1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright 2016-17 IBM Corp.
4  */
5 
6 #define pr_fmt(fmt) "vas: " fmt
7 
8 #include <linux/types.h>
9 #include <linux/mutex.h>
10 #include <linux/slab.h>
11 #include <linux/io.h>
12 #include <linux/log2.h>
13 #include <linux/rcupdate.h>
14 #include <linux/cred.h>
15 #include <linux/sched/mm.h>
16 #include <linux/mmu_context.h>
17 #include <asm/switch_to.h>
18 #include <asm/ppc-opcode.h>
19 #include "vas.h"
20 #include "copy-paste.h"
21 
22 #define CREATE_TRACE_POINTS
23 #include "vas-trace.h"
24 
25 /*
26  * Compute the paste address region for the window @window using the
27  * ->paste_base_addr and ->paste_win_id_shift we got from device tree.
28  */
29 void vas_win_paste_addr(struct vas_window *window, u64 *addr, int *len)
30 {
31 	int winid;
32 	u64 base, shift;
33 
34 	base = window->vinst->paste_base_addr;
35 	shift = window->vinst->paste_win_id_shift;
36 	winid = window->winid;
37 
38 	*addr  = base + (winid << shift);
39 	if (len)
40 		*len = PAGE_SIZE;
41 
42 	pr_debug("Txwin #%d: Paste addr 0x%llx\n", winid, *addr);
43 }
44 
45 static inline void get_hvwc_mmio_bar(struct vas_window *window,
46 			u64 *start, int *len)
47 {
48 	u64 pbaddr;
49 
50 	pbaddr = window->vinst->hvwc_bar_start;
51 	*start = pbaddr + window->winid * VAS_HVWC_SIZE;
52 	*len = VAS_HVWC_SIZE;
53 }
54 
55 static inline void get_uwc_mmio_bar(struct vas_window *window,
56 			u64 *start, int *len)
57 {
58 	u64 pbaddr;
59 
60 	pbaddr = window->vinst->uwc_bar_start;
61 	*start = pbaddr + window->winid * VAS_UWC_SIZE;
62 	*len = VAS_UWC_SIZE;
63 }
64 
65 /*
66  * Map the paste bus address of the given send window into kernel address
67  * space. Unlike MMIO regions (map_mmio_region() below), paste region must
68  * be mapped cache-able and is only applicable to send windows.
69  */
70 static void *map_paste_region(struct vas_window *txwin)
71 {
72 	int len;
73 	void *map;
74 	char *name;
75 	u64 start;
76 
77 	name = kasprintf(GFP_KERNEL, "window-v%d-w%d", txwin->vinst->vas_id,
78 				txwin->winid);
79 	if (!name)
80 		goto free_name;
81 
82 	txwin->paste_addr_name = name;
83 	vas_win_paste_addr(txwin, &start, &len);
84 
85 	if (!request_mem_region(start, len, name)) {
86 		pr_devel("%s(): request_mem_region(0x%llx, %d) failed\n",
87 				__func__, start, len);
88 		goto free_name;
89 	}
90 
91 	map = ioremap_cache(start, len);
92 	if (!map) {
93 		pr_devel("%s(): ioremap_cache(0x%llx, %d) failed\n", __func__,
94 				start, len);
95 		goto free_name;
96 	}
97 
98 	pr_devel("Mapped paste addr 0x%llx to kaddr 0x%p\n", start, map);
99 	return map;
100 
101 free_name:
102 	kfree(name);
103 	return ERR_PTR(-ENOMEM);
104 }
105 
106 static void *map_mmio_region(char *name, u64 start, int len)
107 {
108 	void *map;
109 
110 	if (!request_mem_region(start, len, name)) {
111 		pr_devel("%s(): request_mem_region(0x%llx, %d) failed\n",
112 				__func__, start, len);
113 		return NULL;
114 	}
115 
116 	map = ioremap(start, len);
117 	if (!map) {
118 		pr_devel("%s(): ioremap(0x%llx, %d) failed\n", __func__, start,
119 				len);
120 		return NULL;
121 	}
122 
123 	return map;
124 }
125 
126 static void unmap_region(void *addr, u64 start, int len)
127 {
128 	iounmap(addr);
129 	release_mem_region((phys_addr_t)start, len);
130 }
131 
132 /*
133  * Unmap the paste address region for a window.
