diff options
Diffstat (limited to 'arch/mips/cavium-octeon')
-rw-r--r-- | arch/mips/cavium-octeon/Kconfig | 85 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/Makefile | 16 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/csrc-octeon.c | 58 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/dma-octeon.c | 32 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/executive/Makefile | 13 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/executive/cvmx-bootmem.c | 586 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/executive/cvmx-l2c.c | 734 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/executive/cvmx-sysinfo.c | 116 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/executive/octeon-model.c | 358 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/flash_setup.c | 84 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/octeon-irq.c | 497 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/octeon-memcpy.S | 521 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/serial.c | 136 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/setup.c | 927 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/smp.c | 211 |
15 files changed, 4374 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 @@ | |||
1 | config CAVIUM_OCTEON_SPECIFIC_OPTIONS | ||
2 | bool "Enable Octeon specific options" | ||
3 | depends on CPU_CAVIUM_OCTEON | ||
4 | default "y" | ||
5 | |||
6 | config 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 | |||
16 | config 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 | |||
28 | config 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 | |||
40 | config 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 | |||
47 | config 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 | |||
54 | config 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 | |||
61 | config 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 | |||
68 | config 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 | |||
75 | config 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 | |||
82 | config 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 | |||
12 | obj-y := setup.o serial.o octeon-irq.o csrc-octeon.o | ||
13 | obj-y += dma-octeon.o flash_setup.o | ||
14 | obj-y += octeon-memcpy.o | ||
15 | |||
16 | obj-$(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 | */ | ||
23 | void 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 | |||
41 | static cycle_t octeon_cvmcount_read(void) | ||
42 | { | ||
43 | return read_c0_cvmcount(); | ||
44 | } | ||
45 | |||
46 | static 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 | |||
53 | void __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 | |||
19 | dma_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 | |||
27 | void 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/executive/Makefile b/arch/mips/cavium-octeon/executive/Makefile new file mode 100644 index 000000000000..80d6cb26766b --- /dev/null +++ b/arch/mips/cavium-octeon/executive/Makefile | |||
@@ -0,0 +1,13 @@ | |||
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 | |||
12 | obj-y += cvmx-bootmem.o cvmx-l2c.o cvmx-sysinfo.o octeon-model.o | ||
13 | |||
diff --git a/arch/mips/cavium-octeon/executive/cvmx-bootmem.c b/arch/mips/cavium-octeon/executive/cvmx-bootmem.c new file mode 100644 index 000000000000..4f5a08b37ccd --- /dev/null +++ b/arch/mips/cavium-octeon/executive/cvmx-bootmem.c | |||
@@ -0,0 +1,586 @@ | |||
1 | /***********************license start*************** | ||
2 | * Author: Cavium Networks | ||
3 | * | ||
4 | * Contact: support@caviumnetworks.com | ||
5 | * This file is part of the OCTEON SDK | ||
6 | * | ||
7 | * Copyright (c) 2003-2008 Cavium Networks | ||
8 | * | ||
9 | * This file is free software; you can redistribute it and/or modify | ||
10 | * it under the terms of the GNU General Public License, Version 2, as | ||
11 | * published by the Free Software Foundation. | ||
12 | * | ||
13 | * This file is distributed in the hope that it will be useful, but | ||
14 | * AS-IS and WITHOUT ANY WARRANTY; without even the implied warranty | ||
15 | * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE, TITLE, or | ||
16 | * NONINFRINGEMENT. See the GNU General Public License for more | ||
17 | * details. | ||
18 | * | ||
19 | * You should have received a copy of the GNU General Public License | ||
20 | * along with this file; if not, write to the Free Software | ||
21 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | ||
22 | * or visit http://www.gnu.org/licenses/. | ||
23 | * | ||
24 | * This file may also be available under a different license from Cavium. | ||
25 | * Contact Cavium Networks for more information | ||
26 | ***********************license end**************************************/ | ||
27 | |||
28 | /* | ||
29 | * Simple allocate only memory allocator. Used to allocate memory at | ||
30 | * application start time. | ||
31 | */ | ||
32 | |||
33 | #include <linux/kernel.h> | ||
34 | |||
35 | #include <asm/octeon/cvmx.h> | ||
36 | #include <asm/octeon/cvmx-spinlock.h> | ||
37 | #include <asm/octeon/cvmx-bootmem.h> | ||
38 | |||
39 | /*#define DEBUG */ | ||
40 | |||
41 | |||
42 | static struct cvmx_bootmem_desc *cvmx_bootmem_desc; | ||
43 | |||
44 | /* See header file for descriptions of functions */ | ||
45 | |||
46 | /* | ||
47 | * Wrapper functions are provided for reading/writing the size and | ||
48 | * next block values as these may not be directly addressible (in 32 | ||
49 | * bit applications, for instance.) Offsets of data elements in | ||
50 | * bootmem list, must match cvmx_bootmem_block_header_t. | ||
51 | */ | ||
52 | #define NEXT_OFFSET 0 | ||
53 | #define SIZE_OFFSET 8 | ||
54 | |||
55 | static void cvmx_bootmem_phy_set_size(uint64_t addr, uint64_t size) | ||
56 | { | ||
57 | cvmx_write64_uint64((addr + SIZE_OFFSET) | (1ull << 63), size); | ||
58 | } | ||
59 | |||
60 | static void cvmx_bootmem_phy_set_next(uint64_t addr, uint64_t next) | ||
61 | { | ||
62 | cvmx_write64_uint64((addr + NEXT_OFFSET) | (1ull << 63), next); | ||
63 | } | ||
64 | |||
65 | static uint64_t cvmx_bootmem_phy_get_size(uint64_t addr) | ||
66 | { | ||
67 | return cvmx_read64_uint64((addr + SIZE_OFFSET) | (1ull << 63)); | ||
68 | } | ||
69 | |||
70 | static uint64_t cvmx_bootmem_phy_get_next(uint64_t addr) | ||
71 | { | ||
72 | return cvmx_read64_uint64((addr + NEXT_OFFSET) | (1ull << 63)); | ||
73 | } | ||
74 | |||
75 | void *cvmx_bootmem_alloc_range(uint64_t size, uint64_t alignment, | ||
76 | uint64_t min_addr, uint64_t max_addr) | ||
77 | { | ||
78 | int64_t address; | ||
79 | address = | ||
80 | cvmx_bootmem_phy_alloc(size, min_addr, max_addr, alignment, 0); | ||
81 | |||
82 | if (address > 0) | ||
83 | return cvmx_phys_to_ptr(address); | ||
84 | else | ||
85 | return NULL; | ||
86 | } | ||
87 | |||
88 | void *cvmx_bootmem_alloc_address(uint64_t size, uint64_t address, | ||
89 | uint64_t alignment) | ||
90 | { | ||
91 | return cvmx_bootmem_alloc_range(size, alignment, address, | ||
92 | address + size); | ||
93 | } | ||
94 | |||
95 | void *cvmx_bootmem_alloc(uint64_t size, uint64_t alignment) | ||
96 | { | ||
97 | return cvmx_bootmem_alloc_range(size, alignment, 0, 0); | ||
98 | } | ||
99 | |||
100 | int cvmx_bootmem_free_named(char *name) | ||
101 | { | ||
102 | return cvmx_bootmem_phy_named_block_free(name, 0); | ||
103 | } | ||
104 | |||
105 | struct cvmx_bootmem_named_block_desc *cvmx_bootmem_find_named_block(char *name) | ||
106 | { | ||
107 | return cvmx_bootmem_phy_named_block_find(name, 0); | ||
108 | } | ||
109 | |||
110 | void cvmx_bootmem_lock(void) | ||
111 | { | ||
112 | cvmx_spinlock_lock((cvmx_spinlock_t *) &(cvmx_bootmem_desc->lock)); | ||
113 | } | ||
114 | |||
115 | void cvmx_bootmem_unlock(void) | ||
116 | { | ||
117 | cvmx_spinlock_unlock((cvmx_spinlock_t *) &(cvmx_bootmem_desc->lock)); | ||
118 | } | ||
119 | |||
120 | int cvmx_bootmem_init(void *mem_desc_ptr) | ||
121 | { | ||
122 | /* Here we set the global pointer to the bootmem descriptor | ||
123 | * block. This pointer will be used directly, so we will set | ||
124 | * it up to be directly usable by the application. It is set | ||
125 | * up as follows for the various runtime/ABI combinations: | ||
126 | * | ||
127 | * Linux 64 bit: Set XKPHYS bit | ||
128 | * Linux 32 bit: use mmap to create mapping, use virtual address | ||
129 | * CVMX 64 bit: use physical address directly | ||
130 | * CVMX 32 bit: use physical address directly | ||
131 | * | ||
132 | * Note that the CVMX environment assumes the use of 1-1 TLB | ||
133 | * mappings so that the physical addresses can be used | ||
134 | * directly | ||
135 | */ | ||
136 | if (!cvmx_bootmem_desc) { | ||
137 | #if defined(CVMX_ABI_64) | ||
138 | /* Set XKPHYS bit */ | ||
139 | cvmx_bootmem_desc = cvmx_phys_to_ptr(CAST64(mem_desc_ptr)); | ||
140 | #else | ||
141 | cvmx_bootmem_desc = (struct cvmx_bootmem_desc *) mem_desc_ptr; | ||
142 | #endif | ||
143 | } | ||
144 | |||
145 | return 0; | ||
146 | } | ||
147 | |||
148 | /* | ||
149 | * The cvmx_bootmem_phy* functions below return 64 bit physical | ||
150 | * addresses, and expose more features that the cvmx_bootmem_functions | ||
151 | * above. These are required for full memory space access in 32 bit | ||
152 | * applications, as well as for using some advance features. Most | ||
153 | * applications should not need to use these. | ||
154 | */ | ||
155 | |||
156 | int64_t cvmx_bootmem_phy_alloc(uint64_t req_size, uint64_t address_min, | ||
157 | uint64_t address_max, uint64_t alignment, | ||
158 | uint32_t flags) | ||
159 | { | ||
160 | |||
161 | uint64_t head_addr; | ||
162 | uint64_t ent_addr; | ||
163 | /* points to previous list entry, NULL current entry is head of list */ | ||
164 | uint64_t prev_addr = 0; | ||
165 | uint64_t new_ent_addr = 0; | ||
166 | uint64_t desired_min_addr; | ||
167 | |||
168 | #ifdef DEBUG | ||
169 | cvmx_dprintf("cvmx_bootmem_phy_alloc: req_size: 0x%llx, " | ||
170 | "min_addr: 0x%llx, max_addr: 0x%llx, align: 0x%llx\n", | ||
171 | (unsigned long long)req_size, | ||
172 | (unsigned long long)address_min, | ||
173 | (unsigned long long)address_max, | ||
174 | (unsigned long long)alignment); | ||
175 | #endif | ||
176 | |||
177 | if (cvmx_bootmem_desc->major_version > 3) { | ||
178 | cvmx_dprintf("ERROR: Incompatible bootmem descriptor " | ||
179 | "version: %d.%d at addr: %p\n", | ||
180 | (int)cvmx_bootmem_desc->major_version, | ||
181 | (int)cvmx_bootmem_desc->minor_version, | ||
182 | cvmx_bootmem_desc); | ||
183 | goto error_out; | ||
184 | } | ||
185 | |||
186 | /* | ||
187 | * Do a variety of checks to validate the arguments. The | ||
188 | * allocator code will later assume that these checks have | ||
189 | * been made. We validate that the requested constraints are | ||
190 | * not self-contradictory before we look through the list of | ||
191 | * available memory. | ||
192 | */ | ||
193 | |||
194 | /* 0 is not a valid req_size for this allocator */ | ||
195 | if (!req_size) | ||
196 | goto error_out; | ||
197 | |||
198 | /* Round req_size up to mult of minimum alignment bytes */ | ||
199 | req_size = (req_size + (CVMX_BOOTMEM_ALIGNMENT_SIZE - 1)) & | ||
200 | ~(CVMX_BOOTMEM_ALIGNMENT_SIZE - 1); | ||
201 | |||
202 | /* | ||
203 | * Convert !0 address_min and 0 address_max to special case of | ||
204 | * range that specifies an exact memory block to allocate. Do | ||
205 | * this before other checks and adjustments so that this | ||
206 | * tranformation will be validated. | ||
207 | */ | ||
208 | if (address_min && !address_max) | ||
209 | address_max = address_min + req_size; | ||
210 | else if (!address_min && !address_max) | ||
211 | address_max = ~0ull; /* If no limits given, use max limits */ | ||
212 | |||
213 | |||
214 | /* | ||
215 | * Enforce minimum alignment (this also keeps the minimum free block | ||
216 | * req_size the same as the alignment req_size. | ||
217 | */ | ||
218 | if (alignment < CVMX_BOOTMEM_ALIGNMENT_SIZE) | ||
219 | alignment = CVMX_BOOTMEM_ALIGNMENT_SIZE; | ||
220 | |||
221 | /* | ||
222 | * Adjust address minimum based on requested alignment (round | ||
223 | * up to meet alignment). Do this here so we can reject | ||
224 | * impossible requests up front. (NOP for address_min == 0) | ||
225 | */ | ||
226 | if (alignment) | ||
227 | address_min = __ALIGN_MASK(address_min, (alignment - 1)); | ||
228 | |||
229 | /* | ||
230 | * Reject inconsistent args. We have adjusted these, so this | ||
231 | * may fail due to our internal changes even if this check | ||
232 | * would pass for the values the user supplied. | ||
233 | */ | ||
234 | if (req_size > address_max - address_min) | ||
235 | goto error_out; | ||
236 | |||
237 | /* Walk through the list entries - first fit found is returned */ | ||
238 | |||
239 | if (!(flags & CVMX_BOOTMEM_FLAG_NO_LOCKING)) | ||
240 | cvmx_bootmem_lock(); | ||
241 | head_addr = cvmx_bootmem_desc->head_addr; | ||
242 | ent_addr = head_addr; | ||
243 | for (; ent_addr; | ||
244 | prev_addr = ent_addr, | ||
245 | ent_addr = cvmx_bootmem_phy_get_next(ent_addr)) { | ||
246 | uint64_t usable_base, usable_max; | ||
247 | uint64_t ent_size = cvmx_bootmem_phy_get_size(ent_addr); | ||
248 | |||
249 | if (cvmx_bootmem_phy_get_next(ent_addr) | ||
250 | && ent_addr > cvmx_bootmem_phy_get_next(ent_addr)) { | ||
251 | cvmx_dprintf("Internal bootmem_alloc() error: ent: " | ||
252 | "0x%llx, next: 0x%llx\n", | ||
253 | (unsigned long long)ent_addr, | ||
254 | (unsigned long long) | ||
255 | cvmx_bootmem_phy_get_next(ent_addr)); | ||
256 | goto error_out; | ||
257 | } | ||
258 | |||
259 | /* | ||
260 | * Determine if this is an entry that can satisify the | ||
261 | * request Check to make sure entry is large enough to | ||
262 | * satisfy request. | ||
263 | */ | ||
264 | usable_base = | ||
265 | __ALIGN_MASK(max(address_min, ent_addr), alignment - 1); | ||
266 | usable_max = min(address_max, ent_addr + ent_size); | ||
267 | /* | ||
268 | * We should be able to allocate block at address | ||
269 | * usable_base. | ||
270 | */ | ||
271 | |||
272 | desired_min_addr = usable_base; | ||
273 | /* | ||
274 | * Determine if request can be satisfied from the | ||
275 | * current entry. | ||
276 | */ | ||
277 | if (!((ent_addr + ent_size) > usable_base | ||
