diff options
Diffstat (limited to 'arch/mips/cavium-octeon')
-rw-r--r-- | arch/mips/cavium-octeon/Makefile | 4 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/dma-octeon.c | 311 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/executive/Makefile | 1 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/executive/cvmx-bootmem.c | 104 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/executive/cvmx-helper-errata.c | 73 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/executive/cvmx-helper-jtag.c | 144 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/executive/cvmx-sysinfo.c | 2 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/msi.c | 288 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/octeon-irq.c | 2 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/pci-common.c | 137 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/pci-common.h | 39 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/pci.c | 568 | ||||
-rw-r--r-- | arch/mips/cavium-octeon/pcie.c | 1370 |
13 files changed, 3041 insertions, 2 deletions
diff --git a/arch/mips/cavium-octeon/Makefile b/arch/mips/cavium-octeon/Makefile index d6903c3f3d51..7c0528b0e34c 100644 --- a/arch/mips/cavium-octeon/Makefile +++ b/arch/mips/cavium-octeon/Makefile | |||
@@ -14,5 +14,9 @@ obj-y += dma-octeon.o flash_setup.o | |||
14 | obj-y += octeon-memcpy.o | 14 | obj-y += octeon-memcpy.o |
15 | 15 | ||
16 | obj-$(CONFIG_SMP) += smp.o | 16 | obj-$(CONFIG_SMP) += smp.o |
17 | obj-$(CONFIG_PCI) += pci-common.o | ||
18 | obj-$(CONFIG_PCI) += pci.o | ||
19 | obj-$(CONFIG_PCI) += pcie.o | ||
20 | obj-$(CONFIG_PCI_MSI) += msi.o | ||
17 | 21 | ||
18 | EXTRA_CFLAGS += -Werror | 22 | EXTRA_CFLAGS += -Werror |
diff --git a/arch/mips/cavium-octeon/dma-octeon.c b/arch/mips/cavium-octeon/dma-octeon.c index 01b1ef94b361..627c162a6159 100644 --- a/arch/mips/cavium-octeon/dma-octeon.c +++ b/arch/mips/cavium-octeon/dma-octeon.c | |||
@@ -13,20 +13,327 @@ | |||
13 | */ | 13 | */ |
14 | #include <linux/types.h> | 14 | #include <linux/types.h> |
15 | #include <linux/mm.h> | 15 | #include <linux/mm.h> |
16 | #include <linux/module.h> | ||
17 | #include <linux/string.h> | ||
18 | #include <linux/dma-mapping.h> | ||
19 | #include <linux/platform_device.h> | ||
20 | #include <linux/scatterlist.h> | ||
21 | |||
22 | #include <linux/cache.h> | ||
23 | #include <linux/io.h> | ||
24 | |||
25 | #include <asm/octeon/octeon.h> | ||
26 | #include <asm/octeon/cvmx-npi-defs.h> | ||
27 | #include <asm/octeon/cvmx-pci-defs.h> | ||
16 | 28 | ||
17 | #include <dma-coherence.h> | 29 | #include <dma-coherence.h> |
18 | 30 | ||
31 | #ifdef CONFIG_PCI | ||
32 | #include "pci-common.h" | ||
33 | #endif | ||
34 | |||
35 | #define BAR2_PCI_ADDRESS 0x8000000000ul | ||
36 | |||
37 | struct bar1_index_state { | ||
38 | int16_t ref_count; /* Number of PCI mappings using this index */ | ||
39 | uint16_t address_bits; /* Upper bits of physical address. This is | ||
40 | shifted 22 bits */ | ||
41 | }; | ||
42 | |||
43 | #ifdef CONFIG_PCI | ||
44 | static DEFINE_SPINLOCK(bar1_lock); | ||
45 | static struct bar1_index_state bar1_state[32]; | ||
46 | #endif | ||
47 | |||
19 | dma_addr_t octeon_map_dma_mem(struct device *dev, void *ptr, size_t size) | 48 | dma_addr_t octeon_map_dma_mem(struct device *dev, void *ptr, size_t size) |
20 | { | 49 | { |
50 | #ifndef CONFIG_PCI | ||
21 | /* Without PCI/PCIe this function can be called for Octeon internal | 51 | /* Without PCI/PCIe this function can be called for Octeon internal |
22 | devices such as USB. These devices all support 64bit addressing */ | 52 | devices such as USB. These devices all support 64bit addressing */ |
23 | mb(); | 53 | mb(); |
24 | return virt_to_phys(ptr); | 54 | return virt_to_phys(ptr); |
55 | #else | ||
56 | unsigned long flags; | ||
57 | uint64_t dma_mask; | ||
58 | int64_t start_index; | ||
59 | dma_addr_t result = -1; | ||
60 | uint64_t physical = virt_to_phys(ptr); | ||
61 | int64_t index; | ||
62 | |||
63 | mb(); | ||
64 | /* | ||
65 | * Use the DMA masks to determine the allowed memory | ||
66 | * region. For us it doesn't limit the actual memory, just the | ||
67 | * address visible over PCI. Devices with limits need to use | ||
68 | * lower indexed Bar1 entries. | ||
69 | */ | ||
70 | if (dev) { | ||
71 | dma_mask = dev->coherent_dma_mask; | ||
72 | if (dev->dma_mask) | ||
73 | dma_mask = *dev->dma_mask; | ||
74 | } else { | ||
75 | dma_mask = 0xfffffffful; | ||
76 | } | ||
77 | |||
78 | /* | ||
79 | * Platform devices, such as the internal USB, skip all | ||
80 | * translation and use Octeon physical addresses directly. | ||
81 | */ | ||
82 | if (!dev || dev->bus == &platform_bus_type) | ||
83 | return physical; | ||
84 | |||
85 | switch (octeon_dma_bar_type) { | ||
86 | case OCTEON_DMA_BAR_TYPE_PCIE: | ||
87 | if (unlikely(physical < (16ul << 10))) | ||
88 | panic("dma_map_single: Not allowed to map first 16KB." | ||
89 | " It interferes with BAR0 special area\n"); | ||
90 | else if ((physical + size >= (256ul << 20)) && | ||
91 | (physical < (512ul << 20))) | ||
92 | panic("dma_map_single: Not allowed to map bootbus\n"); | ||
93 | else if ((physical + size >= 0x400000000ull) && | ||
94 | physical < 0x410000000ull) | ||
95 | panic("dma_map_single: " | ||
96 | "Attempt to map illegal memory address 0x%llx\n", | ||
97 | physical); | ||
98 | else if (physical >= 0x420000000ull) | ||
99 | panic("dma_map_single: " | ||
100 | "Attempt to map illegal memory address 0x%llx\n", | ||
101 | physical); | ||
102 | else if ((physical + size >= | ||
103 | (4ull<<30) - (OCTEON_PCI_BAR1_HOLE_SIZE<<20)) | ||
104 | && physical < (4ull<<30)) | ||
105 | pr_warning("dma_map_single: Warning: " | ||
106 | "Mapping memory address that might " | ||
107 | "conflict with devices 0x%llx-0x%llx\n", | ||
108 | physical, physical+size-1); | ||
109 | /* The 2nd 256MB is mapped at 256<<20 instead of 0x410000000 */ | ||
110 | if ((physical >= 0x410000000ull) && physical < 0x420000000ull) | ||
111 | result = physical - 0x400000000ull; | ||
112 | else | ||
113 | result = physical; | ||
114 | if (((result+size-1) & dma_mask) != result+size-1) | ||
115 | panic("dma_map_single: Attempt to map address " | ||
116 | "0x%llx-0x%llx, which can't be accessed " | ||
117 | "according to the dma mask 0x%llx\n", | ||
118 | physical, physical+size-1, dma_mask); | ||
119 | goto done; | ||
120 | |||
121 | case OCTEON_DMA_BAR_TYPE_BIG: | ||
122 | #ifdef CONFIG_64BIT | ||
123 | /* If the device supports 64bit addressing, then use BAR2 */ | ||
124 | if (dma_mask > BAR2_PCI_ADDRESS) { | ||
125 | result = physical + BAR2_PCI_ADDRESS; | ||
126 | goto done; | ||
127 | } | ||
128 | #endif | ||
129 | if (unlikely(physical < (4ul << 10))) { | ||
130 | panic("dma_map_single: Not allowed to map first 4KB. " | ||
131 | "It interferes with BAR0 special area\n"); | ||
132 | } else if (physical < (256ul << 20)) { | ||
133 | if (unlikely(physical + size > (256ul << 20))) | ||
134 | panic("dma_map_single: Requested memory spans " | ||
135 | "Bar0 0:256MB and bootbus\n"); | ||
136 | result = physical; | ||
137 | goto done; | ||
138 | } else if (unlikely(physical < (512ul << 20))) { | ||
139 | panic("dma_map_single: Not allowed to map bootbus\n"); | ||
140 | } else if (physical < (2ul << 30)) { | ||
141 | if (unlikely(physical + size > (2ul << 30))) | ||
142 | panic("dma_map_single: Requested memory spans " | ||
143 | "Bar0 512MB:2GB and BAR1\n"); | ||
144 | result = physical; | ||
145 | goto done; | ||
146 | } else if (physical < (2ul << 30) + (128 << 20)) { | ||
147 | /* Fall through */ | ||
148 | } else if (physical < | ||
149 | (4ul << 30) - (OCTEON_PCI_BAR1_HOLE_SIZE << 20)) { | ||
150 | if (unlikely | ||
151 | (physical + size > | ||
152 | (4ul << 30) - (OCTEON_PCI_BAR1_HOLE_SIZE << 20))) | ||
153 | panic("dma_map_single: Requested memory " | ||
154 | "extends past Bar1 (4GB-%luMB)\n", | ||
155 | OCTEON_PCI_BAR1_HOLE_SIZE); | ||
156 | result = physical; | ||
157 | goto done; | ||
158 | } else if ((physical >= 0x410000000ull) && | ||
159 | (physical < 0x420000000ull)) { | ||
160 | if (unlikely(physical + size > 0x420000000ull)) | ||
161 | panic("dma_map_single: Requested memory spans " | ||
162 | "non existant memory\n"); | ||
163 | /* BAR0 fixed mapping 256MB:512MB -> | ||
164 | * 16GB+256MB:16GB+512MB */ | ||
165 | result = physical - 0x400000000ull; | ||
166 | goto done; | ||
167 | } else { | ||
168 | /* Continued below switch statement */ | ||
169 | } | ||
170 | break; | ||
171 | |||
172 | case OCTEON_DMA_BAR_TYPE_SMALL: | ||
173 | #ifdef CONFIG_64BIT | ||
174 | /* If the device supports 64bit addressing, then use BAR2 */ | ||
175 | if (dma_mask > BAR2_PCI_ADDRESS) { | ||
176 | result = physical + BAR2_PCI_ADDRESS; | ||
177 | goto done; | ||
178 | } | ||
179 | #endif | ||
180 | /* Continued below switch statement */ | ||
181 | break; | ||
182 | |||
183 | default: | ||
184 | panic("dma_map_single: Invalid octeon_dma_bar_type\n"); | ||
185 | } | ||
186 | |||
187 | /* Don't allow mapping to span multiple Bar entries. The hardware guys | ||
188 | won't guarantee that DMA across boards work */ | ||
189 | if (unlikely((physical >> 22) != ((physical + size - 1) >> 22))) | ||
190 | panic("dma_map_single: " | ||
191 | "Requested memory spans more than one Bar1 entry\n"); | ||
192 | |||
193 | if (octeon_dma_bar_type == OCTEON_DMA_BAR_TYPE_BIG) | ||
194 | start_index = 31; | ||
195 | else if (unlikely(dma_mask < (1ul << 27))) | ||
196 | start_index = (dma_mask >> 22); | ||
197 | else | ||
198 | start_index = 31; | ||
199 | |||
200 | /* Only one processor can access the Bar register at once */ | ||
201 | spin_lock_irqsave(&bar1_lock, flags); | ||
202 | |||
203 | /* Look through Bar1 for existing mapping that will work */ | ||
204 | for (index = start_index; index >= 0; index--) { | ||
205 | if ((bar1_state[index].address_bits == physical >> 22) && | ||
206 | (bar1_state[index].ref_count)) { | ||
207 | /* An existing mapping will work, use it */ | ||
208 | bar1_state[index].ref_count++; | ||
209 | if (unlikely(bar1_state[index].ref_count < 0)) | ||
210 | panic("dma_map_single: " | ||
211 | "Bar1[%d] reference count overflowed\n", | ||
212 | (int) index); | ||
213 | result = (index << 22) | (physical & ((1 << 22) - 1)); | ||
214 | /* Large BAR1 is offset at 2GB */ | ||
215 | if (octeon_dma_bar_type == OCTEON_DMA_BAR_TYPE_BIG) | ||
216 | result += 2ul << 30; | ||
217 | goto done_unlock; | ||
218 | } | ||
219 | } | ||
220 | |||
221 | /* No existing mappings, look for a free entry */ | ||
222 | for (index = start_index; index >= 0; index--) { | ||
223 | if (unlikely(bar1_state[index].ref_count == 0)) { | ||
224 | union cvmx_pci_bar1_indexx bar1_index; | ||
225 | /* We have a free entry, use it */ | ||
226 | bar1_state[index].ref_count = 1; | ||
227 | bar1_state[index].address_bits = physical >> 22; | ||
228 | bar1_index.u32 = 0; | ||
229 | /* Address bits[35:22] sent to L2C */ | ||
230 | bar1_index.s.addr_idx = physical >> 22; | ||
231 | /* Don't put PCI accesses in L2. */ | ||
232 | bar1_index.s.ca = 1; | ||
233 | /* Endian Swap Mode */ | ||
234 | bar1_index.s.end_swp = 1; | ||
235 | /* Set '1' when the selected address range is valid. */ | ||
236 | bar1_index.s.addr_v = 1; | ||
237 | octeon_npi_write32(CVMX_NPI_PCI_BAR1_INDEXX(index), | ||
238 | bar1_index.u32); | ||
239 | /* An existing mapping will work, use it */ | ||
240 | result = (index << 22) | (physical & ((1 << 22) - 1)); | ||
241 | /* Large BAR1 is offset at 2GB */ | ||
242 | if (octeon_dma_bar_type == OCTEON_DMA_BAR_TYPE_BIG) | ||
243 | result += 2ul << 30; | ||
244 | goto done_unlock; | ||
245 | } | ||
246 | } | ||
247 | |||
248 | pr_err("dma_map_single: " | ||
249 | "Can't find empty BAR1 index for physical mapping 0x%llx\n", | ||
250 | (unsigned long long) physical); | ||
251 | |||
252 | done_unlock: | ||
253 | spin_unlock_irqrestore(&bar1_lock, flags); | ||
254 | done: | ||
255 | pr_debug("dma_map_single 0x%llx->0x%llx\n", physical, result); | ||
256 | return result; | ||
257 | #endif | ||
25 | } | 258 | } |
26 | 259 | ||
27 | void octeon_unmap_dma_mem(struct device *dev, dma_addr_t dma_addr) | 260 | void octeon_unmap_dma_mem(struct device *dev, dma_addr_t dma_addr) |
28 | { | 261 | { |
29 | /* Without PCI/PCIe this function can be called for Octeon internal | 262 | #ifndef CONFIG_PCI |
30 | * devices such as USB. These devices all support 64bit addressing */ | 263 | /* |
264 | * Without PCI/PCIe this function can be called for Octeon internal | ||
265 | * devices such as USB. These devices all support 64bit addressing. | ||
266 | */ | ||
267 | return; | ||
268 | #else | ||
269 | unsigned long flags; | ||
270 | uint64_t index; | ||
271 | |||
272 | /* | ||
273 | * Platform devices, such as the internal USB, skip all | ||
274 | * translation and use Octeon physical addresses directly. | ||
275 | */ | ||
276 | if (dev->bus == &platform_bus_type) | ||
277 | return; | ||
278 | |||
279 | switch (octeon_dma_bar_type) { | ||
280 | case OCTEON_DMA_BAR_TYPE_PCIE: | ||
281 | /* Nothing to do, all mappings are static */ | ||
282 | goto done; | ||
283 | |||
284 | case OCTEON_DMA_BAR_TYPE_BIG: | ||
285 | #ifdef CONFIG_64BIT | ||
286 | /* Nothing to do for addresses using BAR2 */ | ||
287 | if (dma_addr >= BAR2_PCI_ADDRESS) | ||
288 | goto done; | ||
289 | #endif | ||
290 | if (unlikely(dma_addr < (4ul << 10))) | ||
291 | panic("dma_unmap_single: Unexpect DMA address 0x%llx\n", | ||
292 | dma_addr); | ||
293 | else if (dma_addr < (2ul << 30)) | ||
294 | /* Nothing to do for addresses using BAR0 */ | ||
295 | goto done; | ||
296 | else if (dma_addr < (2ul << 30) + (128ul << 20)) | ||
297 | /* Need to unmap, fall through */ | ||
298 | index = (dma_addr - (2ul << 30)) >> 22; | ||
299 | else if (dma_addr < | ||
300 | (4ul << 30) - (OCTEON_PCI_BAR1_HOLE_SIZE << 20)) | ||
301 | goto done; /* Nothing to do for the rest of BAR1 */ | ||
302 | else | ||
303 | panic("dma_unmap_single: Unexpect DMA address 0x%llx\n", | ||
304 | dma_addr); | ||
305 | /* Continued below switch statement */ | ||
306 | break; | ||
307 | |||
308 | case OCTEON_DMA_BAR_TYPE_SMALL: | ||
309 | #ifdef CONFIG_64BIT | ||
310 | /* Nothing to do for addresses using BAR2 */ | ||
311 | if (dma_addr >= BAR2_PCI_ADDRESS) | ||
312 | goto done; | ||
313 | #endif | ||
314 | index = dma_addr >> 22; | ||
315 | /* Continued below switch statement */ | ||
316 | break; | ||
317 | |||
318 | default: | ||
319 | panic("dma_unmap_single: Invalid octeon_dma_bar_type\n"); | ||
320 | } | ||
321 | |||
322 | if (unlikely(index > 31)) | ||
323 | panic("dma_unmap_single: " | ||
324 | "Attempt to unmap an invalid address (0x%llx)\n", | ||
325 | dma_addr); | ||
326 | |||
327 | spin_lock_irqsave(&bar1_lock, flags); | ||
328 | bar1_state[index].ref_count--; | ||
329 | if (bar1_state[index].ref_count == 0) | ||
330 | octeon_npi_write32(CVMX_NPI_PCI_BAR1_INDEXX(index), 0); | ||
331 | else if (unlikely(bar1_state[index].ref_count < 0)) | ||
332 | panic("dma_unmap_single: Bar1[%u] reference count < 0\n", | ||
333 | (int) index); | ||
334 | spin_unlock_irqrestore(&bar1_lock, flags); | ||
335 | done: | ||
336 | pr_debug("dma_unmap_single 0x%llx\n", dma_addr); | ||
31 | return; | 337 | return; |
338 | #endif | ||
32 | } | 339 | } |
diff --git a/arch/mips/cavium-octeon/executive/Makefile b/arch/mips/cavium-octeon/executive/Makefile index 80d6cb26766b..2fd66db6939e 100644 --- a/arch/mips/cavium-octeon/executive/Makefile +++ b/arch/mips/cavium-octeon/executive/Makefile | |||
@@ -11,3 +11,4 @@ | |||
11 | 11 | ||
12 | obj-y += cvmx-bootmem.o cvmx-l2c.o cvmx-sysinfo.o octeon-model.o | 12 | obj-y += cvmx-bootmem.o cvmx-l2c.o cvmx-sysinfo.o octeon-model.o |
13 | 13 | ||
14 | obj-$(CONFIG_PCI) += cvmx-helper-errata.o cvmx-helper-jtag.o | ||