134  */
135 static void unmap_paste_region(struct vas_window *window)
136 {
137 	int len;
138 	u64 busaddr_start;
139 
140 	if (window->paste_kaddr) {
141 		vas_win_paste_addr(window, &busaddr_start, &len);
142 		unmap_region(window->paste_kaddr, busaddr_start, len);
143 		window->paste_kaddr = NULL;
144 		kfree(window->paste_addr_name);
145 		window->paste_addr_name = NULL;
146 	}
147 }
148 
149 /*
150  * Unmap the MMIO regions for a window. Hold the vas_mutex so we don't
151  * unmap when the window's debugfs dir is in use. This serializes close
152  * of a window even on another VAS instance but since its not a critical
153  * path, just minimize the time we hold the mutex for now. We can add
154  * a per-instance mutex later if necessary.
155  */
156 static void unmap_winctx_mmio_bars(struct vas_window *window)
157 {
158 	int len;
159 	void *uwc_map;
160 	void *hvwc_map;
161 	u64 busaddr_start;
162 
163 	mutex_lock(&vas_mutex);
164 
165 	hvwc_map = window->hvwc_map;
166 	window->hvwc_map = NULL;
167 
168 	uwc_map = window->uwc_map;
169 	window->uwc_map = NULL;
170 
171 	mutex_unlock(&vas_mutex);
172 
173 	if (hvwc_map) {
174 		get_hvwc_mmio_bar(window, &busaddr_start, &len);
175 		unmap_region(hvwc_map, busaddr_start, len);
176 	}
177 
178 	if (uwc_map) {
179 		get_uwc_mmio_bar(window, &busaddr_start, &len);
180 		unmap_region(uwc_map, busaddr_start, len);
181 	}
182 }
183 
184 /*
185  * Find the Hypervisor Window Context (HVWC) MMIO Base Address Region and the
186  * OS/User Window Context (UWC) MMIO Base Address Region for the given window.
187  * Map these bus addresses and save the mapped kernel addresses in @window.
188  */
189 int map_winctx_mmio_bars(struct vas_window *window)
190 {
191 	int len;
192 	u64 start;
193 
194 	get_hvwc_mmio_bar(window, &start, &len);
195 	window->hvwc_map = map_mmio_region("HVWCM_Window", start, len);
196 
197 	get_uwc_mmio_bar(window, &start, &len);
198 	window->uwc_map = map_mmio_region("UWCM_Window", start, len);
199 
200 	if (!window->hvwc_map || !window->uwc_map) {
201 		unmap_winctx_mmio_bars(window);
202 		return -1;
203 	}
204 
205 	return 0;
206 }
207 
208 /*
209  * Reset all valid registers in the HV and OS/User Window Contexts for
210  * the window identified by @window.
211  *
212  * NOTE: We cannot really use a for loop to reset window context. Not all
213  *	 offsets in a window context are valid registers and the valid
214  *	 registers are not sequential. And, we can only write to offsets
215  *	 with valid registers.
216  */
217 void reset_window_regs(struct vas_window *window)
218 {
219 	write_hvwc_reg(window, VREG(LPID), 0ULL);
220 	write_hvwc_reg(window, VREG(PID), 0ULL);
221 	write_hvwc_reg(window, VREG(XLATE_MSR), 0ULL);
222 	write_hvwc_reg(window, VREG(XLATE_LPCR), 0ULL);
223 	write_hvwc_reg(window, VREG(XLATE_CTL), 0ULL);
224 	write_hvwc_reg(window, VREG(AMR), 0ULL);
225 	write_hvwc_reg(window, VREG(SEIDR), 0ULL);
226 	write_hvwc_reg(window, VREG(FAULT_TX_WIN), 0ULL);
227 	write_hvwc_reg(window, VREG(OSU_INTR_SRC_RA), 0ULL);
228 	write_hvwc_reg(window, VREG(HV_INTR_SRC_RA), 0ULL);
229 	write_hvwc_reg(window, VREG(PSWID), 0ULL);
230 	write_hvwc_reg(window, VREG(LFIFO_BAR), 0ULL);
