xref: /openbmc/linux/drivers/dma/ioat/dma.h (revision a5b2c10c)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  * Copyright(c) 2004 - 2009 Intel Corporation. All rights reserved.
4  */
5 #ifndef IOATDMA_H
6 #define IOATDMA_H
7 
8 #include <linux/dmaengine.h>
9 #include <linux/init.h>
10 #include <linux/dmapool.h>
11 #include <linux/cache.h>
12 #include <linux/pci_ids.h>
13 #include <linux/circ_buf.h>
14 #include <linux/interrupt.h>
15 #include "registers.h"
16 #include "hw.h"
17 
18 #define IOAT_DMA_VERSION  "5.00"
19 
20 #define IOAT_DMA_DCA_ANY_CPU		~0
21 
22 #define to_ioatdma_device(dev) container_of(dev, struct ioatdma_device, dma_dev)
23 #define to_dev(ioat_chan) (&(ioat_chan)->ioat_dma->pdev->dev)
24 #define to_pdev(ioat_chan) ((ioat_chan)->ioat_dma->pdev)
25 
26 #define chan_num(ch) ((int)((ch)->reg_base - (ch)->ioat_dma->reg_base) / 0x80)
27 
28 /* ioat hardware assumes at least two sources for raid operations */
29 #define src_cnt_to_sw(x) ((x) + 2)
30 #define src_cnt_to_hw(x) ((x) - 2)
31 #define ndest_to_sw(x) ((x) + 1)
32 #define ndest_to_hw(x) ((x) - 1)
33 #define src16_cnt_to_sw(x) ((x) + 9)
34 #define src16_cnt_to_hw(x) ((x) - 9)
35 
36 /*
37  * workaround for IOAT ver.3.0 null descriptor issue
38  * (channel returns error when size is 0)
39  */
40 #define NULL_DESC_BUFFER_SIZE 1
41 
42 enum ioat_irq_mode {
43 	IOAT_NOIRQ = 0,
44 	IOAT_MSIX,
45 	IOAT_MSI,
46 	IOAT_INTX
47 };
48 
49 /**
50  * struct ioatdma_device - internal representation of a IOAT device
51  * @pdev: PCI-Express device
52  * @reg_base: MMIO register space base address
53  * @completion_pool: DMA buffers for completion ops
54  * @sed_hw_pool: DMA super descriptor pools
55  * @dma_dev: embedded struct dma_device
56  * @version: version of ioatdma device
57  * @msix_entries: irq handlers
58  * @idx: per channel data
59  * @dca: direct cache access context
60  * @irq_mode: interrupt mode (INTX, MSI, MSIX)
61  * @cap: read DMA capabilities register
62  */
63 struct ioatdma_device {
64 	struct pci_dev *pdev;
65 	void __iomem *reg_base;
66 	struct dma_pool *completion_pool;
67 #define MAX_SED_POOLS	5
68 	struct dma_pool *sed_hw_pool[MAX_SED_POOLS];
69 	struct dma_device dma_dev;
70 	u8 version;
71 #define IOAT_MAX_CHANS 4
72 	struct msix_entry msix_entries[IOAT_MAX_CHANS];
73 	struct ioatdma_chan *idx[IOAT_MAX_CHANS];
74 	struct dca_provider *dca;
75 	enum ioat_irq_mode irq_mode;
76 	u32 cap;
77 
78 	/* shadow version for CB3.3 chan reset errata workaround */
79 	u64 msixtba0;
80 	u64 msixdata0;
81 	u32 msixpba;
82 };
83 
84 #define IOAT_MAX_ORDER 16
85 #define IOAT_MAX_DESCS (1 << IOAT_MAX_ORDER)
86 #define IOAT_CHUNK_SIZE (SZ_512K)
87 #define IOAT_DESCS_PER_CHUNK (IOAT_CHUNK_SIZE / IOAT_DESC_SZ)
88 
89 struct ioat_descs {
90 	void *virt;
91 	dma_addr_t hw;
92 };
93 
94 struct ioatdma_chan {
95 	struct dma_chan dma_chan;
96 	void __iomem *reg_base;
97 	dma_addr_t last_completion;
98 	spinlock_t cleanup_lock;
99 	unsigned long state;
100 	#define IOAT_CHAN_DOWN 0
101 	#define IOAT_COMPLETION_ACK 1
102 	#define IOAT_RESET_PENDING 2
103 	#define IOAT_KOBJ_INIT_FAIL 3
104 	#define IOAT_RUN 5
105 	#define IOAT_CHAN_ACTIVE 6
106 	struct timer_list timer;
107 	#define RESET_DELAY msecs_to_jiffies(100)
108 	struct ioatdma_device *ioat_dma;
109 	dma_addr_t completion_dma;
110 	u64 *completion;
111 	struct tasklet_struct cleanup_task;
112 	struct kobject kobj;
113 
114 /* ioat v2 / v3 channel attributes