diff --git a/arch/mips/cavium-octeon/executive/cvmx-bootmem.c b/arch/mips/cavium-octeon/executive/cvmx-bootmem.c index 4f5a08b37ccd..25666da17b22 100644 --- a/arch/mips/cavium-octeon/executive/cvmx-bootmem.c +++ b/arch/mips/cavium-octeon/executive/cvmx-bootmem.c | |||
@@ -31,6 +31,7 @@ | |||
31 | */ | 31 | */ |
32 | 32 | ||
33 | #include <linux/kernel.h> | 33 | #include <linux/kernel.h> |
34 | #include <linux/module.h> | ||
34 | 35 | ||
35 | #include <asm/octeon/cvmx.h> | 36 | #include <asm/octeon/cvmx.h> |
36 | #include <asm/octeon/cvmx-spinlock.h> | 37 | #include <asm/octeon/cvmx-spinlock.h> |
@@ -97,6 +98,33 @@ void *cvmx_bootmem_alloc(uint64_t size, uint64_t alignment) | |||
97 | return cvmx_bootmem_alloc_range(size, alignment, 0, 0); | 98 | return cvmx_bootmem_alloc_range(size, alignment, 0, 0); |
98 | } | 99 | } |
99 | 100 | ||
101 | void *cvmx_bootmem_alloc_named_range(uint64_t size, uint64_t min_addr, | ||
102 | uint64_t max_addr, uint64_t align, | ||
103 | char *name) | ||
104 | { | ||
105 | int64_t addr; | ||
106 | |||
107 | addr = cvmx_bootmem_phy_named_block_alloc(size, min_addr, max_addr, | ||
108 | align, name, 0); | ||
109 | if (addr >= 0) | ||
110 | return cvmx_phys_to_ptr(addr); | ||
111 | else | ||
112 | return NULL; | ||
113 | } | ||
114 | |||
115 | void *cvmx_bootmem_alloc_named_address(uint64_t size, uint64_t address, | ||
116 | char *name) | ||
117 | { | ||
118 | return cvmx_bootmem_alloc_named_range(size, address, address + size, | ||
119 | 0, name); | ||
120 | } | ||
121 | |||
122 | void *cvmx_bootmem_alloc_named(uint64_t size, uint64_t alignment, char *name) | ||
123 | { | ||
124 | return cvmx_bootmem_alloc_named_range(size, 0, 0, alignment, name); | ||
125 | } | ||
126 | EXPORT_SYMBOL(cvmx_bootmem_alloc_named); | ||
127 | |||
100 | int cvmx_bootmem_free_named(char *name) | 128 | int cvmx_bootmem_free_named(char *name) |
101 | { | 129 | { |
102 | return cvmx_bootmem_phy_named_block_free(name, 0); | 130 | return cvmx_bootmem_phy_named_block_free(name, 0); |
@@ -106,6 +134,7 @@ struct cvmx_bootmem_named_block_desc *cvmx_bootmem_find_named_block(char *name) | |||
106 | { | 134 | { |
107 | return cvmx_bootmem_phy_named_block_find(name, 0); | 135 | return cvmx_bootmem_phy_named_block_find(name, 0); |
108 | } | 136 | } |
137 | EXPORT_SYMBOL(cvmx_bootmem_find_named_block); | ||
109 | 138 | ||
110 | void cvmx_bootmem_lock(void) | 139 | void cvmx_bootmem_lock(void) |
111 | { | 140 | { |
@@ -584,3 +613,78 @@ int cvmx_bootmem_phy_named_block_free(char *name, uint32_t flags) | |||
584 | cvmx_bootmem_unlock(); | 613 | cvmx_bootmem_unlock(); |
585 | return named_block_ptr != NULL; /* 0 on failure, 1 on success */ | 614 | return named_block_ptr != NULL; /* 0 on failure, 1 on success */ |
586 | } | 615 | } |
616 | |||
617 | int64_t cvmx_bootmem_phy_named_block_alloc(uint64_t size, uint64_t min_addr, | ||
618 | uint64_t max_addr, | ||
619 | uint64_t alignment, | ||
620 | char *name, | ||
621 | uint32_t flags) | ||
622 | { | ||
623 | int64_t addr_allocated; | ||
624 | struct cvmx_bootmem_named_block_desc *named_block_desc_ptr; | ||
625 | |||
626 | #ifdef DEBUG | ||
627 | cvmx_dprintf("cvmx_bootmem_phy_named_block_alloc: size: 0x%llx, min: " | ||
628 | "0x%llx, max: 0x%llx, align: 0x%llx, name: %s\n", | ||
629 | (unsigned long long)size, | ||
630 | (unsigned long long)min_addr, | ||
631 | (unsigned long long)max_addr, | ||
632 | (unsigned long long)alignment, | ||
633 | name); | ||
634 | #endif | ||
635 | if (cvmx_bootmem_desc->major_version != 3) { | ||
636 | cvmx_dprintf("ERROR: Incompatible bootmem descriptor version: " | ||
637 | "%d.%d at addr: %p\n", | ||
638 | (int)cvmx_bootmem_desc->major_version, | ||
639 | (int)cvmx_bootmem_desc->minor_version, | ||
640 | cvmx_bootmem_desc); | ||
641 | return -1; | ||
642 | } | ||
643 | |||
644 | /* | ||
645 | * Take lock here, as name lookup/block alloc/name add need to | ||
646 | * be atomic. | ||
647 | */ | ||
648 | if (!(flags & CVMX_BOOTMEM_FLAG_NO_LOCKING)) | ||
649 | cvmx_spinlock_lock((cvmx_spinlock_t *)&(cvmx_bootmem_desc->lock)); | ||
650 | |||
651 | /* Get pointer to first available named block descriptor */ | ||
652 | named_block_desc_ptr = | ||
653 | cvmx_bootmem_phy_named_block_find(NULL, | ||
654 | flags | CVMX_BOOTMEM_FLAG_NO_LOCKING); | ||
655 | |||
656 | /* | ||
657 | * Check to see if name already in use, return error if name | ||
658 | * not available or no more room for blocks. | ||
659 | */ | ||
660 | if (cvmx_bootmem_phy_named_block_find(name, | ||
661 | flags | CVMX_BOOTMEM_FLAG_NO_LOCKING) || !named_block_desc_ptr) { | ||
662 | if (!(flags & CVMX_BOOTMEM_FLAG_NO_LOCKING)) | ||
663 | cvmx_spinlock_unlock((cvmx_spinlock_t *)&(cvmx_bootmem_desc->lock)); | ||
664 | return -1; | ||
665 | } | ||
666 | |||
667 | |||
668 | /* | ||
669 | * Round size up to mult of minimum alignment bytes We need | ||
670 | * the actual size allocated to allow for blocks to be | ||
671 | * coallesced when they are freed. The alloc routine does the | ||
672 | * same rounding up on all allocations. | ||
673 | */ | ||
674 | size = __ALIGN_MASK(size, (CVMX_BOOTMEM_ALIGNMENT_SIZE - 1)); | ||
675 | |||
676 | addr_allocated = cvmx_bootmem_phy_alloc(size, min_addr, max_addr, | ||
677 | alignment, | ||
678 | flags | CVMX_BOOTMEM_FLAG_NO_LOCKING); | ||
679 | if (addr_allocated >= 0) { | ||
680 | named_block_desc_ptr->base_addr = addr_allocated; | ||
681 | named_block_desc_ptr->size = size; | ||
682 | strncpy(named_block_desc_ptr->name, name, | ||
683 | cvmx_bootmem_desc->named_block_name_len); | ||
684 | named_block_desc_ptr->name[cvmx_bootmem_desc->named_block_name_len - 1] = 0; | ||
685 | } | ||
686 | |||
687 | if (!(flags & CVMX_BOOTMEM_FLAG_NO_LOCKING)) | ||
688 | cvmx_spinlock_unlock((cvmx_spinlock_t *)&(cvmx_bootmem_desc->lock)); | ||
689 | return addr_allocated; | ||
690 | } | ||
diff --git a/arch/mips/cavium-octeon/executive/cvmx-helper-errata.c b/arch/mips/cavium-octeon/executive/cvmx-helper-errata.c new file mode 100644 index 000000000000..868659e64d4a --- /dev/null +++ b/arch/mips/cavium-octeon/executive/cvmx-helper-errata.c | |||
@@ -0,0 +1,73 @@ | |||
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 | * | ||
30 | * Fixes and workaround for Octeon chip errata. This file | ||
31 | * contains functions called by cvmx-helper to workaround known | ||
32 | * chip errata. For the most part, code doesn't need to call | ||
33 | * these functions directly. | ||
34 | * | ||
35 | */ | ||
36 | #include <linux/module.h> | ||
37 | |||
38 | #include <asm/octeon/octeon.h> | ||
39 | |||
40 | #include <asm/octeon/cvmx-helper-jtag.h> | ||
41 | |||
42 | /** | ||
43 | * Due to errata G-720, the 2nd order CDR circuit on CN52XX pass | ||
44 | * 1 doesn't work properly. The following code disables 2nd order | ||
45 | * CDR for the specified QLM. | ||
46 | * | ||
47 | * @qlm: QLM to disable 2nd order CDR for. | ||
48 | */ | ||
49 | void __cvmx_helper_errata_qlm_disable_2nd_order_cdr(int qlm) | ||
50 | { | ||
51 | int lane; | ||
52 | cvmx_helper_qlm_jtag_init(); | ||
53 | /* We need to load all four lanes of the QLM, a total of 1072 bits */ | ||
54 | for (lane = 0; lane < 4; lane++) { | ||
55 | /* | ||
56 | * Each lane has 268 bits. We need to set | ||
57 | * cfg_cdr_incx<67:64> = 3 and cfg_cdr_secord<77> = | ||
58 | * 1. All other bits are zero. Bits go in LSB first, | ||
59 | * so start off with the zeros for bits <63:0>. | ||
60 | */ | ||
61 | cvmx_helper_qlm_jtag_shift_zeros(qlm, 63 - 0 + 1); | ||
62 | /* cfg_cdr_incx<67:64>=3 */ | ||
63 | cvmx_helper_qlm_jtag_shift(qlm, 67 - 64 + 1, 3); | ||
64 | /* Zeros for bits <76:68> */ | ||
65 | cvmx_helper_qlm_jtag_shift_zeros(qlm, 76 - 68 + 1); | ||
66 | /* cfg_cdr_secord<77>=1 */ | ||
67 | cvmx_helper_qlm_jtag_shift(qlm, 77 - 77 + 1, 1); | ||
68 | /* Zeros for bits <267:78> */ | ||
69 | cvmx_helper_qlm_jtag_shift_zeros(qlm, 267 - 78 + 1); | ||
70 | } | ||
71 | cvmx_helper_qlm_jtag_update(qlm); | ||
72 | } | ||
73 | EXPORT_SYMBOL(__cvmx_helper_errata_qlm_disable_2nd_order_cdr); | ||
diff --git a/arch/mips/cavium-octeon/executive/cvmx-helper-jtag.c b/arch/mips/cavium-octeon/executive/cvmx-helper-jtag.c new file mode 100644 index 000000000000..c1c54890bae0 --- /dev/null +++ b/arch/mips/cavium-octeon/executive/cvmx-helper-jtag.c | |||
@@ -0,0 +1,144 @@ | |||
1 | |||
2 | /***********************license start*************** | ||
3 | * Author: Cavium Networks | ||
4 | * | ||
5 | * Contact: support@caviumnetworks.com | ||
6 | * This file is part of the OCTEON SDK | ||
7 | * | ||
8 | * Copyright (c) 2003-2008 Cavium Networks | ||
9 | * | ||
10 | * This file is free software; you can redistribute it and/or modify | ||
11 | * it under the terms of the GNU General Public License, Version 2, as | ||
12 | * published by the Free Software Foundation. | ||
13 | * | ||
14 | * This file is distributed in the hope that it will be useful, but | ||
15 | * AS-IS and WITHOUT ANY WARRANTY; without even the implied warranty | ||
16 | * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE, TITLE, or | ||
17 | * NONINFRINGEMENT. See the GNU General Public License for more | ||
18 | * details. | ||
19 | * | ||
20 | * You should have received a copy of the GNU General Public License | ||
21 | * along with this file; if not, write to the Free Software | ||
22 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | ||
23 | * or visit http://www.gnu.org/licenses/. | ||
24 | * | ||
25 | * This file may also be available under a different license from Cavium. | ||
26 | * Contact Cavium Networks for more information | ||
27 | ***********************license end**************************************/ | ||
28 | |||
29 | /** | ||
30 | * | ||
31 | * Helper utilities for qlm_jtag. | ||
32 | * | ||
33 | */ | ||
34 | |||
35 | #include <asm/octeon/octeon.h> | ||
36 | #include <asm/octeon/cvmx-helper-jtag.h> | ||
37 | |||
38 | |||
39 | /** | ||
40 | * Initialize the internal QLM JTAG logic to allow programming | ||
41 | * of the JTAG chain by the cvmx_helper_qlm_jtag_*() functions. | ||
42 | * These functions should only be used at the direction of Cavium | ||
43 | * Networks. Programming incorrect values into the JTAG chain | ||
44 | * can cause chip damage. | ||
45 | */ | ||
46 | void cvmx_helper_qlm_jtag_init(void) | ||
47 | { | ||
48 | union cvmx_ciu_qlm_jtgc jtgc; | ||
49 | uint32_t clock_div = 0; | ||
50 | uint32_t divisor = cvmx_sysinfo_get()->cpu_clock_hz / (25 * 1000000); | ||
51 | divisor = (divisor - 1) >> 2; | ||
52 | /* Convert the divisor into a power of 2 shift */ | ||
53 | while (divisor) { | ||
54 | clock_div++; | ||
55 | divisor = divisor >> 1; | ||
56 | } | ||
57 | |||
58 | /* | ||
59 | * Clock divider for QLM JTAG operations. eclk is divided by | ||
60 | * 2^(CLK_DIV + 2) | ||
61 | */ | ||
62 | jtgc.u64 = 0; | ||
63 | jtgc.s.clk_div = clock_div; | ||
64 | jtgc.s.mux_sel = 0; | ||
65 | if (OCTEON_IS_MODEL(OCTEON_CN52XX)) | ||
66 | jtgc.s.bypass = 0x3; | ||
67 | else | ||
68 | jtgc.s.bypass = 0xf; | ||
69 | cvmx_write_csr(CVMX_CIU_QLM_JTGC, jtgc.u64); | ||
70 | cvmx_read_csr(CVMX_CIU_QLM_JTGC); | ||
71 | } | ||
72 | |||
73 | /** | ||
74 | * Write up to 32bits into the QLM jtag chain. Bits are shifted | ||
75 | * into the MSB and out the LSB, so you should shift in the low | ||
76 | * order bits followed by the high order bits. The JTAG chain is | ||
77 | * 4 * 268 bits long, or 1072. | ||
78 | * | ||
79 | * @qlm: QLM to shift value into | ||
80 | * @bits: Number of bits to shift in (1-32). | ||
81 | * @data: Data to shift in. Bit 0 enters the chain first, followed by | ||
82 | * bit 1, etc. | ||
83 | * | ||
84 | * Returns The low order bits of the JTAG chain that shifted out of the | ||
85 | * circle. | ||
86 | */ | ||
87 | uint32_t cvmx_helper_qlm_jtag_shift(int qlm, int bits, uint32_t data) | ||
88 | { | ||
89 | union cvmx_ciu_qlm_jtgd jtgd; | ||
90 | jtgd.u64 = 0; | ||
91 | jtgd.s.shift = 1; | ||
92 | jtgd.s.shft_cnt = bits - 1; | ||
93 | jtgd.s.shft_reg = data; | ||
94 | if (!OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1_X)) | ||
95 | jtgd.s.select = 1 << qlm; | ||
96 | cvmx_write_csr(CVMX_CIU_QLM_JTGD, jtgd.u64); | ||
97 | do { | ||
98 | jtgd.u64 = cvmx_read_csr(CVMX_CIU_QLM_JTGD); | ||
99 | } while (jtgd.s.shift); | ||
100 | return jtgd.s.shft_reg >> (32 - bits); | ||
101 | } | ||
102 | |||
103 | /** | ||
104 | * Shift long sequences of zeros into the QLM JTAG chain. It is | ||
105 | * common to need to shift more than 32 bits of zeros into the | ||
106 | * chain. This function is a convience wrapper around | ||
107 | * cvmx_helper_qlm_jtag_shift() to shift more than 32 bits of | ||
108 | * zeros at a time. | ||
109 | * | ||
110 | * @qlm: QLM to shift zeros into | ||
111 | * @bits: | ||
112 | */ | ||
113 | void cvmx_helper_qlm_jtag_shift_zeros(int qlm, int bits) | ||
114 | { | ||
115 | while (bits > 0) { | ||
116 | int n = bits; | ||
117 | if (n > 32) | ||
118 | n = 32; | ||
119 | cvmx_helper_qlm_jtag_shift(qlm, n, 0); | ||
120 | bits -= n; | ||
121 | } | ||
122 | } | ||
123 | |||
124 | /** | ||
125 | * Program the QLM JTAG chain into all lanes of the QLM. You must | ||
126 | * have already shifted in 268*4, or 1072 bits into the JTAG | ||
127 | * chain. Updating invalid values can possibly cause chip damage. | ||
128 | * | ||
129 | * @qlm: QLM to program | ||
130 | */ | ||
131 | void cvmx_helper_qlm_jtag_update(int qlm) | ||
132 | { | ||
133 | union cvmx_ciu_qlm_jtgd jtgd; | ||
134 | |||
135 | /* Update the new data */ | ||
136 | jtgd.u64 = 0; | ||
137 | jtgd.s.update = 1; | ||
138 | if (!OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1_X)) | ||
139 | jtgd.s.select = 1 << qlm; | ||
140 | cvmx_write_csr(CVMX_CIU_QLM_JTGD, jtgd.u64); | ||
141 | do { | ||
142 | jtgd.u64 = cvmx_read_csr(CVMX_CIU_QLM_JTGD); | ||
143 | } while (jtgd.s.update); | ||
144 | } | ||
diff --git a/arch/mips/cavium-octeon/executive/cvmx-sysinfo.c b/arch/mips/cavium-octeon/executive/cvmx-sysinfo.c index 4812370706a1..e5838890cba5 100644 --- a/arch/mips/cavium-octeon/executive/cvmx-sysinfo.c +++ b/arch/mips/cavium-octeon/executive/cvmx-sysinfo.c | |||
@@ -29,6 +29,7 @@ | |||
29 | * This module provides system/board/application information obtained | 29 | * This module provides system/board/application information obtained |
30 | * by the bootloader. | 30 | * by the bootloader. |
31 | */ | 31 | */ |
32 | #include <linux/module.h> | ||
32 | 33 | ||
33 | #include <asm/octeon/cvmx.h> | 34 | #include <asm/octeon/cvmx.h> |
34 | #include <asm/octeon/cvmx-spinlock.h> | 35 | #include <asm/octeon/cvmx-spinlock.h> |
@@ -69,6 +70,7 @@ struct cvmx_sysinfo *cvmx_sysinfo_get(void) | |||
69 | { | 70 | { |
70 | return &(state.sysinfo); | 71 | return &(state.sysinfo); |
71 | } | 72 | } |
73 | EXPORT_SYMBOL(cvmx_sysinfo_get); | ||
72 | 74 | ||
73 | /** | 75 | /** |
74 | * This function is used in non-simple executive environments (such as | 76 | * This function is used in non-simple executive environments (such as |
diff --git a/arch/mips/cavium-octeon/msi.c b/arch/mips/cavium-octeon/msi.c new file mode 100644 index 000000000000..964b03b75a8f --- /dev/null +++ b/arch/mips/cavium-octeon/msi.c | |||
@@ -0,0 +1,288 @@ | |||
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) 2005-2007 Cavium Networks | ||
7 | */ | ||
8 | #include <linux/kernel.h> | ||
9 | #include <linux/init.h> | ||
10 | #include <linux/msi.h> | ||
11 | #include <linux/spinlock.h> | ||
12 | #include <linux/interrupt.h> | ||
13 | |||
14 | #include <asm/octeon/octeon.h> | ||
15 | #include <asm/octeon/cvmx-npi-defs.h> | ||
16 | #include <asm/octeon/cvmx-pci-defs.h> | ||
17 | #include <asm/octeon/cvmx-npei-defs.h> | ||
18 | #include <asm/octeon/cvmx-pexp-defs.h> | ||
19 | |||
20 | #include "pci-common.h" | ||
21 | |||
22 | /* | ||
23 | * Each bit in msi_free_irq_bitmask represents a MSI interrupt that is | ||
24 | * in use. | ||
25 | */ | ||
26 | static uint64_t msi_free_irq_bitmask; | ||
27 | |||
28 | /* | ||