278 | && ent_addr < address_max | ||
279 | && req_size <= usable_max - usable_base)) | ||
280 | continue; | ||
281 | /* | ||
282 | * We have found an entry that has room to satisfy the | ||
283 | * request, so allocate it from this entry. If end | ||
284 | * CVMX_BOOTMEM_FLAG_END_ALLOC set, then allocate from | ||
285 | * the end of this block rather than the beginning. | ||
286 | */ | ||
287 | if (flags & CVMX_BOOTMEM_FLAG_END_ALLOC) { | ||
288 | desired_min_addr = usable_max - req_size; | ||
289 | /* | ||
290 | * Align desired address down to required | ||
291 | * alignment. | ||
292 | */ | ||
293 | desired_min_addr &= ~(alignment - 1); | ||
294 | } | ||
295 | |||
296 | /* Match at start of entry */ | ||
297 | if (desired_min_addr == ent_addr) { | ||
298 | if (req_size < ent_size) { | ||
299 | /* | ||
300 | * big enough to create a new block | ||
301 | * from top portion of block. | ||
302 | */ | ||
303 | new_ent_addr = ent_addr + req_size; | ||
304 | cvmx_bootmem_phy_set_next(new_ent_addr, | ||
305 | cvmx_bootmem_phy_get_next(ent_addr)); | ||
306 | cvmx_bootmem_phy_set_size(new_ent_addr, | ||
307 | ent_size - | ||
308 | req_size); | ||
309 | |||
310 | /* | ||
311 | * Adjust next pointer as following | ||
312 | * code uses this. | ||
313 | */ | ||
314 | cvmx_bootmem_phy_set_next(ent_addr, | ||
315 | new_ent_addr); | ||
316 | } | ||
317 | |||
318 | /* | ||
319 | * adjust prev ptr or head to remove this | ||
320 | * entry from list. | ||
321 | */ | ||
322 | if (prev_addr) | ||
323 | cvmx_bootmem_phy_set_next(prev_addr, | ||
324 | cvmx_bootmem_phy_get_next(ent_addr)); | ||
325 | else | ||
326 | /* | ||
327 | * head of list being returned, so | ||
328 | * update head ptr. | ||
329 | */ | ||
330 | cvmx_bootmem_desc->head_addr = | ||
331 | cvmx_bootmem_phy_get_next(ent_addr); | ||
332 | |||
333 | if (!(flags & CVMX_BOOTMEM_FLAG_NO_LOCKING)) | ||
334 | cvmx_bootmem_unlock(); | ||
335 | return desired_min_addr; | ||
336 | } | ||
337 | /* | ||
338 | * block returned doesn't start at beginning of entry, | ||
339 | * so we know that we will be splitting a block off | ||
340 | * the front of this one. Create a new block from the | ||
341 | * beginning, add to list, and go to top of loop | ||
342 | * again. | ||
343 | * | ||
344 | * create new block from high portion of | ||
345 | * block, so that top block starts at desired | ||
346 | * addr. | ||
347 | */ | ||
348 | new_ent_addr = desired_min_addr; | ||
349 | cvmx_bootmem_phy_set_next(new_ent_addr, | ||
350 | cvmx_bootmem_phy_get_next | ||
351 | (ent_addr)); | ||
352 | cvmx_bootmem_phy_set_size(new_ent_addr, | ||
353 | cvmx_bootmem_phy_get_size | ||
354 | (ent_addr) - | ||
355 | (desired_min_addr - | ||
356 | ent_addr)); | ||
357 | cvmx_bootmem_phy_set_size(ent_addr, | ||
358 | desired_min_addr - ent_addr); | ||
359 | cvmx_bootmem_phy_set_next(ent_addr, new_ent_addr); | ||
360 | /* Loop again to handle actual alloc from new block */ | ||
361 | } | ||
362 | error_out: | ||
363 | /* We didn't find anything, so return error */ | ||
364 | if (!(flags & CVMX_BOOTMEM_FLAG_NO_LOCKING)) | ||
365 | cvmx_bootmem_unlock(); | ||
366 | return -1; | ||
367 | } | ||
368 | |||
369 | int __cvmx_bootmem_phy_free(uint64_t phy_addr, uint64_t size, uint32_t flags) | ||
370 | { | ||
371 | uint64_t cur_addr; | ||
372 | uint64_t prev_addr = 0; /* zero is invalid */ | ||
373 | int retval = 0; | ||
374 | |||
375 | #ifdef DEBUG | ||
376 | cvmx_dprintf("__cvmx_bootmem_phy_free addr: 0x%llx, size: 0x%llx\n", | ||
377 | (unsigned long long)phy_addr, (unsigned long long)size); | ||
378 | #endif | ||
379 | if (cvmx_bootmem_desc->major_version > 3) { | ||
380 | cvmx_dprintf("ERROR: Incompatible bootmem descriptor " | ||
381 | "version: %d.%d at addr: %p\n", | ||
382 | (int)cvmx_bootmem_desc->major_version, | ||
383 | (int)cvmx_bootmem_desc->minor_version, | ||
384 | cvmx_bootmem_desc); | ||
385 | return 0; | ||
386 | } | ||
387 | |||
388 | /* 0 is not a valid size for this allocator */ | ||
389 | if (!size) | ||
390 | return 0; | ||
391 | |||
392 | if (!(flags & CVMX_BOOTMEM_FLAG_NO_LOCKING)) | ||
393 | cvmx_bootmem_lock(); | ||
394 | cur_addr = cvmx_bootmem_desc->head_addr; | ||
395 | if (cur_addr == 0 || phy_addr < cur_addr) { | ||
396 | /* add at front of list - special case with changing head ptr */ | ||
397 | if (cur_addr && phy_addr + size > cur_addr) | ||
398 | goto bootmem_free_done; /* error, overlapping section */ | ||
399 | else if (phy_addr + size == cur_addr) { | ||
400 | /* Add to front of existing first block */ | ||
401 | cvmx_bootmem_phy_set_next(phy_addr, | ||
402 | cvmx_bootmem_phy_get_next | ||
403 | (cur_addr)); | ||
404 | cvmx_bootmem_phy_set_size(phy_addr, | ||
405 | cvmx_bootmem_phy_get_size | ||
406 | (cur_addr) + size); | ||
407 | cvmx_bootmem_desc->head_addr = phy_addr; | ||
408 | |||
409 | } else { | ||
410 | /* New block before first block. OK if cur_addr is 0 */ | ||
411 | cvmx_bootmem_phy_set_next(phy_addr, cur_addr); | ||
412 | cvmx_bootmem_phy_set_size(phy_addr, size); | ||
413 | cvmx_bootmem_desc->head_addr = phy_addr; | ||
414 | } | ||
415 | retval = 1; | ||
416 | goto bootmem_free_done; | ||
417 | } | ||
418 | |||
419 | /* Find place in list to add block */ | ||
420 | while (cur_addr && phy_addr > cur_addr) { | ||
421 | prev_addr = cur_addr; | ||
422 | cur_addr = cvmx_bootmem_phy_get_next(cur_addr); | ||
423 | } | ||
424 | |||
425 | if (!cur_addr) { | ||
426 | /* | ||
427 | * We have reached the end of the list, add on to end, | ||
428 | * checking to see if we need to combine with last | ||
429 | * block | ||
430 | */ | ||
431 | if (prev_addr + cvmx_bootmem_phy_get_size(prev_addr) == | ||
432 | phy_addr) { | ||
433 | cvmx_bootmem_phy_set_size(prev_addr, | ||
434 | cvmx_bootmem_phy_get_size | ||
435 | (prev_addr) + size); | ||
436 | } else { | ||
437 | cvmx_bootmem_phy_set_next(prev_addr, phy_addr); | ||
438 | cvmx_bootmem_phy_set_size(phy_addr, size); | ||
439 | cvmx_bootmem_phy_set_next(phy_addr, 0); | ||
440 | } | ||
441 | retval = 1; | ||
442 | goto bootmem_free_done; | ||
443 | } else { | ||
444 | /* | ||
445 | * insert between prev and cur nodes, checking for | ||
446 | * merge with either/both. | ||
447 | */ | ||
448 | if (prev_addr + cvmx_bootmem_phy_get_size(prev_addr) == | ||
449 | phy_addr) { | ||
450 | /* Merge with previous */ | ||
451 | cvmx_bootmem_phy_set_size(prev_addr, | ||
452 | cvmx_bootmem_phy_get_size | ||
453 | (prev_addr) + size); | ||
454 | if (phy_addr + size == cur_addr) { | ||
455 | /* Also merge with current */ | ||
456 | cvmx_bootmem_phy_set_size(prev_addr, | ||
457 | cvmx_bootmem_phy_get_size(cur_addr) + | ||
458 | cvmx_bootmem_phy_get_size(prev_addr)); | ||
459 | cvmx_bootmem_phy_set_next(prev_addr, | ||
460 | cvmx_bootmem_phy_get_next(cur_addr)); | ||
461 | } | ||
462 | retval = 1; | ||
463 | goto bootmem_free_done; | ||
464 | } else if (phy_addr + size == cur_addr) { | ||
465 | /* Merge with current */ | ||
466 | cvmx_bootmem_phy_set_size(phy_addr, | ||
467 | cvmx_bootmem_phy_get_size | ||
468 | (cur_addr) + size); | ||
469 | cvmx_bootmem_phy_set_next(phy_addr, | ||
470 | cvmx_bootmem_phy_get_next | ||
471 | (cur_addr)); | ||
472 | cvmx_bootmem_phy_set_next(prev_addr, phy_addr); | ||
473 | retval = 1; | ||
474 | goto bootmem_free_done; | ||
475 | } | ||
476 | |||
477 | /* It is a standalone block, add in between prev and cur */ | ||
478 | cvmx_bootmem_phy_set_size(phy_addr, size); | ||
479 | cvmx_bootmem_phy_set_next(phy_addr, cur_addr); | ||
480 | cvmx_bootmem_phy_set_next(prev_addr, phy_addr); | ||
481 | |||
482 | } | ||
483 | retval = 1; | ||
484 | |||
485 | bootmem_free_done: | ||
486 | if (!(flags & CVMX_BOOTMEM_FLAG_NO_LOCKING)) | ||
487 | cvmx_bootmem_unlock(); | ||
488 | return retval; | ||
489 | |||
490 | } | ||
491 | |||
492 | struct cvmx_bootmem_named_block_desc * | ||
493 | cvmx_bootmem_phy_named_block_find(char *name, uint32_t flags) | ||
494 | { | ||
495 | unsigned int i; | ||
496 | struct cvmx_bootmem_named_block_desc *named_block_array_ptr; | ||
497 | |||
498 | #ifdef DEBUG | ||
499 | cvmx_dprintf("cvmx_bootmem_phy_named_block_find: %s\n", name); | ||
500 | #endif | ||
501 | /* | ||
502 | * Lock the structure to make sure that it is not being | ||
503 | * changed while we are examining it. | ||
504 | */ | ||
505 | if (!(flags & CVMX_BOOTMEM_FLAG_NO_LOCKING)) | ||
506 | cvmx_bootmem_lock(); | ||
507 | |||
508 | /* Use XKPHYS for 64 bit linux */ | ||
509 | named_block_array_ptr = (struct cvmx_bootmem_named_block_desc *) | ||
510 | cvmx_phys_to_ptr(cvmx_bootmem_desc->named_block_array_addr); | ||
511 | |||
512 | #ifdef DEBUG | ||
513 | cvmx_dprintf | ||
514 | ("cvmx_bootmem_phy_named_block_find: named_block_array_ptr: %p\n", | ||
515 | named_block_array_ptr); | ||
516 | #endif | ||
517 | if (cvmx_bootmem_desc->major_version == 3) { | ||
518 | for (i = 0; | ||
519 | i < cvmx_bootmem_desc->named_block_num_blocks; i++) { | ||
520 | if ((name && named_block_array_ptr[i].size | ||
521 | && !strncmp(name, named_block_array_ptr[i].name, | ||
522 | cvmx_bootmem_desc->named_block_name_len | ||
523 | - 1)) | ||
524 | || (!name && !named_block_array_ptr[i].size)) { | ||
525 | if (!(flags & CVMX_BOOTMEM_FLAG_NO_LOCKING)) | ||
526 | cvmx_bootmem_unlock(); | ||
527 | |||
528 | return &(named_block_array_ptr[i]); | ||
529 | } | ||
530 | } | ||
531 | } else { | ||
532 | cvmx_dprintf("ERROR: Incompatible bootmem descriptor " | ||
533 | "version: %d.%d at addr: %p\n", | ||
534 | (int)cvmx_bootmem_desc->major_version, | ||
535 | (int)cvmx_bootmem_desc->minor_version, | ||
536 | cvmx_bootmem_desc); | ||
537 | } | ||
538 | if (!(flags & CVMX_BOOTMEM_FLAG_NO_LOCKING)) | ||
539 | cvmx_bootmem_unlock(); | ||
540 | |||
541 | return NULL; | ||
542 | } | ||
543 | |||
544 | int cvmx_bootmem_phy_named_block_free(char *name, uint32_t flags) | ||
545 | { | ||
546 | struct cvmx_bootmem_named_block_desc *named_block_ptr; | ||
547 | |||
548 | if (cvmx_bootmem_desc->major_version != 3) { | ||
549 | cvmx_dprintf("ERROR: Incompatible bootmem descriptor version: " | ||
550 | "%d.%d at addr: %p\n", | ||
551 | (int)cvmx_bootmem_desc->major_version, | ||
552 | (int)cvmx_bootmem_desc->minor_version, | ||
553 | cvmx_bootmem_desc); | ||
554 | return 0; | ||
555 | } | ||
556 | #ifdef DEBUG | ||
557 | cvmx_dprintf("cvmx_bootmem_phy_named_block_free: %s\n", name); | ||
558 | #endif | ||
559 | |||
560 | /* | ||
561 | * Take lock here, as name lookup/block free/name free need to | ||
562 | * be atomic. | ||
563 | */ | ||
564 | cvmx_bootmem_lock(); | ||
565 | |||
566 | named_block_ptr = | ||
567 | cvmx_bootmem_phy_named_block_find(name, | ||
568 | CVMX_BOOTMEM_FLAG_NO_LOCKING); | ||
569 | if (named_block_ptr) { | ||
570 | #ifdef DEBUG | ||
571 | cvmx_dprintf("cvmx_bootmem_phy_named_block_free: " | ||
572 | "%s, base: 0x%llx, size: 0x%llx\n", | ||
573 | name, | ||
574 | (unsigned long long)named_block_ptr->base_addr, | ||
575 | (unsigned long long)named_block_ptr->size); | ||
576 | #endif | ||
577 | __cvmx_bootmem_phy_free(named_block_ptr->base_addr, | ||
578 | named_block_ptr->size, | ||
579 | CVMX_BOOTMEM_FLAG_NO_LOCKING); | ||
580 | named_block_ptr->size = 0; | ||
581 | /* Set size to zero to indicate block not used. */ | ||
582 | } | ||
583 | |||
584 | cvmx_bootmem_unlock(); | ||
585 | return named_block_ptr != NULL; /* 0 on failure, 1 on success */ | ||
586 | } | ||
diff --git a/arch/mips/cavium-octeon/executive/cvmx-l2c.c b/arch/mips/cavium-octeon/executive/cvmx-l2c.c new file mode 100644 index 000000000000..6abe56f1e097 --- /dev/null +++ b/arch/mips/cavium-octeon/executive/cvmx-l2c.c | |||
@@ -0,0 +1,734 @@ | |||
1 | /***********************license start*************** | ||
2 | * Author: Cavium Networks | ||
3 | * | ||
4 | * Contact: support@caviumnetworks.com | ||
5 | * This file is part of the OCTEON SDK | ||
6 | * | ||
7 | * Copyright (c) 2003-2008 Cavium Networks | ||
8 | * | ||
9 | * This file is free software; you can redistribute it and/or modify | ||
10 | * it under the terms of the GNU General Public License, Version 2, as | ||
11 | * published by the Free Software Foundation. | ||
12 | * | ||
13 | * This file is distributed in the hope that it will be useful, but | ||
14 | * AS-IS and WITHOUT ANY WARRANTY; without even the implied warranty | ||
15 | * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE, TITLE, or | ||
16 | * NONINFRINGEMENT. See the GNU General Public License for more | ||
17 | * details. | ||
18 | * | ||
19 | * You should have received a copy of the GNU General Public License | ||
20 | * along with this file; if not, write to the Free Software | ||
21 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | ||
22 | * or visit http://www.gnu.org/licenses/. | ||
23 | * | ||
24 | * This file may also be available under a different license from Cavium. | ||
25 | * Contact Cavium Networks for more information | ||
26 | ***********************license end**************************************/ | ||
27 | |||
28 | /* | ||
29 | * Implementation of the Level 2 Cache (L2C) control, measurement, and | ||
30 | * debugging facilities. | ||
31 | */ | ||
32 | |||
33 | #include <asm/octeon/cvmx.h> | ||
34 | #include <asm/octeon/cvmx-l2c.h> | ||
35 | #include <asm/octeon/cvmx-spinlock.h> | ||
36 | |||
37 | /* | ||
38 | * This spinlock is used internally to ensure that only one core is | ||
39 | * performing certain L2 operations at a time. | ||
40 | * | ||
41 | * NOTE: This only protects calls from within a single application - | ||
42 | * if multiple applications or operating systems are running, then it | ||
43 | * is up to the user program to coordinate between them. | ||
44 | */ | ||
45 | static cvmx_spinlock_t cvmx_l2c_spinlock; | ||
46 | |||
47 | static inline int l2_size_half(void) | ||
48 | { | ||
49 | uint64_t val = cvmx_read_csr(CVMX_L2D_FUS3); | ||
50 | return !!(val & (1ull << 34)); | ||
51 | } | ||
52 | |||
53 | int cvmx_l2c_get_core_way_partition(uint32_t core) | ||
54 | { | ||
55 | uint32_t field; | ||
56 | |||
57 | /* Validate the core number */ | ||
58 | if (core >= cvmx_octeon_num_cores()) | ||
59 | return -1; | ||
60 | |||
61 | /* | ||
62 | * Use the lower two bits of the coreNumber to determine the | ||
63 | * bit offset of the UMSK[] field in the L2C_SPAR register. | ||
64 | */ | ||
65 | field = (core & 0x3) * 8; | ||