115  * @xfercap_log; log2 of channel max transfer length (for fast division)
116  * @head: allocated index
117  * @issued: hardware notification point
118  * @tail: cleanup index
119  * @dmacount: identical to 'head' except for occasionally resetting to zero
120  * @alloc_order: log2 of the number of allocated descriptors
121  * @produce: number of descriptors to produce at submit time
122  * @ring: software ring buffer implementation of hardware ring
123  * @prep_lock: serializes descriptor preparation (producers)
124  */
125 	size_t xfercap_log;
126 	u16 head;
127 	u16 issued;
128 	u16 tail;
129 	u16 dmacount;
130 	u16 alloc_order;
131 	u16 produce;
132 	struct ioat_ring_ent **ring;
133 	spinlock_t prep_lock;
134 	struct ioat_descs descs[IOAT_MAX_DESCS / IOAT_DESCS_PER_CHUNK];
135 	int desc_chunks;
136 	int intr_coalesce;
137 	int prev_intr_coalesce;
138 };
139 
140 struct ioat_sysfs_entry {
141 	struct attribute attr;
142 	ssize_t (*show)(struct dma_chan *, char *);
143 	ssize_t (*store)(struct dma_chan *, const char *, size_t);
144 };
145 
146 /**
147  * struct ioat_sed_ent - wrapper around super extended hardware descriptor
148  * @hw: hardware SED
149  * @dma: dma address for the SED
150  * @parent: point to the dma descriptor that's the parent
151  * @hw_pool: descriptor pool index
152  */
153 struct ioat_sed_ent {
154 	struct ioat_sed_raw_descriptor *hw;
155 	dma_addr_t dma;
156 	struct ioat_ring_ent *parent;
157 	unsigned int hw_pool;
158 };
159 
160 /**
161  * struct ioat_ring_ent - wrapper around hardware descriptor
162  * @hw: hardware DMA descriptor (for memcpy)
163  * @xor: hardware xor descriptor
164  * @xor_ex: hardware xor extension descriptor
165  * @pq: hardware pq descriptor
166  * @pq_ex: hardware pq extension descriptor
167  * @pqu: hardware pq update descriptor
168  * @raw: hardware raw (un-typed) descriptor
169  * @txd: the generic software descriptor for all engines
170  * @len: total transaction length for unmap
171  * @result: asynchronous result of validate operations
172  * @id: identifier for debug
173  * @sed: pointer to super extended descriptor sw desc
174  */
175 
176 struct ioat_ring_ent {
177 	union {
178 		struct ioat_dma_descriptor *hw;
179 		struct ioat_xor_descriptor *xor;
180 		struct ioat_xor_ext_descriptor *xor_ex;
181 		struct ioat_pq_descriptor *pq;
182 		struct ioat_pq_ext_descriptor *pq_ex;
183 		struct ioat_pq_update_descriptor *pqu;
184 		struct ioat_raw_descriptor *raw;
185 	};
186 	size_t len;
187 	struct dma_async_tx_descriptor txd;
188 	enum sum_check_flags *result;
189 	#ifdef DEBUG
190 	int id;
191 	#endif
192 	struct ioat_sed_ent *sed;
193 };
194 
195 extern const struct sysfs_ops ioat_sysfs_ops;
196 extern struct ioat_sysfs_entry ioat_version_attr;
197 extern struct ioat_sysfs_entry ioat_cap_attr;
198 extern int ioat_pending_level;
199 extern int ioat_ring_alloc_order;
200 extern struct kobj_type ioat_ktype;
201 extern struct kmem_cache *ioat_cache;
202 extern int ioat_ring_max_alloc_order;
203 extern struct kmem_cache *ioat_sed_cache;
204 
205 static inline struct ioatdma_chan *to_ioat_chan(struct dma_chan *c)
206 {
207 	return container_of(c, struct ioatdma_chan, dma_chan);
208 }
209 
210 /* wrapper around hardware descriptor format + additional software fields */
211 #ifdef DEBUG
212 #define set_desc_id(desc, i) ((desc)->id = (i))
213 #define desc_id(desc) ((desc)->id)
214 #else
215 #define set_desc_id(desc, i)
216 #define desc_id(desc) (0)
217 #endif
218 
219 static inline void
