1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __POWERNV_PCI_H
3 #define __POWERNV_PCI_H
4
5 #include <linux/compiler.h> /* for __printf */
6 #include <linux/iommu.h>
7 #include <asm/iommu.h>
8 #include <asm/msi_bitmap.h>
9
10 struct pci_dn;
11
12 enum pnv_phb_type {
13 PNV_PHB_IODA1 = 0,
14 PNV_PHB_IODA2 = 1,
15 PNV_PHB_NPU_NVLINK = 2,
16 PNV_PHB_NPU_OCAPI = 3,
17 };
18
19 /* Precise PHB model for error management */
20 enum pnv_phb_model {
21 PNV_PHB_MODEL_UNKNOWN,
22 PNV_PHB_MODEL_P7IOC,
23 PNV_PHB_MODEL_PHB3,
24 PNV_PHB_MODEL_NPU,
25 PNV_PHB_MODEL_NPU2,
26 };
27
28 #define PNV_PCI_DIAG_BUF_SIZE 8192
29 #define PNV_IODA_PE_DEV (1 << 0) /* PE has single PCI device */
30 #define PNV_IODA_PE_BUS (1 << 1) /* PE has primary PCI bus */
31 #define PNV_IODA_PE_BUS_ALL (1 << 2) /* PE has subordinate buses */
32 #define PNV_IODA_PE_MASTER (1 << 3) /* Master PE in compound case */
33 #define PNV_IODA_PE_SLAVE (1 << 4) /* Slave PE in compound case */
34 #define PNV_IODA_PE_VF (1 << 5) /* PE for one VF */
35
36 /*
37 * A brief note on PNV_IODA_PE_BUS_ALL
38 *
39 * This is needed because of the behaviour of PCIe-to-PCI bridges. The PHB uses
40 * the Requester ID field of the PCIe request header to determine the device
41 * (and PE) that initiated a DMA. In legacy PCI individual memory read/write
42 * requests aren't tagged with the RID. To work around this the PCIe-to-PCI
43 * bridge will use (secondary_bus_no << 8) | 0x00 as the RID on the PCIe side.
44 *
45 * PCIe-to-X bridges have a similar issue even though PCI-X requests also have
46 * a RID in the transaction header. The PCIe-to-X bridge is permitted to "take
47 * ownership" of a transaction by a PCI-X device when forwarding it to the PCIe
48 * side of the bridge.
49 *
50 * To work around these problems we use the BUS_ALL flag since every subordinate
51 * bus of the bridge should go into the same PE.
52 */
53
54 /* Indicates operations are frozen for a PE: MMIO in PESTA & DMA in PESTB. */
55 #define PNV_IODA_STOPPED_STATE 0x8000000000000000
56
57 /* Data associated with a PE, including IOMMU tracking etc.. */
58 struct pnv_phb;
59 struct pnv_ioda_pe {
60 unsigned long flags;
61 struct pnv_phb *phb;
62 int device_count;
63
64 /* A PE can be associated with a single device or an
65 * entire bus (& children). In the former case, pdev
66 * is populated, in the later case, pbus is.
67 */
68 #ifdef CONFIG_PCI_IOV
69 struct pci_dev *parent_dev;
70 #endif
71 struct pci_dev *pdev;
72 struct pci_bus *pbus;
73
74 /* Effective RID (device RID for a device PE and base bus
75 * RID with devfn 0 for a bus PE)
76 */
77 unsigned int rid;
78
79 /* PE number */
80 unsigned int pe_number;
81
82 /* "Base" iommu table, ie, 4K TCEs, 32-bit DMA */
83 struct iommu_table_group table_group;
84 struct npu_comp *npucomp;
85
86 /* 64-bit TCE bypass region */
87 bool tce_bypass_enabled;
88 uint64_t tce_bypass_base;
89
90 /*
91 * Used to track whether we've done DMA setup for this PE or not. We
92 * want to defer allocating TCE tables, etc until we've added a
93 * non-bridge device to the PE.