66 | |||
67 | /* | ||
68 | * Return the UMSK[] field from the appropriate L2C_SPAR | ||
69 | * register based on the coreNumber. | ||
70 | */ | ||
71 | |||
72 | switch (core & 0xC) { | ||
73 | case 0x0: | ||
74 | return (cvmx_read_csr(CVMX_L2C_SPAR0) & (0xFF << field)) >> | ||
75 | field; | ||
76 | case 0x4: | ||
77 | return (cvmx_read_csr(CVMX_L2C_SPAR1) & (0xFF << field)) >> | ||
78 | field; | ||
79 | case 0x8: | ||
80 | return (cvmx_read_csr(CVMX_L2C_SPAR2) & (0xFF << field)) >> | ||
81 | field; | ||
82 | case 0xC: | ||
83 | return (cvmx_read_csr(CVMX_L2C_SPAR3) & (0xFF << field)) >> | ||
84 | field; | ||
85 | } | ||
86 | return 0; | ||
87 | } | ||
88 | |||
89 | int cvmx_l2c_set_core_way_partition(uint32_t core, uint32_t mask) | ||
90 | { | ||
91 | uint32_t field; | ||
92 | uint32_t valid_mask; | ||
93 | |||
94 | valid_mask = (0x1 << cvmx_l2c_get_num_assoc()) - 1; | ||
95 | |||
96 | mask &= valid_mask; | ||
97 | |||
98 | /* A UMSK setting which blocks all L2C Ways is an error. */ | ||
99 | if (mask == valid_mask) | ||
100 | return -1; | ||
101 | |||
102 | /* Validate the core number */ | ||
103 | if (core >= cvmx_octeon_num_cores()) | ||
104 | return -1; | ||
105 | |||
106 | /* Check to make sure current mask & new mask don't block all ways */ | ||
107 | if (((mask | cvmx_l2c_get_core_way_partition(core)) & valid_mask) == | ||
108 | valid_mask) | ||
109 | return -1; | ||
110 | |||
111 | /* Use the lower two bits of core to determine the bit offset of the | ||
112 | * UMSK[] field in the L2C_SPAR register. | ||
113 | */ | ||
114 | field = (core & 0x3) * 8; | ||
115 | |||
116 | /* Assign the new mask setting to the UMSK[] field in the appropriate | ||
117 | * L2C_SPAR register based on the core_num. | ||
118 | * | ||
119 | */ | ||
120 | switch (core & 0xC) { | ||
121 | case 0x0: | ||
122 | cvmx_write_csr(CVMX_L2C_SPAR0, | ||
123 | (cvmx_read_csr(CVMX_L2C_SPAR0) & | ||
124 | ~(0xFF << field)) | mask << field); | ||
125 | break; | ||
126 | case 0x4: | ||
127 | cvmx_write_csr(CVMX_L2C_SPAR1, | ||
128 | (cvmx_read_csr(CVMX_L2C_SPAR1) & | ||
129 | ~(0xFF << field)) | mask << field); | ||
130 | break; | ||
131 | case 0x8: | ||
132 | cvmx_write_csr(CVMX_L2C_SPAR2, | ||
133 | (cvmx_read_csr(CVMX_L2C_SPAR2) & | ||
134 | ~(0xFF << field)) | mask << field); | ||
135 | break; | ||
136 | case 0xC: | ||
137 | cvmx_write_csr(CVMX_L2C_SPAR3, | ||
138 | (cvmx_read_csr(CVMX_L2C_SPAR3) & | ||
139 | ~(0xFF << field)) | mask << field); | ||
140 | break; | ||
141 | } | ||
142 | return 0; | ||
143 | } | ||
144 | |||
145 | int cvmx_l2c_set_hw_way_partition(uint32_t mask) | ||
146 | { | ||
147 | uint32_t valid_mask; | ||
148 | |||
149 | valid_mask = 0xff; | ||
150 | |||
151 | if (OCTEON_IS_MODEL(OCTEON_CN58XX) || OCTEON_IS_MODEL(OCTEON_CN38XX)) { | ||
152 | if (l2_size_half()) | ||
153 | valid_mask = 0xf; | ||
154 | } else if (l2_size_half()) | ||
155 | valid_mask = 0x3; | ||
156 | |||
157 | mask &= valid_mask; | ||
158 | |||
159 | /* A UMSK setting which blocks all L2C Ways is an error. */ | ||
160 | if (mask == valid_mask) | ||
161 | return -1; | ||
162 | /* Check to make sure current mask & new mask don't block all ways */ | ||
163 | if (((mask | cvmx_l2c_get_hw_way_partition()) & valid_mask) == | ||
164 | valid_mask) | ||
165 | return -1; | ||
166 | |||
167 | cvmx_write_csr(CVMX_L2C_SPAR4, | ||
168 | (cvmx_read_csr(CVMX_L2C_SPAR4) & ~0xFF) | mask); | ||
169 | return 0; | ||
170 | } | ||
171 | |||
172 | int cvmx_l2c_get_hw_way_partition(void) | ||
173 | { | ||
174 | return cvmx_read_csr(CVMX_L2C_SPAR4) & (0xFF); | ||
175 | } | ||
176 | |||
177 | void cvmx_l2c_config_perf(uint32_t counter, enum cvmx_l2c_event event, | ||
178 | uint32_t clear_on_read) | ||
179 | { | ||
180 | union cvmx_l2c_pfctl pfctl; | ||
181 | |||
182 | pfctl.u64 = cvmx_read_csr(CVMX_L2C_PFCTL); | ||
183 | |||
184 | switch (counter) { | ||
185 | case 0: | ||
186 | pfctl.s.cnt0sel = event; | ||
187 | pfctl.s.cnt0ena = 1; | ||
188 | if (!cvmx_octeon_is_pass1()) | ||
189 | pfctl.s.cnt0rdclr = clear_on_read; | ||
190 | break; | ||
191 | case 1: | ||
192 | pfctl.s.cnt1sel = event; | ||
193 | pfctl.s.cnt1ena = 1; | ||
194 | if (!cvmx_octeon_is_pass1()) | ||
195 | pfctl.s.cnt1rdclr = clear_on_read; | ||
196 | break; | ||
197 | case 2: | ||
198 | pfctl.s.cnt2sel = event; | ||
199 | pfctl.s.cnt2ena = 1; | ||
200 | if (!cvmx_octeon_is_pass1()) | ||
201 | pfctl.s.cnt2rdclr = clear_on_read; | ||
202 | break; | ||
203 | case 3: | ||
204 | default: | ||
205 | pfctl.s.cnt3sel = event; | ||
206 | pfctl.s.cnt3ena = 1; | ||
207 | if (!cvmx_octeon_is_pass1()) | ||
208 | pfctl.s.cnt3rdclr = clear_on_read; | ||
209 | break; | ||
210 | } | ||
211 | |||
212 | cvmx_write_csr(CVMX_L2C_PFCTL, pfctl.u64); | ||
213 | } | ||
214 | |||
215 | uint64_t cvmx_l2c_read_perf(uint32_t counter) | ||
216 | { | ||
217 | switch (counter) { | ||
218 | case 0: | ||
219 | return cvmx_read_csr(CVMX_L2C_PFC0); | ||
220 | case 1: | ||
221 | return cvmx_read_csr(CVMX_L2C_PFC1); | ||
222 | case 2: | ||
223 | return cvmx_read_csr(CVMX_L2C_PFC2); | ||
224 | case 3: | ||
225 | default: | ||
226 | return cvmx_read_csr(CVMX_L2C_PFC3); | ||
227 | } | ||
228 | } | ||
229 | |||
230 | /** | ||
231 | * @INTERNAL | ||
232 | * Helper function use to fault in cache lines for L2 cache locking | ||
233 | * | ||
234 | * @addr: Address of base of memory region to read into L2 cache | ||
235 | * @len: Length (in bytes) of region to fault in | ||
236 | */ | ||
237 | static void fault_in(uint64_t addr, int len) | ||
238 | { | ||
239 | volatile char *ptr; | ||
240 | volatile char dummy; | ||
241 | /* | ||
242 | * Adjust addr and length so we get all cache lines even for | ||
243 | * small ranges spanning two cache lines | ||
244 | */ | ||
245 | len += addr & CVMX_CACHE_LINE_MASK; | ||
246 | addr &= ~CVMX_CACHE_LINE_MASK; | ||
247 | ptr = (volatile char *)cvmx_phys_to_ptr(addr); | ||
248 | /* | ||
249 | * Invalidate L1 cache to make sure all loads result in data | ||
250 | * being in L2. | ||
251 | */ | ||
252 | CVMX_DCACHE_INVALIDATE; | ||
253 | while (len > 0) { | ||
254 | dummy += *ptr; | ||
255 | len -= CVMX_CACHE_LINE_SIZE; | ||
256 | ptr += CVMX_CACHE_LINE_SIZE; | ||
257 | } | ||
258 | } | ||
259 | |||
260 | int cvmx_l2c_lock_line(uint64_t addr) | ||
261 | { | ||
262 | int retval = 0; | ||
263 | union cvmx_l2c_dbg l2cdbg; | ||
264 | union cvmx_l2c_lckbase lckbase; | ||
265 | union cvmx_l2c_lckoff lckoff; | ||
266 | union cvmx_l2t_err l2t_err; | ||
267 | l2cdbg.u64 = 0; | ||
268 | lckbase.u64 = 0; | ||
269 | lckoff.u64 = 0; | ||
270 | |||
271 | cvmx_spinlock_lock(&cvmx_l2c_spinlock); | ||
272 | |||
273 | /* Clear l2t error bits if set */ | ||
274 | l2t_err.u64 = cvmx_read_csr(CVMX_L2T_ERR); | ||
275 | l2t_err.s.lckerr = 1; | ||
276 | l2t_err.s.lckerr2 = 1; | ||
277 | cvmx_write_csr(CVMX_L2T_ERR, l2t_err.u64); | ||
278 | |||
279 | addr &= ~CVMX_CACHE_LINE_MASK; | ||
280 | |||
281 | /* Set this core as debug core */ | ||
282 | l2cdbg.s.ppnum = cvmx_get_core_num(); | ||
283 | CVMX_SYNC; | ||
284 | cvmx_write_csr(CVMX_L2C_DBG, l2cdbg.u64); | ||
285 | cvmx_read_csr(CVMX_L2C_DBG); | ||
286 | |||
287 | lckoff.s.lck_offset = 0; /* Only lock 1 line at a time */ | ||
288 | cvmx_write_csr(CVMX_L2C_LCKOFF, lckoff.u64); | ||
289 | cvmx_read_csr(CVMX_L2C_LCKOFF); | ||
290 | |||
291 | if (((union cvmx_l2c_cfg) (cvmx_read_csr(CVMX_L2C_CFG))).s.idxalias) { | ||
292 | int alias_shift = | ||
293 | CVMX_L2C_IDX_ADDR_SHIFT + 2 * CVMX_L2_SET_BITS - 1; | ||
294 | uint64_t addr_tmp = | ||
295 | addr ^ (addr & ((1 << alias_shift) - 1)) >> | ||
296 | CVMX_L2_SET_BITS; | ||
297 | lckbase.s.lck_base = addr_tmp >> 7; | ||
298 | } else { | ||
299 | lckbase.s.lck_base = addr >> 7; | ||
300 | } | ||
301 | |||
302 | lckbase.s.lck_ena = 1; | ||
303 | cvmx_write_csr(CVMX_L2C_LCKBASE, lckbase.u64); | ||
304 | cvmx_read_csr(CVMX_L2C_LCKBASE); /* Make sure it gets there */ | ||
305 | |||
306 | fault_in(addr, CVMX_CACHE_LINE_SIZE); | ||
307 | |||
308 | lckbase.s.lck_ena = 0; | ||
309 | cvmx_write_csr(CVMX_L2C_LCKBASE, lckbase.u64); | ||
310 | cvmx_read_csr(CVMX_L2C_LCKBASE); /* Make sure it gets there */ | ||
311 | |||
312 | /* Stop being debug core */ | ||
313 | cvmx_write_csr(CVMX_L2C_DBG, 0); | ||
314 | cvmx_read_csr(CVMX_L2C_DBG); | ||
315 | |||
316 | l2t_err.u64 = cvmx_read_csr(CVMX_L2T_ERR); | ||
317 | if (l2t_err.s.lckerr || l2t_err.s.lckerr2) | ||
318 | retval = 1; /* We were unable to lock the line */ | ||
319 | |||
320 | cvmx_spinlock_unlock(&cvmx_l2c_spinlock); | ||
321 | |||
322 | return retval; | ||
323 | } | ||
324 | |||
325 | int cvmx_l2c_lock_mem_region(uint64_t start, uint64_t len) | ||
326 | { | ||
327 | int retval = 0; | ||
328 | |||
329 | /* Round start/end to cache line boundaries */ | ||
330 | len += start & CVMX_CACHE_LINE_MASK; | ||
331 | start &= ~CVMX_CACHE_LINE_MASK; | ||
332 | len = (len + CVMX_CACHE_LINE_MASK) & ~CVMX_CACHE_LINE_MASK; | ||
333 | |||
334 | while (len) { | ||
335 | retval += cvmx_l2c_lock_line(start); | ||
336 | start += CVMX_CACHE_LINE_SIZE; | ||
337 | len -= CVMX_CACHE_LINE_SIZE; | ||
338 | } | ||
339 | |||
340 | return retval; | ||
341 | } | ||
342 | |||
343 | void cvmx_l2c_flush(void) | ||
344 | { | ||
345 | uint64_t assoc, set; | ||
346 | uint64_t n_assoc, n_set; | ||
347 | union cvmx_l2c_dbg l2cdbg; | ||
348 | |||
349 | cvmx_spinlock_lock(&cvmx_l2c_spinlock); | ||
350 | |||
351 | l2cdbg.u64 = 0; | ||
352 | if (!OCTEON_IS_MODEL(OCTEON_CN30XX)) | ||
353 | l2cdbg.s.ppnum = cvmx_get_core_num(); | ||
354 | l2cdbg.s.finv = 1; | ||
355 | n_set = CVMX_L2_SETS; | ||
356 | n_assoc = l2_size_half() ? (CVMX_L2_ASSOC / 2) : CVMX_L2_ASSOC; | ||
357 | for (set = 0; set < n_set; set++) { | ||
358 | for (assoc = 0; assoc < n_assoc; assoc++) { | ||
359 | l2cdbg.s.set = assoc; | ||
360 | /* Enter debug mode, and make sure all other | ||
361 | ** writes complete before we enter debug | ||
362 | ** mode */ | ||
363 | CVMX_SYNCW; | ||
364 | cvmx_write_csr(CVMX_L2C_DBG, l2cdbg.u64); | ||
365 | cvmx_read_csr(CVMX_L2C_DBG); | ||
366 | |||
367 | CVMX_PREPARE_FOR_STORE(CVMX_ADD_SEG | ||
368 | (CVMX_MIPS_SPACE_XKPHYS, | ||
369 | set * CVMX_CACHE_LINE_SIZE), 0); | ||
370 | CVMX_SYNCW; /* Push STF out to L2 */ | ||
371 | /* Exit debug mode */ | ||
372 | CVMX_SYNC; | ||
373 | cvmx_write_csr(CVMX_L2C_DBG, 0); | ||
374 | cvmx_read_csr(CVMX_L2C_DBG); | ||
375 | } | ||
376 | } | ||
377 | |||
378 | cvmx_spinlock_unlock(&cvmx_l2c_spinlock); | ||
379 | } | ||
380 | |||
381 | int cvmx_l2c_unlock_line(uint64_t address) | ||
382 | { | ||
383 | int assoc; | ||
384 | union cvmx_l2c_tag tag; | ||
385 | union cvmx_l2c_dbg l2cdbg; | ||
386 | uint32_t tag_addr; | ||
387 | |||
388 | uint32_t index = cvmx_l2c_address_to_index(address); | ||
389 | |||
390 | cvmx_spinlock_lock(&cvmx_l2c_spinlock); | ||
391 | /* Compute portion of address that is stored in tag */ | ||
392 | tag_addr = | ||
393 | ((address >> CVMX_L2C_TAG_ADDR_ALIAS_SHIFT) & | ||
394 | ((1 << CVMX_L2C_TAG_ADDR_ALIAS_SHIFT) - 1)); | ||
395 | for (assoc = 0; assoc < CVMX_L2_ASSOC; assoc++) { | ||
396 | tag = cvmx_get_l2c_tag(assoc, index); | ||
397 | |||
398 | if (tag.s.V && (tag.s.addr == tag_addr)) { | ||
399 | l2cdbg.u64 = 0; | ||
400 | l2cdbg.s.ppnum = cvmx_get_core_num(); | ||
401 | l2cdbg.s.set = assoc; | ||
402 | l2cdbg.s.finv = 1; | ||
403 | |||
404 | CVMX_SYNC; | ||
405 | /* Enter debug mode */ | ||
406 | cvmx_write_csr(CVMX_L2C_DBG, l2cdbg.u64); | ||
407 | cvmx_read_csr(CVMX_L2C_DBG); | ||
408 | |||
409 | CVMX_PREPARE_FOR_STORE(CVMX_ADD_SEG | ||
410 | (CVMX_MIPS_SPACE_XKPHYS, | ||
411 | address), 0); | ||
412 | CVMX_SYNC; | ||
413 | /* Exit debug mode */ | ||
414 | cvmx_write_csr(CVMX_L2C_DBG, 0); | ||
415 | cvmx_read_csr(CVMX_L2C_DBG); | ||
416 | cvmx_spinlock_unlock(&cvmx_l2c_spinlock); | ||
417 | return tag.s.L; | ||
418 | } | ||
419 | } | ||
420 | cvmx_spinlock_unlock(&cvmx_l2c_spinlock); | ||
421 | return 0; | ||
422 | } | ||
423 | |||
424 | int cvmx_l2c_unlock_mem_region(uint64_t start, uint64_t len) | ||
425 | { | ||
426 | int num_unlocked = 0; | ||
427 | /* Round start/end to cache line boundaries */ | ||
428 | len += start & CVMX_CACHE_LINE_MASK; | ||
429 | start &= ~CVMX_CACHE_LINE_MASK; | ||
430 | len = (len + CVMX_CACHE_LINE_MASK) & ~CVMX_CACHE_LINE_MASK; | ||
431 | while (len > 0) { | ||
432 | num_unlocked += cvmx_l2c_unlock_line(start); | ||
433 | start += CVMX_CACHE_LINE_SIZE; | ||
434 | len -= CVMX_CACHE_LINE_SIZE; | ||
435 | } | ||
436 | |||
437 | return num_unlocked; | ||
438 | } | ||
439 | |||
440 | /* | ||
441 | * Internal l2c tag types. These are converted to a generic structure | ||
442 | * that can be used on all chips. | ||
443 | */ | ||
444 | union __cvmx_l2c_tag { | ||
445 | uint64_t u64; | ||
446 | struct cvmx_l2c_tag_cn50xx { | ||
447 | uint64_t reserved:40; | ||
448 | uint64_t V:1; /* Line valid */ | ||
449 | uint64_t D:1; /* Line dirty */ | ||
450 | uint64_t L:1; /* Line locked */ | ||
451 | uint64_t U:1; /* Use, LRU eviction */ | ||
452 | uint64_t addr:20; /* Phys mem addr (33..14) */ | ||
453 | } cn50xx; | ||
454 | struct cvmx_l2c_tag_cn30xx { | ||
455 | uint64_t reserved:41; | ||
456 | uint64_t V:1; /* Line valid */ | ||
457 | uint64_t D:1; /* Line dirty */ | ||
458 | uint64_t L:1; /* Line locked */ | ||
459 | uint64_t U:1; /* Use, LRU eviction */ | ||
460 | uint64_t addr:19; /* Phys mem addr (33..15) */ | ||
461 | } cn30xx; | ||
462 | struct cvmx_l2c_tag_cn31xx { | ||
463 | uint64_t reserved:42; | ||
464 | uint64_t V:1; /* Line valid */ | ||
465 | uint64_t D:1; /* Line dirty */ | ||
466 | uint64_t L:1; /* Line locked */ | ||
467 | uint64_t U:1; /* Use, LRU eviction */ | ||
468 | uint64_t addr:18; /* Phys mem addr (33..16) */ | ||
469 | } cn31xx; | ||
470 | struct cvmx_l2c_tag_cn38xx { | ||
471 | uint64_t reserved:43; | ||
472 | uint64_t V:1; /* Line valid */ | ||
473 | uint64_t D:1; /* Line dirty */ | ||
474 | uint64_t L:1; /* Line locked */ | ||
475 | uint64_t U:1; /* Use, LRU eviction */ | ||
476 | uint64_t addr:17; /* Phys mem addr (33..17) */ | ||
477 | } cn38xx; | ||
478 | struct cvmx_l2c_tag_cn58xx { | ||
479 | uint64_t reserved:44; | ||
480 | uint64_t V:1; /* Line valid */ | ||
481 | uint64_t D:1; /* Line dirty */ | ||
482 | uint64_t L:1; /* Line locked */ | ||
483 | uint64_t U:1; /* Use, LRU eviction */ | ||