29 | * Each bit in msi_multiple_irq_bitmask tells that the device using | ||
30 | * this bit in msi_free_irq_bitmask is also using the next bit. This | ||
31 | * is used so we can disable all of the MSI interrupts when a device | ||
32 | * uses multiple. | ||
33 | */ | ||
34 | static uint64_t msi_multiple_irq_bitmask; | ||
35 | |||
36 | /* | ||
37 | * This lock controls updates to msi_free_irq_bitmask and | ||
38 | * msi_multiple_irq_bitmask. | ||
39 | */ | ||
40 | static DEFINE_SPINLOCK(msi_free_irq_bitmask_lock); | ||
41 | |||
42 | |||
43 | /** | ||
44 | * Called when a driver request MSI interrupts instead of the | ||
45 | * legacy INT A-D. This routine will allocate multiple interrupts | ||
46 | * for MSI devices that support them. A device can override this by | ||
47 | * programming the MSI control bits [6:4] before calling | ||
48 | * pci_enable_msi(). | ||
49 | * | ||
50 | * @param dev Device requesting MSI interrupts | ||
51 | * @param desc MSI descriptor | ||
52 | * | ||
53 | * Returns 0 on success. | ||
54 | */ | ||
55 | int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc) | ||
56 | { | ||
57 | struct msi_msg msg; | ||
58 | uint16_t control; | ||
59 | int configured_private_bits; | ||
60 | int request_private_bits; | ||
61 | int irq; | ||
62 | int irq_step; | ||
63 | uint64_t search_mask; | ||
64 | |||
65 | /* | ||
66 | * Read the MSI config to figure out how many IRQs this device | ||
67 | * wants. Most devices only want 1, which will give | ||
68 | * configured_private_bits and request_private_bits equal 0. | ||
69 | */ | ||
70 | pci_read_config_word(dev, desc->msi_attrib.pos + PCI_MSI_FLAGS, | ||
71 | &control); | ||
72 | |||
73 | /* | ||
74 | * If the number of private bits has been configured then use | ||
75 | * that value instead of the requested number. This gives the | ||
76 | * driver the chance to override the number of interrupts | ||
77 | * before calling pci_enable_msi(). | ||
78 | */ | ||
79 | configured_private_bits = (control & PCI_MSI_FLAGS_QSIZE) >> 4; | ||
80 | if (configured_private_bits == 0) { | ||
81 | /* Nothing is configured, so use the hardware requested size */ | ||
82 | request_private_bits = (control & PCI_MSI_FLAGS_QMASK) >> 1; | ||
83 | } else { | ||
84 | /* | ||
85 | * Use the number of configured bits, assuming the | ||
86 | * driver wanted to override the hardware request | ||
87 | * value. | ||
88 | */ | ||
89 | request_private_bits = configured_private_bits; | ||
90 | } | ||
91 | |||
92 | /* | ||
93 | * The PCI 2.3 spec mandates that there are at most 32 | ||
94 | * interrupts. If this device asks for more, only give it one. | ||
95 | */ | ||
96 | if (request_private_bits > 5) | ||
97 | request_private_bits = 0; | ||
98 | |||
99 | try_only_one: | ||
100 | /* | ||
101 | * The IRQs have to be aligned on a power of two based on the | ||
102 | * number being requested. | ||
103 | */ | ||
104 | irq_step = 1 << request_private_bits; | ||
105 | |||
106 | /* Mask with one bit for each IRQ */ | ||
107 | search_mask = (1 << irq_step) - 1; | ||
108 | |||
109 | /* | ||
110 | * We're going to search msi_free_irq_bitmask_lock for zero | ||
111 | * bits. This represents an MSI interrupt number that isn't in | ||
112 | * use. | ||
113 | */ | ||
114 | spin_lock(&msi_free_irq_bitmask_lock); | ||
115 | for (irq = 0; irq < 64; irq += irq_step) { | ||
116 | if ((msi_free_irq_bitmask & (search_mask << irq)) == 0) { | ||
117 | msi_free_irq_bitmask |= search_mask << irq; | ||
118 | msi_multiple_irq_bitmask |= (search_mask >> 1) << irq; | ||
119 | break; | ||
120 | } | ||
121 | } | ||
122 | spin_unlock(&msi_free_irq_bitmask_lock); | ||
123 | |||
124 | /* Make sure the search for available interrupts didn't fail */ | ||
125 | if (irq >= 64) { | ||
126 | if (request_private_bits) { | ||
127 | pr_err("arch_setup_msi_irq: Unable to find %d free " | ||
128 | "interrupts, trying just one", | ||
129 | 1 << request_private_bits); | ||
130 | request_private_bits = 0; | ||
131 | goto try_only_one; | ||
132 | } else | ||
133 | panic("arch_setup_msi_irq: Unable to find a free MSI " | ||
134 | "interrupt"); | ||
135 | } | ||
136 | |||
137 | /* MSI interrupts start at logical IRQ OCTEON_IRQ_MSI_BIT0 */ | ||
138 | irq += OCTEON_IRQ_MSI_BIT0; | ||
139 | |||
140 | switch (octeon_dma_bar_type) { | ||
141 | case OCTEON_DMA_BAR_TYPE_SMALL: | ||
142 | /* When not using big bar, Bar 0 is based at 128MB */ | ||
143 | msg.address_lo = | ||
144 | ((128ul << 20) + CVMX_PCI_MSI_RCV) & 0xffffffff; | ||
145 | msg.address_hi = ((128ul << 20) + CVMX_PCI_MSI_RCV) >> 32; | ||
146 | case OCTEON_DMA_BAR_TYPE_BIG: | ||
147 | /* When using big bar, Bar 0 is based at 0 */ | ||
148 | msg.address_lo = (0 + CVMX_PCI_MSI_RCV) & 0xffffffff; | ||
149 | msg.address_hi = (0 + CVMX_PCI_MSI_RCV) >> 32; | ||
150 | break; | ||
151 | case OCTEON_DMA_BAR_TYPE_PCIE: | ||
152 | /* When using PCIe, Bar 0 is based at 0 */ | ||
153 | /* FIXME CVMX_NPEI_MSI_RCV* other than 0? */ | ||
154 | msg.address_lo = (0 + CVMX_NPEI_PCIE_MSI_RCV) & 0xffffffff; | ||
155 | msg.address_hi = (0 + CVMX_NPEI_PCIE_MSI_RCV) >> 32; | ||
156 | break; | ||
157 | default: | ||
158 | panic("arch_setup_msi_irq: Invalid octeon_dma_bar_type\n"); | ||
159 | } | ||
160 | msg.data = irq - OCTEON_IRQ_MSI_BIT0; | ||
161 | |||
162 | /* Update the number of IRQs the device has available to it */ | ||
163 | control &= ~PCI_MSI_FLAGS_QSIZE; | ||
164 | control |= request_private_bits << 4; | ||
165 | pci_write_config_word(dev, desc->msi_attrib.pos + PCI_MSI_FLAGS, | ||
166 | control); | ||
167 | |||
168 | set_irq_msi(irq, desc); | ||
169 | write_msi_msg(irq, &msg); | ||
170 | return 0; | ||
171 | } | ||
172 | |||
173 | |||
174 | /** | ||
175 | * Called when a device no longer needs its MSI interrupts. All | ||
176 | * MSI interrupts for the device are freed. | ||
177 | * | ||
178 | * @irq: The devices first irq number. There may be multple in sequence. | ||
179 | */ | ||
180 | void arch_teardown_msi_irq(unsigned int irq) | ||
181 | { | ||
182 | int number_irqs; | ||
183 | uint64_t bitmask; | ||
184 | |||
185 | if ((irq < OCTEON_IRQ_MSI_BIT0) || (irq > OCTEON_IRQ_MSI_BIT63)) | ||
186 | panic("arch_teardown_msi_irq: Attempted to teardown illegal " | ||
187 | "MSI interrupt (%d)", irq); | ||
188 | irq -= OCTEON_IRQ_MSI_BIT0; | ||
189 | |||
190 | /* | ||
191 | * Count the number of IRQs we need to free by looking at the | ||
192 | * msi_multiple_irq_bitmask. Each bit set means that the next | ||
193 | * IRQ is also owned by this device. | ||
194 | */ | ||
195 | number_irqs = 0; | ||
196 | while ((irq+number_irqs < 64) && | ||
197 | (msi_multiple_irq_bitmask & (1ull << (irq + number_irqs)))) | ||
198 | number_irqs++; | ||
199 | number_irqs++; | ||
200 | /* Mask with one bit for each IRQ */ | ||
201 | bitmask = (1 << number_irqs) - 1; | ||
202 | /* Shift the mask to the correct bit location */ | ||
203 | bitmask <<= irq; | ||
204 | if ((msi_free_irq_bitmask & bitmask) != bitmask) | ||
205 | panic("arch_teardown_msi_irq: Attempted to teardown MSI " | ||
206 | "interrupt (%d) not in use", irq); | ||
207 | |||
208 | /* Checks are done, update the in use bitmask */ | ||
209 | spin_lock(&msi_free_irq_bitmask_lock); | ||
210 | msi_free_irq_bitmask &= ~bitmask; | ||
211 | msi_multiple_irq_bitmask &= ~bitmask; | ||
212 | spin_unlock(&msi_free_irq_bitmask_lock); | ||
213 | } | ||
214 | |||
215 | |||
216 | /** | ||
217 | * Called by the interrupt handling code when an MSI interrupt | ||
218 | * occurs. | ||
219 | * | ||
220 | * @param cpl | ||
221 | * @param dev_id | ||
222 | * | ||
223 | * @return | ||
224 | */ | ||
225 | static irqreturn_t octeon_msi_interrupt(int cpl, void *dev_id) | ||
226 | { | ||
227 | uint64_t msi_bits; | ||
228 | int irq; | ||
229 | |||
230 | if (octeon_dma_bar_type == OCTEON_DMA_BAR_TYPE_PCIE) | ||
231 | msi_bits = cvmx_read_csr(CVMX_PEXP_NPEI_MSI_RCV0); | ||
232 | else | ||
233 | msi_bits = cvmx_read_csr(CVMX_NPI_NPI_MSI_RCV); | ||
234 | irq = fls64(msi_bits); | ||
235 | if (irq) { | ||
236 | irq += OCTEON_IRQ_MSI_BIT0 - 1; | ||
237 | if (irq_desc[irq].action) { | ||
238 | do_IRQ(irq); | ||
239 | return IRQ_HANDLED; | ||
240 | } else { | ||
241 | pr_err("Spurious MSI interrupt %d\n", irq); | ||
242 | if (octeon_has_feature(OCTEON_FEATURE_PCIE)) { | ||
243 | /* These chips have PCIe */ | ||
244 | cvmx_write_csr(CVMX_PEXP_NPEI_MSI_RCV0, | ||
245 | 1ull << (irq - | ||
246 | OCTEON_IRQ_MSI_BIT0)); | ||
247 | } else { | ||
248 | /* These chips have PCI */ | ||
249 | cvmx_write_csr(CVMX_NPI_NPI_MSI_RCV, | ||
250 | 1ull << (irq - | ||
251 | OCTEON_IRQ_MSI_BIT0)); | ||
252 | } | ||
253 | } | ||
254 | } | ||
255 | return IRQ_NONE; | ||
256 | } | ||
257 | |||
258 | |||
259 | /** | ||
260 | * Initializes the MSI interrupt handling code | ||
261 | * | ||
262 | * @return | ||
263 | */ | ||
264 | int octeon_msi_initialize(void) | ||
265 | { | ||
266 | int r; | ||
267 | if (octeon_has_feature(OCTEON_FEATURE_PCIE)) { | ||
268 | r = request_irq(OCTEON_IRQ_PCI_MSI0, octeon_msi_interrupt, | ||
269 | IRQF_SHARED, | ||
270 | "MSI[0:63]", octeon_msi_interrupt); | ||
271 | } else if (octeon_is_pci_host()) { | ||
272 | r = request_irq(OCTEON_IRQ_PCI_MSI0, octeon_msi_interrupt, | ||
273 | IRQF_SHARED, | ||
274 | "MSI[0:15]", octeon_msi_interrupt); | ||
275 | r += request_irq(OCTEON_IRQ_PCI_MSI1, octeon_msi_interrupt, | ||
276 | IRQF_SHARED, | ||
277 | "MSI[16:31]", octeon_msi_interrupt); | ||
278 | r += request_irq(OCTEON_IRQ_PCI_MSI2, octeon_msi_interrupt, | ||
279 | IRQF_SHARED, | ||
280 | "MSI[32:47]", octeon_msi_interrupt); | ||
281 | r += request_irq(OCTEON_IRQ_PCI_MSI3, octeon_msi_interrupt, | ||
282 | IRQF_SHARED, | ||
283 | "MSI[48:63]", octeon_msi_interrupt); | ||
284 | } | ||
285 | return 0; | ||
286 | } | ||
287 | |||
288 | subsys_initcall(octeon_msi_initialize); | ||
diff --git a/arch/mips/cavium-octeon/octeon-irq.c b/arch/mips/cavium-octeon/octeon-irq.c index d3a0c8154bec..8dfa009e0070 100644 --- a/arch/mips/cavium-octeon/octeon-irq.c +++ b/arch/mips/cavium-octeon/octeon-irq.c | |||
@@ -10,6 +10,8 @@ | |||
10 | #include <linux/hardirq.h> | 10 | #include <linux/hardirq.h> |
11 | 11 | ||
12 | #include <asm/octeon/octeon.h> | 12 | #include <asm/octeon/octeon.h> |
13 | #include <asm/octeon/cvmx-pexp-defs.h> | ||
14 | #include <asm/octeon/cvmx-npi-defs.h> | ||
13 | 15 | ||
14 | DEFINE_RWLOCK(octeon_irq_ciu0_rwlock); | 16 | DEFINE_RWLOCK(octeon_irq_ciu0_rwlock); |
15 | DEFINE_RWLOCK(octeon_irq_ciu1_rwlock); | 17 | DEFINE_RWLOCK(octeon_irq_ciu1_rwlock); |
diff --git a/arch/mips/cavium-octeon/pci-common.c b/arch/mips/cavium-octeon/pci-common.c new file mode 100644 index 000000000000..cd029f88da7f --- /dev/null +++ b/arch/mips/cavium-octeon/pci-common.c | |||
@@ -0,0 +1,137 @@ | |||
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) 2005-2007 Cavium Networks | ||
7 | */ | ||
8 | #include <linux/kernel.h> | ||
9 | #include <linux/init.h> | ||
10 | #include <linux/pci.h> | ||
11 | #include <linux/interrupt.h> | ||
12 | #include <linux/time.h> | ||
13 | #include <linux/delay.h> | ||
14 | #include "pci-common.h" | ||
15 | |||
16 | typeof(pcibios_map_irq) *octeon_pcibios_map_irq; | ||
17 | enum octeon_dma_bar_type octeon_dma_bar_type = OCTEON_DMA_BAR_TYPE_INVALID; | ||
18 | |||
19 | /** | ||
20 | * Map a PCI device to the appropriate interrupt line | ||
21 | * | ||
22 | * @param dev The Linux PCI device structure for the device to map | ||
23 | * @param slot The slot number for this device on __BUS 0__. Linux | ||
24 | * enumerates through all the bridges and figures out the | ||
25 | * slot on Bus 0 where this device eventually hooks to. | ||
26 | * @param pin The PCI interrupt pin read from the device, then swizzled | ||
27 | * as it goes through each bridge. | ||
28 | * @return Interrupt number for the device | ||
29 | */ | ||
30 | int __init pcibios_map_irq(const struct pci_dev *dev, u8 slot, u8 pin) | ||
31 | { | ||
32 | if (octeon_pcibios_map_irq) | ||
33 | return octeon_pcibios_map_irq(dev, slot, pin); | ||
34 | else | ||
35 | panic("octeon_pcibios_map_irq doesn't point to a " | ||
36 | "pcibios_map_irq() function"); | ||
37 | } | ||
38 | |||
39 | |||
40 | /** | ||
41 | * Called to perform platform specific PCI setup | ||
42 | * | ||
43 | * @param dev | ||
44 | * @return | ||
45 | */ | ||
46 | int pcibios_plat_dev_init(struct pci_dev *dev) | ||
47 | { | ||
48 | uint16_t config; | ||
49 | uint32_t dconfig; | ||
50 | int pos; | ||
51 | /* | ||
52 | * Force the Cache line setting to 64 bytes. The standard | ||
53 | * Linux bus scan doesn't seem to set it. Octeon really has | ||
54 | * 128 byte lines, but Intel bridges get really upset if you | ||
55 | * try and set values above 64 bytes. Value is specified in | ||
56 | * 32bit words. | ||
57 | */ | ||
58 | pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, 64 / 4); | ||
59 | /* Set latency timers for all devices */ | ||
60 | pci_write_config_byte(dev, PCI_LATENCY_TIMER, 48); | ||
61 | |||
62 | /* Enable reporting System errors and parity errors on all devices */ | ||
63 | /* Enable parity checking and error reporting */ | ||
64 | pci_read_config_word(dev, PCI_COMMAND, &config); | ||
65 | config |= PCI_COMMAND_PARITY | PCI_COMMAND_SERR; | ||
66 | pci_write_config_word(dev, PCI_COMMAND, config); | ||
67 | |||
68 | if (dev->subordinate) { | ||
69 | /* Set latency timers on sub bridges */ | ||
70 | pci_write_config_byte(dev, PCI_SEC_LATENCY_TIMER, 48); | ||
71 | /* More bridge error detection */ | ||
72 | pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &config); | ||
73 | config |= PCI_BRIDGE_CTL_PARITY | PCI_BRIDGE_CTL_SERR; | ||
74 | pci_write_config_word(dev, PCI_BRIDGE_CONTROL, config); | ||
75 | } | ||
76 | |||
77 | /* Enable the PCIe normal error reporting */ | ||
78 | pos = pci_find_capability(dev, PCI_CAP_ID_EXP); | ||
79 | if (pos) { | ||
80 | /* Update Device Control */ | ||
81 | pci_read_config_word(dev, pos + PCI_EXP_DEVCTL, &config); | ||
82 | /* Correctable Error Reporting */ | ||
83 | config |= PCI_EXP_DEVCTL_CERE; | ||
84 | /* Non-Fatal Error Reporting */ | ||
85 | config |= PCI_EXP_DEVCTL_NFERE; | ||
86 | /* Fatal Error Reporting */ | ||
87 | config |= PCI_EXP_DEVCTL_FERE; | ||
88 | /* Unsupported Request */ | ||
89 | config |= PCI_EXP_DEVCTL_URRE; | ||
90 | pci_write_config_word(dev, pos + PCI_EXP_DEVCTL, config); | ||
91 | } | ||
92 | |||
93 | /* Find the Advanced Error Reporting capability */ | ||
94 | pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR); | ||
95 | if (pos) { | ||
96 | /* Clear Uncorrectable Error Status */ | ||
97 | pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, | ||
98 | &dconfig); | ||
99 | pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, | ||
100 | dconfig); | ||
101 | /* Enable reporting of all uncorrectable errors */ | ||
102 | /* Uncorrectable Error Mask - turned on bits disable errors */ | ||
103 | pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, 0); | ||
104 | /* | ||
105 | * Leave severity at HW default. This only controls if | ||
106 | * errors are reported as uncorrectable or | ||
107 | * correctable, not if the error is reported. | ||
108 | */ | ||
109 | /* PCI_ERR_UNCOR_SEVER - Uncorrectable Error Severity */ | ||
110 | /* Clear Correctable Error Status */ | ||
111 | pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &dconfig); | ||
112 | pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, dconfig); | ||
113 | /* Enable reporting of all correctable errors */ | ||
114 | /* Correctable Error Mask - turned on bits disable errors */ | ||
115 | pci_write_config_dword(dev, pos + PCI_ERR_COR_MASK, 0); | ||
116 | /* Advanced Error Capabilities */ | ||
117 | pci_read_config_dword(dev, pos + PCI_ERR_CAP, &dconfig); | ||
118 | /* ECRC Generation Enable */ | ||
119 | if (config & PCI_ERR_CAP_ECRC_GENC) | ||
120 | config |= PCI_ERR_CAP_ECRC_GENE; | ||
121 | /* ECRC Check Enable */ | ||