94 */
95 bool dma_setup_done;
96
97 /* MSIs. MVE index is identical for 32 and 64 bit MSI
98 * and -1 if not supported. (It's actually identical to the
99 * PE number)
100 */
101 int mve_number;
102
103 /* PEs in compound case */
104 struct pnv_ioda_pe *master;
105 struct list_head slaves;
106
107 /* Link in list of PE#s */
108 struct list_head list;
109 };
110
111 #define PNV_PHB_FLAG_EEH (1 << 0)
112
113 struct pnv_phb {
114 struct pci_controller *hose;
115 enum pnv_phb_type type;
116 enum pnv_phb_model model;
117 u64 hub_id;
118 u64 opal_id;
119 int flags;
120 void __iomem *regs;
121 u64 regs_phys;
122 int initialized;
123 spinlock_t lock;
124
125 #ifdef CONFIG_DEBUG_FS
126 int has_dbgfs;
127 struct dentry *dbgfs;
128 #endif
129
130 unsigned int msi_base;
131 unsigned int msi32_support;
132 struct msi_bitmap msi_bmp;
133 int (*msi_setup)(struct pnv_phb *phb, struct pci_dev *dev,
134 unsigned int hwirq, unsigned int virq,
135 unsigned int is_64, struct msi_msg *msg);
136 int (*init_m64)(struct pnv_phb *phb);
137 int (*get_pe_state)(struct pnv_phb *phb, int pe_no);
138 void (*freeze_pe)(struct pnv_phb *phb, int pe_no);
139 int (*unfreeze_pe)(struct pnv_phb *phb, int pe_no, int opt);
140
141 struct {
142 /* Global bridge info */
143 unsigned int total_pe_num;
144 unsigned int reserved_pe_idx;
145 unsigned int root_pe_idx;
146
147 /* 32-bit MMIO window */
148 unsigned int m32_size;
149 unsigned int m32_segsize;
150 unsigned int m32_pci_base;
151
152 /* 64-bit MMIO window */
153 unsigned int m64_bar_idx;
154 unsigned long m64_size;
155 unsigned long m64_segsize;
156 unsigned long m64_base;
157 #define MAX_M64_BARS 64
158 unsigned long m64_bar_alloc;
159
160 /* IO ports */
161 unsigned int io_size;
162 unsigned int io_segsize;
163 unsigned int io_pci_base;
164
165 /* PE allocation */
166 struct mutex pe_alloc_mutex;
167 unsigned long *pe_alloc;
168 struct pnv_ioda_pe *pe_array;
169
170 /* M32 & IO segment maps */
171 unsigned int *m64_segmap;
172 unsigned int *m32_segmap;
173 unsigned int *io_segmap;
174
175 /* DMA32 segment maps - IODA1 only */
176 unsigned int dma32_count;
177 unsigned int *dma32_segmap;
178
179 /* IRQ chip */
180 int irq_chip_init;
181 struct irq_chip irq_chip;
182
183 /* Sorted list of used PE's based
184 * on the sequence of creation
185 */
186 struct list_head pe_list;
187 struct mutex pe_list_mutex;
188
189 /* Reverse map of PEs, indexed by {bus, devfn} */
190 unsigned int pe_rmap[0x10000];
191 } ioda;
192
193 /* PHB and hub diagnostics */
194 unsigned int diag_data_size;
195 u8 *diag_data;
196 };
197
198
199 /* IODA PE management */
200
pnv_pci_is_m64(struct pnv_phb * phb,struct resource * r)201 static inline bool pnv_pci_is_m64(struct pnv_phb *phb, struct resource *r)
202 {
203 /*
204 * WARNING: We cannot rely on the resource flags. The Linux PCI
205 * allocation code sometimes decides to put a 64-bit prefetchable
206 * BAR in the 32-bit window, so we have to compare the addresses.
207 *
208 * For simplicity we only test resource start.
209 */
210 return (r->start >= phb->ioda.m64_base &&
211 r->start < (phb->ioda.m64_base + phb->ioda.m64_size));
212 }
213
pnv_pci_is_m64_flags(unsigned long resource_flags)214 static inline bool pnv_pci_is_m64_flags(unsigned long resource_flags)
215 {
216 unsigned long flags = (IORESOURCE_MEM_64 | IORESOURCE_PREFETCH);
217
218 return (resource_flags & flags) == flags;
219 }
220
221 int pnv_ioda_configure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe);
222 int pnv_ioda_deconfigure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe);
223
224 void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe);
225 void pnv_pci_ioda2_release_pe_dma(struct pnv_ioda_pe *pe);
226
227 struct pnv_ioda_pe *pnv_ioda_alloc_pe(struct pnv_phb *phb, int count);
228 void pnv_ioda_free_pe(struct pnv_ioda_pe *pe);
229
230 #ifdef CONFIG_PCI_IOV
231 /*
232 * For SR-IOV we want to put each VF's MMIO resource in to a separate PE.
233 * This requires a bit of acrobatics with the MMIO -> PE configuration
234 * and this structure is used to keep track of it all.
235 */
236 struct pnv_iov_data {
237 /* number of VFs enabled */
238 u16 num_vfs;
239
240 /* pointer to the array of VF PEs. num_vfs long*/
241 struct pnv_ioda_pe *vf_pe_arr;
242
243 /* Did we map the VF BAR with single-PE IODA BARs? */
244 bool m64_single_mode[PCI_SRIOV_NUM_BARS];
245
246 /*
247 * True if we're using any segmented windows. In that case we need
248 * shift the start of the IOV resource the segment corresponding to
249 * the allocated PE.
250 */
251 bool need_shift;
252
253 /*
254 * Bit mask used to track which m64 windows are used to map the
255 * SR-IOV BARs for this device.
256 */
257 DECLARE_BITMAP(used_m64_bar_mask, MAX_M64_BARS);
258
259 /*
260 * If we map the SR-IOV BARs with a segmented window then
261 * parts of that window will be "claimed" by other PEs.
262 *
263 * "holes" here is used to reserve the leading portion
264 * of the window that is used by other (non VF) PEs.
