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_IODA2,
14 PNV_PHB_NPU_OCAPI,
15 };
16
17 /* Precise PHB model for error management */
18 enum pnv_phb_model {
19 PNV_PHB_MODEL_UNKNOWN,
20 PNV_PHB_MODEL_P7IOC,
21 PNV_PHB_MODEL_PHB3,
22 };
23
24 #define PNV_PCI_DIAG_BUF_SIZE 8192
25 #define PNV_IODA_PE_DEV (1 << 0) /* PE has single PCI device */
26 #define PNV_IODA_PE_BUS (1 << 1) /* PE has primary PCI bus */
27 #define PNV_IODA_PE_BUS_ALL (1 << 2) /* PE has subordinate buses */
28 #define PNV_IODA_PE_MASTER (1 << 3) /* Master PE in compound case */
29 #define PNV_IODA_PE_SLAVE (1 << 4) /* Slave PE in compound case */
30 #define PNV_IODA_PE_VF (1 << 5) /* PE for one VF */
31
32 /*
33 * A brief note on PNV_IODA_PE_BUS_ALL
34 *
35 * This is needed because of the behaviour of PCIe-to-PCI bridges. The PHB uses
36 * the Requester ID field of the PCIe request header to determine the device
37 * (and PE) that initiated a DMA. In legacy PCI individual memory read/write
38 * requests aren't tagged with the RID. To work around this the PCIe-to-PCI
39 * bridge will use (secondary_bus_no << 8) | 0x00 as the RID on the PCIe side.
40 *
41 * PCIe-to-X bridges have a similar issue even though PCI-X requests also have
42 * a RID in the transaction header. The PCIe-to-X bridge is permitted to "take
43 * ownership" of a transaction by a PCI-X device when forwarding it to the PCIe
44 * side of the bridge.
45 *
46 * To work around these problems we use the BUS_ALL flag since every subordinate
47 * bus of the bridge should go into the same PE.
48 */
49
50 /* Indicates operations are frozen for a PE: MMIO in PESTA & DMA in PESTB. */
51 #define PNV_IODA_STOPPED_STATE 0x8000000000000000
52
53 /* Data associated with a PE, including IOMMU tracking etc.. */
54 struct pnv_phb;
55 struct pnv_ioda_pe {
56 unsigned long flags;
57 struct pnv_phb *phb;
58 int device_count;
59
60 /* A PE can be associated with a single device or an
61 * entire bus (& children). In the former case, pdev
62 * is populated, in the later case, pbus is.
63 */
64 #ifdef CONFIG_PCI_IOV
65 struct pci_dev *parent_dev;
66 #endif
67 struct pci_dev *pdev;
68 struct pci_bus *pbus;
69
70 /* Effective RID (device RID for a device PE and base bus
71 * RID with devfn 0 for a bus PE)
72 */
73 unsigned int rid;
74
75 /* PE number */
76 unsigned int pe_number;
77
78 /* "Base" iommu table, ie, 4K TCEs, 32-bit DMA */
79 struct iommu_table_group table_group;
80
81 /* 64-bit TCE bypass region */
82 bool tce_bypass_enabled;
83 uint64_t tce_bypass_base;
84
85 /*
86 * Used to track whether we've done DMA setup for this PE or not. We
87 * want to defer allocating TCE tables, etc until we've added a
88 * non-bridge device to the PE.