122 | if (config & PCI_ERR_CAP_ECRC_CHKC) | ||
123 | config |= PCI_ERR_CAP_ECRC_CHKE; | ||
124 | pci_write_config_dword(dev, pos + PCI_ERR_CAP, dconfig); | ||
125 | /* PCI_ERR_HEADER_LOG - Header Log Register (16 bytes) */ | ||
126 | /* Report all errors to the root complex */ | ||
127 | pci_write_config_dword(dev, pos + PCI_ERR_ROOT_COMMAND, | ||
128 | PCI_ERR_ROOT_CMD_COR_EN | | ||
129 | PCI_ERR_ROOT_CMD_NONFATAL_EN | | ||
130 | PCI_ERR_ROOT_CMD_FATAL_EN); | ||
131 | /* Clear the Root status register */ | ||
132 | pci_read_config_dword(dev, pos + PCI_ERR_ROOT_STATUS, &dconfig); | ||
133 | pci_write_config_dword(dev, pos + PCI_ERR_ROOT_STATUS, dconfig); | ||
134 | } | ||
135 | |||
136 | return 0; | ||
137 | } | ||
diff --git a/arch/mips/cavium-octeon/pci-common.h b/arch/mips/cavium-octeon/pci-common.h new file mode 100644 index 000000000000..74ae79991e45 --- /dev/null +++ b/arch/mips/cavium-octeon/pci-common.h | |||
@@ -0,0 +1,39 @@ | |||
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) 2005-2007 Cavium Networks | ||
7 | */ | ||
8 | #ifndef __OCTEON_PCI_COMMON_H__ | ||
9 | #define __OCTEON_PCI_COMMON_H__ | ||
10 | |||
11 | #include <linux/pci.h> | ||
12 | |||
13 | /* Some PCI cards require delays when accessing config space. */ | ||
14 | #define PCI_CONFIG_SPACE_DELAY 10000 | ||
15 | |||
16 | /* pcibios_map_irq() is defined inside pci-common.c. All it does is call the | ||
17 | Octeon specific version pointed to by this variable. This function needs to | ||
18 | change for PCI or PCIe based hosts */ | ||
19 | extern typeof(pcibios_map_irq) *octeon_pcibios_map_irq; | ||
20 | |||
21 | /* The following defines are only used when octeon_dma_bar_type = | ||
22 | OCTEON_DMA_BAR_TYPE_BIG */ | ||
23 | #define OCTEON_PCI_BAR1_HOLE_BITS 5 | ||
24 | #define OCTEON_PCI_BAR1_HOLE_SIZE (1ul<<(OCTEON_PCI_BAR1_HOLE_BITS+3)) | ||
25 | |||
26 | enum octeon_dma_bar_type { | ||
27 | OCTEON_DMA_BAR_TYPE_INVALID, | ||
28 | OCTEON_DMA_BAR_TYPE_SMALL, | ||
29 | OCTEON_DMA_BAR_TYPE_BIG, | ||
30 | OCTEON_DMA_BAR_TYPE_PCIE | ||
31 | }; | ||
32 | |||
33 | /** | ||
34 | * This is a variable to tell the DMA mapping system in dma-octeon.c | ||
35 | * how to map PCI DMA addresses. | ||
36 | */ | ||
37 | extern enum octeon_dma_bar_type octeon_dma_bar_type; | ||
38 | |||
39 | #endif | ||
diff --git a/arch/mips/cavium-octeon/pci.c b/arch/mips/cavium-octeon/pci.c new file mode 100644 index 000000000000..67c0ff5e92f1 --- /dev/null +++ b/arch/mips/cavium-octeon/pci.c | |||
@@ -0,0 +1,568 @@ | |||
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) 2005-2007 Cavium Networks | ||
7 | */ | ||
8 | #include <linux/kernel.h> | ||
9 | #include <linux/init.h> | ||
10 | #include <linux/pci.h> | ||
11 | #include <linux/interrupt.h> | ||
12 | #include <linux/time.h> | ||
13 | #include <linux/delay.h> | ||
14 | |||
15 | #include <asm/time.h> | ||
16 | |||
17 | #include <asm/octeon/octeon.h> | ||
18 | #include <asm/octeon/cvmx-npi-defs.h> | ||
19 | #include <asm/octeon/cvmx-pci-defs.h> | ||
20 | |||
21 | #include "pci-common.h" | ||
22 | |||
23 | #define USE_OCTEON_INTERNAL_ARBITER | ||
24 | |||
25 | /* | ||
26 | * Octeon's PCI controller uses did=3, subdid=2 for PCI IO | ||
27 | * addresses. Use PCI endian swapping 1 so no address swapping is | ||
28 | * necessary. The Linux io routines will endian swap the data. | ||
29 | */ | ||
30 | #define OCTEON_PCI_IOSPACE_BASE 0x80011a0400000000ull | ||
31 | #define OCTEON_PCI_IOSPACE_SIZE (1ull<<32) | ||
32 | |||
33 | /* Octeon't PCI controller uses did=3, subdid=3 for PCI memory. */ | ||
34 | #define OCTEON_PCI_MEMSPACE_OFFSET (0x00011b0000000000ull) | ||
35 | |||
36 | /** | ||
37 | * This is the bit decoding used for the Octeon PCI controller addresses | ||
38 | */ | ||
39 | union octeon_pci_address { | ||
40 | uint64_t u64; | ||
41 | struct { | ||
42 | uint64_t upper:2; | ||
43 | uint64_t reserved:13; | ||
44 | uint64_t io:1; | ||
45 | uint64_t did:5; | ||
46 | uint64_t subdid:3; | ||
47 | uint64_t reserved2:4; | ||
48 | uint64_t endian_swap:2; | ||
49 | uint64_t reserved3:10; | ||
50 | uint64_t bus:8; | ||
51 | uint64_t dev:5; | ||
52 | uint64_t func:3; | ||
53 | uint64_t reg:8; | ||
54 | } s; | ||
55 | }; | ||
56 | |||
57 | /** | ||
58 | * Return the mapping of PCI device number to IRQ line. Each | ||
59 | * character in the return string represents the interrupt | ||
60 | * line for the device at that position. Device 1 maps to the | ||
61 | * first character, etc. The characters A-D are used for PCI | ||
62 | * interrupts. | ||
63 | * | ||
64 | * Returns PCI interrupt mapping | ||
65 | */ | ||
66 | const char *octeon_get_pci_interrupts(void) | ||
67 | { | ||
68 | /* | ||
69 | * Returning an empty string causes the interrupts to be | ||
70 | * routed based on the PCI specification. From the PCI spec: | ||
71 | * | ||
72 | * INTA# of Device Number 0 is connected to IRQW on the system | ||
73 | * board. (Device Number has no significance regarding being | ||
74 | * located on the system board or in a connector.) INTA# of | ||
75 | * Device Number 1 is connected to IRQX on the system | ||
76 | * board. INTA# of Device Number 2 is connected to IRQY on the | ||
77 | * system board. INTA# of Device Number 3 is connected to IRQZ | ||
78 | * on the system board. The table below describes how each | ||
79 | * agent's INTx# lines are connected to the system board | ||
80 | * interrupt lines. The following equation can be used to | ||
81 | * determine to which INTx# signal on the system board a given | ||
82 | * device's INTx# line(s) is connected. | ||
83 | * | ||
84 | * MB = (D + I) MOD 4 MB = System board Interrupt (IRQW = 0, | ||
85 | * IRQX = 1, IRQY = 2, and IRQZ = 3) D = Device Number I = | ||
86 | * Interrupt Number (INTA# = 0, INTB# = 1, INTC# = 2, and | ||
87 | * INTD# = 3) | ||
88 | */ | ||
89 | switch (octeon_bootinfo->board_type) { | ||
90 | case CVMX_BOARD_TYPE_NAO38: | ||
91 | /* This is really the NAC38 */ | ||
92 | return "AAAAADABAAAAAAAAAAAAAAAAAAAAAAAA"; | ||
93 | case CVMX_BOARD_TYPE_THUNDER: | ||
94 | return ""; | ||
95 | case CVMX_BOARD_TYPE_EBH3000: | ||
96 | return ""; | ||
97 | case CVMX_BOARD_TYPE_EBH3100: | ||
98 | case CVMX_BOARD_TYPE_CN3010_EVB_HS5: | ||
99 | case CVMX_BOARD_TYPE_CN3005_EVB_HS5: | ||
100 | return "AAABAAAAAAAAAAAAAAAAAAAAAAAAAAAA"; | ||
101 | case CVMX_BOARD_TYPE_BBGW_REF: | ||
102 | return "AABCD"; | ||
103 | default: | ||
104 | return ""; | ||
105 | } | ||
106 | } | ||
107 | |||
108 | /** | ||
109 | * Map a PCI device to the appropriate interrupt line | ||
110 | * | ||
111 | * @dev: The Linux PCI device structure for the device to map | ||
112 | * @slot: The slot number for this device on __BUS 0__. Linux | ||
113 | * enumerates through all the bridges and figures out the | ||
114 | * slot on Bus 0 where this device eventually hooks to. | ||
115 | * @pin: The PCI interrupt pin read from the device, then swizzled | ||
116 | * as it goes through each bridge. | ||
117 | * Returns Interrupt number for the device | ||
118 | */ | ||
119 | int __init octeon_pci_pcibios_map_irq(const struct pci_dev *dev, | ||
120 | u8 slot, u8 pin) | ||
121 | { | ||
122 | int irq_num; | ||
123 | const char *interrupts; | ||
124 | int dev_num; | ||
125 | |||
126 | /* Get the board specific interrupt mapping */ | ||
127 | interrupts = octeon_get_pci_interrupts(); | ||
128 | |||
129 | dev_num = dev->devfn >> 3; | ||
130 | if (dev_num < strlen(interrupts)) | ||
131 | irq_num = ((interrupts[dev_num] - 'A' + pin - 1) & 3) + | ||
132 | OCTEON_IRQ_PCI_INT0; | ||
133 | else | ||
134 | irq_num = ((slot + pin - 3) & 3) + OCTEON_IRQ_PCI_INT0; | ||
135 | return irq_num; | ||
136 | } | ||
137 | |||
138 | |||
139 | /** | ||
140 | * Read a value from configuration space | ||
141 | * | ||
142 | */ | ||
143 | static int octeon_read_config(struct pci_bus *bus, unsigned int devfn, | ||
144 | int reg, int size, u32 *val) | ||
145 | { | ||
146 | union octeon_pci_address pci_addr; | ||
147 | |||
148 | pci_addr.u64 = 0; | ||
149 | pci_addr.s.upper = 2; | ||
150 | pci_addr.s.io = 1; | ||
151 | pci_addr.s.did = 3; | ||
152 | pci_addr.s.subdid = 1; | ||
153 | pci_addr.s.endian_swap = 1; | ||
154 | pci_addr.s.bus = bus->number; | ||
155 | pci_addr.s.dev = devfn >> 3; | ||
156 | pci_addr.s.func = devfn & 0x7; | ||
157 | pci_addr.s.reg = reg; | ||
158 | |||
159 | #if PCI_CONFIG_SPACE_DELAY | ||
160 | udelay(PCI_CONFIG_SPACE_DELAY); | ||
161 | #endif | ||
162 | switch (size) { | ||
163 | case 4: | ||
164 | *val = le32_to_cpu(cvmx_read64_uint32(pci_addr.u64)); | ||
165 | return PCIBIOS_SUCCESSFUL; | ||
166 | case 2: | ||
167 | *val = le16_to_cpu(cvmx_read64_uint16(pci_addr.u64)); | ||
168 | return PCIBIOS_SUCCESSFUL; | ||
169 | case 1: | ||
170 | *val = cvmx_read64_uint8(pci_addr.u64); | ||
171 | return PCIBIOS_SUCCESSFUL; | ||
172 | } | ||
173 | return PCIBIOS_FUNC_NOT_SUPPORTED; | ||
174 | } | ||
175 | |||
176 | |||
177 | /** | ||
178 | * Write a value to PCI configuration space | ||
179 | * | ||
180 | * @bus: | ||
181 | * @devfn: | ||
182 | * @reg: | ||
183 | * @size: | ||
184 | * @val: | ||
185 | * Returns | ||
186 | */ | ||
187 | static int octeon_write_config(struct pci_bus *bus, unsigned int devfn, | ||
188 | int reg, int size, u32 val) | ||
189 | { | ||
190 | union octeon_pci_address pci_addr; | ||
191 | |||
192 | pci_addr.u64 = 0; | ||
193 | pci_addr.s.upper = 2; | ||
194 | pci_addr.s.io = 1; | ||
195 | pci_addr.s.did = 3; | ||
196 | pci_addr.s.subdid = 1; | ||
197 | pci_addr.s.endian_swap = 1; | ||
198 | pci_addr.s.bus = bus->number; | ||
199 | pci_addr.s.dev = devfn >> 3; | ||
200 | pci_addr.s.func = devfn & 0x7; | ||
201 | pci_addr.s.reg = reg; | ||
202 | |||
203 | #if PCI_CONFIG_SPACE_DELAY | ||
204 | udelay(PCI_CONFIG_SPACE_DELAY); | ||
205 | #endif | ||
206 | switch (size) { | ||
207 | case 4: | ||
208 | cvmx_write64_uint32(pci_addr.u64, cpu_to_le32(val)); | ||
209 | return PCIBIOS_SUCCESSFUL; | ||
210 | case 2: | ||
211 | cvmx_write64_uint16(pci_addr.u64, cpu_to_le16(val)); | ||
212 | return PCIBIOS_SUCCESSFUL; | ||
213 | case 1: | ||
214 | cvmx_write64_uint8(pci_addr.u64, val); | ||
215 | return PCIBIOS_SUCCESSFUL; | ||
216 | } | ||
217 | return PCIBIOS_FUNC_NOT_SUPPORTED; | ||
218 | } | ||
219 | |||
220 | |||
221 | static struct pci_ops octeon_pci_ops = { | ||
222 | octeon_read_config, | ||
223 | octeon_write_config, | ||
224 | }; | ||
225 | |||
226 | static struct resource octeon_pci_mem_resource = { | ||
227 | .start = 0, | ||
228 | .end = 0, | ||
229 | .name = "Octeon PCI MEM", | ||
230 | .flags = IORESOURCE_MEM, | ||
231 | }; | ||
232 | |||
233 | /* | ||
234 | * PCI ports must be above 16KB so the ISA bus filtering in the PCI-X to PCI | ||
235 | * bridge | ||
236 | */ | ||
237 | static struct resource octeon_pci_io_resource = { | ||
238 | .start = 0x4000, | ||
239 | .end = OCTEON_PCI_IOSPACE_SIZE - 1, | ||
240 | .name = "Octeon PCI IO", | ||
241 | .flags = IORESOURCE_IO, | ||
242 | }; | ||
243 | |||
244 | static struct pci_controller octeon_pci_controller = { | ||
245 | .pci_ops = &octeon_pci_ops, | ||
246 | .mem_resource = &octeon_pci_mem_resource, | ||
247 | .mem_offset = OCTEON_PCI_MEMSPACE_OFFSET, | ||
248 | .io_resource = &octeon_pci_io_resource, | ||
249 | .io_offset = 0, | ||
250 | .io_map_base = OCTEON_PCI_IOSPACE_BASE, | ||
251 | }; | ||
252 | |||
253 | |||
254 | /** | ||
255 | * Low level initialize the Octeon PCI controller | ||
256 | * | ||
257 | * Returns | ||
258 | */ | ||
259 | static void octeon_pci_initialize(void) | ||
260 | { | ||
261 | union cvmx_pci_cfg01 cfg01; | ||
262 | union cvmx_npi_ctl_status ctl_status; | ||
263 | union cvmx_pci_ctl_status_2 ctl_status_2; | ||
264 | union cvmx_pci_cfg19 cfg19; | ||
265 | union cvmx_pci_cfg16 cfg16; | ||
266 | union cvmx_pci_cfg22 cfg22; | ||
267 | union cvmx_pci_cfg56 cfg56; | ||
268 | |||
269 | /* Reset the PCI Bus */ | ||
270 | cvmx_write_csr(CVMX_CIU_SOFT_PRST, 0x1); | ||
271 | cvmx_read_csr(CVMX_CIU_SOFT_PRST); | ||
272 | |||
273 | udelay(2000); /* Hold PCI reset for 2 ms */ | ||
274 | |||
275 | ctl_status.u64 = 0; /* cvmx_read_csr(CVMX_NPI_CTL_STATUS); */ | ||
276 | ctl_status.s.max_word = 1; | ||
277 | ctl_status.s.timer = 1; | ||
278 | cvmx_write_csr(CVMX_NPI_CTL_STATUS, ctl_status.u64); | ||
279 | |||
280 | /* Deassert PCI reset and advertize PCX Host Mode Device Capability | ||
281 | (64b) */ | ||
282 | cvmx_write_csr(CVMX_CIU_SOFT_PRST, 0x4); | ||
283 | cvmx_read_csr(CVMX_CIU_SOFT_PRST); | ||
284 | |||
285 | udelay(2000); /* Wait 2 ms after deasserting PCI reset */ | ||
286 | |||
287 | ctl_status_2.u32 = 0; | ||
288 | ctl_status_2.s.tsr_hwm = 1; /* Initializes to 0. Must be set | ||
289 | before any PCI reads. */ | ||
290 | ctl_status_2.s.bar2pres = 1; /* Enable BAR2 */ | ||
291 | ctl_status_2.s.bar2_enb = 1; | ||
292 | ctl_status_2.s.bar2_cax = 1; /* Don't use L2 */ | ||
293 | ctl_status_2.s.bar2_esx = 1; | ||
294 | ctl_status_2.s.pmo_amod = 1; /* Round robin priority */ | ||
295 | if (octeon_dma_bar_type == OCTEON_DMA_BAR_TYPE_BIG) { | ||
296 | /* BAR1 hole */ | ||
297 | ctl_status_2.s.bb1_hole = OCTEON_PCI_BAR1_HOLE_BITS; | ||
298 | ctl_status_2.s.bb1_siz = 1; /* BAR1 is 2GB */ | ||
299 | ctl_status_2.s.bb_ca = 1; /* Don't use L2 with big bars */ | ||
300 | ctl_status_2.s.bb_es = 1; /* Big bar in byte swap mode */ | ||
301 | ctl_status_2.s.bb1 = 1; /* BAR1 is big */ | ||
302 | ctl_status_2.s.bb0 = 1; /* BAR0 is big */ | ||
303 | } | ||
304 | |||
305 | octeon_npi_write32(CVMX_NPI_PCI_CTL_STATUS_2, ctl_status_2.u32); | ||
306 | udelay(2000); /* Wait 2 ms before doing PCI reads */ | ||
307 | |||
308 | ctl_status_2.u32 = octeon_npi_read32(CVMX_NPI_PCI_CTL_STATUS_2); | ||
309 | pr_notice("PCI Status: %s %s-bit\n", | ||
310 | ctl_status_2.s.ap_pcix ? "PCI-X" : "PCI", | ||
311 | ctl_status_2.s.ap_64ad ? "64" : "32"); | ||
312 | |||
313 | if (OCTEON_IS_MODEL(OCTEON_CN58XX) || OCTEON_IS_MODEL(OCTEON_CN50XX)) { | ||
314 | union cvmx_pci_cnt_reg cnt_reg_start; | ||
315 | union cvmx_pci_cnt_reg cnt_reg_end; | ||
316 | unsigned long cycles, pci_clock; | ||
317 | |||
318 | cnt_reg_start.u64 = cvmx_read_csr(CVMX_NPI_PCI_CNT_REG); | ||
319 | cycles = read_c0_cvmcount(); | ||
320 | udelay(1000); | ||
321 | cnt_reg_end.u64 = cvmx_read_csr(CVMX_NPI_PCI_CNT_REG); | ||
322 | cycles = read_c0_cvmcount() - cycles; | ||
323 | pci_clock = (cnt_reg_end.s.pcicnt - cnt_reg_start.s.pcicnt) / | ||
324 | (cycles / (mips_hpt_frequency / 1000000)); | ||
325 | pr_notice("PCI Clock: %lu MHz\n", pci_clock); | ||
326 | } | ||
327 | |||
328 | /* | ||
329 | * TDOMC must be set to one in PCI mode. TDOMC should be set to 4 | ||
330 | * in PCI-X mode to allow four oustanding splits. Otherwise, | ||
331 | * should not change from its reset value. Don't write PCI_CFG19 | ||
332 | * in PCI mode (0x82000001 reset value), write it to 0x82000004 | ||
333 | * after PCI-X mode is known. MRBCI,MDWE,MDRE -> must be zero. | ||
334 | * MRBCM -> must be one. | ||
335 | */ | ||
336 | if (ctl_status_2.s.ap_pcix) { | ||
337 | cfg19.u32 = 0; | ||
338 | /* | ||
339 | * Target Delayed/Split request outstanding maximum | ||
340 | * count. [1..31] and 0=32. NOTE: If the user | ||
341 | * programs these bits beyond the Designed Maximum | ||
342 | * outstanding count, then the designed maximum table | ||
343 | * depth will be used instead. No additional | ||
344 | * Deferred/Split transactions will be accepted if | ||
345 | * this outstanding maximum count is | ||
346 | * reached. Furthermore, no additional deferred/split | ||
347 | * transactions will be accepted if the I/O delay/ I/O | ||
348 | * Split Request outstanding maximum is reached. | ||
349 | */ | ||
350 | cfg19.s.tdomc = 4; | ||
351 | /* | ||
352 | * Master Deferred Read Request Outstanding Max Count | ||
353 | * (PCI only). CR4C[26:24] Max SAC cycles MAX DAC | ||
354 | * cycles 000 8 4 001 1 0 010 2 1 011 3 1 100 4 2 101 | ||
355 | * 5 2 110 6 3 111 7 3 For example, if these bits are | ||