484 | uint64_t addr:16; /* Phys mem addr (33..18) */ | ||
485 | } cn58xx; | ||
486 | struct cvmx_l2c_tag_cn58xx cn56xx; /* 2048 sets */ | ||
487 | struct cvmx_l2c_tag_cn31xx cn52xx; /* 512 sets */ | ||
488 | }; | ||
489 | |||
490 | /** | ||
491 | * @INTERNAL | ||
492 | * Function to read a L2C tag. This code make the current core | ||
493 | * the 'debug core' for the L2. This code must only be executed by | ||
494 | * 1 core at a time. | ||
495 | * | ||
496 | * @assoc: Association (way) of the tag to dump | ||
497 | * @index: Index of the cacheline | ||
498 | * | ||
499 | * Returns The Octeon model specific tag structure. This is | ||
500 | * translated by a wrapper function to a generic form that is | ||
501 | * easier for applications to use. | ||
502 | */ | ||
503 | static union __cvmx_l2c_tag __read_l2_tag(uint64_t assoc, uint64_t index) | ||
504 | { | ||
505 | |||
506 | uint64_t debug_tag_addr = (((1ULL << 63) | (index << 7)) + 96); | ||
507 | uint64_t core = cvmx_get_core_num(); | ||
508 | union __cvmx_l2c_tag tag_val; | ||
509 | uint64_t dbg_addr = CVMX_L2C_DBG; | ||
510 | unsigned long flags; | ||
511 | |||
512 | union cvmx_l2c_dbg debug_val; | ||
513 | debug_val.u64 = 0; | ||
514 | /* | ||
515 | * For low core count parts, the core number is always small enough | ||
516 | * to stay in the correct field and not set any reserved bits. | ||
517 | */ | ||
518 | debug_val.s.ppnum = core; | ||
519 | debug_val.s.l2t = 1; | ||
520 | debug_val.s.set = assoc; | ||
521 | /* | ||
522 | * Make sure core is quiet (no prefetches, etc.) before | ||
523 | * entering debug mode. | ||
524 | */ | ||
525 | CVMX_SYNC; | ||
526 | /* Flush L1 to make sure debug load misses L1 */ | ||
527 | CVMX_DCACHE_INVALIDATE; | ||
528 | |||
529 | local_irq_save(flags); | ||
530 | |||
531 | /* | ||
532 | * The following must be done in assembly as when in debug | ||
533 | * mode all data loads from L2 return special debug data, not | ||
534 | * normal memory contents. Also, interrupts must be | ||
535 | * disabled, since if an interrupt occurs while in debug mode | ||
536 | * the ISR will get debug data from all its memory reads | ||
537 | * instead of the contents of memory | ||
538 | */ | ||
539 | |||
540 | asm volatile (".set push \n" | ||
541 | " .set mips64 \n" | ||
542 | " .set noreorder \n" | ||
543 | /* Enter debug mode, wait for store */ | ||
544 | " sd %[dbg_val], 0(%[dbg_addr]) \n" | ||
545 | " ld $0, 0(%[dbg_addr]) \n" | ||
546 | /* Read L2C tag data */ | ||
547 | " ld %[tag_val], 0(%[tag_addr]) \n" | ||
548 | /* Exit debug mode, wait for store */ | ||
549 | " sd $0, 0(%[dbg_addr]) \n" | ||
550 | " ld $0, 0(%[dbg_addr]) \n" | ||
551 | /* Invalidate dcache to discard debug data */ | ||
552 | " cache 9, 0($0) \n" | ||
553 | " .set pop" : | ||
554 | [tag_val] "=r"(tag_val.u64) : [dbg_addr] "r"(dbg_addr), | ||
555 | [dbg_val] "r"(debug_val.u64), | ||
556 | [tag_addr] "r"(debug_tag_addr) : "memory"); | ||
557 | |||
558 | local_irq_restore(flags); | ||
559 | return tag_val; | ||
560 | |||
561 | } | ||
562 | |||
563 | union cvmx_l2c_tag cvmx_l2c_get_tag(uint32_t association, uint32_t index) | ||
564 | { | ||
565 | union __cvmx_l2c_tag tmp_tag; | ||
566 | union cvmx_l2c_tag tag; | ||
567 | tag.u64 = 0; | ||
568 | |||
569 | if ((int)association >= cvmx_l2c_get_num_assoc()) { | ||
570 | cvmx_dprintf | ||
571 | ("ERROR: cvmx_get_l2c_tag association out of range\n"); | ||
572 | return tag; | ||
573 | } | ||
574 | if ((int)index >= cvmx_l2c_get_num_sets()) { | ||
575 | cvmx_dprintf("ERROR: cvmx_get_l2c_tag " | ||
576 | "index out of range (arg: %d, max: %d\n", | ||
577 | index, cvmx_l2c_get_num_sets()); | ||
578 | return tag; | ||
579 | } | ||
580 | /* __read_l2_tag is intended for internal use only */ | ||
581 | tmp_tag = __read_l2_tag(association, index); | ||
582 | |||
583 | /* | ||
584 | * Convert all tag structure types to generic version, as it | ||
585 | * can represent all models. | ||
586 | */ | ||
587 | if (OCTEON_IS_MODEL(OCTEON_CN58XX) || OCTEON_IS_MODEL(OCTEON_CN56XX)) { | ||
588 | tag.s.V = tmp_tag.cn58xx.V; | ||
589 | tag.s.D = tmp_tag.cn58xx.D; | ||
590 | tag.s.L = tmp_tag.cn58xx.L; | ||
591 | tag.s.U = tmp_tag.cn58xx.U; | ||
592 | tag.s.addr = tmp_tag.cn58xx.addr; | ||
593 | } else if (OCTEON_IS_MODEL(OCTEON_CN38XX)) { | ||
594 | tag.s.V = tmp_tag.cn38xx.V; | ||
595 | tag.s.D = tmp_tag.cn38xx.D; | ||
596 | tag.s.L = tmp_tag.cn38xx.L; | ||
597 | tag.s.U = tmp_tag.cn38xx.U; | ||
598 | tag.s.addr = tmp_tag.cn38xx.addr; | ||
599 | } else if (OCTEON_IS_MODEL(OCTEON_CN31XX) | ||
600 | || OCTEON_IS_MODEL(OCTEON_CN52XX)) { | ||
601 | tag.s.V = tmp_tag.cn31xx.V; | ||
602 | tag.s.D = tmp_tag.cn31xx.D; | ||
603 | tag.s.L = tmp_tag.cn31xx.L; | ||
604 | tag.s.U = tmp_tag.cn31xx.U; | ||
605 | tag.s.addr = tmp_tag.cn31xx.addr; | ||
606 | } else if (OCTEON_IS_MODEL(OCTEON_CN30XX)) { | ||
607 | tag.s.V = tmp_tag.cn30xx.V; | ||
608 | tag.s.D = tmp_tag.cn30xx.D; | ||
609 | tag.s.L = tmp_tag.cn30xx.L; | ||
610 | tag.s.U = tmp_tag.cn30xx.U; | ||
611 | tag.s.addr = tmp_tag.cn30xx.addr; | ||
612 | } else if (OCTEON_IS_MODEL(OCTEON_CN50XX)) { | ||
613 | tag.s.V = tmp_tag.cn50xx.V; | ||
614 | tag.s.D = tmp_tag.cn50xx.D; | ||
615 | tag.s.L = tmp_tag.cn50xx.L; | ||
616 | tag.s.U = tmp_tag.cn50xx.U; | ||
617 | tag.s.addr = tmp_tag.cn50xx.addr; | ||
618 | } else { | ||
619 | cvmx_dprintf("Unsupported OCTEON Model in %s\n", __func__); | ||
620 | } | ||
621 | |||
622 | return tag; | ||
623 | } | ||
624 | |||
625 | uint32_t cvmx_l2c_address_to_index(uint64_t addr) | ||
626 | { | ||
627 | uint64_t idx = addr >> CVMX_L2C_IDX_ADDR_SHIFT; | ||
628 | union cvmx_l2c_cfg l2c_cfg; | ||
629 | l2c_cfg.u64 = cvmx_read_csr(CVMX_L2C_CFG); | ||
630 | |||
631 | if (l2c_cfg.s.idxalias) { | ||
632 | idx ^= | ||
633 | ((addr & CVMX_L2C_ALIAS_MASK) >> | ||
634 | CVMX_L2C_TAG_ADDR_ALIAS_SHIFT); | ||
635 | } | ||
636 | idx &= CVMX_L2C_IDX_MASK; | ||
637 | return idx; | ||
638 | } | ||
639 | |||
640 | int cvmx_l2c_get_cache_size_bytes(void) | ||
641 | { | ||
642 | return cvmx_l2c_get_num_sets() * cvmx_l2c_get_num_assoc() * | ||
643 | CVMX_CACHE_LINE_SIZE; | ||
644 | } | ||
645 | |||
646 | /** | ||
647 | * Return log base 2 of the number of sets in the L2 cache | ||
648 | * Returns | ||
649 | */ | ||
650 | int cvmx_l2c_get_set_bits(void) | ||
651 | { | ||
652 | int l2_set_bits; | ||
653 | if (OCTEON_IS_MODEL(OCTEON_CN56XX) || OCTEON_IS_MODEL(OCTEON_CN58XX)) | ||
654 | l2_set_bits = 11; /* 2048 sets */ | ||
655 | else if (OCTEON_IS_MODEL(OCTEON_CN38XX)) | ||
656 | l2_set_bits = 10; /* 1024 sets */ | ||
657 | else if (OCTEON_IS_MODEL(OCTEON_CN31XX) | ||
658 | || OCTEON_IS_MODEL(OCTEON_CN52XX)) | ||
659 | l2_set_bits = 9; /* 512 sets */ | ||
660 | else if (OCTEON_IS_MODEL(OCTEON_CN30XX)) | ||
661 | l2_set_bits = 8; /* 256 sets */ | ||
662 | else if (OCTEON_IS_MODEL(OCTEON_CN50XX)) | ||
663 | l2_set_bits = 7; /* 128 sets */ | ||
664 | else { | ||
665 | cvmx_dprintf("Unsupported OCTEON Model in %s\n", __func__); | ||
666 | l2_set_bits = 11; /* 2048 sets */ | ||
667 | } | ||
668 | return l2_set_bits; | ||
669 | |||
670 | } | ||
671 | |||
672 | /* Return the number of sets in the L2 Cache */ | ||
673 | int cvmx_l2c_get_num_sets(void) | ||
674 | { | ||
675 | return 1 << cvmx_l2c_get_set_bits(); | ||
676 | } | ||
677 | |||
678 | /* Return the number of associations in the L2 Cache */ | ||
679 | int cvmx_l2c_get_num_assoc(void) | ||
680 | { | ||
681 | int l2_assoc; | ||
682 | if (OCTEON_IS_MODEL(OCTEON_CN56XX) || | ||
683 | OCTEON_IS_MODEL(OCTEON_CN52XX) || | ||
684 | OCTEON_IS_MODEL(OCTEON_CN58XX) || | ||
685 | OCTEON_IS_MODEL(OCTEON_CN50XX) || OCTEON_IS_MODEL(OCTEON_CN38XX)) | ||
686 | l2_assoc = 8; | ||
687 | else if (OCTEON_IS_MODEL(OCTEON_CN31XX) || | ||
688 | OCTEON_IS_MODEL(OCTEON_CN30XX)) | ||
689 | l2_assoc = 4; | ||
690 | else { | ||
691 | cvmx_dprintf("Unsupported OCTEON Model in %s\n", __func__); | ||
692 | l2_assoc = 8; | ||
693 | } | ||
694 | |||
695 | /* Check to see if part of the cache is disabled */ | ||
696 | if (cvmx_fuse_read(265)) | ||
697 | l2_assoc = l2_assoc >> 2; | ||
698 | else if (cvmx_fuse_read(264)) | ||
699 | l2_assoc = l2_assoc >> 1; | ||
700 | |||
701 | return l2_assoc; | ||
702 | } | ||
703 | |||
704 | /** | ||
705 | * Flush a line from the L2 cache | ||
706 | * This should only be called from one core at a time, as this routine | ||
707 | * sets the core to the 'debug' core in order to flush the line. | ||
708 | * | ||
709 | * @assoc: Association (or way) to flush | ||
710 | * @index: Index to flush | ||
711 | */ | ||
712 | void cvmx_l2c_flush_line(uint32_t assoc, uint32_t index) | ||
713 | { | ||
714 | union cvmx_l2c_dbg l2cdbg; | ||
715 | |||
716 | l2cdbg.u64 = 0; | ||
717 | l2cdbg.s.ppnum = cvmx_get_core_num(); | ||
718 | l2cdbg.s.finv = 1; | ||
719 | |||
720 | l2cdbg.s.set = assoc; | ||
721 | /* | ||
722 | * Enter debug mode, and make sure all other writes complete | ||
723 | * before we enter debug mode. | ||
724 | */ | ||
725 | asm volatile ("sync" : : : "memory"); | ||
726 | cvmx_write_csr(CVMX_L2C_DBG, l2cdbg.u64); | ||
727 | cvmx_read_csr(CVMX_L2C_DBG); | ||
728 | |||
729 | CVMX_PREPARE_FOR_STORE(((1ULL << 63) + (index) * 128), 0); | ||
730 | /* Exit debug mode */ | ||
731 | asm volatile ("sync" : : : "memory"); | ||
732 | cvmx_write_csr(CVMX_L2C_DBG, 0); | ||
733 | cvmx_read_csr(CVMX_L2C_DBG); | ||
734 | } | ||
diff --git a/arch/mips/cavium-octeon/executive/cvmx-sysinfo.c b/arch/mips/cavium-octeon/executive/cvmx-sysinfo.c new file mode 100644 index 000000000000..4812370706a1 --- /dev/null +++ b/arch/mips/cavium-octeon/executive/cvmx-sysinfo.c | |||
@@ -0,0 +1,116 @@ | |||
1 | /***********************license start*************** | ||
2 | * Author: Cavium Networks | ||
3 | * | ||
4 | * Contact: support@caviumnetworks.com | ||
5 | * This file is part of the OCTEON SDK | ||
6 | * | ||
7 | * Copyright (c) 2003-2008 Cavium Networks | ||
8 | * | ||
9 | * This file is free software; you can redistribute it and/or modify | ||
10 | * it under the terms of the GNU General Public License, Version 2, as | ||
11 | * published by the Free Software Foundation. | ||
12 | * | ||
13 | * This file is distributed in the hope that it will be useful, but | ||
14 | * AS-IS and WITHOUT ANY WARRANTY; without even the implied warranty | ||
15 | * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE, TITLE, or | ||
16 | * NONINFRINGEMENT. See the GNU General Public License for more | ||
17 | * details. | ||
18 | * | ||
19 | * You should have received a copy of the GNU General Public License | ||
20 | * along with this file; if not, write to the Free Software | ||
21 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | ||
22 | * or visit http://www.gnu.org/licenses/. | ||
23 | * | ||
24 | * This file may also be available under a different license from Cavium. | ||
25 | * Contact Cavium Networks for more information | ||
26 | ***********************license end**************************************/ | ||
27 | |||
28 | /* | ||
29 | * This module provides system/board/application information obtained | ||
30 | * by the bootloader. | ||
31 | */ | ||
32 | |||
33 | #include <asm/octeon/cvmx.h> | ||
34 | #include <asm/octeon/cvmx-spinlock.h> | ||
35 | #include <asm/octeon/cvmx-sysinfo.h> | ||
36 | |||
37 | /** | ||
38 | * This structure defines the private state maintained by sysinfo module. | ||
39 | * | ||
40 | */ | ||
41 | static struct { | ||
42 | struct cvmx_sysinfo sysinfo; /* system information */ | ||
43 | cvmx_spinlock_t lock; /* mutex spinlock */ | ||
44 | |||
45 | } state = { | ||
46 | .lock = CVMX_SPINLOCK_UNLOCKED_INITIALIZER | ||
47 | }; | ||
48 | |||
49 | |||
50 | /* | ||
51 | * Global variables that define the min/max of the memory region set | ||
52 | * up for 32 bit userspace access. | ||
53 | */ | ||
54 | uint64_t linux_mem32_min; | ||
55 | uint64_t linux_mem32_max; | ||
56 | uint64_t linux_mem32_wired; | ||
57 | uint64_t linux_mem32_offset; | ||
58 | |||
59 | /** | ||
60 | * This function returns the application information as obtained | ||
61 | * by the bootloader. This provides the core mask of the cores | ||
62 | * running the same application image, as well as the physical | ||
63 | * memory regions available to the core. | ||
64 | * | ||
65 | * Returns Pointer to the boot information structure | ||
66 | * | ||
67 | */ | ||
68 | struct cvmx_sysinfo *cvmx_sysinfo_get(void) | ||
69 | { | ||
70 | return &(state.sysinfo); | ||
71 | } | ||
72 | |||
73 | /** | ||
74 | * This function is used in non-simple executive environments (such as | ||
75 | * Linux kernel, u-boot, etc.) to configure the minimal fields that | ||
76 | * are required to use simple executive files directly. | ||
77 | * | ||
78 | * Locking (if required) must be handled outside of this | ||
79 | * function | ||
80 | * | ||
81 | * @phy_mem_desc_ptr: | ||
82 | * Pointer to global physical memory descriptor | ||
83 | * (bootmem descriptor) @board_type: Octeon board | ||
84 | * type enumeration | ||
85 | * | ||
86 | * @board_rev_major: | ||
87 | * Board major revision | ||
88 | * @board_rev_minor: | ||
89 | * Board minor revision | ||
90 | * @cpu_clock_hz: | ||
91 | * CPU clock freqency in hertz | ||
92 | * | ||
93 | * Returns 0: Failure | ||
94 | * 1: success | ||
95 | */ | ||
96 | int cvmx_sysinfo_minimal_initialize(void *phy_mem_desc_ptr, | ||
97 | uint16_t board_type, | ||
98 | uint8_t board_rev_major, | ||
99 | uint8_t board_rev_minor, | ||
100 | uint32_t cpu_clock_hz) | ||
101 | { | ||
102 | |||
103 | /* The sysinfo structure was already initialized */ | ||
104 | if (state.sysinfo.board_type) | ||
105 | return 0; | ||
106 | |||
107 | memset(&(state.sysinfo), 0x0, sizeof(state.sysinfo)); | ||
108 | state.sysinfo.phy_mem_desc_ptr = phy_mem_desc_ptr; | ||
109 | state.sysinfo.board_type = board_type; | ||
110 | state.sysinfo.board_rev_major = board_rev_major; | ||
111 | state.sysinfo.board_rev_minor = board_rev_minor; | ||
112 | state.sysinfo.cpu_clock_hz = cpu_clock_hz; | ||
113 | |||
114 | return 1; | ||
115 | } | ||
116 | |||
diff --git a/arch/mips/cavium-octeon/executive/octeon-model.c b/arch/mips/cavium-octeon/executive/octeon-model.c new file mode 100644 index 000000000000..9afc3794ed1b --- /dev/null +++ b/arch/mips/cavium-octeon/executive/octeon-model.c | |||