89 */
90 bool dma_setup_done;
91
92 /* MSIs. MVE index is identical for 32 and 64 bit MSI
93 * and -1 if not supported. (It's actually identical to the
94 * PE number)
95 */
96 int mve_number;
97
98 /* PEs in compound case */
99 struct pnv_ioda_pe *master;
100 struct list_head slaves;
101
102 /* Link in list of PE#s */
103 struct list_head list;
104 };
105
106 #define PNV_PHB_FLAG_EEH (1 << 0)
107
108 struct pnv_phb {
109 struct pci_controller *hose;
110 enum pnv_phb_type type;
111 enum pnv_phb_model model;
112 u64 hub_id;
113 u64 opal_id;
114 int flags;
115 void __iomem *regs;
116 u64 regs_phys;
117 spinlock_t lock;
118
119 #ifdef CONFIG_DEBUG_FS
120 int has_dbgfs;
121 struct dentry *dbgfs;
122 #endif
123
124 unsigned int msi_base;
125 struct msi_bitmap msi_bmp;
126 int (*init_m64)(struct pnv_phb *phb);
127 int (*get_pe_state)(struct pnv_phb *phb, int pe_no);
128 void (*freeze_pe)(struct pnv_phb *phb, int pe_no);
129 int (*unfreeze_pe)(struct pnv_phb *phb, int pe_no, int opt);
130
131 struct {
132 /* Global bridge info */
133 unsigned int total_pe_num;
134 unsigned int reserved_pe_idx;
135 unsigned int root_pe_idx;
136
137 /* 32-bit MMIO window */
138 unsigned int m32_size;
139 unsigned int m32_segsize;
140 unsigned int m32_pci_base;
141
142 /* 64-bit MMIO window */
143 unsigned int m64_bar_idx;
144 unsigned long m64_size;
145 unsigned long m64_segsize;
146 unsigned long m64_base;
147 #define MAX_M64_BARS 64
148 unsigned long m64_bar_alloc;
149
150 /* IO ports */
151 unsigned int io_size;
152 unsigned int io_segsize;
153 unsigned int io_pci_base;
154
155 /* PE allocation */
156 struct mutex pe_alloc_mutex;
157 unsigned long *pe_alloc;
158 struct pnv_ioda_pe *pe_array;
159
160 /* M32 & IO segment maps */
161 unsigned int *m64_segmap;
162 unsigned int *m32_segmap;
163 unsigned int *io_segmap;
164
165 /* IRQ chip */
166 struct irq_chip irq_chip;
167
168 /* Sorted list of used PE's based
169 * on the sequence of creation
170 */
171 struct list_head pe_list;
172 struct mutex pe_list_mutex;
173
174 /* Reverse map of PEs, indexed by {bus, devfn} */
175 unsigned int pe_rmap[0x10000];
176 } ioda;
177
178 /* PHB and hub diagnostics */
179 unsigned int diag_data_size;
180 u8 *diag_data;
181 };
182
183
184 /* IODA PE management */
185
pnv_pci_is_m64(struct pnv_phb * phb,struct resource * r)186 static inline bool pnv_pci_is_m64(struct pnv_phb *phb, struct resource *r)
187 {
188 /*
189 * WARNING: We cannot rely on the resource flags. The Linux PCI
190 * allocation code sometimes decides to put a 64-bit prefetchable
191 * BAR in the 32-bit window, so we have to compare the addresses.
192 *
193 * For simplicity we only test resource start.
194 */
195 return (r->start >= phb->ioda.m64_base &&
196 r->start < (phb->ioda.m64_base + phb->ioda.m64_size));
197 }
198
pnv_pci_is_m64_flags(unsigned long resource_flags)199 static inline bool pnv_pci_is_m64_flags(unsigned long resource_flags)
200 {
201 unsigned long flags = (IORESOURCE_MEM_64 | IORESOURCE_PREFETCH);
202
203 return (resource_flags & flags) == flags;
204 }
205
206 int pnv_ioda_configure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe);
207 int pnv_ioda_deconfigure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe);
208
209 void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe);
210 void pnv_pci_ioda2_release_pe_dma(struct pnv_ioda_pe *pe);
211
212 struct pnv_ioda_pe *pnv_ioda_alloc_pe(struct pnv_phb *phb, int count);
213 void pnv_ioda_free_pe(struct pnv_ioda_pe *pe);
214
215 #ifdef CONFIG_PCI_IOV
216 /*
217 * For SR-IOV we want to put each VF's MMIO resource in to a separate PE.
218 * This requires a bit of acrobatics with the MMIO -> PE configuration
219 * and this structure is used to keep track of it all.
220 */
221 struct pnv_iov_data {
222 /* number of VFs enabled */
223 u16 num_vfs;
224
225 /* pointer to the array of VF PEs. num_vfs long*/
226 struct pnv_ioda_pe *vf_pe_arr;
227
228 /* Did we map the VF BAR with single-PE IODA BARs? */
229 bool m64_single_mode[PCI_SRIOV_NUM_BARS];
230
231 /*
232 * True if we're using any segmented windows. In that case we need
233 * shift the start of the IOV resource the segment corresponding to
234 * the allocated PE.
235 */
236 bool need_shift;
237
238 /*
239 * Bit mask used to track which m64 windows are used to map the
240 * SR-IOV BARs for this device.
241 */
242 DECLARE_BITMAP(used_m64_bar_mask, MAX_M64_BARS);
243
244 /*
245 * If we map the SR-IOV BARs with a segmented window then
246 * parts of that window will be "claimed" by other PEs.
247 *
248 * "holes" here is used to reserve the leading portion
249 * of the window that is used by other (non VF) PEs.