356 | * programmed to 100, the core can support 2 DAC | ||
357 | * cycles, 4 SAC cycles or a combination of 1 DAC and | ||
358 | * 2 SAC cycles. NOTE: For the PCI-X maximum | ||
359 | * outstanding split transactions, refer to | ||
360 | * CRE0[22:20]. | ||
361 | */ | ||
362 | cfg19.s.mdrrmc = 2; | ||
363 | /* | ||
364 | * Master Request (Memory Read) Byte Count/Byte Enable | ||
365 | * select. 0 = Byte Enables valid. In PCI mode, a | ||
366 | * burst transaction cannot be performed using Memory | ||
367 | * Read command=4?h6. 1 = DWORD Byte Count valid | ||
368 | * (default). In PCI Mode, the memory read byte | ||
369 | * enables are automatically generated by the | ||
370 | * core. Note: N3 Master Request transaction sizes are | ||
371 | * always determined through the | ||
372 | * am_attr[<35:32>|<7:0>] field. | ||
373 | */ | ||
374 | cfg19.s.mrbcm = 1; | ||
375 | octeon_npi_write32(CVMX_NPI_PCI_CFG19, cfg19.u32); | ||
376 | } | ||
377 | |||
378 | |||
379 | cfg01.u32 = 0; | ||
380 | cfg01.s.msae = 1; /* Memory Space Access Enable */ | ||
381 | cfg01.s.me = 1; /* Master Enable */ | ||
382 | cfg01.s.pee = 1; /* PERR# Enable */ | ||
383 | cfg01.s.see = 1; /* System Error Enable */ | ||
384 | cfg01.s.fbbe = 1; /* Fast Back to Back Transaction Enable */ | ||
385 | |||
386 | octeon_npi_write32(CVMX_NPI_PCI_CFG01, cfg01.u32); | ||
387 | |||
388 | #ifdef USE_OCTEON_INTERNAL_ARBITER | ||
389 | /* | ||
390 | * When OCTEON is a PCI host, most systems will use OCTEON's | ||
391 | * internal arbiter, so must enable it before any PCI/PCI-X | ||
392 | * traffic can occur. | ||
393 | */ | ||
394 | { | ||
395 | union cvmx_npi_pci_int_arb_cfg pci_int_arb_cfg; | ||
396 | |||
397 | pci_int_arb_cfg.u64 = 0; | ||
398 | pci_int_arb_cfg.s.en = 1; /* Internal arbiter enable */ | ||
399 | cvmx_write_csr(CVMX_NPI_PCI_INT_ARB_CFG, pci_int_arb_cfg.u64); | ||
400 | } | ||
401 | #endif /* USE_OCTEON_INTERNAL_ARBITER */ | ||
402 | |||
403 | /* | ||
404 | * Preferrably written to 1 to set MLTD. [RDSATI,TRTAE, | ||
405 | * TWTAE,TMAE,DPPMR -> must be zero. TILT -> must not be set to | ||
406 | * 1..7. | ||
407 | */ | ||
408 | cfg16.u32 = 0; | ||
409 | cfg16.s.mltd = 1; /* Master Latency Timer Disable */ | ||
410 | octeon_npi_write32(CVMX_NPI_PCI_CFG16, cfg16.u32); | ||
411 | |||
412 | /* | ||
413 | * Should be written to 0x4ff00. MTTV -> must be zero. | ||
414 | * FLUSH -> must be 1. MRV -> should be 0xFF. | ||
415 | */ | ||
416 | cfg22.u32 = 0; | ||
417 | /* Master Retry Value [1..255] and 0=infinite */ | ||
418 | cfg22.s.mrv = 0xff; | ||
419 | /* | ||
420 | * AM_DO_FLUSH_I control NOTE: This bit MUST BE ONE for proper | ||
421 | * N3K operation. | ||
422 | */ | ||
423 | cfg22.s.flush = 1; | ||
424 | octeon_npi_write32(CVMX_NPI_PCI_CFG22, cfg22.u32); | ||
425 | |||
426 | /* | ||
427 | * MOST Indicates the maximum number of outstanding splits (in -1 | ||
428 | * notation) when OCTEON is in PCI-X mode. PCI-X performance is | ||
429 | * affected by the MOST selection. Should generally be written | ||
430 | * with one of 0x3be807, 0x2be807, 0x1be807, or 0x0be807, | ||
431 | * depending on the desired MOST of 3, 2, 1, or 0, respectively. | ||
432 | */ | ||
433 | cfg56.u32 = 0; | ||
434 | cfg56.s.pxcid = 7; /* RO - PCI-X Capability ID */ | ||
435 | cfg56.s.ncp = 0xe8; /* RO - Next Capability Pointer */ | ||
436 | cfg56.s.dpere = 1; /* Data Parity Error Recovery Enable */ | ||
437 | cfg56.s.roe = 1; /* Relaxed Ordering Enable */ | ||
438 | cfg56.s.mmbc = 1; /* Maximum Memory Byte Count | ||
439 | [0=512B,1=1024B,2=2048B,3=4096B] */ | ||
440 | cfg56.s.most = 3; /* Maximum outstanding Split transactions [0=1 | ||
441 | .. 7=32] */ | ||
442 | |||
443 | octeon_npi_write32(CVMX_NPI_PCI_CFG56, cfg56.u32); | ||
444 | |||
445 | /* | ||
446 | * Affects PCI performance when OCTEON services reads to its | ||
447 | * BAR1/BAR2. Refer to Section 10.6.1. The recommended values are | ||
448 | * 0x22, 0x33, and 0x33 for PCI_READ_CMD_6, PCI_READ_CMD_C, and | ||
449 | * PCI_READ_CMD_E, respectively. Unfortunately due to errata DDR-700, | ||
450 | * these values need to be changed so they won't possibly prefetch off | ||
451 | * of the end of memory if PCI is DMAing a buffer at the end of | ||
452 | * memory. Note that these values differ from their reset values. | ||
453 | */ | ||
454 | octeon_npi_write32(CVMX_NPI_PCI_READ_CMD_6, 0x21); | ||
455 | octeon_npi_write32(CVMX_NPI_PCI_READ_CMD_C, 0x31); | ||
456 | octeon_npi_write32(CVMX_NPI_PCI_READ_CMD_E, 0x31); | ||
457 | } | ||
458 | |||
459 | |||
460 | /** | ||
461 | * Initialize the Octeon PCI controller | ||
462 | * | ||
463 | * Returns | ||
464 | */ | ||
465 | static int __init octeon_pci_setup(void) | ||
466 | { | ||
467 | union cvmx_npi_mem_access_subidx mem_access; | ||
468 | int index; | ||
469 | |||
470 | /* Only these chips have PCI */ | ||
471 | if (octeon_has_feature(OCTEON_FEATURE_PCIE)) | ||
472 | return 0; | ||
473 | |||
474 | /* Point pcibios_map_irq() to the PCI version of it */ | ||
475 | octeon_pcibios_map_irq = octeon_pci_pcibios_map_irq; | ||
476 | |||
477 | /* Only use the big bars on chips that support it */ | ||
478 | if (OCTEON_IS_MODEL(OCTEON_CN31XX) || | ||
479 | OCTEON_IS_MODEL(OCTEON_CN38XX_PASS2) || | ||
480 | OCTEON_IS_MODEL(OCTEON_CN38XX_PASS1)) | ||
481 | octeon_dma_bar_type = OCTEON_DMA_BAR_TYPE_SMALL; | ||
482 | else | ||
483 | octeon_dma_bar_type = OCTEON_DMA_BAR_TYPE_BIG; | ||
484 | |||
485 | /* PCI I/O and PCI MEM values */ | ||
486 | set_io_port_base(OCTEON_PCI_IOSPACE_BASE); | ||
487 | ioport_resource.start = 0; | ||
488 | ioport_resource.end = OCTEON_PCI_IOSPACE_SIZE - 1; | ||
489 | if (!octeon_is_pci_host()) { | ||
490 | pr_notice("Not in host mode, PCI Controller not initialized\n"); | ||
491 | return 0; | ||
492 | } | ||
493 | |||
494 | pr_notice("%s Octeon big bar support\n", | ||
495 | (octeon_dma_bar_type == | ||
496 | OCTEON_DMA_BAR_TYPE_BIG) ? "Enabling" : "Disabling"); | ||
497 | |||
498 | octeon_pci_initialize(); | ||
499 | |||
500 | mem_access.u64 = 0; | ||
501 | mem_access.s.esr = 1; /* Endian-Swap on read. */ | ||
502 | mem_access.s.esw = 1; /* Endian-Swap on write. */ | ||
503 | mem_access.s.nsr = 0; /* No-Snoop on read. */ | ||
504 | mem_access.s.nsw = 0; /* No-Snoop on write. */ | ||
505 | mem_access.s.ror = 0; /* Relax Read on read. */ | ||
506 | mem_access.s.row = 0; /* Relax Order on write. */ | ||
507 | mem_access.s.ba = 0; /* PCI Address bits [63:36]. */ | ||
508 | cvmx_write_csr(CVMX_NPI_MEM_ACCESS_SUBID3, mem_access.u64); | ||
509 | |||
510 | /* | ||
511 | * Remap the Octeon BAR 2 above all 32 bit devices | ||
512 | * (0x8000000000ul). This is done here so it is remapped | ||
513 | * before the readl()'s below. We don't want BAR2 overlapping | ||
514 | * with BAR0/BAR1 during these reads. | ||
515 | */ | ||
516 | octeon_npi_write32(CVMX_NPI_PCI_CFG08, 0); | ||
517 | octeon_npi_write32(CVMX_NPI_PCI_CFG09, 0x80); | ||
518 | |||
519 | /* Disable the BAR1 movable mappings */ | ||
520 | for (index = 0; index < 32; index++) | ||
521 | octeon_npi_write32(CVMX_NPI_PCI_BAR1_INDEXX(index), 0); | ||
522 | |||
523 | if (octeon_dma_bar_type == OCTEON_DMA_BAR_TYPE_BIG) { | ||
524 | /* Remap the Octeon BAR 0 to 0-2GB */ | ||
525 | octeon_npi_write32(CVMX_NPI_PCI_CFG04, 0); | ||
526 | octeon_npi_write32(CVMX_NPI_PCI_CFG05, 0); | ||
527 | |||
528 | /* | ||
529 | * Remap the Octeon BAR 1 to map 2GB-4GB (minus the | ||
530 | * BAR 1 hole). | ||
531 | */ | ||
532 | octeon_npi_write32(CVMX_NPI_PCI_CFG06, 2ul << 30); | ||
533 | octeon_npi_write32(CVMX_NPI_PCI_CFG07, 0); | ||
534 | |||
535 | /* Devices go after BAR1 */ | ||
536 | octeon_pci_mem_resource.start = | ||
537 | OCTEON_PCI_MEMSPACE_OFFSET + (4ul << 30) - | ||
538 | (OCTEON_PCI_BAR1_HOLE_SIZE << 20); | ||
539 | octeon_pci_mem_resource.end = | ||
540 | octeon_pci_mem_resource.start + (1ul << 30); | ||
541 | } else { | ||
542 | /* Remap the Octeon BAR 0 to map 128MB-(128MB+4KB) */ | ||
543 | octeon_npi_write32(CVMX_NPI_PCI_CFG04, 128ul << 20); | ||
544 | octeon_npi_write32(CVMX_NPI_PCI_CFG05, 0); | ||
545 | |||
546 | /* Remap the Octeon BAR 1 to map 0-128MB */ | ||
547 | octeon_npi_write32(CVMX_NPI_PCI_CFG06, 0); | ||
548 | octeon_npi_write32(CVMX_NPI_PCI_CFG07, 0); | ||
549 | |||
550 | /* Devices go after BAR0 */ | ||
551 | octeon_pci_mem_resource.start = | ||
552 | OCTEON_PCI_MEMSPACE_OFFSET + (128ul << 20) + | ||
553 | (4ul << 10); | ||
554 | octeon_pci_mem_resource.end = | ||
555 | octeon_pci_mem_resource.start + (1ul << 30); | ||
556 | } | ||
557 | |||
558 | register_pci_controller(&octeon_pci_controller); | ||
559 | |||
560 | /* | ||
561 | * Clear any errors that might be pending from before the bus | ||
562 | * was setup properly. | ||
563 | */ | ||
564 | cvmx_write_csr(CVMX_NPI_PCI_INT_SUM2, -1); | ||
565 | return 0; | ||
566 | } | ||
567 | |||
568 | arch_initcall(octeon_pci_setup); | ||
diff --git a/arch/mips/cavium-octeon/pcie.c b/arch/mips/cavium-octeon/pcie.c new file mode 100644 index 000000000000..49d14081b3b5 --- /dev/null +++ b/arch/mips/cavium-octeon/pcie.c | |||
@@ -0,0 +1,1370 @@ | |||
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, 2008 Cavium Networks | ||
7 | */ | ||
8 | #include <linux/kernel.h> | ||
9 | #include <linux/init.h> | ||
10 | #include <linux/pci.h> | ||
11 | #include <linux/interrupt.h> | ||
12 | #include <linux/time.h> | ||
13 | #include <linux/delay.h> | ||
14 | |||
15 | #include <asm/octeon/octeon.h> | ||
16 | #include <asm/octeon/cvmx-npei-defs.h> | ||
17 | #include <asm/octeon/cvmx-pciercx-defs.h> | ||
18 | #include <asm/octeon/cvmx-pescx-defs.h> | ||
19 | #include <asm/octeon/cvmx-pexp-defs.h> | ||
20 | #include <asm/octeon/cvmx-helper-errata.h> | ||
21 | |||
22 | #include "pci-common.h" | ||
23 | |||
24 | union cvmx_pcie_address { | ||
25 | uint64_t u64; | ||
26 | struct { | ||
27 | uint64_t upper:2; /* Normally 2 for XKPHYS */ | ||
28 | uint64_t reserved_49_61:13; /* Must be zero */ | ||
29 | uint64_t io:1; /* 1 for IO space access */ | ||
30 | uint64_t did:5; /* PCIe DID = 3 */ | ||
31 | uint64_t subdid:3; /* PCIe SubDID = 1 */ | ||
32 | uint64_t reserved_36_39:4; /* Must be zero */ | ||
33 | uint64_t es:2; /* Endian swap = 1 */ | ||
34 | uint64_t port:2; /* PCIe port 0,1 */ | ||
35 | uint64_t reserved_29_31:3; /* Must be zero */ | ||
36 | /* | ||
37 | * Selects the type of the configuration request (0 = type 0, | ||
38 | * 1 = type 1). | ||
39 | */ | ||
40 | uint64_t ty:1; | ||
41 | /* Target bus number sent in the ID in the request. */ | ||
42 | uint64_t bus:8; | ||
43 | /* | ||
44 | * Target device number sent in the ID in the | ||
45 | * request. Note that Dev must be zero for type 0 | ||
46 | * configuration requests. | ||
47 | */ | ||
48 | uint64_t dev:5; | ||
49 | /* Target function number sent in the ID in the request. */ | ||
50 | uint64_t func:3; | ||
51 | /* | ||
52 | * Selects a register in the configuration space of | ||
53 | * the target. | ||
54 | */ | ||
55 | uint64_t reg:12; | ||
56 | } config; | ||
57 | struct { | ||
58 | uint64_t upper:2; /* Normally 2 for XKPHYS */ | ||
59 | uint64_t reserved_49_61:13; /* Must be zero */ | ||
60 | uint64_t io:1; /* 1 for IO space access */ | ||
61 | uint64_t did:5; /* PCIe DID = 3 */ | ||
62 | uint64_t subdid:3; /* PCIe SubDID = 2 */ | ||
63 | uint64_t reserved_36_39:4; /* Must be zero */ | ||
64 | uint64_t es:2; /* Endian swap = 1 */ | ||
65 | uint64_t port:2; /* PCIe port 0,1 */ | ||
66 | uint64_t address:32; /* PCIe IO address */ | ||
67 | } io; | ||
68 | struct { | ||
69 | uint64_t upper:2; /* Normally 2 for XKPHYS */ | ||
70 | uint64_t reserved_49_61:13; /* Must be zero */ | ||
71 | uint64_t io:1; /* 1 for IO space access */ | ||
72 | uint64_t did:5; /* PCIe DID = 3 */ | ||
73 | uint64_t subdid:3; /* PCIe SubDID = 3-6 */ | ||
74 | uint64_t reserved_36_39:4; /* Must be zero */ | ||
75 | uint64_t address:36; /* PCIe Mem address */ | ||
76 | } mem; | ||
77 | }; | ||
78 | |||
79 | /** | ||
80 | * Return the Core virtual base address for PCIe IO access. IOs are | ||
81 | * read/written as an offset from this address. | ||
82 | * | ||
83 | * @pcie_port: PCIe port the IO is for | ||
84 | * | ||
85 | * Returns 64bit Octeon IO base address for read/write | ||
86 | */ | ||
87 | static inline uint64_t cvmx_pcie_get_io_base_address(int pcie_port) | ||
88 | { | ||
89 | union cvmx_pcie_address pcie_addr; | ||
90 | pcie_addr.u64 = 0; | ||
91 | pcie_addr.io.upper = 0; | ||
92 | pcie_addr.io.io = 1; | ||
93 | pcie_addr.io.did = 3; | ||
94 | pcie_addr.io.subdid = 2; | ||
95 | pcie_addr.io.es = 1; | ||
96 | pcie_addr.io.port = pcie_port; | ||
97 | return pcie_addr.u64; | ||
98 | } | ||
99 | |||
100 | /** | ||
101 | * Size of the IO address region returned at address | ||
102 | * cvmx_pcie_get_io_base_address() | ||
103 | * | ||
104 | * @pcie_port: PCIe port the IO is for | ||
105 | * | ||
106 | * Returns Size of the IO window | ||
107 | */ | ||
108 | static inline uint64_t cvmx_pcie_get_io_size(int pcie_port) | ||
109 | { | ||
110 | return 1ull << 32; | ||
111 | } | ||
112 | |||
113 | /** | ||
114 | * Return the Core virtual base address for PCIe MEM access. Memory is | ||
115 | * read/written as an offset from this address. | ||
116 | * | ||
117 | * @pcie_port: PCIe port the IO is for | ||
118 | * | ||
119 | * Returns 64bit Octeon IO base address for read/write | ||
120 | */ | ||
121 | static inline uint64_t cvmx_pcie_get_mem_base_address(int pcie_port) | ||
122 | { | ||
123 | union cvmx_pcie_address pcie_addr; | ||
124 | pcie_addr.u64 = 0; | ||
125 | pcie_addr.mem.upper = 0; | ||
126 | pcie_addr.mem.io = 1; | ||
127 | pcie_addr.mem.did = 3; | ||
128 | pcie_addr.mem.subdid = 3 + pcie_port; | ||
129 | return pcie_addr.u64; | ||
130 | } | ||
131 | |||
132 | /** | ||
133 | * Size of the Mem address region returned at address | ||
134 | * cvmx_pcie_get_mem_base_address() | ||
135 | * | ||
136 | * @pcie_port: PCIe port the IO is for | ||
137 | * | ||
138 | * Returns Size of the Mem window | ||
139 | */ | ||
140 | static inline uint64_t cvmx_pcie_get_mem_size(int pcie_port) | ||
141 | { | ||
142 | return 1ull << 36; | ||
143 | } | ||
144 | |||
145 | /** | ||
146 | * Read a PCIe config space register indirectly. This is used for | ||
147 | * registers of the form PCIEEP_CFG??? and PCIERC?_CFG???. | ||
148 | * | ||
149 | * @pcie_port: PCIe port to read from | ||
150 | * @cfg_offset: Address to read | ||
151 | * | ||
152 | * Returns Value read | ||
153 | */ | ||
154 | static uint32_t cvmx_pcie_cfgx_read(int pcie_port, uint32_t cfg_offset) | ||
155 | { | ||
156 | union cvmx_pescx_cfg_rd pescx_cfg_rd; | ||
157 | pescx_cfg_rd.u64 = 0; | ||
158 | pescx_cfg_rd.s.addr = cfg_offset; | ||
159 | cvmx_write_csr(CVMX_PESCX_CFG_RD(pcie_port), pescx_cfg_rd.u64); | ||
160 | pescx_cfg_rd.u64 = cvmx_read_csr(CVMX_PESCX_CFG_RD(pcie_port)); | ||
161 | return pescx_cfg_rd.s.data; | ||
162 | } | ||
163 | |||
164 | /** | ||
165 | * Write a PCIe config space register indirectly. This is used for | ||
166 | * registers of the form PCIEEP_CFG??? and PCIERC?_CFG???. | ||
167 | * | ||
168 | * @pcie_port: PCIe port to write to | ||
169 | * @cfg_offset: Address to write | ||
170 | * @val: Value to write | ||
171 | */ | ||
172 | static void cvmx_pcie_cfgx_write(int pcie_port, uint32_t cfg_offset, | ||
173 | uint32_t val) | ||
174 | { | ||
175 | union cvmx_pescx_cfg_wr pescx_cfg_wr; | ||
176 | pescx_cfg_wr.u64 = 0; | ||
177 | pescx_cfg_wr.s.addr = cfg_offset; | ||
178 | pescx_cfg_wr.s.data = val; | ||
179 | cvmx_write_csr(CVMX_PESCX_CFG_WR(pcie_port), pescx_cfg_wr.u64); | ||
180 | } | ||
181 | |||
182 | /** | ||
183 | * Build a PCIe config space request address for a device | ||
184 | * | ||
185 | * @pcie_port: PCIe port to access | ||
186 | * @bus: Sub bus | ||
187 | * @dev: Device ID | ||
188 | * @fn: Device sub function | ||
189 | * @reg: Register to access | ||
