(cpu);
110 mb();
111
112 count = 10000;
113 while (octeon_processor_sp && count) {
114 /* Waiting for processor to get the SP and GP */
115 udelay(1);
116 count--;
117 }
118 if (count == 0)
119 pr_err("Secondary boot timeout\n");
120}
121
122/**
123 * After we've done initial boot, this function is called to allow the
124 * board code to clean up state, if needed
125 */
126static void octeon_init_secondary(void)
127{
128 const int coreid = cvmx_get_core_num();
129 union cvmx_ciu_intx_sum0 interrupt_enable;
130
131 octeon_check_cpu_bist();
132 octeon_init_cvmcount();
133 /*
134 pr_info("SMP: CPU%d (CoreId %lu) started\n", cpu, coreid);
135 */
136 /* Enable Mailbox interrupts to this core. These are the only
137 interrupts allowed on line 3 */
138 cvmx_write_csr(CVMX_CIU_MBOX_CLRX(coreid), 0xffffffff);
139 interrupt_enable.u64 = 0;
140 interrupt_enable.s.mbox = 0x3;
141 cvmx_write_csr(CVMX_CIU_INTX_EN0((coreid * 2)), interrupt_enable.u64);
142 cvmx_write_csr(CVMX_CIU_INTX_EN0((coreid * 2 + 1)), 0);
143 cvmx_write_csr(CVMX_CIU_INTX_EN1((coreid * 2)), 0);
144 cvmx_write_csr(CVMX_CIU_INTX_EN1((coreid * 2 + 1)), 0);
145 /* Enable core interrupt processing for 2,3 and 7 */
146 set_c0_status(0x8c01);
147}
148
149/**
150 * Callout to firmware before smp_init
151 *
152 */
153void octeon_prepare_cpus(unsigned int max_cpus)
154{
155 cvmx_write_csr(CVMX_CIU_MBOX_CLRX(cvmx_get_core_num()), 0xffffffff);
156 if (request_irq(OCTEON_IRQ_MBOX0, mailbox_interrupt, IRQF_SHARED,
157 "mailbox0", mailbox_interrupt)) {
158 panic("Cannot request_irq(OCTEON_IRQ_MBOX0)\n");
159 }
160 if (request_irq(OCTEON_IRQ_MBOX1, mailbox_interrupt, IRQF_SHARED,
161 "mailbox1", mailbox_interrupt)) {
162 panic("Cannot request_irq(OCTEON_IRQ_MBOX1)\n");
163 }
164}
165
166/**
167 * Last chance for the board code to finish SMP initialization before
168 * the CPU is "online".
169 */
170static void octeon_smp_finish(void)
171{
172#ifdef CONFIG_CAVIUM_GDB
173 unsigned long tmp;
174 /* Pulse MCD0 signal on Ctrl-C to stop all the cores. Also set the MCD0
175 to be not masked by this core so we know the signal is received by
176 someone */
177 asm volatile ("dmfc0 %0, $22\n"
178 "ori %0, %0, 0x9100\n" "dmtc0 %0, $22\n" : "=r" (tmp));
179#endif
180
181 octeon_user_io_init();
182
183 /* to generate the first CPU timer interrupt */
184 write_c0_compare(read_c0_count() + mips_hpt_frequency / HZ);
185}
186
187/**
188 * Hook for after all CPUs are online
189 */
190static void octeon_cpus_done(void)
191{
192#ifdef CONFIG_CAVIUM_GDB
193 unsigned long tmp;
194 /* Pulse MCD0 signal on Ctrl-C to stop all the cores. Also set the MCD0
195 to be not masked by this core so we know the signal is received by
196 someone */
197 asm volatile ("dmfc0 %0, $22\n"
198 "ori %0, %0, 0x9100\n" "dmtc0 %0, $22\n" : "=r" (tmp));
199#endif
200}
201
202struct plat_smp_ops octeon_smp_ops = {
203 .send_ipi_single = octeon_send_ipi_single,
204 .send_ipi_mask = octeon_send_ipi_mask,
205 .init_secondary = octeon_init_secondary,
206 .smp_finish = octeon_smp_finish,
207 .cpus_done = octeon_cpus_done,
208 .boot_secondary = octeon_boot_secondary,
209 .smp_setup = octeon_smp_setup,
210 .prepare_cpus = octeon_prepare_cpus,
211};
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-27 14:34:50 -0500