250 */
251 struct resource holes[PCI_SRIOV_NUM_BARS];
252 };
253
pnv_iov_get(struct pci_dev * pdev)254 static inline struct pnv_iov_data *pnv_iov_get(struct pci_dev *pdev)
255 {
256 return pdev->dev.archdata.iov_data;
257 }
258
259 void pnv_pci_ioda_fixup_iov(struct pci_dev *pdev);
260 resource_size_t pnv_pci_iov_resource_alignment(struct pci_dev *pdev, int resno);
261
262 int pnv_pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs);
263 int pnv_pcibios_sriov_disable(struct pci_dev *pdev);
264 #endif /* CONFIG_PCI_IOV */
265
266 extern struct pci_ops pnv_pci_ops;
267
268 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
269 unsigned char *log_buff);
270 int pnv_pci_cfg_read(struct pci_dn *pdn,
271 int where, int size, u32 *val);
272 int pnv_pci_cfg_write(struct pci_dn *pdn,
273 int where, int size, u32 val);
274 extern struct iommu_table *pnv_pci_table_alloc(int nid);
275
276 extern void pnv_pci_init_ioda2_phb(struct device_node *np);
277 extern void pnv_pci_init_npu2_opencapi_phb(struct device_node *np);
278 extern void pnv_pci_reset_secondary_bus(struct pci_dev *dev);
279 extern int pnv_eeh_phb_reset(struct pci_controller *hose, int option);
280
281 extern struct pnv_ioda_pe *pnv_pci_bdfn_to_pe(struct pnv_phb *phb, u16 bdfn);
282 extern struct pnv_ioda_pe *pnv_ioda_get_pe(struct pci_dev *dev);
283 extern unsigned long pnv_pci_ioda2_get_table_size(__u32 page_shift,
284 __u64 window_size, __u32 levels);
285 extern int pnv_eeh_post_init(void);
286
287 __printf(3, 4)
288 extern void pe_level_printk(const struct pnv_ioda_pe *pe, const char *level,
289 const char *fmt, ...);
290 #define pe_err(pe, fmt, ...) \
291 pe_level_printk(pe, KERN_ERR, fmt, ##__VA_ARGS__)
292 #define pe_warn(pe, fmt, ...) \
293 pe_level_printk(pe, KERN_WARNING, fmt, ##__VA_ARGS__)
294 #define pe_info(pe, fmt, ...) \
295 pe_level_printk(pe, KERN_INFO, fmt, ##__VA_ARGS__)
296
297 /* pci-ioda-tce.c */
298 #define POWERNV_IOMMU_DEFAULT_LEVELS 2
299 #define POWERNV_IOMMU_MAX_LEVELS 5
300
301 extern int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
302 unsigned long uaddr, enum dma_data_direction direction,
303 unsigned long attrs);
304 extern void pnv_tce_free(struct iommu_table *tbl, long index, long npages);
305 extern int pnv_tce_xchg(struct iommu_table *tbl, long index,
306 unsigned long *hpa, enum dma_data_direction *direction);
307 extern __be64 *pnv_tce_useraddrptr(struct iommu_table *tbl, long index,
308 bool alloc);
309 extern unsigned long pnv_tce_get(struct iommu_table *tbl, long index);
310
311 extern long pnv_pci_ioda2_table_alloc_pages(int nid, __u64 bus_offset,
312 __u32 page_shift, __u64 window_size, __u32 levels,
313 bool alloc_userspace_copy, struct iommu_table *tbl);
314 extern void pnv_pci_ioda2_table_free_pages(struct iommu_table *tbl);
315
316 extern long pnv_pci_link_table_and_group(int node, int num,
317 struct iommu_table *tbl,
318 struct iommu_table_group *table_group);
319 extern void pnv_pci_unlink_table_and_group(struct iommu_table *tbl,
320 struct iommu_table_group *table_group);
321 extern void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
322 void *tce_mem, u64 tce_size,
323 u64 dma_offset, unsigned int page_shift);
324
325 extern unsigned long pnv_ioda_parse_tce_sizes(struct pnv_phb *phb);
326
pci_bus_to_pnvhb(struct pci_bus * bus)327 static inline struct pnv_phb *pci_bus_to_pnvhb(struct pci_bus *bus)
328 {
329 struct pci_controller *hose = bus->sysdata;
330
331 if (hose)
332 return hose->private_data;
333
334 return NULL;
335 }
336
337 #endif /* __POWERNV_PCI_H */
338