265 */
266 struct resource holes[PCI_SRIOV_NUM_BARS];
267 };
268
pnv_iov_get(struct pci_dev * pdev)269 static inline struct pnv_iov_data *pnv_iov_get(struct pci_dev *pdev)
270 {
271 return pdev->dev.archdata.iov_data;
272 }
273
274 void pnv_pci_ioda_fixup_iov(struct pci_dev *pdev);
275 resource_size_t pnv_pci_iov_resource_alignment(struct pci_dev *pdev, int resno);
276
277 int pnv_pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs);
278 int pnv_pcibios_sriov_disable(struct pci_dev *pdev);
279 #endif /* CONFIG_PCI_IOV */
280
281 extern struct pci_ops pnv_pci_ops;
282
283 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
284 unsigned char *log_buff);
285 int pnv_pci_cfg_read(struct pci_dn *pdn,
286 int where, int size, u32 *val);
287 int pnv_pci_cfg_write(struct pci_dn *pdn,
288 int where, int size, u32 val);
289 extern struct iommu_table *pnv_pci_table_alloc(int nid);
290
291 extern void pnv_pci_init_ioda_hub(struct device_node *np);
292 extern void pnv_pci_init_ioda2_phb(struct device_node *np);
293 extern void pnv_pci_init_npu_phb(struct device_node *np);
294 extern void pnv_pci_init_npu2_opencapi_phb(struct device_node *np);
295 extern void pnv_npu2_map_lpar(struct pnv_ioda_pe *gpe, unsigned long msr);
296 extern void pnv_pci_reset_secondary_bus(struct pci_dev *dev);
297 extern int pnv_eeh_phb_reset(struct pci_controller *hose, int option);
298
299 extern int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type);
300 extern void pnv_teardown_msi_irqs(struct pci_dev *pdev);
301 extern struct pnv_ioda_pe *pnv_pci_bdfn_to_pe(struct pnv_phb *phb, u16 bdfn);
302 extern struct pnv_ioda_pe *pnv_ioda_get_pe(struct pci_dev *dev);
303 extern void pnv_set_msi_irq_chip(struct pnv_phb *phb, unsigned int virq);
304 extern unsigned long pnv_pci_ioda2_get_table_size(__u32 page_shift,
305 __u64 window_size, __u32 levels);
306 extern int pnv_eeh_post_init(void);
307
308 __printf(3, 4)
309 extern void pe_level_printk(const struct pnv_ioda_pe *pe, const char *level,
310 const char *fmt, ...);
311 #define pe_err(pe, fmt, ...) \
312 pe_level_printk(pe, KERN_ERR, fmt, ##__VA_ARGS__)
313 #define pe_warn(pe, fmt, ...) \
314 pe_level_printk(pe, KERN_WARNING, fmt, ##__VA_ARGS__)
315 #define pe_info(pe, fmt, ...) \
316 pe_level_printk(pe, KERN_INFO, fmt, ##__VA_ARGS__)
317
318 /* Nvlink functions */
319 extern void pnv_npu_try_dma_set_bypass(struct pci_dev *gpdev, bool bypass);
320 extern void pnv_pci_ioda2_tce_invalidate_entire(struct pnv_phb *phb, bool rm);
321 extern void pnv_pci_npu_setup_iommu_groups(void);
322
323 /* pci-ioda-tce.c */
324 #define POWERNV_IOMMU_DEFAULT_LEVELS 2
325 #define POWERNV_IOMMU_MAX_LEVELS 5
326
327 extern int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
328 unsigned long uaddr, enum dma_data_direction direction,
329 unsigned long attrs);
330 extern void pnv_tce_free(struct iommu_table *tbl, long index, long npages);
331 extern int pnv_tce_xchg(struct iommu_table *tbl, long index,
332 unsigned long *hpa, enum dma_data_direction *direction,
333 bool alloc);
334 extern __be64 *pnv_tce_useraddrptr(struct iommu_table *tbl, long index,
335 bool alloc);
336 extern unsigned long pnv_tce_get(struct iommu_table *tbl, long index);
337
338 extern long pnv_pci_ioda2_table_alloc_pages(int nid, __u64 bus_offset,
339 __u32 page_shift, __u64 window_size, __u32 levels,
340 bool alloc_userspace_copy, struct iommu_table *tbl);
341 extern void pnv_pci_ioda2_table_free_pages(struct iommu_table *tbl);
342
343 extern long pnv_pci_link_table_and_group(int node, int num,
344 struct iommu_table *tbl,
345 struct iommu_table_group *table_group);
346 extern void pnv_pci_unlink_table_and_group(struct iommu_table *tbl,
347 struct iommu_table_group *table_group);
348 extern void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
349 void *tce_mem, u64 tce_size,
350 u64 dma_offset, unsigned int page_shift);
351
352 extern unsigned long pnv_ioda_parse_tce_sizes(struct pnv_phb *phb);
353
pci_bus_to_pnvhb(struct pci_bus * bus)354 static inline struct pnv_phb *pci_bus_to_pnvhb(struct pci_bus *bus)
355 {
356 struct pci_controller *hose = bus->sysdata;
357
358 if (hose)
359 return hose->private_data;
360
361 return NULL;
362 }
363
364 #endif /* __POWERNV_PCI_H */
365