@@ -0,0 +1,358 @@ | |||
1 | /***********************license start*************** | ||
2 | * Author: Cavium Networks | ||
3 | * | ||
4 | * Contact: support@caviumnetworks.com | ||
5 | * This file is part of the OCTEON SDK | ||
6 | * | ||
7 | * Copyright (c) 2003-2008 Cavium Networks | ||
8 | * | ||
9 | * This file is free software; you can redistribute it and/or modify | ||
10 | * it under the terms of the GNU General Public License, Version 2, as | ||
11 | * published by the Free Software Foundation. | ||
12 | * | ||
13 | * This file is distributed in the hope that it will be useful, but | ||
14 | * AS-IS and WITHOUT ANY WARRANTY; without even the implied warranty | ||
15 | * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE, TITLE, or | ||
16 | * NONINFRINGEMENT. See the GNU General Public License for more | ||
17 | * details. | ||
18 | * | ||
19 | * You should have received a copy of the GNU General Public License | ||
20 | * along with this file; if not, write to the Free Software | ||
21 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | ||
22 | * or visit http://www.gnu.org/licenses/. | ||
23 | * | ||
24 | * This file may also be available under a different license from Cavium. | ||
25 | * Contact Cavium Networks for more information | ||
26 | ***********************license end**************************************/ | ||
27 | |||
28 | /* | ||
29 | * File defining functions for working with different Octeon | ||
30 | * models. | ||
31 | */ | ||
32 | #include <asm/octeon/octeon.h> | ||
33 | |||
34 | /** | ||
35 | * Given the chip processor ID from COP0, this function returns a | ||
36 | * string representing the chip model number. The string is of the | ||
37 | * form CNXXXXpX.X-FREQ-SUFFIX. | ||
38 | * - XXXX = The chip model number | ||
39 | * - X.X = Chip pass number | ||
40 | * - FREQ = Current frequency in Mhz | ||
41 | * - SUFFIX = NSP, EXP, SCP, SSP, or CP | ||
42 | * | ||
43 | * @chip_id: Chip ID | ||
44 | * | ||
45 | * Returns Model string | ||
46 | */ | ||
47 | const char *octeon_model_get_string(uint32_t chip_id) | ||
48 | { | ||
49 | static char buffer[32]; | ||
50 | return octeon_model_get_string_buffer(chip_id, buffer); | ||
51 | } | ||
52 | |||
53 | /* | ||
54 | * Version of octeon_model_get_string() that takes buffer as argument, | ||
55 | * as running early in u-boot static/global variables don't work when | ||
56 | * running from flash. | ||
57 | */ | ||
58 | const char *octeon_model_get_string_buffer(uint32_t chip_id, char *buffer) | ||
59 | { | ||
60 | const char *family; | ||
61 | const char *core_model; | ||
62 | char pass[4]; | ||
63 | int clock_mhz; | ||
64 | const char *suffix; | ||
65 | union cvmx_l2d_fus3 fus3; | ||
66 | int num_cores; | ||
67 | union cvmx_mio_fus_dat2 fus_dat2; | ||
68 | union cvmx_mio_fus_dat3 fus_dat3; | ||
69 | char fuse_model[10]; | ||
70 | uint32_t fuse_data = 0; | ||
71 | |||
72 | fus3.u64 = cvmx_read_csr(CVMX_L2D_FUS3); | ||
73 | fus_dat2.u64 = cvmx_read_csr(CVMX_MIO_FUS_DAT2); | ||
74 | fus_dat3.u64 = cvmx_read_csr(CVMX_MIO_FUS_DAT3); | ||
75 | |||
76 | num_cores = cvmx_octeon_num_cores(); | ||
77 | |||
78 | /* Make sure the non existant devices look disabled */ | ||
79 | switch ((chip_id >> 8) & 0xff) { | ||
80 | case 6: /* CN50XX */ | ||
81 | case 2: /* CN30XX */ | ||
82 | fus_dat3.s.nodfa_dte = 1; | ||
83 | fus_dat3.s.nozip = 1; | ||
84 | break; | ||
85 | case 4: /* CN57XX or CN56XX */ | ||
86 | fus_dat3.s.nodfa_dte = 1; | ||
87 | break; | ||
88 | default: | ||
89 | break; | ||
90 | } | ||
91 | |||
92 | /* Make a guess at the suffix */ | ||
93 | /* NSP = everything */ | ||
94 | /* EXP = No crypto */ | ||
95 | /* SCP = No DFA, No zip */ | ||
96 | /* CP = No DFA, No crypto, No zip */ | ||
97 | if (fus_dat3.s.nodfa_dte) { | ||
98 | if (fus_dat2.s.nocrypto) | ||
99 | suffix = "CP"; | ||
100 | else | ||
101 | suffix = "SCP"; | ||
102 | } else if (fus_dat2.s.nocrypto) | ||
103 | suffix = "EXP"; | ||
104 | else | ||
105 | suffix = "NSP"; | ||
106 | |||
107 | /* | ||
108 | * Assume pass number is encoded using <5:3><2:0>. Exceptions | ||
109 | * will be fixed later. | ||
110 | */ | ||
111 | sprintf(pass, "%u.%u", ((chip_id >> 3) & 7) + 1, chip_id & 7); | ||
112 | |||
113 | /* | ||
114 | * Use the number of cores to determine the last 2 digits of | ||
115 | * the model number. There are some exceptions that are fixed | ||
116 | * later. | ||
117 | */ | ||
118 | switch (num_cores) { | ||
119 | case 16: | ||
120 | core_model = "60"; | ||
121 | break; | ||
122 | case 15: | ||
123 | core_model = "58"; | ||
124 | break; | ||
125 | case 14: | ||
126 | core_model = "55"; | ||
127 | break; | ||
128 | case 13: | ||
129 | core_model = "52"; | ||
130 | break; | ||
131 | case 12: | ||
132 | core_model = "50"; | ||
133 | break; | ||
134 | case 11: | ||
135 | core_model = "48"; | ||
136 | break; | ||
137 | case 10: | ||
138 | core_model = "45"; | ||
139 | break; | ||
140 | case 9: | ||
141 | core_model = "42"; | ||
142 | break; | ||
143 | case 8: | ||
144 | core_model = "40"; | ||
145 | break; | ||
146 | case 7: | ||
147 | core_model = "38"; | ||
148 | break; | ||
149 | case 6: | ||
150 | core_model = "34"; | ||
151 | break; | ||
152 | case 5: | ||
153 | core_model = "32"; | ||
154 | break; | ||
155 | case 4: | ||
156 | core_model = "30"; | ||
157 | break; | ||
158 | case 3: | ||
159 | core_model = "25"; | ||
160 | break; | ||
161 | case 2: | ||
162 | core_model = "20"; | ||
163 | break; | ||
164 | case 1: | ||
165 | core_model = "10"; | ||
166 | break; | ||
167 | default: | ||
168 | core_model = "XX"; | ||
169 | break; | ||
170 | } | ||
171 | |||
172 | /* Now figure out the family, the first two digits */ | ||
173 | switch ((chip_id >> 8) & 0xff) { | ||
174 | case 0: /* CN38XX, CN37XX or CN36XX */ | ||
175 | if (fus3.cn38xx.crip_512k) { | ||
176 | /* | ||
177 | * For some unknown reason, the 16 core one is | ||
178 | * called 37 instead of 36. | ||
179 | */ | ||
180 | if (num_cores >= 16) | ||
181 | family = "37"; | ||
182 | else | ||
183 | family = "36"; | ||
184 | } else | ||
185 | family = "38"; | ||
186 | /* | ||
187 | * This series of chips didn't follow the standard | ||
188 | * pass numbering. | ||
189 | */ | ||
190 | switch (chip_id & 0xf) { | ||
191 | case 0: | ||
192 | strcpy(pass, "1.X"); | ||
193 | break; | ||
194 | case 1: | ||
195 | strcpy(pass, "2.X"); | ||
196 | break; | ||
197 | case 3: | ||
198 | strcpy(pass, "3.X"); | ||
199 | break; | ||
200 | default: | ||
201 | strcpy(pass, "X.X"); | ||
202 | break; | ||
203 | } | ||
204 | break; | ||
205 | case 1: /* CN31XX or CN3020 */ | ||
206 | if ((chip_id & 0x10) || fus3.cn31xx.crip_128k) | ||
207 | family = "30"; | ||
208 | else | ||
209 | family = "31"; | ||
210 | /* | ||
211 | * This series of chips didn't follow the standard | ||
212 | * pass numbering. | ||
213 | */ | ||
214 | switch (chip_id & 0xf) { | ||
215 | case 0: | ||
216 | strcpy(pass, "1.0"); | ||
217 | break; | ||
218 | case 2: | ||
219 | strcpy(pass, "1.1"); | ||
220 | break; | ||
221 | default: | ||
222 | strcpy(pass, "X.X"); | ||
223 | break; | ||
224 | } | ||
225 | break; | ||
226 | case 2: /* CN3010 or CN3005 */ | ||
227 | family = "30"; | ||
228 | /* A chip with half cache is an 05 */ | ||
229 | if (fus3.cn30xx.crip_64k) | ||
230 | core_model = "05"; | ||
231 | /* | ||
232 | * This series of chips didn't follow the standard | ||
233 | * pass numbering. | ||
234 | */ | ||
235 | switch (chip_id & 0xf) { | ||
236 | case 0: | ||
237 | strcpy(pass, "1.0"); | ||
238 | break; | ||
239 | case 2: | ||
240 | strcpy(pass, "1.1"); | ||
241 | break; | ||
242 | default: | ||
243 | strcpy(pass, "X.X"); | ||
244 | break; | ||
245 | } | ||
246 | break; | ||
247 | case 3: /* CN58XX */ | ||
248 | family = "58"; | ||
249 | /* Special case. 4 core, no crypto */ | ||
250 | if ((num_cores == 4) && fus_dat2.cn38xx.nocrypto) | ||
251 | core_model = "29"; | ||
252 | |||
253 | /* Pass 1 uses different encodings for pass numbers */ | ||
254 | if ((chip_id & 0xFF) < 0x8) { | ||
255 | switch (chip_id & 0x3) { | ||
256 | case 0: | ||
257 | strcpy(pass, "1.0"); | ||
258 | break; | ||
259 | case 1: | ||
260 | strcpy(pass, "1.1"); | ||
261 | break; | ||
262 | case 3: | ||
263 | strcpy(pass, "1.2"); | ||
264 | break; | ||
265 | default: | ||
266 | strcpy(pass, "1.X"); | ||
267 | break; | ||
268 | } | ||
269 | } | ||
270 | break; | ||
271 | case 4: /* CN57XX, CN56XX, CN55XX, CN54XX */ | ||
272 | if (fus_dat2.cn56xx.raid_en) { | ||
273 | if (fus3.cn56xx.crip_1024k) | ||
274 | family = "55"; | ||
275 | else | ||
276 | family = "57"; | ||
277 | if (fus_dat2.cn56xx.nocrypto) | ||
278 | suffix = "SP"; | ||
279 | else | ||
280 | suffix = "SSP"; | ||
281 | } else { | ||
282 | if (fus_dat2.cn56xx.nocrypto) | ||
283 | suffix = "CP"; | ||
284 | else { | ||
285 | suffix = "NSP"; | ||
286 | if (fus_dat3.s.nozip) | ||
287 | suffix = "SCP"; | ||
288 | } | ||
289 | if (fus3.cn56xx.crip_1024k) | ||
290 | family = "54"; | ||
291 | else | ||
292 | family = "56"; | ||
293 | } | ||
294 | break; | ||
295 | case 6: /* CN50XX */ | ||
296 | family = "50"; | ||
297 | break; | ||
298 | case 7: /* CN52XX */ | ||
299 | if (fus3.cn52xx.crip_256k) | ||
300 | family = "51"; | ||
301 | else | ||
302 | family = "52"; | ||
303 | break; | ||
304 | default: | ||
305 | family = "XX"; | ||
306 | core_model = "XX"; | ||
307 | strcpy(pass, "X.X"); | ||
308 | suffix = "XXX"; | ||
309 | break; | ||
310 | } | ||
311 | |||
312 | clock_mhz = octeon_get_clock_rate() / 1000000; | ||
313 | |||
314 | if (family[0] != '3') { | ||
315 | /* Check for model in fuses, overrides normal decode */ | ||
316 | /* This is _not_ valid for Octeon CN3XXX models */ | ||
317 | fuse_data |= cvmx_fuse_read_byte(51); | ||
318 | fuse_data = fuse_data << 8; | ||
319 | fuse_data |= cvmx_fuse_read_byte(50); | ||
320 | fuse_data = fuse_data << 8; | ||
321 | fuse_data |= cvmx_fuse_read_byte(49); | ||
322 | fuse_data = fuse_data << 8; | ||
323 | fuse_data |= cvmx_fuse_read_byte(48); | ||
324 | if (fuse_data & 0x7ffff) { | ||
325 | int model = fuse_data & 0x3fff; | ||
326 | int suffix = (fuse_data >> 14) & 0x1f; | ||
327 | if (suffix && model) { | ||
328 | /* | ||
329 | * Have both number and suffix in | ||
330 | * fuses, so both | ||
331 | */ | ||
332 | sprintf(fuse_model, "%d%c", | ||
333 | model, 'A' + suffix - 1); | ||
334 | core_model = ""; | ||
335 | family = fuse_model; | ||
336 | } else if (suffix && !model) { | ||
337 | /* | ||
338 | * Only have suffix, so add suffix to | ||
339 | * 'normal' model number. | ||
340 | */ | ||
341 | sprintf(fuse_model, "%s%c", core_model, | ||
342 | 'A' + suffix - 1); | ||
343 | core_model = fuse_model; | ||
344 | } else { | ||
345 | /* | ||
346 | * Don't have suffix, so just use | ||
347 | * model from fuses. | ||
348 | */ | ||
349 | sprintf(fuse_model, "%d", model); | ||
350 | core_model = ""; | ||
351 | family = fuse_model; | ||
352 | } | ||
353 | } | ||
354 | } | ||
355 | sprintf(buffer, "CN%s%sp%s-%d-%s", | ||
356 | family, core_model, pass, clock_mhz, suffix); | ||
357 | return buffer; | ||
358 | } | ||
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 | |||
17 | static struct map_info flash_map; | ||
18 | static struct mtd_info *mymtd; | ||
19 | #ifdef CONFIG_MTD_PARTITIONS | ||
20 | static int nr_parts; | ||
21 | static struct mtd_partition *parts; | ||
22 | static 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 | */ | ||
36 | static 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 | |||
84 | late_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 | |||
14 | DEFINE_RWLOCK(octeon_irq_ciu0_rwlock); | ||
15 | DEFINE_RWLOCK(octeon_irq_ciu1_rwlock); | ||
16 | DEFINE_SPINLOCK(octeon_irq_msi_lock); | ||
17 | |||
18 | static 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 | |||
32 | static 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 | |||
54 | static 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 | |||
68 | static 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 | |||
81 | static 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 | |||
91 | static 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 | |||
100 | static 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 | |||
116 | static 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 | |||
126 | static 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 | |||
148 | static 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 | ||
180 | static 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 | |||
205 | static 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 | |||
217 | static 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 | |||
231 | static 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 | |||
241 | static 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 | |||
263 | static 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 | ||
295 | static 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 | |||
321 | static 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 | |||
334 | static 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 | |||
350 | static void octeon_irq_msi_eoi(unsigned int irq) | ||
351 | { | ||
352 | /* Nothing needed */ | ||
353 | } | ||
354 | |||
355 | static 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 | |||
383 | static 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 | |||
405 | static 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 | |||
414 | void __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 | |||
457 | asmlinkage 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) \ | ||
74 | 9: 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 | ||
173 | LEAF(memcpy) /* a0=dst a1=src a2=len */ | ||
174 | move v0, dst /* return value */ | ||
175 | __memcpy: | ||
176 | FEXPORT(__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 | ||
203 | 2: pref 0, 256(src) # We must not prefetch invalid addresses | ||
204 | # | ||
205 | # This is where we loop if we can't prefetch anymore | ||
206 | 1: | ||
207 | EXC( LOAD t0, UNIT(0)(src), l_exc) | ||
208 | EXC( LOAD t1, UNIT(1)(src), l_exc_copy) | ||
209 | EXC( LOAD t2, UNIT(2)(src), l_exc_copy) | ||
210 | EXC( LOAD t3, UNIT(3)(src), l_exc_copy) | ||
211 | SUB len, len, 16*NBYTES | ||
212 | EXC( STORE t0, UNIT(0)(dst), s_exc_p16u) | ||
213 | EXC( STORE t1, UNIT(1)(dst), s_exc_p15u) | ||