220 __dump_desc_dbg(struct ioatdma_chan *ioat_chan, struct ioat_dma_descriptor *hw,
221 		struct dma_async_tx_descriptor *tx, int id)
222 {
223 	struct device *dev = to_dev(ioat_chan);
224 
225 	dev_dbg(dev, "desc[%d]: (%#llx->%#llx) cookie: %d flags: %#x"
226 		" ctl: %#10.8x (op: %#x int_en: %d compl: %d)\n", id,
227 		(unsigned long long) tx->phys,
228 		(unsigned long long) hw->next, tx->cookie, tx->flags,
229 		hw->ctl, hw->ctl_f.op, hw->ctl_f.int_en, hw->ctl_f.compl_write);
230 }
231 
232 #define dump_desc_dbg(c, d) \
233 	({ if (d) __dump_desc_dbg(c, d->hw, &d->txd, desc_id(d)); 0; })
234 
235 static inline struct ioatdma_chan *
236 ioat_chan_by_index(struct ioatdma_device *ioat_dma, int index)
237 {
238 	return ioat_dma->idx[index];
239 }
240 
241 static inline u64 ioat_chansts(struct ioatdma_chan *ioat_chan)
242 {
243 	return readq(ioat_chan->reg_base + IOAT_CHANSTS_OFFSET);
244 }
245 
246 static inline u64 ioat_chansts_to_addr(u64 status)
247 {
248 	return status & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_ADDR;
249 }
250 
251 static inline u32 ioat_chanerr(struct ioatdma_chan *ioat_chan)
252 {
253 	return readl(ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
254 }
255 
256 static inline void ioat_suspend(struct ioatdma_chan *ioat_chan)
257 {
258 	u8 ver = ioat_chan->ioat_dma->version;
259 
260 	writeb(IOAT_CHANCMD_SUSPEND,
261 	       ioat_chan->reg_base + IOAT_CHANCMD_OFFSET(ver));
262 }
263 
264 static inline void ioat_reset(struct ioatdma_chan *ioat_chan)
265 {
266 	u8 ver = ioat_chan->ioat_dma->version;
267 
268 	writeb(IOAT_CHANCMD_RESET,
269 	       ioat_chan->reg_base + IOAT_CHANCMD_OFFSET(ver));
270 }
271 
272 static inline bool ioat_reset_pending(struct ioatdma_chan *ioat_chan)
273 {
274 	u8 ver = ioat_chan->ioat_dma->version;
275 	u8 cmd;
276 
277 	cmd = readb(ioat_chan->reg_base + IOAT_CHANCMD_OFFSET(ver));
278 	return (cmd & IOAT_CHANCMD_RESET) == IOAT_CHANCMD_RESET;
279 }
280 
281 static inline bool is_ioat_active(unsigned long status)
282 {
283 	return ((status & IOAT_CHANSTS_STATUS) == IOAT_CHANSTS_ACTIVE);
284 }
285 
286 static inline bool is_ioat_idle(unsigned long status)
287 {
288 	return ((status & IOAT_CHANSTS_STATUS) == IOAT_CHANSTS_DONE);
289 }
290 
291 static inline bool is_ioat_halted(unsigned long status)
292 {
293 	return ((status & IOAT_CHANSTS_STATUS) == IOAT_CHANSTS_HALTED);
294 }
295 
296 static inline bool is_ioat_suspended(unsigned long status)
297 {
298 	return ((status & IOAT_CHANSTS_STATUS) == IOAT_CHANSTS_SUSPENDED);
299 }
300 
301 /* channel was fatally programmed */
302 static inline bool is_ioat_bug(unsigned long err)
303 {
304 	return !!err;
305 }
306 
307 
308 static inline u32 ioat_ring_size(struct ioatdma_chan *ioat_chan)
309 {
310 	return 1 << ioat_chan->alloc_order;
311 }
312 
313 /* count of descriptors in flight with the engine */
314 static inline u16 ioat_ring_active(struct ioatdma_chan *ioat_chan)
315 {
316 	return CIRC_CNT(ioat_chan->head, ioat_chan->tail,
317 			ioat_ring_size(ioat_chan));
318 }
319 
320 /* count of descriptors pending submission to hardware */
321 static inline u16 ioat_ring_pending(struct ioatdma_chan *ioat_chan)
322 {
323 	return CIRC_CNT(ioat_chan->head, ioat_chan->issued,
324 			ioat_ring_size(ioat_chan));
325 }
326 
327 static inline u32 ioat_ring_space(struct ioatdma_chan *ioat_chan)
328 {
329 	return ioat_ring_size(ioat_chan) - ioat_ring_active(ioat_chan);
330 }
331 
332 static inline u16