190 | * | ||
191 | * Returns 64bit Octeon IO address | ||
192 | */ | ||
193 | static inline uint64_t __cvmx_pcie_build_config_addr(int pcie_port, int bus, | ||
194 | int dev, int fn, int reg) | ||
195 | { | ||
196 | union cvmx_pcie_address pcie_addr; | ||
197 | union cvmx_pciercx_cfg006 pciercx_cfg006; | ||
198 | |||
199 | pciercx_cfg006.u32 = | ||
200 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG006(pcie_port)); | ||
201 | if ((bus <= pciercx_cfg006.s.pbnum) && (dev != 0)) | ||
202 | return 0; | ||
203 | |||
204 | pcie_addr.u64 = 0; | ||
205 | pcie_addr.config.upper = 2; | ||
206 | pcie_addr.config.io = 1; | ||
207 | pcie_addr.config.did = 3; | ||
208 | pcie_addr.config.subdid = 1; | ||
209 | pcie_addr.config.es = 1; | ||
210 | pcie_addr.config.port = pcie_port; | ||
211 | pcie_addr.config.ty = (bus > pciercx_cfg006.s.pbnum); | ||
212 | pcie_addr.config.bus = bus; | ||
213 | pcie_addr.config.dev = dev; | ||
214 | pcie_addr.config.func = fn; | ||
215 | pcie_addr.config.reg = reg; | ||
216 | return pcie_addr.u64; | ||
217 | } | ||
218 | |||
219 | /** | ||
220 | * Read 8bits from a Device's config space | ||
221 | * | ||
222 | * @pcie_port: PCIe port the device is on | ||
223 | * @bus: Sub bus | ||
224 | * @dev: Device ID | ||
225 | * @fn: Device sub function | ||
226 | * @reg: Register to access | ||
227 | * | ||
228 | * Returns Result of the read | ||
229 | */ | ||
230 | static uint8_t cvmx_pcie_config_read8(int pcie_port, int bus, int dev, | ||
231 | int fn, int reg) | ||
232 | { | ||
233 | uint64_t address = | ||
234 | __cvmx_pcie_build_config_addr(pcie_port, bus, dev, fn, reg); | ||
235 | if (address) | ||
236 | return cvmx_read64_uint8(address); | ||
237 | else | ||
238 | return 0xff; | ||
239 | } | ||
240 | |||
241 | /** | ||
242 | * Read 16bits from a Device's config space | ||
243 | * | ||
244 | * @pcie_port: PCIe port the device is on | ||
245 | * @bus: Sub bus | ||
246 | * @dev: Device ID | ||
247 | * @fn: Device sub function | ||
248 | * @reg: Register to access | ||
249 | * | ||
250 | * Returns Result of the read | ||
251 | */ | ||
252 | static uint16_t cvmx_pcie_config_read16(int pcie_port, int bus, int dev, | ||
253 | int fn, int reg) | ||
254 | { | ||
255 | uint64_t address = | ||
256 | __cvmx_pcie_build_config_addr(pcie_port, bus, dev, fn, reg); | ||
257 | if (address) | ||
258 | return le16_to_cpu(cvmx_read64_uint16(address)); | ||
259 | else | ||
260 | return 0xffff; | ||
261 | } | ||
262 | |||
263 | /** | ||
264 | * Read 32bits from a Device's config space | ||
265 | * | ||
266 | * @pcie_port: PCIe port the device is on | ||
267 | * @bus: Sub bus | ||
268 | * @dev: Device ID | ||
269 | * @fn: Device sub function | ||
270 | * @reg: Register to access | ||
271 | * | ||
272 | * Returns Result of the read | ||
273 | */ | ||
274 | static uint32_t cvmx_pcie_config_read32(int pcie_port, int bus, int dev, | ||
275 | int fn, int reg) | ||
276 | { | ||
277 | uint64_t address = | ||
278 | __cvmx_pcie_build_config_addr(pcie_port, bus, dev, fn, reg); | ||
279 | if (address) | ||
280 | return le32_to_cpu(cvmx_read64_uint32(address)); | ||
281 | else | ||
282 | return 0xffffffff; | ||
283 | } | ||
284 | |||
285 | /** | ||
286 | * Write 8bits to a Device's config space | ||
287 | * | ||
288 | * @pcie_port: PCIe port the device is on | ||
289 | * @bus: Sub bus | ||
290 | * @dev: Device ID | ||
291 | * @fn: Device sub function | ||
292 | * @reg: Register to access | ||
293 | * @val: Value to write | ||
294 | */ | ||
295 | static void cvmx_pcie_config_write8(int pcie_port, int bus, int dev, int fn, | ||
296 | int reg, uint8_t val) | ||
297 | { | ||
298 | uint64_t address = | ||
299 | __cvmx_pcie_build_config_addr(pcie_port, bus, dev, fn, reg); | ||
300 | if (address) | ||
301 | cvmx_write64_uint8(address, val); | ||
302 | } | ||
303 | |||
304 | /** | ||
305 | * Write 16bits to a Device's config space | ||
306 | * | ||
307 | * @pcie_port: PCIe port the device is on | ||
308 | * @bus: Sub bus | ||
309 | * @dev: Device ID | ||
310 | * @fn: Device sub function | ||
311 | * @reg: Register to access | ||
312 | * @val: Value to write | ||
313 | */ | ||
314 | static void cvmx_pcie_config_write16(int pcie_port, int bus, int dev, int fn, | ||
315 | int reg, uint16_t val) | ||
316 | { | ||
317 | uint64_t address = | ||
318 | __cvmx_pcie_build_config_addr(pcie_port, bus, dev, fn, reg); | ||
319 | if (address) | ||
320 | cvmx_write64_uint16(address, cpu_to_le16(val)); | ||
321 | } | ||
322 | |||
323 | /** | ||
324 | * Write 32bits to a Device's config space | ||
325 | * | ||
326 | * @pcie_port: PCIe port the device is on | ||
327 | * @bus: Sub bus | ||
328 | * @dev: Device ID | ||
329 | * @fn: Device sub function | ||
330 | * @reg: Register to access | ||
331 | * @val: Value to write | ||
332 | */ | ||
333 | static void cvmx_pcie_config_write32(int pcie_port, int bus, int dev, int fn, | ||
334 | int reg, uint32_t val) | ||
335 | { | ||
336 | uint64_t address = | ||
337 | __cvmx_pcie_build_config_addr(pcie_port, bus, dev, fn, reg); | ||
338 | if (address) | ||
339 | cvmx_write64_uint32(address, cpu_to_le32(val)); | ||
340 | } | ||
341 | |||
342 | /** | ||
343 | * Initialize the RC config space CSRs | ||
344 | * | ||
345 | * @pcie_port: PCIe port to initialize | ||
346 | */ | ||
347 | static void __cvmx_pcie_rc_initialize_config_space(int pcie_port) | ||
348 | { | ||
349 | union cvmx_pciercx_cfg030 pciercx_cfg030; | ||
350 | union cvmx_npei_ctl_status2 npei_ctl_status2; | ||
351 | union cvmx_pciercx_cfg070 pciercx_cfg070; | ||
352 | union cvmx_pciercx_cfg001 pciercx_cfg001; | ||
353 | union cvmx_pciercx_cfg032 pciercx_cfg032; | ||
354 | union cvmx_pciercx_cfg006 pciercx_cfg006; | ||
355 | union cvmx_pciercx_cfg008 pciercx_cfg008; | ||
356 | union cvmx_pciercx_cfg009 pciercx_cfg009; | ||
357 | union cvmx_pciercx_cfg010 pciercx_cfg010; | ||
358 | union cvmx_pciercx_cfg011 pciercx_cfg011; | ||
359 | union cvmx_pciercx_cfg035 pciercx_cfg035; | ||
360 | union cvmx_pciercx_cfg075 pciercx_cfg075; | ||
361 | union cvmx_pciercx_cfg034 pciercx_cfg034; | ||
362 | |||
363 | /* Max Payload Size (PCIE*_CFG030[MPS]) */ | ||
364 | /* Max Read Request Size (PCIE*_CFG030[MRRS]) */ | ||
365 | /* Relaxed-order, no-snoop enables (PCIE*_CFG030[RO_EN,NS_EN] */ | ||
366 | /* Error Message Enables (PCIE*_CFG030[CE_EN,NFE_EN,FE_EN,UR_EN]) */ | ||
367 | pciercx_cfg030.u32 = | ||
368 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG030(pcie_port)); | ||
369 | /* | ||
370 | * Max payload size = 128 bytes for best Octeon DMA | ||
371 | * performance. | ||
372 | */ | ||
373 | pciercx_cfg030.s.mps = 0; | ||
374 | /* | ||
375 | * Max read request size = 128 bytes for best Octeon DMA | ||
376 | * performance. | ||
377 | */ | ||
378 | pciercx_cfg030.s.mrrs = 0; | ||
379 | /* Enable relaxed ordering. */ | ||
380 | pciercx_cfg030.s.ro_en = 1; | ||
381 | /* Enable no snoop. */ | ||
382 | pciercx_cfg030.s.ns_en = 1; | ||
383 | /* Correctable error reporting enable. */ | ||
384 | pciercx_cfg030.s.ce_en = 1; | ||
385 | /* Non-fatal error reporting enable. */ | ||
386 | pciercx_cfg030.s.nfe_en = 1; | ||
387 | /* Fatal error reporting enable. */ | ||
388 | pciercx_cfg030.s.fe_en = 1; | ||
389 | /* Unsupported request reporting enable. */ | ||
390 | pciercx_cfg030.s.ur_en = 1; | ||
391 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG030(pcie_port), | ||
392 | pciercx_cfg030.u32); | ||
393 | |||
394 | /* | ||
395 | * Max Payload Size (NPEI_CTL_STATUS2[MPS]) must match | ||
396 | * PCIE*_CFG030[MPS] | ||
397 | * | ||
398 | * Max Read Request Size (NPEI_CTL_STATUS2[MRRS]) must not | ||
399 | * exceed PCIE*_CFG030[MRRS]. | ||
400 | */ | ||
401 | npei_ctl_status2.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_CTL_STATUS2); | ||
402 | /* Max payload size = 128 bytes for best Octeon DMA performance */ | ||
403 | npei_ctl_status2.s.mps = 0; | ||
404 | /* Max read request size = 128 bytes for best Octeon DMA performance */ | ||
405 | npei_ctl_status2.s.mrrs = 0; | ||
406 | cvmx_write_csr(CVMX_PEXP_NPEI_CTL_STATUS2, npei_ctl_status2.u64); | ||
407 | |||
408 | /* ECRC Generation (PCIE*_CFG070[GE,CE]) */ | ||
409 | pciercx_cfg070.u32 = | ||
410 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG070(pcie_port)); | ||
411 | pciercx_cfg070.s.ge = 1; /* ECRC generation enable. */ | ||
412 | pciercx_cfg070.s.ce = 1; /* ECRC check enable. */ | ||
413 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG070(pcie_port), | ||
414 | pciercx_cfg070.u32); | ||
415 | |||
416 | /* | ||
417 | * Access Enables (PCIE*_CFG001[MSAE,ME]) ME and MSAE should | ||
418 | * always be set. | ||
419 | * | ||
420 | * Interrupt Disable (PCIE*_CFG001[I_DIS]) System Error | ||
421 | * Message Enable (PCIE*_CFG001[SEE]) | ||
422 | */ | ||
423 | pciercx_cfg001.u32 = | ||
424 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG001(pcie_port)); | ||
425 | pciercx_cfg001.s.msae = 1; /* Memory space enable. */ | ||
426 | pciercx_cfg001.s.me = 1; /* Bus master enable. */ | ||
427 | pciercx_cfg001.s.i_dis = 1; /* INTx assertion disable. */ | ||
428 | pciercx_cfg001.s.see = 1; /* SERR# enable */ | ||
429 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG001(pcie_port), | ||
430 | pciercx_cfg001.u32); | ||
431 | |||
432 | /* Advanced Error Recovery Message Enables */ | ||
433 | /* (PCIE*_CFG066,PCIE*_CFG067,PCIE*_CFG069) */ | ||
434 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG066(pcie_port), 0); | ||
435 | /* Use CVMX_PCIERCX_CFG067 hardware default */ | ||
436 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG069(pcie_port), 0); | ||
437 | |||
438 | /* Active State Power Management (PCIE*_CFG032[ASLPC]) */ | ||
439 | pciercx_cfg032.u32 = | ||
440 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG032(pcie_port)); | ||
441 | pciercx_cfg032.s.aslpc = 0; /* Active state Link PM control. */ | ||
442 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG032(pcie_port), | ||
443 | pciercx_cfg032.u32); | ||
444 | |||
445 | /* Entrance Latencies (PCIE*_CFG451[L0EL,L1EL]) */ | ||
446 | |||
447 | /* | ||
448 | * Link Width Mode (PCIERCn_CFG452[LME]) - Set during | ||
449 | * cvmx_pcie_rc_initialize_link() | ||
450 | * | ||
451 | * Primary Bus Number (PCIERCn_CFG006[PBNUM]) | ||
452 | * | ||
453 | * We set the primary bus number to 1 so IDT bridges are | ||
454 | * happy. They don't like zero. | ||
455 | */ | ||
456 | pciercx_cfg006.u32 = 0; | ||
457 | pciercx_cfg006.s.pbnum = 1; | ||
458 | pciercx_cfg006.s.sbnum = 1; | ||
459 | pciercx_cfg006.s.subbnum = 1; | ||
460 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG006(pcie_port), | ||
461 | pciercx_cfg006.u32); | ||
462 | |||
463 | /* | ||
464 | * Memory-mapped I/O BAR (PCIERCn_CFG008) | ||
465 | * Most applications should disable the memory-mapped I/O BAR by | ||
466 | * setting PCIERCn_CFG008[ML_ADDR] < PCIERCn_CFG008[MB_ADDR] | ||
467 | */ | ||
468 | pciercx_cfg008.u32 = 0; | ||
469 | pciercx_cfg008.s.mb_addr = 0x100; | ||
470 | pciercx_cfg008.s.ml_addr = 0; | ||
471 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG008(pcie_port), | ||
472 | pciercx_cfg008.u32); | ||
473 | |||
474 | /* | ||
475 | * Prefetchable BAR (PCIERCn_CFG009,PCIERCn_CFG010,PCIERCn_CFG011) | ||
476 | * Most applications should disable the prefetchable BAR by setting | ||
477 | * PCIERCn_CFG011[UMEM_LIMIT],PCIERCn_CFG009[LMEM_LIMIT] < | ||
478 | * PCIERCn_CFG010[UMEM_BASE],PCIERCn_CFG009[LMEM_BASE] | ||
479 | */ | ||
480 | pciercx_cfg009.u32 = | ||
481 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG009(pcie_port)); | ||
482 | pciercx_cfg010.u32 = | ||
483 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG010(pcie_port)); | ||
484 | pciercx_cfg011.u32 = | ||
485 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG011(pcie_port)); | ||
486 | pciercx_cfg009.s.lmem_base = 0x100; | ||
487 | pciercx_cfg009.s.lmem_limit = 0; | ||
488 | pciercx_cfg010.s.umem_base = 0x100; | ||
489 | pciercx_cfg011.s.umem_limit = 0; | ||
490 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG009(pcie_port), | ||
491 | pciercx_cfg009.u32); | ||
492 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG010(pcie_port), | ||
493 | pciercx_cfg010.u32); | ||
494 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG011(pcie_port), | ||
495 | pciercx_cfg011.u32); | ||
496 | |||
497 | /* | ||
498 | * System Error Interrupt Enables (PCIERCn_CFG035[SECEE,SEFEE,SENFEE]) | ||
499 | * PME Interrupt Enables (PCIERCn_CFG035[PMEIE]) | ||
500 | */ | ||
501 | pciercx_cfg035.u32 = | ||
502 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG035(pcie_port)); | ||
503 | /* System error on correctable error enable. */ | ||
504 | pciercx_cfg035.s.secee = 1; | ||
505 | /* System error on fatal error enable. */ | ||
506 | pciercx_cfg035.s.sefee = 1; | ||
507 | /* System error on non-fatal error enable. */ | ||
508 | pciercx_cfg035.s.senfee = 1; | ||
509 | /* PME interrupt enable. */ | ||
510 | pciercx_cfg035.s.pmeie = 1; | ||
511 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG035(pcie_port), | ||
512 | pciercx_cfg035.u32); | ||
513 | |||
514 | /* | ||
515 | * Advanced Error Recovery Interrupt Enables | ||
516 | * (PCIERCn_CFG075[CERE,NFERE,FERE]) | ||
517 | */ | ||
518 | pciercx_cfg075.u32 = | ||
519 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG075(pcie_port)); | ||
520 | /* Correctable error reporting enable. */ | ||
521 | pciercx_cfg075.s.cere = 1; | ||
522 | /* Non-fatal error reporting enable. */ | ||
523 | pciercx_cfg075.s.nfere = 1; | ||
524 | /* Fatal error reporting enable. */ | ||
525 | pciercx_cfg075.s.fere = 1; | ||
526 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG075(pcie_port), | ||
527 | pciercx_cfg075.u32); | ||
528 | |||
529 | /* HP Interrupt Enables (PCIERCn_CFG034[HPINT_EN], | ||
530 | * PCIERCn_CFG034[DLLS_EN,CCINT_EN]) | ||
531 | */ | ||
532 | pciercx_cfg034.u32 = | ||
533 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG034(pcie_port)); | ||
534 | /* Hot-plug interrupt enable. */ | ||
535 | pciercx_cfg034.s.hpint_en = 1; | ||
536 | /* Data Link Layer state changed enable */ | ||
537 | pciercx_cfg034.s.dlls_en = 1; | ||
538 | /* Command completed interrupt enable. */ | ||
539 | pciercx_cfg034.s.ccint_en = 1; | ||
540 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG034(pcie_port), | ||
541 | pciercx_cfg034.u32); | ||
542 | } | ||
543 | |||
544 | /** | ||
545 | * Initialize a host mode PCIe link. This function takes a PCIe | ||
546 | * port from reset to a link up state. Software can then begin | ||
547 | * configuring the rest of the link. | ||
548 | * | ||
549 | * @pcie_port: PCIe port to initialize | ||
550 | * | ||
551 | * Returns Zero on success | ||
552 | */ | ||
553 | static int __cvmx_pcie_rc_initialize_link(int pcie_port) | ||
554 | { | ||
555 | uint64_t start_cycle; | ||
556 | union cvmx_pescx_ctl_status pescx_ctl_status; | ||
557 | union cvmx_pciercx_cfg452 pciercx_cfg452; | ||
558 | union cvmx_pciercx_cfg032 pciercx_cfg032; | ||
559 | union cvmx_pciercx_cfg448 pciercx_cfg448; | ||
560 | |||
561 | /* Set the lane width */ | ||
562 | pciercx_cfg452.u32 = | ||
563 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG452(pcie_port)); | ||
564 | pescx_ctl_status.u64 = cvmx_read_csr(CVMX_PESCX_CTL_STATUS(pcie_port)); | ||
565 | if (pescx_ctl_status.s.qlm_cfg == 0) { | ||
566 | /* We're in 8 lane (56XX) or 4 lane (54XX) mode */ | ||
567 | pciercx_cfg452.s.lme = 0xf; | ||
568 | } else { | ||
569 | /* We're in 4 lane (56XX) or 2 lane (52XX) mode */ | ||
570 | pciercx_cfg452.s.lme = 0x7; | ||
571 | } | ||
572 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG452(pcie_port), | ||
573 | pciercx_cfg452.u32); | ||
574 | |||
575 | /* | ||
576 | * CN52XX pass 1.x has an errata where length mismatches on UR | ||
577 | * responses can cause bus errors on 64bit memory | ||
578 | * reads. Turning off length error checking fixes this. | ||