214 | EXC( STORE t2, UNIT(2)(dst), s_exc_p14u) | ||
215 | EXC( STORE t3, UNIT(3)(dst), s_exc_p13u) | ||
216 | EXC( LOAD t0, UNIT(4)(src), l_exc_copy) | ||
217 | EXC( LOAD t1, UNIT(5)(src), l_exc_copy) | ||
218 | EXC( LOAD t2, UNIT(6)(src), l_exc_copy) | ||
219 | EXC( LOAD t3, UNIT(7)(src), l_exc_copy) | ||
220 | EXC( STORE t0, UNIT(4)(dst), s_exc_p12u) | ||
221 | EXC( STORE t1, UNIT(5)(dst), s_exc_p11u) | ||
222 | EXC( STORE t2, UNIT(6)(dst), s_exc_p10u) | ||
223 | ADD src, src, 16*NBYTES | ||
224 | EXC( STORE t3, UNIT(7)(dst), s_exc_p9u) | ||
225 | ADD dst, dst, 16*NBYTES | ||
226 | EXC( LOAD t0, UNIT(-8)(src), l_exc_copy) | ||
227 | EXC( LOAD t1, UNIT(-7)(src), l_exc_copy) | ||
228 | EXC( LOAD t2, UNIT(-6)(src), l_exc_copy) | ||
229 | EXC( LOAD t3, UNIT(-5)(src), l_exc_copy) | ||
230 | EXC( STORE t0, UNIT(-8)(dst), s_exc_p8u) | ||
231 | EXC( STORE t1, UNIT(-7)(dst), s_exc_p7u) | ||
232 | EXC( STORE t2, UNIT(-6)(dst), s_exc_p6u) | ||
233 | EXC( STORE t3, UNIT(-5)(dst), s_exc_p5u) | ||
234 | EXC( LOAD t0, UNIT(-4)(src), l_exc_copy) | ||
235 | EXC( LOAD t1, UNIT(-3)(src), l_exc_copy) | ||
236 | EXC( LOAD t2, UNIT(-2)(src), l_exc_copy) | ||
237 | EXC( LOAD t3, UNIT(-1)(src), l_exc_copy) | ||
238 | EXC( STORE t0, UNIT(-4)(dst), s_exc_p4u) | ||
239 | EXC( STORE t1, UNIT(-3)(dst), s_exc_p3u) | ||
240 | EXC( STORE t2, UNIT(-2)(dst), s_exc_p2u) | ||
241 | EXC( 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 | # | ||
250 | cleanup_both_aligned: | ||
251 | beqz len, done | ||
252 | sltu t0, len, 8*NBYTES | ||
253 | bnez t0, less_than_8units | ||
254 | nop | ||
255 | EXC( LOAD t0, UNIT(0)(src), l_exc) | ||
256 | EXC( LOAD t1, UNIT(1)(src), l_exc_copy) | ||
257 | EXC( LOAD t2, UNIT(2)(src), l_exc_copy) | ||
258 | EXC( LOAD t3, UNIT(3)(src), l_exc_copy) | ||
259 | SUB len, len, 8*NBYTES | ||
260 | EXC( STORE t0, UNIT(0)(dst), s_exc_p8u) | ||
261 | EXC( STORE t1, UNIT(1)(dst), s_exc_p7u) | ||
262 | EXC( STORE t2, UNIT(2)(dst), s_exc_p6u) | ||
263 | EXC( STORE t3, UNIT(3)(dst), s_exc_p5u) | ||
264 | EXC( LOAD t0, UNIT(4)(src), l_exc_copy) | ||
265 | EXC( LOAD t1, UNIT(5)(src), l_exc_copy) | ||
266 | EXC( LOAD t2, UNIT(6)(src), l_exc_copy) | ||
267 | EXC( LOAD t3, UNIT(7)(src), l_exc_copy) | ||
268 | EXC( STORE t0, UNIT(4)(dst), s_exc_p4u) | ||
269 | EXC( STORE t1, UNIT(5)(dst), s_exc_p3u) | ||
270 | EXC( STORE t2, UNIT(6)(dst), s_exc_p2u) | ||
271 | EXC( 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 | # | ||
278 | less_than_8units: | ||
279 | sltu t0, len, 4*NBYTES | ||
280 | bnez t0, less_than_4units | ||
281 | nop | ||
282 | EXC( LOAD t0, UNIT(0)(src), l_exc) | ||
283 | EXC( LOAD t1, UNIT(1)(src), l_exc_copy) | ||
284 | EXC( LOAD t2, UNIT(2)(src), l_exc_copy) | ||
285 | EXC( LOAD t3, UNIT(3)(src), l_exc_copy) | ||
286 | SUB len, len, 4*NBYTES | ||
287 | EXC( STORE t0, UNIT(0)(dst), s_exc_p4u) | ||
288 | EXC( STORE t1, UNIT(1)(dst), s_exc_p3u) | ||
289 | EXC( STORE t2, UNIT(2)(dst), s_exc_p2u) | ||
290 | EXC( 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 | # | ||
298 | less_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 | # | ||
305 | EXC( LOAD t0, 0(src), l_exc) | ||
306 | SUB len, len, NBYTES | ||
307 | sltu t1, len, 8 | ||
308 | EXC( 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 | # | ||
315 | EXC( LOAD t0, 0(src), l_exc) | ||
316 | SUB len, len, NBYTES | ||
317 | sltu t1, len, 8 | ||
318 | EXC( 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 | # | ||
325 | EXC( 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 | ||
330 | EXC( STORE t0, -8(dst), s_exc_p1u) | ||
331 | |||
332 | src_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 | ||
337 | 1: | ||
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 | */ | ||
344 | EXC( LDFIRST t0, FIRST(0)(src), l_exc) | ||
345 | EXC( LDFIRST t1, FIRST(1)(src), l_exc_copy) | ||
346 | SUB len, len, 4*NBYTES | ||
347 | EXC( LDREST t0, REST(0)(src), l_exc_copy) | ||
348 | EXC( LDREST t1, REST(1)(src), l_exc_copy) | ||
349 | EXC( LDFIRST t2, FIRST(2)(src), l_exc_copy) | ||
350 | EXC( LDFIRST t3, FIRST(3)(src), l_exc_copy) | ||
351 | EXC( LDREST t2, REST(2)(src), l_exc_copy) | ||
352 | EXC( LDREST t3, REST(3)(src), l_exc_copy) | ||
353 | ADD src, src, 4*NBYTES | ||
354 | EXC( STORE t0, UNIT(0)(dst), s_exc_p4u) | ||
355 | EXC( STORE t1, UNIT(1)(dst), s_exc_p3u) | ||
356 | EXC( STORE t2, UNIT(2)(dst), s_exc_p2u) | ||
357 | EXC( STORE t3, UNIT(3)(dst), s_exc_p1u) | ||
358 | bne len, rem, 1b | ||
359 | ADD dst, dst, 4*NBYTES | ||
360 | |||
361 | cleanup_src_unaligned: | ||
362 | beqz len, done | ||
363 | and rem, len, NBYTES-1 # rem = len % NBYTES | ||
364 | beq rem, len, copy_bytes | ||
365 | nop | ||
366 | 1: | ||
367 | EXC( LDFIRST t0, FIRST(0)(src), l_exc) | ||
368 | EXC( LDREST t0, REST(0)(src), l_exc_copy) | ||
369 | SUB len, len, NBYTES | ||
370 | EXC( STORE t0, 0(dst), s_exc_p1u) | ||
371 | ADD src, src, NBYTES | ||
372 | bne len, rem, 1b | ||
373 | ADD dst, dst, NBYTES | ||
374 | |||
375 | copy_bytes_checklen: | ||
376 | beqz len, done | ||
377 | nop | ||
378 | copy_bytes: | ||
379 | /* 0 < len < NBYTES */ | ||
380 | #define COPY_BYTE(N) \ | ||
381 | EXC( lb t0, N(src), l_exc); \ | ||
382 | SUB len, len, 1; \ | ||
383 | beqz len, done; \ | ||
384 | EXC( 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 | ||
394 | EXC( lb t0, NBYTES-2(src), l_exc) | ||
395 | SUB len, len, 1 | ||
396 | jr ra | ||
397 | EXC( sb t0, NBYTES-2(dst), s_exc_p1) | ||
398 | done: | ||
399 | jr ra | ||
400 | nop | ||
401 | END(memcpy) | ||
402 | |||
403 | l_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) | ||
417 | 1: | ||
418 | EXC( 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 | ||
423 | l_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 | ||
442 | 1: 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) \ | ||
451 | s_exc_p ## n ## u: \ | ||
452 | jr ra; \ | ||
453 | ADD len, len, n*NBYTES | ||
454 | |||
455 | SEXC(16) | ||
456 | SEXC(15) | ||
457 | SEXC(14) | ||
458 | SEXC(13) | ||
459 | SEXC(12) | ||
460 | SEXC(11) | ||
461 | SEXC(10) | ||
462 | SEXC(9) | ||
463 | SEXC(8) | ||
464 | SEXC(7) | ||
465 | SEXC(6) | ||
466 | SEXC(5) | ||
467 | SEXC(4) | ||
468 | SEXC(3) | ||
469 | SEXC(2) | ||
470 | SEXC(1) | ||
471 | |||
472 | s_exc_p1: | ||
473 | jr ra | ||
474 | ADD len, len, 1 | ||
475 | s_exc: | ||
476 | jr ra | ||
477 | nop | ||
478 | |||
479 | .align 5 | ||
480 | LEAF(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 */ | ||
492 | LEAF(__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 | |||
499 | r_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 | |||
507 | r_out: | ||
508 | jr ra | ||
509 | move a2, zero | ||
510 | |||
511 | r_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 | |||
27 | unsigned 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 | |||
38 | void 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 | ||
53 | static struct plat_serial8250_port octeon_uart8250_data[OCTEON_MAX_UARTS + 1]; | ||
54 | static 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 | |||
62 | static 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 | |||
73 | static 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 | |||
136 | device_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..5f4e49ba4713 --- /dev/null +++ b/arch/mips/cavium-octeon/setup.c | |||
@@ -0,0 +1,927 @@ | |||
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/tty.h> | ||
19 | #include <linux/time.h> | ||
20 | #include <linux/platform_device.h> | ||
21 | #include <linux/serial_core.h> | ||
22 | #include <linux/serial_8250.h> | ||
23 | |||
24 | #include <asm/processor.h> | ||
25 | #include <asm/reboot.h> | ||
26 | #include <asm/smp-ops.h> | ||
27 | #include <asm/system.h> | ||
28 | #include <asm/irq_cpu.h> | ||
29 | #include <asm/mipsregs.h> | ||
30 | #include <asm/bootinfo.h> | ||
31 | #include <asm/sections.h> | ||
32 | #include <asm/time.h> | ||
33 | |||
34 | #include <asm/octeon/octeon.h> | ||
35 | |||
36 | #ifdef CONFIG_CAVIUM_DECODE_RSL | ||
37 | extern void cvmx_interrupt_rsl_decode(void); | ||
38 | extern int __cvmx_interrupt_ecc_report_single_bit_errors; | ||
39 | extern void cvmx_interrupt_rsl_enable(void); | ||
40 | #endif | ||
41 | |||
42 | extern struct plat_smp_ops octeon_smp_ops; | ||
43 | |||
44 | #ifdef CONFIG_PCI | ||
45 | extern void pci_console_init(const char *arg); | ||
46 | #endif | ||
47 | |||
48 | #ifdef CONFIG_CAVIUM_RESERVE32 | ||
49 | extern uint64_t octeon_reserve32_memory; | ||
50 | #endif | ||
51 | static unsigned long long MAX_MEMORY = 512ull << 20; | ||
52 | |||
53 | struct octeon_boot_descriptor *octeon_boot_desc_ptr; | ||
54 | |||
55 | struct cvmx_bootinfo *octeon_bootinfo; | ||
56 | EXPORT_SYMBOL(octeon_bootinfo); | ||
57 | |||
58 | #ifdef CONFIG_CAVIUM_RESERVE32 | ||
59 | uint64_t octeon_reserve32_memory; | ||
60 | EXPORT_SYMBOL(octeon_reserve32_memory); | ||
61 | #endif | ||
62 | |||
63 | static int octeon_uart; | ||
64 | |||
65 | extern asmlinkage void handle_int(void); | ||
66 | extern asmlinkage void plat_irq_dispatch(void); | ||
67 | |||
68 | /** | ||
69 | * Return non zero if we are currently running in the Octeon simulator | ||
70 | * | ||
71 | * Returns | ||
72 | */ | ||
73 | int octeon_is_simulation(void) | ||
74 | { | ||
75 | return octeon_bootinfo->board_type == CVMX_BOARD_TYPE_SIM; | ||
76 | } | ||
77 | EXPORT_SYMBOL(octeon_is_simulation); | ||
78 | |||
79 | /** | ||
80 | * Return true if Octeon is in PCI Host mode. This means | ||
81 | * Linux can control the PCI bus. | ||
82 | * | ||
83 | * Returns Non zero if Octeon in host mode. | ||
84 | */ | ||
85 | int octeon_is_pci_host(void) | ||
86 | { | ||
87 | #ifdef CONFIG_PCI | ||
88 | return octeon_bootinfo->config_flags & CVMX_BOOTINFO_CFG_FLAG_PCI_HOST; | ||
89 | #else | ||
90 | return 0; | ||
91 | #endif | ||
92 | } | ||
93 | |||
94 | /** | ||
95 | * Get the clock rate of Octeon | ||
96 | * | ||
97 | * Returns Clock rate in HZ | ||
98 | */ | ||
99 | uint64_t octeon_get_clock_rate(void) | ||
100 | { | ||
101 | if (octeon_is_simulation()) | ||
102 | octeon_bootinfo->eclock_hz = 6000000; | ||
103 | return octeon_bootinfo->eclock_hz; | ||
104 | } | ||
105 | EXPORT_SYMBOL(octeon_get_clock_rate); | ||
106 | |||
107 | /** | ||
108 | * Write to the LCD display connected to the bootbus. This display | ||
109 | * exists on most Cavium evaluation boards. If it doesn't exist, then | ||
110 | * this function doesn't do anything. | ||
111 | * | ||
112 | * @s: String to write | ||
113 | */ | ||
114 | void octeon_write_lcd(const char *s) | ||
115 | { | ||
116 | if (octeon_bootinfo->led_display_base_addr) { | ||
117 | void __iomem *lcd_address = | ||
118 | ioremap_nocache(octeon_bootinfo->led_display_base_addr, | ||
119 | 8); | ||
120 | int i; | ||
121 | for (i = 0; i < 8; i++, s++) { | ||
122 | if (*s) | ||
123 | iowrite8(*s, lcd_address + i); | ||
124 | else | ||
125 | iowrite8(' ', lcd_address + i); | ||
126 | } | ||
127 | iounmap(lcd_address); | ||
128 | } | ||
129 | } | ||
130 | |||
131 | /** | ||
132 | * Return the console uart passed by the bootloader | ||
133 | * | ||
134 | * Returns uart (0 or 1) | ||
135 | */ | ||
136 | int octeon_get_boot_uart(void) | ||
137 | { | ||
138 | int uart; | ||
139 | #ifdef CONFIG_CAVIUM_OCTEON_2ND_KERNEL | ||
140 | uart = 1; | ||
141 | #else | ||
142 | uart = (octeon_boot_desc_ptr->flags & OCTEON_BL_FLAG_CONSOLE_UART1) ? | ||
143 | 1 : 0; | ||
144 | #endif | ||
145 | return uart; | ||
146 | } | ||
147 | |||
148 | /** | ||
149 | * Get the coremask Linux was booted on. | ||
150 | * | ||
151 | * Returns Core mask | ||
152 | */ | ||
153 | int octeon_get_boot_coremask(void) | ||
154 | { | ||
155 | return octeon_boot_desc_ptr->core_mask; | ||
156 | } | ||
157 | |||
158 | /** | ||
159 | * Check the hardware BIST results for a CPU | ||
160 | */ | ||
161 | void octeon_check_cpu_bist(void) | ||
162 | { | ||
163 | const int coreid = cvmx_get_core_num(); | ||
164 | unsigned long long mask; | ||
165 | unsigned long long bist_val; | ||
166 | |||
167 | /* Check BIST results for COP0 registers */ | ||
168 | mask = 0x1f00000000ull; | ||
169 | bist_val = read_octeon_c0_icacheerr(); | ||
170 | if (bist_val & mask) | ||
171 | pr_err("Core%d BIST Failure: CacheErr(icache) = 0x%llx\n", | ||
172 | coreid, bist_val); | ||
173 | |||
174 | bist_val = read_octeon_c0_dcacheerr(); | ||
175 | if (bist_val & 1) | ||
176 | pr_err("Core%d L1 Dcache parity error: " | ||
177 | "CacheErr(dcache) = 0x%llx\n", | ||
178 | coreid, bist_val); | ||
179 | |||
180 | mask = 0xfc00000000000000ull; | ||
181 | bist_val = read_c0_cvmmemctl(); | ||
182 | if (bist_val & mask) | ||
183 | pr_err("Core%d BIST Failure: COP0_CVM_MEM_CTL = 0x%llx\n", | ||
184 | coreid, bist_val); | ||
185 | |||
186 | write_octeon_c0_dcacheerr(0); | ||
187 | } | ||
188 | |||
189 | #ifdef CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB | ||
190 | /** | ||
191 | * Called on every core to setup the wired tlb entry needed | ||
192 | * if CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB is set. | ||
193 | * | ||
194 | */ | ||
195 | static void octeon_hal_setup_per_cpu_reserved32(void *unused) | ||
196 | { | ||
197 | /* | ||
198 | * The config has selected to wire the reserve32 memory for all | ||
199 | * userspace applications. We need to put a wired TLB entry in for each | ||
200 | * 512MB of reserve32 memory. We only handle double 256MB pages here, | ||
201 | * so reserve32 must be multiple of 512MB. | ||
202 | */ | ||
203 | uint32_t size = CONFIG_CAVIUM_RESERVE32; | ||
204 | uint32_t entrylo0 = | ||
205 | 0x7 | ((octeon_reserve32_memory & ((1ul << 40) - 1)) >> 6); | ||
206 | uint32_t entrylo1 = entrylo0 + (256 << 14); | ||
207 | uint32_t entryhi = (0x80000000UL - (CONFIG_CAVIUM_RESERVE32 << 20)); | ||
208 | while (size >= 512) { | ||
209 | #if 0 | ||
210 | pr_info("CPU%d: Adding double wired TLB entry for 0x%lx\n", | ||
211 | smp_processor_id(), entryhi); | ||
212 | #endif | ||
213 | add_wired_entry(entrylo0, entrylo1, entryhi, PM_256M); | ||
214 | entrylo0 += 512 << 14; | ||
215 | entrylo1 += 512 << 14; | ||
216 | entryhi += 512 << 20; | ||
217 | size -= 512; | ||
218 | } | ||
219 | } | ||
220 | #endif /* CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB */ | ||
221 | |||
222 | /** | ||