333 ioat_xferlen_to_descs(struct ioatdma_chan *ioat_chan, size_t len)
334 {
335 	u16 num_descs = len >> ioat_chan->xfercap_log;
336 
337 	num_descs += !!(len & ((1 << ioat_chan->xfercap_log) - 1));
338 	return num_descs;
339 }
340 
341 static inline struct ioat_ring_ent *
342 ioat_get_ring_ent(struct ioatdma_chan *ioat_chan, u16 idx)
343 {
344 	return ioat_chan->ring[idx & (ioat_ring_size(ioat_chan) - 1)];
345 }
346 
347 static inline void
348 ioat_set_chainaddr(struct ioatdma_chan *ioat_chan, u64 addr)
349 {
350 	writel(addr & 0x00000000FFFFFFFF,
351 	       ioat_chan->reg_base + IOAT2_CHAINADDR_OFFSET_LOW);
352 	writel(addr >> 32,
353 	       ioat_chan->reg_base + IOAT2_CHAINADDR_OFFSET_HIGH);
354 }
355 
356 /* IOAT Prep functions */
357 struct dma_async_tx_descriptor *
358 ioat_dma_prep_memcpy_lock(struct dma_chan *c, dma_addr_t dma_dest,
359 			   dma_addr_t dma_src, size_t len, unsigned long flags);
360 struct dma_async_tx_descriptor *
361 ioat_prep_interrupt_lock(struct dma_chan *c, unsigned long flags);
362 struct dma_async_tx_descriptor *
363 ioat_prep_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
364 	       unsigned int src_cnt, size_t len, unsigned long flags);
365 struct dma_async_tx_descriptor *
366 ioat_prep_xor_val(struct dma_chan *chan, dma_addr_t *src,
367 		    unsigned int src_cnt, size_t len,
368 		    enum sum_check_flags *result, unsigned long flags);
369 struct dma_async_tx_descriptor *
370 ioat_prep_pq(struct dma_chan *chan, dma_addr_t *dst, dma_addr_t *src,
371 	      unsigned int src_cnt, const unsigned char *scf, size_t len,
372 	      unsigned long flags);
373 struct dma_async_tx_descriptor *
374 ioat_prep_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
375 		  unsigned int src_cnt, const unsigned char *scf, size_t len,
376 		  enum sum_check_flags *pqres, unsigned long flags);
377 struct dma_async_tx_descriptor *
378 ioat_prep_pqxor(struct dma_chan *chan, dma_addr_t dst, dma_addr_t *src,
379 		 unsigned int src_cnt, size_t len, unsigned long flags);
380 struct dma_async_tx_descriptor *
381 ioat_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src,
382 		     unsigned int src_cnt, size_t len,
383 		     enum sum_check_flags *result, unsigned long flags);
384 
385 /* IOAT Operation functions */
386 irqreturn_t ioat_dma_do_interrupt(int irq, void *data);
387 irqreturn_t ioat_dma_do_interrupt_msix(int irq, void *data);
388 struct ioat_ring_ent **
389 ioat_alloc_ring(struct dma_chan *c, int order, gfp_t flags);
390 void ioat_start_null_desc(struct ioatdma_chan *ioat_chan);
391 void ioat_free_ring_ent(struct ioat_ring_ent *desc, struct dma_chan *chan);
392 int ioat_reset_hw(struct ioatdma_chan *ioat_chan);
393 enum dma_status
394 ioat_tx_status(struct dma_chan *c, dma_cookie_t cookie,
395 		struct dma_tx_state *txstate);
396 void ioat_cleanup_event(unsigned long data);
397 void ioat_timer_event(struct timer_list *t);
398 int ioat_check_space_lock(struct ioatdma_chan *ioat_chan, int num_descs);
399 void ioat_issue_pending(struct dma_chan *chan);
400 
401 /* IOAT Init functions */
402 bool is_bwd_ioat(struct pci_dev *pdev);
403 struct dca_provider *ioat_dca_init(struct pci_dev *pdev, void __iomem *iobase);
404 void ioat_kobject_add(struct ioatdma_device *ioat_dma, struct kobj_type *type);
405 void ioat_kobject_del(struct ioatdma_device *ioat_dma);
406 int ioat_dma_setup_interrupts(struct ioatdma_device *ioat_dma);
407 void ioat_stop(struct ioatdma_chan *ioat_chan);
408 #endif /* IOATDMA_H */
409