579 | */ | ||
580 | if (OCTEON_IS_MODEL(OCTEON_CN52XX_PASS1_X)) { | ||
581 | union cvmx_pciercx_cfg455 pciercx_cfg455; | ||
582 | pciercx_cfg455.u32 = | ||
583 | cvmx_pcie_cfgx_read(pcie_port, | ||
584 | CVMX_PCIERCX_CFG455(pcie_port)); | ||
585 | pciercx_cfg455.s.m_cpl_len_err = 1; | ||
586 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG455(pcie_port), | ||
587 | pciercx_cfg455.u32); | ||
588 | } | ||
589 | |||
590 | /* Lane swap needs to be manually enabled for CN52XX */ | ||
591 | if (OCTEON_IS_MODEL(OCTEON_CN52XX) && (pcie_port == 1)) { | ||
592 | pescx_ctl_status.s.lane_swp = 1; | ||
593 | cvmx_write_csr(CVMX_PESCX_CTL_STATUS(pcie_port), | ||
594 | pescx_ctl_status.u64); | ||
595 | } | ||
596 | |||
597 | /* Bring up the link */ | ||
598 | pescx_ctl_status.u64 = cvmx_read_csr(CVMX_PESCX_CTL_STATUS(pcie_port)); | ||
599 | pescx_ctl_status.s.lnk_enb = 1; | ||
600 | cvmx_write_csr(CVMX_PESCX_CTL_STATUS(pcie_port), pescx_ctl_status.u64); | ||
601 | |||
602 | /* | ||
603 | * CN52XX pass 1.0: Due to a bug in 2nd order CDR, it needs to | ||
604 | * be disabled. | ||
605 | */ | ||
606 | if (OCTEON_IS_MODEL(OCTEON_CN52XX_PASS1_0)) | ||
607 | __cvmx_helper_errata_qlm_disable_2nd_order_cdr(0); | ||
608 | |||
609 | /* Wait for the link to come up */ | ||
610 | cvmx_dprintf("PCIe: Waiting for port %d link\n", pcie_port); | ||
611 | start_cycle = cvmx_get_cycle(); | ||
612 | do { | ||
613 | if (cvmx_get_cycle() - start_cycle > | ||
614 | 2 * cvmx_sysinfo_get()->cpu_clock_hz) { | ||
615 | cvmx_dprintf("PCIe: Port %d link timeout\n", | ||
616 | pcie_port); | ||
617 | return -1; | ||
618 | } | ||
619 | cvmx_wait(10000); | ||
620 | pciercx_cfg032.u32 = | ||
621 | cvmx_pcie_cfgx_read(pcie_port, | ||
622 | CVMX_PCIERCX_CFG032(pcie_port)); | ||
623 | } while (pciercx_cfg032.s.dlla == 0); | ||
624 | |||
625 | /* Display the link status */ | ||
626 | cvmx_dprintf("PCIe: Port %d link active, %d lanes\n", pcie_port, | ||
627 | pciercx_cfg032.s.nlw); | ||
628 | |||
629 | /* | ||
630 | * Update the Replay Time Limit. Empirically, some PCIe | ||
631 | * devices take a little longer to respond than expected under | ||
632 | * load. As a workaround for this we configure the Replay Time | ||
633 | * Limit to the value expected for a 512 byte MPS instead of | ||
634 | * our actual 256 byte MPS. The numbers below are directly | ||
635 | * from the PCIe spec table 3-4. | ||
636 | */ | ||
637 | pciercx_cfg448.u32 = | ||
638 | cvmx_pcie_cfgx_read(pcie_port, CVMX_PCIERCX_CFG448(pcie_port)); | ||
639 | switch (pciercx_cfg032.s.nlw) { | ||
640 | case 1: /* 1 lane */ | ||
641 | pciercx_cfg448.s.rtl = 1677; | ||
642 | break; | ||
643 | case 2: /* 2 lanes */ | ||
644 | pciercx_cfg448.s.rtl = 867; | ||
645 | break; | ||
646 | case 4: /* 4 lanes */ | ||
647 | pciercx_cfg448.s.rtl = 462; | ||
648 | break; | ||
649 | case 8: /* 8 lanes */ | ||
650 | pciercx_cfg448.s.rtl = 258; | ||
651 | break; | ||
652 | } | ||
653 | cvmx_pcie_cfgx_write(pcie_port, CVMX_PCIERCX_CFG448(pcie_port), | ||
654 | pciercx_cfg448.u32); | ||
655 | |||
656 | return 0; | ||
657 | } | ||
658 | |||
659 | /** | ||
660 | * Initialize a PCIe port for use in host(RC) mode. It doesn't | ||
661 | * enumerate the bus. | ||
662 | * | ||
663 | * @pcie_port: PCIe port to initialize | ||
664 | * | ||
665 | * Returns Zero on success | ||
666 | */ | ||
667 | static int cvmx_pcie_rc_initialize(int pcie_port) | ||
668 | { | ||
669 | int i; | ||
670 | union cvmx_ciu_soft_prst ciu_soft_prst; | ||
671 | union cvmx_pescx_bist_status pescx_bist_status; | ||
672 | union cvmx_pescx_bist_status2 pescx_bist_status2; | ||
673 | union cvmx_npei_ctl_status npei_ctl_status; | ||
674 | union cvmx_npei_mem_access_ctl npei_mem_access_ctl; | ||
675 | union cvmx_npei_mem_access_subidx mem_access_subid; | ||
676 | union cvmx_npei_dbg_data npei_dbg_data; | ||
677 | union cvmx_pescx_ctl_status2 pescx_ctl_status2; | ||
678 | |||
679 | /* | ||
680 | * Make sure we aren't trying to setup a target mode interface | ||
681 | * in host mode. | ||
682 | */ | ||
683 | npei_ctl_status.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_CTL_STATUS); | ||
684 | if ((pcie_port == 0) && !npei_ctl_status.s.host_mode) { | ||
685 | cvmx_dprintf("PCIe: ERROR: cvmx_pcie_rc_initialize() called " | ||
686 | "on port0, but port0 is not in host mode\n"); | ||
687 | return -1; | ||
688 | } | ||
689 | |||
690 | /* | ||
691 | * Make sure a CN52XX isn't trying to bring up port 1 when it | ||
692 | * is disabled. | ||
693 | */ | ||
694 | if (OCTEON_IS_MODEL(OCTEON_CN52XX)) { | ||
695 | npei_dbg_data.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_DBG_DATA); | ||
696 | if ((pcie_port == 1) && npei_dbg_data.cn52xx.qlm0_link_width) { | ||
697 | cvmx_dprintf("PCIe: ERROR: cvmx_pcie_rc_initialize() " | ||
698 | "called on port1, but port1 is " | ||
699 | "disabled\n"); | ||
700 | return -1; | ||
701 | } | ||
702 | } | ||
703 | |||
704 | /* | ||
705 | * PCIe switch arbitration mode. '0' == fixed priority NPEI, | ||
706 | * PCIe0, then PCIe1. '1' == round robin. | ||
707 | */ | ||
708 | npei_ctl_status.s.arb = 1; | ||
709 | /* Allow up to 0x20 config retries */ | ||
710 | npei_ctl_status.s.cfg_rtry = 0x20; | ||
711 | /* | ||
712 | * CN52XX pass1.x has an errata where P0_NTAGS and P1_NTAGS | ||
713 | * don't reset. | ||
714 | */ | ||
715 | if (OCTEON_IS_MODEL(OCTEON_CN52XX_PASS1_X)) { | ||
716 | npei_ctl_status.s.p0_ntags = 0x20; | ||
717 | npei_ctl_status.s.p1_ntags = 0x20; | ||
718 | } | ||
719 | cvmx_write_csr(CVMX_PEXP_NPEI_CTL_STATUS, npei_ctl_status.u64); | ||
720 | |||
721 | /* Bring the PCIe out of reset */ | ||
722 | if (cvmx_sysinfo_get()->board_type == CVMX_BOARD_TYPE_EBH5200) { | ||
723 | /* | ||
724 | * The EBH5200 board swapped the PCIe reset lines on | ||
725 | * the board. As a workaround for this bug, we bring | ||
726 | * both PCIe ports out of reset at the same time | ||
727 | * instead of on separate calls. So for port 0, we | ||
728 | * bring both out of reset and do nothing on port 1. | ||
729 | */ | ||
730 | if (pcie_port == 0) { | ||
731 | ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST); | ||
732 | /* | ||
733 | * After a chip reset the PCIe will also be in | ||
734 | * reset. If it isn't, most likely someone is | ||
735 | * trying to init it again without a proper | ||
736 | * PCIe reset. | ||
737 | */ | ||
738 | if (ciu_soft_prst.s.soft_prst == 0) { | ||
739 | /* Reset the ports */ | ||
740 | ciu_soft_prst.s.soft_prst = 1; | ||
741 | cvmx_write_csr(CVMX_CIU_SOFT_PRST, | ||
742 | ciu_soft_prst.u64); | ||
743 | ciu_soft_prst.u64 = | ||
744 | cvmx_read_csr(CVMX_CIU_SOFT_PRST1); | ||
745 | ciu_soft_prst.s.soft_prst = 1; | ||
746 | cvmx_write_csr(CVMX_CIU_SOFT_PRST1, | ||
747 | ciu_soft_prst.u64); | ||
748 | /* Wait until pcie resets the ports. */ | ||
749 | udelay(2000); | ||
750 | } | ||
751 | ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST1); | ||
752 | ciu_soft_prst.s.soft_prst = 0; | ||
753 | cvmx_write_csr(CVMX_CIU_SOFT_PRST1, ciu_soft_prst.u64); | ||
754 | ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST); | ||
755 | ciu_soft_prst.s.soft_prst = 0; | ||
756 | cvmx_write_csr(CVMX_CIU_SOFT_PRST, ciu_soft_prst.u64); | ||
757 | } | ||
758 | } else { | ||
759 | /* | ||
760 | * The normal case: The PCIe ports are completely | ||
761 | * separate and can be brought out of reset | ||
762 | * independently. | ||
763 | */ | ||
764 | if (pcie_port) | ||
765 | ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST1); | ||
766 | else | ||
767 | ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST); | ||
768 | /* | ||
769 | * After a chip reset the PCIe will also be in | ||
770 | * reset. If it isn't, most likely someone is trying | ||
771 | * to init it again without a proper PCIe reset. | ||
772 | */ | ||
773 | if (ciu_soft_prst.s.soft_prst == 0) { | ||
774 | /* Reset the port */ | ||
775 | ciu_soft_prst.s.soft_prst = 1; | ||
776 | if (pcie_port) | ||
777 | cvmx_write_csr(CVMX_CIU_SOFT_PRST1, | ||
778 | ciu_soft_prst.u64); | ||
779 | else | ||
780 | cvmx_write_csr(CVMX_CIU_SOFT_PRST, | ||
781 | ciu_soft_prst.u64); | ||
782 | /* Wait until pcie resets the ports. */ | ||
783 | udelay(2000); | ||
784 | } | ||
785 | if (pcie_port) { | ||
786 | ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST1); | ||
787 | ciu_soft_prst.s.soft_prst = 0; | ||
788 | cvmx_write_csr(CVMX_CIU_SOFT_PRST1, ciu_soft_prst.u64); | ||
789 | } else { | ||
790 | ciu_soft_prst.u64 = cvmx_read_csr(CVMX_CIU_SOFT_PRST); | ||
791 | ciu_soft_prst.s.soft_prst = 0; | ||
792 | cvmx_write_csr(CVMX_CIU_SOFT_PRST, ciu_soft_prst.u64); | ||
793 | } | ||
794 | } | ||
795 | |||
796 | /* | ||
797 | * Wait for PCIe reset to complete. Due to errata PCIE-700, we | ||
798 | * don't poll PESCX_CTL_STATUS2[PCIERST], but simply wait a | ||
799 | * fixed number of cycles. | ||
800 | */ | ||
801 | cvmx_wait(400000); | ||
802 | |||
803 | /* PESCX_BIST_STATUS2[PCLK_RUN] was missing on pass 1 of CN56XX and | ||
804 | CN52XX, so we only probe it on newer chips */ | ||
805 | if (!OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1_X) | ||
806 | && !OCTEON_IS_MODEL(OCTEON_CN52XX_PASS1_X)) { | ||
807 | /* Clear PCLK_RUN so we can check if the clock is running */ | ||
808 | pescx_ctl_status2.u64 = | ||
809 | cvmx_read_csr(CVMX_PESCX_CTL_STATUS2(pcie_port)); | ||
810 | pescx_ctl_status2.s.pclk_run = 1; | ||
811 | cvmx_write_csr(CVMX_PESCX_CTL_STATUS2(pcie_port), | ||
812 | pescx_ctl_status2.u64); | ||
813 | /* | ||
814 | * Now that we cleared PCLK_RUN, wait for it to be set | ||
815 | * again telling us the clock is running. | ||
816 | */ | ||
817 | if (CVMX_WAIT_FOR_FIELD64(CVMX_PESCX_CTL_STATUS2(pcie_port), | ||
818 | union cvmx_pescx_ctl_status2, | ||
819 | pclk_run, ==, 1, 10000)) { | ||
820 | cvmx_dprintf("PCIe: Port %d isn't clocked, skipping.\n", | ||
821 | pcie_port); | ||
822 | return -1; | ||
823 | } | ||
824 | } | ||
825 | |||
826 | /* | ||
827 | * Check and make sure PCIe came out of reset. If it doesn't | ||
828 | * the board probably hasn't wired the clocks up and the | ||
829 | * interface should be skipped. | ||
830 | */ | ||
831 | pescx_ctl_status2.u64 = | ||
832 | cvmx_read_csr(CVMX_PESCX_CTL_STATUS2(pcie_port)); | ||
833 | if (pescx_ctl_status2.s.pcierst) { | ||
834 | cvmx_dprintf("PCIe: Port %d stuck in reset, skipping.\n", | ||
835 | pcie_port); | ||
836 | return -1; | ||
837 | } | ||
838 | |||
839 | /* | ||
840 | * Check BIST2 status. If any bits are set skip this interface. This | ||
841 | * is an attempt to catch PCIE-813 on pass 1 parts. | ||
842 | */ | ||
843 | pescx_bist_status2.u64 = | ||
844 | cvmx_read_csr(CVMX_PESCX_BIST_STATUS2(pcie_port)); | ||
845 | if (pescx_bist_status2.u64) { | ||
846 | cvmx_dprintf("PCIe: Port %d BIST2 failed. Most likely this " | ||
847 | "port isn't hooked up, skipping.\n", | ||
848 | pcie_port); | ||
849 | return -1; | ||
850 | } | ||
851 | |||
852 | /* Check BIST status */ | ||
853 | pescx_bist_status.u64 = | ||
854 | cvmx_read_csr(CVMX_PESCX_BIST_STATUS(pcie_port)); | ||
855 | if (pescx_bist_status.u64) | ||
856 | cvmx_dprintf("PCIe: BIST FAILED for port %d (0x%016llx)\n", | ||
857 | pcie_port, CAST64(pescx_bist_status.u64)); | ||
858 | |||
859 | /* Initialize the config space CSRs */ | ||
860 | __cvmx_pcie_rc_initialize_config_space(pcie_port); | ||
861 | |||
862 | /* Bring the link up */ | ||
863 | if (__cvmx_pcie_rc_initialize_link(pcie_port)) { | ||
864 | cvmx_dprintf | ||
865 | ("PCIe: ERROR: cvmx_pcie_rc_initialize_link() failed\n"); | ||
866 | return -1; | ||
867 | } | ||
868 | |||
869 | /* Store merge control (NPEI_MEM_ACCESS_CTL[TIMER,MAX_WORD]) */ | ||
870 | npei_mem_access_ctl.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_MEM_ACCESS_CTL); | ||
871 | /* Allow 16 words to combine */ | ||
872 | npei_mem_access_ctl.s.max_word = 0; | ||
873 | /* Wait up to 127 cycles for more data */ | ||
874 | npei_mem_access_ctl.s.timer = 127; | ||
875 | cvmx_write_csr(CVMX_PEXP_NPEI_MEM_ACCESS_CTL, npei_mem_access_ctl.u64); | ||
876 | |||
877 | /* Setup Mem access SubDIDs */ | ||
878 | mem_access_subid.u64 = 0; | ||
879 | /* Port the request is sent to. */ | ||
880 | mem_access_subid.s.port = pcie_port; | ||
881 | /* Due to an errata on pass 1 chips, no merging is allowed. */ | ||
882 | mem_access_subid.s.nmerge = 1; | ||
883 | /* Endian-swap for Reads. */ | ||
884 | mem_access_subid.s.esr = 1; | ||
885 | /* Endian-swap for Writes. */ | ||
886 | mem_access_subid.s.esw = 1; | ||
887 | /* No Snoop for Reads. */ | ||
888 | mem_access_subid.s.nsr = 1; | ||
889 | /* No Snoop for Writes. */ | ||
890 | mem_access_subid.s.nsw = 1; | ||
891 | /* Disable Relaxed Ordering for Reads. */ | ||
892 | mem_access_subid.s.ror = 0; | ||
893 | /* Disable Relaxed Ordering for Writes. */ | ||
894 | mem_access_subid.s.row = 0; | ||
895 | /* PCIe Adddress Bits <63:34>. */ | ||
896 | mem_access_subid.s.ba = 0; | ||
897 | |||
898 | /* | ||
899 | * Setup mem access 12-15 for port 0, 16-19 for port 1, | ||
900 | * supplying 36 bits of address space. | ||
901 | */ | ||
902 | for (i = 12 + pcie_port * 4; i < 16 + pcie_port * 4; i++) { | ||
903 | cvmx_write_csr(CVMX_PEXP_NPEI_MEM_ACCESS_SUBIDX(i), | ||
904 | mem_access_subid.u64); | ||
905 | /* Set each SUBID to extend the addressable range */ | ||
906 | mem_access_subid.s.ba += 1; | ||
907 | } | ||
908 | |||
909 | /* | ||
910 | * Disable the peer to peer forwarding register. This must be | ||
911 | * setup by the OS after it enumerates the bus and assigns | ||
912 | * addresses to the PCIe busses. | ||
913 | */ | ||
914 | for (i = 0; i < 4; i++) { | ||
915 | cvmx_write_csr(CVMX_PESCX_P2P_BARX_START(i, pcie_port), -1); | ||
916 | cvmx_write_csr(CVMX_PESCX_P2P_BARX_END(i, pcie_port), -1); | ||
917 | } | ||
918 | |||
919 | /* Set Octeon's BAR0 to decode 0-16KB. It overlaps with Bar2 */ | ||
920 | cvmx_write_csr(CVMX_PESCX_P2N_BAR0_START(pcie_port), 0); | ||
921 | |||
922 | /* | ||
923 | * Disable Octeon's BAR1. It isn't needed in RC mode since | ||
924 | * BAR2 maps all of memory. BAR2 also maps 256MB-512MB into | ||
925 | * the 2nd 256MB of memory. | ||
926 | */ | ||
927 | cvmx_write_csr(CVMX_PESCX_P2N_BAR1_START(pcie_port), -1); | ||
928 | |||
929 | /* | ||
930 | * Set Octeon's BAR2 to decode 0-2^39. Bar0 and Bar1 take | ||
931 | * precedence where they overlap. It also overlaps with the | ||
932 | * device addresses, so make sure the peer to peer forwarding | ||
933 | * is set right. | ||
934 | */ | ||
935 | cvmx_write_csr(CVMX_PESCX_P2N_BAR2_START(pcie_port), 0); | ||
936 | |||
937 | /* | ||
938 | * Setup BAR2 attributes | ||
939 | * | ||
940 | * Relaxed Ordering (NPEI_CTL_PORTn[PTLP_RO,CTLP_RO, WAIT_COM]) | ||
941 | * - PTLP_RO,CTLP_RO should normally be set (except for debug). | ||
942 | * - WAIT_COM=0 will likely work for all applications. | ||
943 | * | ||
944 | * Load completion relaxed ordering (NPEI_CTL_PORTn[WAITL_COM]). | ||
945 | */ | ||
946 | if (pcie_port) { | ||
947 | union cvmx_npei_ctl_port1 npei_ctl_port; | ||
948 | npei_ctl_port.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_CTL_PORT1); | ||
949 | npei_ctl_port.s.bar2_enb = 1; | ||
950 | npei_ctl_port.s.bar2_esx = 1; | ||
951 | npei_ctl_port.s.bar2_cax = 0; | ||
952 | npei_ctl_port.s.ptlp_ro = 1; | ||
953 | npei_ctl_port.s.ctlp_ro = 1; | ||
954 | npei_ctl_port.s.wait_com = 0; | ||
955 | npei_ctl_port.s.waitl_com = 0; | ||
956 | cvmx_write_csr(CVMX_PEXP_NPEI_CTL_PORT1, npei_ctl_port.u64); | ||
957 | } else { | ||
958 | union cvmx_npei_ctl_port0 npei_ctl_port; | ||
959 | npei_ctl_port.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_CTL_PORT0); | ||
960 | npei_ctl_port.s.bar2_enb = 1; | ||
961 | npei_ctl_port.s.bar2_esx = 1; | ||
962 | npei_ctl_port.s.bar2_cax = 0; | ||
963 | npei_ctl_port.s.ptlp_ro = 1; | ||
964 | npei_ctl_port.s.ctlp_ro = 1; | ||
965 | npei_ctl_port.s.wait_com = 0; | ||