223 | * Called to release the named block which was used to made sure | ||
224 | * that nobody used the memory for something else during | ||
225 | * init. Now we'll free it so userspace apps can use this | ||
226 | * memory region with bootmem_alloc. | ||
227 | * | ||
228 | * This function is called only once from prom_free_prom_memory(). | ||
229 | */ | ||
230 | void octeon_hal_setup_reserved32(void) | ||
231 | { | ||
232 | #ifdef CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB | ||
233 | on_each_cpu(octeon_hal_setup_per_cpu_reserved32, NULL, 0, 1); | ||
234 | #endif | ||
235 | } | ||
236 | |||
237 | /** | ||
238 | * Reboot Octeon | ||
239 | * | ||
240 | * @command: Command to pass to the bootloader. Currently ignored. | ||
241 | */ | ||
242 | static void octeon_restart(char *command) | ||
243 | { | ||
244 | /* Disable all watchdogs before soft reset. They don't get cleared */ | ||
245 | #ifdef CONFIG_SMP | ||
246 | int cpu; | ||
247 | for_each_online_cpu(cpu) | ||
248 | cvmx_write_csr(CVMX_CIU_WDOGX(cpu_logical_map(cpu)), 0); | ||
249 | #else | ||
250 | cvmx_write_csr(CVMX_CIU_WDOGX(cvmx_get_core_num()), 0); | ||
251 | #endif | ||
252 | |||
253 | mb(); | ||
254 | while (1) | ||
255 | cvmx_write_csr(CVMX_CIU_SOFT_RST, 1); | ||
256 | } | ||
257 | |||
258 | |||
259 | /** | ||
260 | * Permanently stop a core. | ||
261 | * | ||
262 | * @arg: Ignored. | ||
263 | */ | ||
264 | static void octeon_kill_core(void *arg) | ||
265 | { | ||
266 | mb(); | ||
267 | if (octeon_is_simulation()) { | ||
268 | /* The simulator needs the watchdog to stop for dead cores */ | ||
269 | cvmx_write_csr(CVMX_CIU_WDOGX(cvmx_get_core_num()), 0); | ||
270 | /* A break instruction causes the simulator stop a core */ | ||
271 | asm volatile ("sync\nbreak"); | ||
272 | } | ||
273 | } | ||
274 | |||
275 | |||
276 | /** | ||
277 | * Halt the system | ||
278 | */ | ||
279 | static void octeon_halt(void) | ||
280 | { | ||
281 | smp_call_function(octeon_kill_core, NULL, 0); | ||
282 | |||
283 | switch (octeon_bootinfo->board_type) { | ||
284 | case CVMX_BOARD_TYPE_NAO38: | ||
285 | /* Driving a 1 to GPIO 12 shuts off this board */ | ||
286 | cvmx_write_csr(CVMX_GPIO_BIT_CFGX(12), 1); | ||
287 | cvmx_write_csr(CVMX_GPIO_TX_SET, 0x1000); | ||
288 | break; | ||
289 | default: | ||
290 | octeon_write_lcd("PowerOff"); | ||
291 | break; | ||
292 | } | ||
293 | |||
294 | octeon_kill_core(NULL); | ||
295 | } | ||
296 | |||
297 | #if 0 | ||
298 | /** | ||
299 | * Platform time init specifics. | ||
300 | * Returns | ||
301 | */ | ||
302 | void __init plat_time_init(void) | ||
303 | { | ||
304 | /* Nothing special here, but we are required to have one */ | ||
305 | } | ||
306 | |||
307 | #endif | ||
308 | |||
309 | /** | ||
310 | * Handle all the error condition interrupts that might occur. | ||
311 | * | ||
312 | */ | ||
313 | #ifdef CONFIG_CAVIUM_DECODE_RSL | ||
314 | static irqreturn_t octeon_rlm_interrupt(int cpl, void *dev_id) | ||
315 | { | ||
316 | cvmx_interrupt_rsl_decode(); | ||
317 | return IRQ_HANDLED; | ||
318 | } | ||
319 | #endif | ||
320 | |||
321 | /** | ||
322 | * Return a string representing the system type | ||
323 | * | ||
324 | * Returns | ||
325 | */ | ||
326 | const char *octeon_board_type_string(void) | ||
327 | { | ||
328 | static char name[80]; | ||
329 | sprintf(name, "%s (%s)", | ||
330 | cvmx_board_type_to_string(octeon_bootinfo->board_type), | ||
331 | octeon_model_get_string(read_c0_prid())); | ||
332 | return name; | ||
333 | } | ||
334 | |||
335 | const char *get_system_type(void) | ||
336 | __attribute__ ((alias("octeon_board_type_string"))); | ||
337 | |||
338 | void octeon_user_io_init(void) | ||
339 | { | ||
340 | union octeon_cvmemctl cvmmemctl; | ||
341 | union cvmx_iob_fau_timeout fau_timeout; | ||
342 | union cvmx_pow_nw_tim nm_tim; | ||
343 | uint64_t cvmctl; | ||
344 | |||
345 | /* Get the current settings for CP0_CVMMEMCTL_REG */ | ||
346 | cvmmemctl.u64 = read_c0_cvmmemctl(); | ||
347 | /* R/W If set, marked write-buffer entries time out the same | ||
348 | * as as other entries; if clear, marked write-buffer entries | ||
349 | * use the maximum timeout. */ | ||
350 | cvmmemctl.s.dismarkwblongto = 1; | ||
351 | /* R/W If set, a merged store does not clear the write-buffer | ||
352 | * entry timeout state. */ | ||
353 | cvmmemctl.s.dismrgclrwbto = 0; | ||
354 | /* R/W Two bits that are the MSBs of the resultant CVMSEG LM | ||
355 | * word location for an IOBDMA. The other 8 bits come from the | ||
356 | * SCRADDR field of the IOBDMA. */ | ||
357 | cvmmemctl.s.iobdmascrmsb = 0; | ||
358 | /* R/W If set, SYNCWS and SYNCS only order marked stores; if | ||
359 | * clear, SYNCWS and SYNCS only order unmarked | ||
360 | * stores. SYNCWSMARKED has no effect when DISSYNCWS is | ||
361 | * set. */ | ||
362 | cvmmemctl.s.syncwsmarked = 0; | ||
363 | /* R/W If set, SYNCWS acts as SYNCW and SYNCS acts as SYNC. */ | ||
364 | cvmmemctl.s.dissyncws = 0; | ||
365 | /* R/W If set, no stall happens on write buffer full. */ | ||
366 | if (OCTEON_IS_MODEL(OCTEON_CN38XX_PASS2)) | ||
367 | cvmmemctl.s.diswbfst = 1; | ||
368 | else | ||
369 | cvmmemctl.s.diswbfst = 0; | ||
370 | /* R/W If set (and SX set), supervisor-level loads/stores can | ||
371 | * use XKPHYS addresses with <48>==0 */ | ||
372 | cvmmemctl.s.xkmemenas = 0; | ||
373 | |||
374 | /* R/W If set (and UX set), user-level loads/stores can use | ||
375 | * XKPHYS addresses with VA<48>==0 */ | ||
376 | cvmmemctl.s.xkmemenau = 0; | ||
377 | |||
378 | /* R/W If set (and SX set), supervisor-level loads/stores can | ||
379 | * use XKPHYS addresses with VA<48>==1 */ | ||
380 | cvmmemctl.s.xkioenas = 0; | ||
381 | |||
382 | /* R/W If set (and UX set), user-level loads/stores can use | ||
383 | * XKPHYS addresses with VA<48>==1 */ | ||
384 | cvmmemctl.s.xkioenau = 0; | ||
385 | |||
386 | /* R/W If set, all stores act as SYNCW (NOMERGE must be set | ||
387 | * when this is set) RW, reset to 0. */ | ||
388 | cvmmemctl.s.allsyncw = 0; | ||
389 | |||
390 | /* R/W If set, no stores merge, and all stores reach the | ||
391 | * coherent bus in order. */ | ||
392 | cvmmemctl.s.nomerge = 0; | ||
393 | /* R/W Selects the bit in the counter used for DID time-outs 0 | ||
394 | * = 231, 1 = 230, 2 = 229, 3 = 214. Actual time-out is | ||
395 | * between 1x and 2x this interval. For example, with | ||
396 | * DIDTTO=3, expiration interval is between 16K and 32K. */ | ||
397 | cvmmemctl.s.didtto = 0; | ||
398 | /* R/W If set, the (mem) CSR clock never turns off. */ | ||
399 | cvmmemctl.s.csrckalwys = 0; | ||
400 | /* R/W If set, mclk never turns off. */ | ||
401 | cvmmemctl.s.mclkalwys = 0; | ||
402 | /* R/W Selects the bit in the counter used for write buffer | ||
403 | * flush time-outs (WBFLT+11) is the bit position in an | ||
404 | * internal counter used to determine expiration. The write | ||
405 | * buffer expires between 1x and 2x this interval. For | ||
406 | * example, with WBFLT = 0, a write buffer expires between 2K | ||
407 | * and 4K cycles after the write buffer entry is allocated. */ | ||
408 | cvmmemctl.s.wbfltime = 0; | ||
409 | /* R/W If set, do not put Istream in the L2 cache. */ | ||
410 | cvmmemctl.s.istrnol2 = 0; | ||
411 | /* R/W The write buffer threshold. */ | ||
412 | cvmmemctl.s.wbthresh = 10; | ||
413 | /* R/W If set, CVMSEG is available for loads/stores in | ||
414 | * kernel/debug mode. */ | ||
415 | #if CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE > 0 | ||
416 | cvmmemctl.s.cvmsegenak = 1; | ||
417 | #else | ||
418 | cvmmemctl.s.cvmsegenak = 0; | ||
419 | #endif | ||
420 | /* R/W If set, CVMSEG is available for loads/stores in | ||
421 | * supervisor mode. */ | ||
422 | cvmmemctl.s.cvmsegenas = 0; | ||
423 | /* R/W If set, CVMSEG is available for loads/stores in user | ||
424 | * mode. */ | ||
425 | cvmmemctl.s.cvmsegenau = 0; | ||
426 | /* R/W Size of local memory in cache blocks, 54 (6912 bytes) | ||
427 | * is max legal value. */ | ||
428 | cvmmemctl.s.lmemsz = CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE; | ||
429 | |||
430 | |||
431 | if (smp_processor_id() == 0) | ||
432 | pr_notice("CVMSEG size: %d cache lines (%d bytes)\n", | ||
433 | CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE, | ||
434 | CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE * 128); | ||
435 | |||
436 | write_c0_cvmmemctl(cvmmemctl.u64); | ||
437 | |||
438 | /* Move the performance counter interrupts to IRQ 6 */ | ||
439 | cvmctl = read_c0_cvmctl(); | ||
440 | cvmctl &= ~(7 << 7); | ||
441 | cvmctl |= 6 << 7; | ||
442 | write_c0_cvmctl(cvmctl); | ||
443 | |||
444 | /* Set a default for the hardware timeouts */ | ||
445 | fau_timeout.u64 = 0; | ||
446 | fau_timeout.s.tout_val = 0xfff; | ||
447 | /* Disable tagwait FAU timeout */ | ||
448 | fau_timeout.s.tout_enb = 0; | ||
449 | cvmx_write_csr(CVMX_IOB_FAU_TIMEOUT, fau_timeout.u64); | ||
450 | |||
451 | nm_tim.u64 = 0; | ||
452 | /* 4096 cycles */ | ||
453 | nm_tim.s.nw_tim = 3; | ||
454 | cvmx_write_csr(CVMX_POW_NW_TIM, nm_tim.u64); | ||
455 | |||
456 | write_octeon_c0_icacheerr(0); | ||
457 | write_c0_derraddr1(0); | ||
458 | } | ||
459 | |||
460 | /** | ||
461 | * Early entry point for arch setup | ||
462 | */ | ||
463 | void __init prom_init(void) | ||
464 | { | ||
465 | struct cvmx_sysinfo *sysinfo; | ||
466 | const int coreid = cvmx_get_core_num(); | ||
467 | int i; | ||
468 | int argc; | ||
469 | struct uart_port octeon_port; | ||
470 | #ifdef CONFIG_CAVIUM_RESERVE32 | ||
471 | int64_t addr = -1; | ||
472 | #endif | ||
473 | /* | ||
474 | * The bootloader passes a pointer to the boot descriptor in | ||
475 | * $a3, this is available as fw_arg3. | ||
476 | */ | ||
477 | octeon_boot_desc_ptr = (struct octeon_boot_descriptor *)fw_arg3; | ||
478 | octeon_bootinfo = | ||
479 | cvmx_phys_to_ptr(octeon_boot_desc_ptr->cvmx_desc_vaddr); | ||
480 | cvmx_bootmem_init(cvmx_phys_to_ptr(octeon_bootinfo->phy_mem_desc_addr)); | ||
481 | |||
482 | /* | ||
483 | * Only enable the LED controller if we're running on a CN38XX, CN58XX, | ||
484 | * or CN56XX. The CN30XX and CN31XX don't have an LED controller. | ||
485 | */ | ||
486 | if (!octeon_is_simulation() && | ||
487 | octeon_has_feature(OCTEON_FEATURE_LED_CONTROLLER)) { | ||
488 | cvmx_write_csr(CVMX_LED_EN, 0); | ||
489 | cvmx_write_csr(CVMX_LED_PRT, 0); | ||
490 | cvmx_write_csr(CVMX_LED_DBG, 0); | ||
491 | cvmx_write_csr(CVMX_LED_PRT_FMT, 0); | ||
492 | cvmx_write_csr(CVMX_LED_UDD_CNTX(0), 32); | ||
493 | cvmx_write_csr(CVMX_LED_UDD_CNTX(1), 32); | ||
494 | cvmx_write_csr(CVMX_LED_UDD_DATX(0), 0); | ||
495 | cvmx_write_csr(CVMX_LED_UDD_DATX(1), 0); | ||
496 | cvmx_write_csr(CVMX_LED_EN, 1); | ||
497 | } | ||
498 | #ifdef CONFIG_CAVIUM_RESERVE32 | ||
499 | /* | ||
500 | * We need to temporarily allocate all memory in the reserve32 | ||
501 | * region. This makes sure the kernel doesn't allocate this | ||
502 | * memory when it is getting memory from the | ||
503 | * bootloader. Later, after the memory allocations are | ||
504 | * complete, the reserve32 will be freed. | ||
505 | */ | ||
506 | #ifdef CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB | ||
507 | if (CONFIG_CAVIUM_RESERVE32 & 0x1ff) | ||
508 | pr_err("CAVIUM_RESERVE32 isn't a multiple of 512MB. " | ||
509 | "This is required if CAVIUM_RESERVE32_USE_WIRED_TLB " | ||
510 | "is set\n"); | ||
511 | else | ||
512 | addr = cvmx_bootmem_phy_named_block_alloc(CONFIG_CAVIUM_RESERVE32 << 20, | ||
513 | 0, 0, 512 << 20, | ||
514 | "CAVIUM_RESERVE32", 0); | ||
515 | #else | ||
516 | /* | ||
517 | * Allocate memory for RESERVED32 aligned on 2MB boundary. This | ||
518 | * is in case we later use hugetlb entries with it. | ||
519 | */ | ||
520 | addr = cvmx_bootmem_phy_named_block_alloc(CONFIG_CAVIUM_RESERVE32 << 20, | ||
521 | 0, 0, 2 << 20, | ||
522 | "CAVIUM_RESERVE32", 0); | ||
523 | #endif | ||
524 | if (addr < 0) | ||
525 | pr_err("Failed to allocate CAVIUM_RESERVE32 memory area\n"); | ||
526 | else | ||
527 | octeon_reserve32_memory = addr; | ||
528 | #endif | ||
529 | |||
530 | #ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2 | ||
531 | if (cvmx_read_csr(CVMX_L2D_FUS3) & (3ull << 34)) { | ||
532 | pr_info("Skipping L2 locking due to reduced L2 cache size\n"); | ||
533 | } else { | ||
534 | uint32_t ebase = read_c0_ebase() & 0x3ffff000; | ||
535 | #ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2_TLB | ||
536 | /* TLB refill */ | ||
537 | cvmx_l2c_lock_mem_region(ebase, 0x100); | ||
538 | #endif | ||
539 | #ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2_EXCEPTION | ||
540 | /* General exception */ | ||
541 | cvmx_l2c_lock_mem_region(ebase + 0x180, 0x80); | ||
542 | #endif | ||
543 | #ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2_LOW_LEVEL_INTERRUPT | ||
544 | /* Interrupt handler */ | ||
545 | cvmx_l2c_lock_mem_region(ebase + 0x200, 0x80); | ||
546 | #endif | ||
547 | #ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2_INTERRUPT | ||
548 | cvmx_l2c_lock_mem_region(__pa_symbol(handle_int), 0x100); | ||
549 | cvmx_l2c_lock_mem_region(__pa_symbol(plat_irq_dispatch), 0x80); | ||
550 | #endif | ||
551 | #ifdef CONFIG_CAVIUM_OCTEON_LOCK_L2_MEMCPY | ||
552 | cvmx_l2c_lock_mem_region(__pa_symbol(memcpy), 0x480); | ||
553 | #endif | ||
554 | } | ||
555 | #endif | ||
556 | |||
557 | sysinfo = cvmx_sysinfo_get(); | ||
558 | memset(sysinfo, 0, sizeof(*sysinfo)); | ||
559 | sysinfo->system_dram_size = octeon_bootinfo->dram_size << 20; | ||
560 | sysinfo->phy_mem_desc_ptr = | ||
561 | cvmx_phys_to_ptr(octeon_bootinfo->phy_mem_desc_addr); | ||
562 | sysinfo->core_mask = octeon_bootinfo->core_mask; | ||
563 | sysinfo->exception_base_addr = octeon_bootinfo->exception_base_addr; | ||
564 | sysinfo->cpu_clock_hz = octeon_bootinfo->eclock_hz; | ||
565 | sysinfo->dram_data_rate_hz = octeon_bootinfo->dclock_hz * 2; | ||
566 | sysinfo->board_type = octeon_bootinfo->board_type; | ||
567 | sysinfo->board_rev_major = octeon_bootinfo->board_rev_major; | ||
568 | sysinfo->board_rev_minor = octeon_bootinfo->board_rev_minor; | ||
569 | memcpy(sysinfo->mac_addr_base, octeon_bootinfo->mac_addr_base, | ||
570 | sizeof(sysinfo->mac_addr_base)); | ||
571 | sysinfo->mac_addr_count = octeon_bootinfo->mac_addr_count; | ||
572 | memcpy(sysinfo->board_serial_number, | ||
573 | octeon_bootinfo->board_serial_number, | ||
574 | sizeof(sysinfo->board_serial_number)); | ||
575 | sysinfo->compact_flash_common_base_addr = | ||
576 | octeon_bootinfo->compact_flash_common_base_addr; | ||
577 | sysinfo->compact_flash_attribute_base_addr = | ||
578 | octeon_bootinfo->compact_flash_attribute_base_addr; | ||
579 | sysinfo->led_display_base_addr = octeon_bootinfo->led_display_base_addr; | ||
580 | sysinfo->dfa_ref_clock_hz = octeon_bootinfo->dfa_ref_clock_hz; | ||
581 | sysinfo->bootloader_config_flags = octeon_bootinfo->config_flags; | ||
582 | |||
583 | |||
584 | octeon_check_cpu_bist(); | ||
585 | |||
586 | octeon_uart = octeon_get_boot_uart(); | ||
587 | |||
588 | /* | ||
589 | * Disable All CIU Interrupts. The ones we need will be | ||
590 | * enabled later. Read the SUM register so we know the write | ||
591 | * completed. | ||
592 | */ | ||
593 | cvmx_write_csr(CVMX_CIU_INTX_EN0((coreid * 2)), 0); | ||
594 | cvmx_write_csr(CVMX_CIU_INTX_EN0((coreid * 2 + 1)), 0); | ||
595 | cvmx_write_csr(CVMX_CIU_INTX_EN1((coreid * 2)), 0); | ||
596 | cvmx_write_csr(CVMX_CIU_INTX_EN1((coreid * 2 + 1)), 0); | ||
597 | cvmx_read_csr(CVMX_CIU_INTX_SUM0((coreid * 2))); | ||
598 | |||
599 | #ifdef CONFIG_SMP | ||
600 | octeon_write_lcd("LinuxSMP"); | ||
601 | #else | ||
602 | octeon_write_lcd("Linux"); | ||
603 | #endif | ||
604 | |||
605 | #ifdef CONFIG_CAVIUM_GDB | ||
606 | /* | ||
607 | * When debugging the linux kernel, force the cores to enter | ||
608 | * the debug exception handler to break in. | ||
609 | */ | ||
610 | if (octeon_get_boot_debug_flag()) { | ||
611 | cvmx_write_csr(CVMX_CIU_DINT, 1 << cvmx_get_core_num()); | ||
612 | cvmx_read_csr(CVMX_CIU_DINT); | ||
613 | } | ||
614 | #endif | ||
615 | |||
616 | /* | ||
617 | * BIST should always be enabled when doing a soft reset. L2 | ||
618 | * Cache locking for instance is not cleared unless BIST is | ||
619 | * enabled. Unfortunately due to a chip errata G-200 for | ||
620 | * Cn38XX and CN31XX, BIST msut be disabled on these parts. | ||
621 | */ | ||
622 | if (OCTEON_IS_MODEL(OCTEON_CN38XX_PASS2) || | ||
623 | OCTEON_IS_MODEL(OCTEON_CN31XX)) | ||
624 | cvmx_write_csr(CVMX_CIU_SOFT_BIST, 0); | ||
625 | else | ||
626 | cvmx_write_csr(CVMX_CIU_SOFT_BIST, 1); | ||
627 | |||
628 | /* Default to 64MB in the simulator to speed things up */ | ||
629 | if (octeon_is_simulation()) | ||
630 | MAX_MEMORY = 64ull << 20; | ||
631 | |||
632 | arcs_cmdline[0] = 0; | ||
633 | argc = octeon_boot_desc_ptr->argc; | ||
634 | for (i = 0; i < argc; i++) { | ||
635 | const char *arg = | ||
636 | cvmx_phys_to_ptr(octeon_boot_desc_ptr->argv[i]); | ||
637 | if ((strncmp(arg, "MEM=", 4) == 0) || | ||
638 | (strncmp(arg, "mem=", 4) == 0)) { | ||
639 | sscanf(arg + 4, "%llu", &MAX_MEMORY); | ||
640 | MAX_MEMORY <<= 20; | ||
641 | if (MAX_MEMORY == 0) | ||
642 | MAX_MEMORY = 32ull << 30; | ||
643 | } else if (strcmp(arg, "ecc_verbose") == 0) { | ||
644 | #ifdef CONFIG_CAVIUM_REPORT_SINGLE_BIT_ECC | ||
645 | __cvmx_interrupt_ecc_report_single_bit_errors = 1; | ||
646 | pr_notice("Reporting of single bit ECC errors is " | ||
647 | "turned on\n"); | ||
648 | #endif | ||
649 | } else if (strlen(arcs_cmdline) + strlen(arg) + 1 < | ||
650 | sizeof(arcs_cmdline) - 1) { | ||
651 | strcat(arcs_cmdline, " "); | ||
652 | strcat(arcs_cmdline, arg); | ||
653 | } | ||
654 | } | ||
655 | |||
656 | if (strstr(arcs_cmdline, "console=") == NULL) { | ||
657 | #ifdef CONFIG_GDB_CONSOLE | ||
658 | strcat(arcs_cmdline, " console=gdb"); | ||
659 | #else | ||
660 | #ifdef CONFIG_CAVIUM_OCTEON_2ND_KERNEL | ||
661 | strcat(arcs_cmdline, " console=ttyS0,115200"); | ||
662 | #else | ||
663 | if (octeon_uart == 1) | ||
664 | strcat(arcs_cmdline, " console=ttyS1,115200"); | ||
665 | else | ||
666 | strcat(arcs_cmdline, " console=ttyS0,115200"); | ||
667 | #endif | ||
668 | #endif | ||
669 | } | ||
670 | |||
671 | if (octeon_is_simulation()) { | ||
672 | /* | ||
673 | * The simulator uses a mtdram device pre filled with | ||
674 | * the filesystem. Also specify the calibration delay | ||
675 | * to avoid calculating it every time. | ||
676 | */ | ||
677 | strcat(arcs_cmdline, " rw root=1f00" | ||
678 | " lpj=60176 slram=root,0x40000000,+1073741824"); | ||
679 | } | ||
680 | |||
681 | mips_hpt_frequency = octeon_get_clock_rate(); | ||
682 | |||
683 | octeon_init_cvmcount(); | ||
684 | |||
685 | _machine_restart = octeon_restart; | ||
686 | _machine_halt = octeon_halt; | ||
687 | |||
688 | memset(&octeon_port, 0, sizeof(octeon_port)); | ||
689 | /* | ||
690 | * For early_serial_setup we don't set the port type or | ||
691 | * UPF_FIXED_TYPE. | ||
692 | */ | ||
693 | octeon_port.flags = ASYNC_SKIP_TEST | UPF_SHARE_IRQ; | ||
694 | octeon_port.iotype = UPIO_MEM; | ||
695 | /* I/O addresses are every 8 bytes */ | ||
696 | octeon_port.regshift = 3; | ||
697 | /* Clock rate of the chip */ | ||
698 | octeon_port.uartclk = mips_hpt_frequency; | ||
699 | octeon_port.fifosize = 64; | ||
700 | octeon_port.mapbase = 0x0001180000000800ull + (1024 * octeon_uart); | ||
701 | octeon_port.membase = cvmx_phys_to_ptr(octeon_port.mapbase); | ||
702 | octeon_port.serial_in = octeon_serial_in; | ||
703 | octeon_port.serial_out = octeon_serial_out; | ||
704 | #ifdef CONFIG_CAVIUM_OCTEON_2ND_KERNEL | ||
705 | octeon_port.line = 0; | ||
706 | #else | ||
707 | octeon_port.line = octeon_uart; | ||
708 | #endif | ||
709 | octeon_port.irq = 42 + octeon_uart; | ||
710 | early_serial_setup(&octeon_port); | ||
711 | |||
712 | octeon_user_io_init(); | ||
713 | register_smp_ops(&octeon_smp_ops); | ||
714 | } | ||
715 | |||
716 | void __init plat_mem_setup(void) | ||
717 | { | ||
718 | uint64_t mem_alloc_size; | ||
719 | uint64_t total; | ||
720 | int64_t memory; | ||
721 | |||
722 | total = 0; | ||
723 | |||
724 | /* First add the init memory we will be returning. */ | ||
725 | memory = __pa_symbol(&__init_begin) & PAGE_MASK; | ||
726 | mem_alloc_size = (__pa_symbol(&__init_end) & PAGE_MASK) - memory; | ||
727 | if (mem_alloc_size > 0) { | ||
728 | add_memory_region(memory, mem_alloc_size, BOOT_MEM_RAM); | ||
729 | total += mem_alloc_size; | ||
730 | } | ||
731 | |||
732 | /* | ||
733 | * The Mips memory init uses the first memory location for | ||
734 | * some memory vectors. When SPARSEMEM is in use, it doesn't | ||
735 | * verify that the size is big enough for the final | ||
736 | * vectors. Making the smallest chuck 4MB seems to be enough | ||
737 | * to consistantly work. | ||
738 | */ | ||
739 | mem_alloc_size = 4 << 20; | ||
740 | if (mem_alloc_size > MAX_MEMORY) | ||
741 | mem_alloc_size = MAX_MEMORY; | ||
742 | |||
743 | /* | ||
744 | * When allocating memory, we want incrementing addresses from | ||
745 | * bootmem_alloc so the code in add_memory_region can merge | ||
746 | * regions next to each other. | ||
747 | */ | ||
748 | cvmx_bootmem_lock(); | ||
749 | while ((boot_mem_map.nr_map < BOOT_MEM_MAP_MAX) | ||
750 | && (total < MAX_MEMORY)) { | ||
751 | #if defined(CONFIG_64BIT) || defined(CONFIG_64BIT_PHYS_ADDR) | ||
752 | memory = cvmx_bootmem_phy_alloc(mem_alloc_size, | ||
753 | __pa_symbol(&__init_end), -1, | ||
754 | 0x100000, | ||
755 | CVMX_BOOTMEM_FLAG_NO_LOCKING); | ||
756 | #elif defined(CONFIG_HIGHMEM) | ||
757 | memory = cvmx_bootmem_phy_alloc(mem_alloc_size, 0, 1ull << 31, | ||
758 | 0x100000, | ||
759 | CVMX_BOOTMEM_FLAG_NO_LOCKING); | ||
760 | #else | ||
761 | memory = cvmx_bootmem_phy_alloc(mem_alloc_size, 0, 512 << 20, | ||
762 | 0x100000, | ||
763 | CVMX_BOOTMEM_FLAG_NO_LOCKING); | ||
764 | #endif | ||
765 | if (memory >= 0) { | ||
766 | /* | ||
767 | * This function automatically merges address | ||
768 | * regions next to each other if they are | ||
769 | * received in incrementing order. | ||
770 | */ | ||
771 | add_memory_region(memory, mem_alloc_size, BOOT_MEM_RAM); | ||
772 | total += mem_alloc_size; | ||
773 | } else { | ||
774 | break; | ||
775 | } | ||
776 | } | ||
777 | cvmx_bootmem_unlock(); | ||
778 | |||
779 | #ifdef CONFIG_CAVIUM_RESERVE32 | ||
780 | /* | ||
781 | * Now that we've allocated the kernel memory it is safe to | ||
782 | * free the reserved region. We free it here so that builtin | ||
783 | * drivers can use the memory. | ||
784 | */ | ||
785 | if (octeon_reserve32_memory) | ||
786 | cvmx_bootmem_free_named("CAVIUM_RESERVE32"); | ||
787 | #endif /* CONFIG_CAVIUM_RESERVE32 */ | ||
788 | |||
789 | if (total == 0) | ||
790 | panic("Unable to allocate memory from " | ||
791 | "cvmx_bootmem_phy_alloc\n"); | ||
792 | } | ||
793 | |||
794 | |||
795 | int prom_putchar(char c) | ||
796 | { | ||
797 | uint64_t lsrval; | ||
798 | |||
799 | /* Spin until there is room */ | ||
800 | do { | ||
801 | lsrval = cvmx_read_csr(CVMX_MIO_UARTX_LSR(octeon_uart)); | ||
802 | } while ((lsrval & 0x20) == 0); | ||
803 | |||
804 | /* Write the byte */ | ||
805 | cvmx_write_csr(CVMX_MIO_UARTX_THR(octeon_uart), c); | ||
806 | return 1; | ||
807 | } | ||
808 | |||
809 | void prom_free_prom_memory(void) | ||
810 | { | ||
811 | #ifdef CONFIG_CAVIUM_DECODE_RSL | ||
812 | cvmx_interrupt_rsl_enable(); | ||
813 | |||
814 | /* Add an interrupt handler for general failures. */ | ||
815 | if (request_irq(OCTEON_IRQ_RML, octeon_rlm_interrupt, IRQF_SHARED, | ||
816 | "RML/RSL", octeon_rlm_interrupt)) { | ||
817 | panic("Unable to request_irq(OCTEON_IRQ_RML)\n"); | ||
818 | } | ||
819 | #endif | ||
820 | |||
821 | /* This call is here so that it is performed after any TLB | ||
822 | initializations. It needs to be after these in case the | ||
823 | CONFIG_CAVIUM_RESERVE32_USE_WIRED_TLB option is set */ | ||
824 | octeon_hal_setup_reserved32(); | ||
825 | } | ||
826 | |||
827 | static struct octeon_cf_data octeon_cf_data; | ||
828 | |||
829 | static int __init octeon_cf_device_init(void) | ||
830 | { | ||
831 | union cvmx_mio_boot_reg_cfgx mio_boot_reg_cfg; | ||
832 | unsigned long base_ptr, region_base, region_size; | ||
833 | struct platform_device *pd; | ||
834 | struct resource cf_resources[3]; | ||
835 | unsigned int num_resources; | ||
836 | int i; | ||
837 | int ret = 0; | ||
838 | |||
839 | /* Setup octeon-cf platform device if present. */ | ||
840 | base_ptr = 0; | ||
841 | if (octeon_bootinfo->major_version == 1 | ||
842 | && octeon_bootinfo->minor_version >= 1) { | ||
843 | if (octeon_bootinfo->compact_flash_common_base_addr) | ||
844 | base_ptr = | ||
845 | octeon_bootinfo->compact_flash_common_base_addr; | ||
846 | } else { | ||
847 | base_ptr = 0x1d000800; | ||
848 | } | ||
849 | |||
850 | if (!base_ptr) | ||
851 | return ret; | ||
852 | |||
853 | /* Find CS0 region. */ | ||
854 | for (i = 0; i < 8; i++) { | ||
855 | mio_boot_reg_cfg.u64 = cvmx_read_csr(CVMX_MIO_BOOT_REG_CFGX(i)); | ||
856 | region_base = mio_boot_reg_cfg.s.base << 16; | ||
857 | region_size = (mio_boot_reg_cfg.s.size + 1) << 16; | ||
858 | if (mio_boot_reg_cfg.s.en && base_ptr >= region_base | ||
859 | && base_ptr < region_base + region_size) | ||
860 | break; | ||
861 | } | ||
862 | if (i >= 7) { | ||
863 | /* i and i + 1 are CS0 and CS1, both must be less than 8. */ | ||
864 | goto out; | ||
865 | } | ||
866 | octeon_cf_data.base_region = i; | ||
867 | octeon_cf_data.is16bit = mio_boot_reg_cfg.s.width; | ||
868 | octeon_cf_data.base_region_bias = base_ptr - region_base; | ||
869 | memset(cf_resources, 0, sizeof(cf_resources)); | ||
870 | num_resources = 0; | ||
871 | cf_resources[num_resources].flags = IORESOURCE_MEM; | ||
872 | cf_resources[num_resources].start = region_base; | ||
873 | cf_resources[num_resources].end = region_base + region_size - 1; | ||
874 | num_resources++; | ||
875 | |||
876 | |||
877 | if (!(base_ptr & 0xfffful)) { | ||
878 | /* | ||
879 | * Boot loader signals availability of DMA (true_ide | ||
880 | * mode) by setting low order bits of base_ptr to | ||
881 | * zero. | ||
882 | */ | ||
883 | |||
884 | /* Asume that CS1 immediately follows. */ | ||
885 | mio_boot_reg_cfg.u64 = | ||
886 | cvmx_read_csr(CVMX_MIO_BOOT_REG_CFGX(i + 1)); | ||
887 | region_base = mio_boot_reg_cfg.s.base << 16; | ||
888 | region_size = (mio_boot_reg_cfg.s.size + 1) << 16; | ||
889 | if (!mio_boot_reg_cfg.s.en) | ||
890 | goto out; | ||
891 | |||
892 | cf_resources[num_resources].flags = IORESOURCE_MEM; | ||
893 | cf_resources[num_resources].start = region_base; | ||
894 | cf_resources[num_resources].end = region_base + region_size - 1; | ||
895 | num_resources++; | ||
896 | |||
897 | octeon_cf_data.dma_engine = 0; | ||
898 | cf_resources[num_resources].flags = IORESOURCE_IRQ; | ||
899 | cf_resources[num_resources].start = OCTEON_IRQ_BOOTDMA; | ||
900 | cf_resources[num_resources].end = OCTEON_IRQ_BOOTDMA; | ||
901 | num_resources++; | ||
902 | } else { | ||
903 | octeon_cf_data.dma_engine = -1; | ||
904 | } | ||
905 | |||
906 | pd = platform_device_alloc("pata_octeon_cf", -1); | ||
907 | if (!pd) { | ||
908 | ret = -ENOMEM; | ||
909 | goto out; | ||
910 | } | ||
911 | pd->dev.platform_data = &octeon_cf_data; | ||
912 | |||
913 | ret = platform_device_add_resources(pd, cf_resources, num_resources); | ||
914 | if (ret) | ||
915 | goto fail; | ||
916 | |||
917 | ret = platform_device_add(pd); | ||
918 | if (ret) | ||
919 | goto fail; | ||
920 | |||
921 | return ret; | ||
922 | fail: | ||
923 | platform_device_put(pd); | ||
924 | out: | ||
925 | return ret; | ||
926 | } | ||
927 | device_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 | |||
22 | volatile unsigned long octeon_processor_boot = 0xff; | ||
23 | volatile unsigned long octeon_processor_sp; | ||
24 | volatile unsigned long octeon_processor_gp; | ||
25 | |||
26 | static 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 | */ | ||
51 | void 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 | |||
61 | static 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 | */ | ||
72 | static 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 | */ | ||
100 | static 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 | */ | ||
126 | static 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 | */ | ||
153 | void 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 | */ | ||
170 | static 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 | */ | ||
190 | static 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 | |||
202 | struct 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 | }; | ||