966 | npei_ctl_port.s.waitl_com = 0; | ||
967 | cvmx_write_csr(CVMX_PEXP_NPEI_CTL_PORT0, npei_ctl_port.u64); | ||
968 | } | ||
969 | return 0; | ||
970 | } | ||
971 | |||
972 | |||
973 | /* Above was cvmx-pcie.c, below original pcie.c */ | ||
974 | |||
975 | |||
976 | /** | ||
977 | * Map a PCI device to the appropriate interrupt line | ||
978 | * | ||
979 | * @param dev The Linux PCI device structure for the device to map | ||
980 | * @param slot The slot number for this device on __BUS 0__. Linux | ||
981 | * enumerates through all the bridges and figures out the | ||
982 | * slot on Bus 0 where this device eventually hooks to. | ||
983 | * @param pin The PCI interrupt pin read from the device, then swizzled | ||
984 | * as it goes through each bridge. | ||
985 | * @return Interrupt number for the device | ||
986 | */ | ||
987 | int __init octeon_pcie_pcibios_map_irq(const struct pci_dev *dev, | ||
988 | u8 slot, u8 pin) | ||
989 | { | ||
990 | /* | ||
991 | * The EBH5600 board with the PCI to PCIe bridge mistakenly | ||
992 | * wires the first slot for both device id 2 and interrupt | ||
993 | * A. According to the PCI spec, device id 2 should be C. The | ||
994 | * following kludge attempts to fix this. | ||
995 | */ | ||
996 | if (strstr(octeon_board_type_string(), "EBH5600") && | ||
997 | dev->bus && dev->bus->parent) { | ||
998 | /* | ||
999 | * Iterate all the way up the device chain and find | ||
1000 | * the root bus. | ||
1001 | */ | ||
1002 | while (dev->bus && dev->bus->parent) | ||
1003 | dev = to_pci_dev(dev->bus->bridge); | ||
1004 | /* If the root bus is number 0 and the PEX 8114 is the | ||
1005 | * root, assume we are behind the miswired bus. We | ||
1006 | * need to correct the swizzle level by two. Yuck. | ||
1007 | */ | ||
1008 | if ((dev->bus->number == 0) && | ||
1009 | (dev->vendor == 0x10b5) && (dev->device == 0x8114)) { | ||
1010 | /* | ||
1011 | * The pin field is one based, not zero. We | ||
1012 | * need to swizzle it by minus two. | ||
1013 | */ | ||
1014 | pin = ((pin - 3) & 3) + 1; | ||
1015 | } | ||
1016 | } | ||
1017 | /* | ||
1018 | * The -1 is because pin starts with one, not zero. It might | ||
1019 | * be that this equation needs to include the slot number, but | ||
1020 | * I don't have hardware to check that against. | ||
1021 | */ | ||
1022 | return pin - 1 + OCTEON_IRQ_PCI_INT0; | ||
1023 | } | ||
1024 | |||
1025 | /** | ||
1026 | * Read a value from configuration space | ||
1027 | * | ||
1028 | * @param bus | ||
1029 | * @param devfn | ||
1030 | * @param reg | ||
1031 | * @param size | ||
1032 | * @param val | ||
1033 | * @return | ||
1034 | */ | ||
1035 | static inline int octeon_pcie_read_config(int pcie_port, struct pci_bus *bus, | ||
1036 | unsigned int devfn, int reg, int size, | ||
1037 | u32 *val) | ||
1038 | { | ||
1039 | union octeon_cvmemctl cvmmemctl; | ||
1040 | union octeon_cvmemctl cvmmemctl_save; | ||
1041 | int bus_number = bus->number; | ||
1042 | |||
1043 | /* | ||
1044 | * We need to force the bus number to be zero on the root | ||
1045 | * bus. Linux numbers the 2nd root bus to start after all | ||
1046 | * buses on root 0. | ||
1047 | */ | ||
1048 | if (bus->parent == NULL) | ||
1049 | bus_number = 0; | ||
1050 | |||
1051 | /* | ||
1052 | * PCIe only has a single device connected to Octeon. It is | ||
1053 | * always device ID 0. Don't bother doing reads for other | ||
1054 | * device IDs on the first segment. | ||
1055 | */ | ||
1056 | if ((bus_number == 0) && (devfn >> 3 != 0)) | ||
1057 | return PCIBIOS_FUNC_NOT_SUPPORTED; | ||
1058 | |||
1059 | /* | ||
1060 | * The following is a workaround for the CN57XX, CN56XX, | ||
1061 | * CN55XX, and CN54XX errata with PCIe config reads from non | ||
1062 | * existent devices. These chips will hang the PCIe link if a | ||
1063 | * config read is performed that causes a UR response. | ||
1064 | */ | ||
1065 | if (OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1) || | ||
1066 | OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1_1)) { | ||
1067 | /* | ||
1068 | * For our EBH5600 board, port 0 has a bridge with two | ||
1069 | * PCI-X slots. We need a new special checks to make | ||
1070 | * sure we only probe valid stuff. The PCIe->PCI-X | ||
1071 | * bridge only respondes to device ID 0, function | ||
1072 | * 0-1 | ||
1073 | */ | ||
1074 | if ((bus_number == 0) && (devfn >= 2)) | ||
1075 | return PCIBIOS_FUNC_NOT_SUPPORTED; | ||
1076 | /* | ||
1077 | * The PCI-X slots are device ID 2,3. Choose one of | ||
1078 | * the below "if" blocks based on what is plugged into | ||
1079 | * the board. | ||
1080 | */ | ||
1081 | #if 1 | ||
1082 | /* Use this option if you aren't using either slot */ | ||
1083 | if (bus_number == 1) | ||
1084 | return PCIBIOS_FUNC_NOT_SUPPORTED; | ||
1085 | #elif 0 | ||
1086 | /* | ||
1087 | * Use this option if you are using the first slot but | ||
1088 | * not the second. | ||
1089 | */ | ||
1090 | if ((bus_number == 1) && (devfn >> 3 != 2)) | ||
1091 | return PCIBIOS_FUNC_NOT_SUPPORTED; | ||
1092 | #elif 0 | ||
1093 | /* | ||
1094 | * Use this option if you are using the second slot | ||
1095 | * but not the first. | ||
1096 | */ | ||
1097 | if ((bus_number == 1) && (devfn >> 3 != 3)) | ||
1098 | return PCIBIOS_FUNC_NOT_SUPPORTED; | ||
1099 | #elif 0 | ||
1100 | /* Use this opion if you are using both slots */ | ||
1101 | if ((bus_number == 1) && | ||
1102 | !((devfn == (2 << 3)) || (devfn == (3 << 3)))) | ||
1103 | return PCIBIOS_FUNC_NOT_SUPPORTED; | ||
1104 | #endif | ||
1105 | |||
1106 | /* | ||
1107 | * Shorten the DID timeout so bus errors for PCIe | ||
1108 | * config reads from non existent devices happen | ||
1109 | * faster. This allows us to continue booting even if | ||
1110 | * the above "if" checks are wrong. Once one of these | ||
1111 | * errors happens, the PCIe port is dead. | ||
1112 | */ | ||
1113 | cvmmemctl_save.u64 = __read_64bit_c0_register($11, 7); | ||
1114 | cvmmemctl.u64 = cvmmemctl_save.u64; | ||
1115 | cvmmemctl.s.didtto = 2; | ||
1116 | __write_64bit_c0_register($11, 7, cvmmemctl.u64); | ||
1117 | } | ||
1118 | |||
1119 | switch (size) { | ||
1120 | case 4: | ||
1121 | *val = cvmx_pcie_config_read32(pcie_port, bus_number, | ||
1122 | devfn >> 3, devfn & 0x7, reg); | ||
1123 | break; | ||
1124 | case 2: | ||
1125 | *val = cvmx_pcie_config_read16(pcie_port, bus_number, | ||
1126 | devfn >> 3, devfn & 0x7, reg); | ||
1127 | break; | ||
1128 | case 1: | ||
1129 | *val = cvmx_pcie_config_read8(pcie_port, bus_number, devfn >> 3, | ||
1130 | devfn & 0x7, reg); | ||
1131 | break; | ||
1132 | default: | ||
1133 | return PCIBIOS_FUNC_NOT_SUPPORTED; | ||
1134 | } | ||
1135 | |||
1136 | if (OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1) || | ||
1137 | OCTEON_IS_MODEL(OCTEON_CN56XX_PASS1_1)) | ||
1138 | __write_64bit_c0_register($11, 7, cvmmemctl_save.u64); | ||
1139 | return PCIBIOS_SUCCESSFUL; | ||
1140 | } | ||
1141 | |||
1142 | static int octeon_pcie0_read_config(struct pci_bus *bus, unsigned int devfn, | ||
1143 | int reg, int size, u32 *val) | ||
1144 | { | ||
1145 | return octeon_pcie_read_config(0, bus, devfn, reg, size, val); | ||
1146 | } | ||
1147 | |||
1148 | static int octeon_pcie1_read_config(struct pci_bus *bus, unsigned int devfn, | ||
1149 | int reg, int size, u32 *val) | ||
1150 | { | ||
1151 | return octeon_pcie_read_config(1, bus, devfn, reg, size, val); | ||
1152 | } | ||
1153 | |||
1154 | |||
1155 | |||
1156 | /** | ||
1157 | * Write a value to PCI configuration space | ||
1158 | * | ||
1159 | * @param bus | ||
1160 | * @param devfn | ||
1161 | * @param reg | ||
1162 | * @param size | ||
1163 | * @param val | ||
1164 | * @return | ||
1165 | */ | ||
1166 | static inline int octeon_pcie_write_config(int pcie_port, struct pci_bus *bus, | ||
1167 | unsigned int devfn, int reg, | ||
1168 | int size, u32 val) | ||
1169 | { | ||
1170 | int bus_number = bus->number; | ||
1171 | /* | ||
1172 | * We need to force the bus number to be zero on the root | ||
1173 | * bus. Linux numbers the 2nd root bus to start after all | ||
1174 | * busses on root 0. | ||
1175 | */ | ||
1176 | if (bus->parent == NULL) | ||
1177 | bus_number = 0; | ||
1178 | |||
1179 | switch (size) { | ||
1180 | case 4: | ||
1181 | cvmx_pcie_config_write32(pcie_port, bus_number, devfn >> 3, | ||
1182 | devfn & 0x7, reg, val); | ||
1183 | return PCIBIOS_SUCCESSFUL; | ||
1184 | case 2: | ||
1185 | cvmx_pcie_config_write16(pcie_port, bus_number, devfn >> 3, | ||
1186 | devfn & 0x7, reg, val); | ||
1187 | return PCIBIOS_SUCCESSFUL; | ||
1188 | case 1: | ||
1189 | cvmx_pcie_config_write8(pcie_port, bus_number, devfn >> 3, | ||
1190 | devfn & 0x7, reg, val); | ||
1191 | return PCIBIOS_SUCCESSFUL; | ||
1192 | } | ||
1193 | #if PCI_CONFIG_SPACE_DELAY | ||
1194 | udelay(PCI_CONFIG_SPACE_DELAY); | ||
1195 | #endif | ||
1196 | return PCIBIOS_FUNC_NOT_SUPPORTED; | ||
1197 | } | ||
1198 | |||
1199 | static int octeon_pcie0_write_config(struct pci_bus *bus, unsigned int devfn, | ||
1200 | int reg, int size, u32 val) | ||
1201 | { | ||
1202 | return octeon_pcie_write_config(0, bus, devfn, reg, size, val); | ||
1203 | } | ||
1204 | |||
1205 | static int octeon_pcie1_write_config(struct pci_bus *bus, unsigned int devfn, | ||
1206 | int reg, int size, u32 val) | ||
1207 | { | ||
1208 | return octeon_pcie_write_config(1, bus, devfn, reg, size, val); | ||
1209 | } | ||
1210 | |||
1211 | static struct pci_ops octeon_pcie0_ops = { | ||
1212 | octeon_pcie0_read_config, | ||
1213 | octeon_pcie0_write_config, | ||
1214 | }; | ||
1215 | |||
1216 | static struct resource octeon_pcie0_mem_resource = { | ||
1217 | .name = "Octeon PCIe0 MEM", | ||
1218 | .flags = IORESOURCE_MEM, | ||
1219 | }; | ||
1220 | |||
1221 | static struct resource octeon_pcie0_io_resource = { | ||
1222 | .name = "Octeon PCIe0 IO", | ||
1223 | .flags = IORESOURCE_IO, | ||
1224 | }; | ||
1225 | |||
1226 | static struct pci_controller octeon_pcie0_controller = { | ||
1227 | .pci_ops = &octeon_pcie0_ops, | ||
1228 | .mem_resource = &octeon_pcie0_mem_resource, | ||
1229 | .io_resource = &octeon_pcie0_io_resource, | ||
1230 | }; | ||
1231 | |||
1232 | static struct pci_ops octeon_pcie1_ops = { | ||
1233 | octeon_pcie1_read_config, | ||
1234 | octeon_pcie1_write_config, | ||
1235 | }; | ||
1236 | |||
1237 | static struct resource octeon_pcie1_mem_resource = { | ||
1238 | .name = "Octeon PCIe1 MEM", | ||
1239 | .flags = IORESOURCE_MEM, | ||
1240 | }; | ||
1241 | |||
1242 | static struct resource octeon_pcie1_io_resource = { | ||
1243 | .name = "Octeon PCIe1 IO", | ||
1244 | .flags = IORESOURCE_IO, | ||
1245 | }; | ||
1246 | |||
1247 | static struct pci_controller octeon_pcie1_controller = { | ||
1248 | .pci_ops = &octeon_pcie1_ops, | ||
1249 | .mem_resource = &octeon_pcie1_mem_resource, | ||
1250 | .io_resource = &octeon_pcie1_io_resource, | ||
1251 | }; | ||
1252 | |||
1253 | |||
1254 | /** | ||
1255 | * Initialize the Octeon PCIe controllers | ||
1256 | * | ||
1257 | * @return | ||
1258 | */ | ||
1259 | static int __init octeon_pcie_setup(void) | ||
1260 | { | ||
1261 | union cvmx_npei_ctl_status npei_ctl_status; | ||
1262 | int result; | ||
1263 | |||
1264 | /* These chips don't have PCIe */ | ||
1265 | if (!octeon_has_feature(OCTEON_FEATURE_PCIE)) | ||
1266 | return 0; | ||
1267 | |||
1268 | /* Point pcibios_map_irq() to the PCIe version of it */ | ||
1269 | octeon_pcibios_map_irq = octeon_pcie_pcibios_map_irq; | ||
1270 | |||
1271 | /* Use the PCIe based DMA mappings */ | ||
1272 | octeon_dma_bar_type = OCTEON_DMA_BAR_TYPE_PCIE; | ||
1273 | |||
1274 | /* | ||
1275 | * PCIe I/O range. It is based on port 0 but includes up until | ||
1276 | * port 1's end. | ||
1277 | */ | ||
1278 | set_io_port_base(CVMX_ADD_IO_SEG(cvmx_pcie_get_io_base_address(0))); | ||
1279 | ioport_resource.start = 0; | ||
1280 | ioport_resource.end = | ||
1281 | cvmx_pcie_get_io_base_address(1) - | ||
1282 | cvmx_pcie_get_io_base_address(0) + cvmx_pcie_get_io_size(1) - 1; | ||
1283 | |||
1284 | npei_ctl_status.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_CTL_STATUS); | ||
1285 | if (npei_ctl_status.s.host_mode) { | ||
1286 | pr_notice("PCIe: Initializing port 0\n"); | ||
1287 | result = cvmx_pcie_rc_initialize(0); | ||
1288 | if (result == 0) { | ||
1289 | /* Memory offsets are physical addresses */ | ||
1290 | octeon_pcie0_controller.mem_offset = | ||
1291 | cvmx_pcie_get_mem_base_address(0); | ||
1292 | /* IO offsets are Mips virtual addresses */ | ||
1293 | octeon_pcie0_controller.io_map_base = | ||
1294 | CVMX_ADD_IO_SEG(cvmx_pcie_get_io_base_address | ||
1295 | (0)); | ||
1296 | octeon_pcie0_controller.io_offset = 0; | ||
1297 | /* | ||
1298 | * To keep things similar to PCI, we start | ||
1299 | * device addresses at the same place as PCI | ||
1300 | * uisng big bar support. This normally | ||
1301 | * translates to 4GB-256MB, which is the same | ||
1302 | * as most x86 PCs. | ||
1303 | */ | ||
1304 | octeon_pcie0_controller.mem_resource->start = | ||
1305 | cvmx_pcie_get_mem_base_address(0) + | ||
1306 | (4ul << 30) - (OCTEON_PCI_BAR1_HOLE_SIZE << 20); | ||
1307 | octeon_pcie0_controller.mem_resource->end = | ||
1308 | cvmx_pcie_get_mem_base_address(0) + | ||
1309 | cvmx_pcie_get_mem_size(0) - 1; | ||
1310 | /* | ||
1311 | * Ports must be above 16KB for the ISA bus | ||
1312 | * filtering in the PCI-X to PCI bridge. | ||
1313 | */ | ||
1314 | octeon_pcie0_controller.io_resource->start = 4 << 10; | ||
1315 | octeon_pcie0_controller.io_resource->end = | ||
1316 | cvmx_pcie_get_io_size(0) - 1; | ||
1317 | register_pci_controller(&octeon_pcie0_controller); | ||
1318 | } | ||
1319 | } else { | ||
1320 | pr_notice("PCIe: Port 0 in endpoint mode, skipping.\n"); | ||
1321 | } | ||
1322 | |||
1323 | /* Skip the 2nd port on CN52XX if port 0 is in 4 lane mode */ | ||
1324 | if (OCTEON_IS_MODEL(OCTEON_CN52XX)) { | ||
1325 | union cvmx_npei_dbg_data npei_dbg_data; | ||
1326 | npei_dbg_data.u64 = cvmx_read_csr(CVMX_PEXP_NPEI_DBG_DATA); | ||
1327 | if (npei_dbg_data.cn52xx.qlm0_link_width) | ||
1328 | return 0; | ||
1329 | } | ||
1330 | |||
1331 | pr_notice("PCIe: Initializing port 1\n"); | ||
1332 | result = cvmx_pcie_rc_initialize(1); | ||
1333 | if (result == 0) { | ||
1334 | /* Memory offsets are physical addresses */ | ||
1335 | octeon_pcie1_controller.mem_offset = | ||
1336 | cvmx_pcie_get_mem_base_address(1); | ||
1337 | /* IO offsets are Mips virtual addresses */ | ||
1338 | octeon_pcie1_controller.io_map_base = | ||
1339 | CVMX_ADD_IO_SEG(cvmx_pcie_get_io_base_address(1)); | ||
1340 | octeon_pcie1_controller.io_offset = | ||
1341 | cvmx_pcie_get_io_base_address(1) - | ||
1342 | cvmx_pcie_get_io_base_address(0); | ||
1343 | /* | ||
1344 | * To keep things similar to PCI, we start device | ||
1345 | * addresses at the same place as PCI uisng big bar | ||
1346 | * support. This normally translates to 4GB-256MB, | ||
1347 | * which is the same as most x86 PCs. | ||
1348 | */ | ||
1349 | octeon_pcie1_controller.mem_resource->start = | ||
1350 | cvmx_pcie_get_mem_base_address(1) + (4ul << 30) - | ||
1351 | (OCTEON_PCI_BAR1_HOLE_SIZE << 20); | ||
1352 | octeon_pcie1_controller.mem_resource->end = | ||
1353 | cvmx_pcie_get_mem_base_address(1) + | ||
1354 | cvmx_pcie_get_mem_size(1) - 1; | ||
1355 | /* | ||
1356 | * Ports must be above 16KB for the ISA bus filtering | ||
1357 | * in the PCI-X to PCI bridge. | ||
1358 | */ | ||
1359 | octeon_pcie1_controller.io_resource->start = | ||
1360 | cvmx_pcie_get_io_base_address(1) - | ||
1361 | cvmx_pcie_get_io_base_address(0); | ||
1362 | octeon_pcie1_controller.io_resource->end = | ||
1363 | octeon_pcie1_controller.io_resource->start + | ||
1364 | cvmx_pcie_get_io_size(1) - 1; | ||
1365 | register_pci_controller(&octeon_pcie1_controller); | ||
1366 | } | ||
1367 | return 0; | ||
1368 | } | ||
1369 | |||
1370 | arch_initcall(octeon_pcie_setup); | ||