231 	write_hvwc_reg(window, VREG(LDATA_STAMP_CTL), 0ULL);
232 	write_hvwc_reg(window, VREG(LDMA_CACHE_CTL), 0ULL);
233 	write_hvwc_reg(window, VREG(LRFIFO_PUSH), 0ULL);
234 	write_hvwc_reg(window, VREG(CURR_MSG_COUNT), 0ULL);
235 	write_hvwc_reg(window, VREG(LNOTIFY_AFTER_COUNT), 0ULL);
236 	write_hvwc_reg(window, VREG(LRX_WCRED), 0ULL);
237 	write_hvwc_reg(window, VREG(LRX_WCRED_ADDER), 0ULL);
238 	write_hvwc_reg(window, VREG(TX_WCRED), 0ULL);
239 	write_hvwc_reg(window, VREG(TX_WCRED_ADDER), 0ULL);
240 	write_hvwc_reg(window, VREG(LFIFO_SIZE), 0ULL);
241 	write_hvwc_reg(window, VREG(WINCTL), 0ULL);
242 	write_hvwc_reg(window, VREG(WIN_STATUS), 0ULL);
243 	write_hvwc_reg(window, VREG(WIN_CTX_CACHING_CTL), 0ULL);
244 	write_hvwc_reg(window, VREG(TX_RSVD_BUF_COUNT), 0ULL);
245 	write_hvwc_reg(window, VREG(LRFIFO_WIN_PTR), 0ULL);
246 	write_hvwc_reg(window, VREG(LNOTIFY_CTL), 0ULL);
247 	write_hvwc_reg(window, VREG(LNOTIFY_PID), 0ULL);
248 	write_hvwc_reg(window, VREG(LNOTIFY_LPID), 0ULL);
249 	write_hvwc_reg(window, VREG(LNOTIFY_TID), 0ULL);
250 	write_hvwc_reg(window, VREG(LNOTIFY_SCOPE), 0ULL);
251 	write_hvwc_reg(window, VREG(NX_UTIL_ADDER), 0ULL);
252 
253 	/* Skip read-only registers: NX_UTIL and NX_UTIL_SE */
254 
255 	/*
256 	 * The send and receive window credit adder registers are also
257 	 * accessible from HVWC and have been initialized above. We don't
258 	 * need to initialize from the OS/User Window Context, so skip
259 	 * following calls:
260 	 *
261 	 *	write_uwc_reg(window, VREG(TX_WCRED_ADDER), 0ULL);
262 	 *	write_uwc_reg(window, VREG(LRX_WCRED_ADDER), 0ULL);
263 	 */
264 }
265 
266 /*
267  * Initialize window context registers related to Address Translation.
268  * These registers are common to send/receive windows although they
269  * differ for user/kernel windows. As we resolve the TODOs we may
270  * want to add fields to vas_winctx and move the initialization to
271  * init_vas_winctx_regs().
272  */
273 static void init_xlate_regs(struct vas_window *window, bool user_win)
274 {
275 	u64 lpcr, val;
276 
277 	/*
278 	 * MSR_TA, MSR_US are false for both kernel and user.
279 	 * MSR_DR and MSR_PR are false for kernel.
280 	 */
281 	val = 0ULL;
282 	val = SET_FIELD(VAS_XLATE_MSR_HV, val, 1);
283 	val = SET_FIELD(VAS_XLATE_MSR_SF, val, 1);
284 	if (user_win) {
285 		val = SET_FIELD(VAS_XLATE_MSR_DR, val, 1);
286 		val = SET_FIELD(VAS_XLATE_MSR_PR, val, 1);
287 	}
288 	write_hvwc_reg(window, VREG(XLATE_MSR), val);
289 
290 	lpcr = mfspr(SPRN_LPCR);
291 	val = 0ULL;
292 	/*
293 	 * NOTE: From Section 5.7.8.1 Segment Lookaside Buffer of the
294 	 *	 Power ISA, v3.0B, Page size encoding is 0 = 4KB, 5 = 64KB.
295 	 *
296 	 * NOTE: From Section 1.3.1, Address Translation Context of the
297 	 *	 Nest MMU Workbook, LPCR_SC should be 0 for Power9.
298 	 */
299 	val = SET_FIELD(VAS_XLATE_LPCR_PAGE_SIZE, val, 5);
300 	val = SET_FIELD(VAS_XLATE_LPCR_ISL, val, lpcr & LPCR_ISL);
301 	val = SET_FIELD(VAS_XLATE_LPCR_TC, val, lpcr & LPCR_TC);
302 	val = SET_FIELD(VAS_XLATE_LPCR_SC, val, 0);
303 	write_hvwc_reg(window, VREG(XLATE_LPCR), val);
304 
305 	/*
306 	 * Section 1.3.1 (Address translation Context) of NMMU workbook.
307 	 *	0b00	Hashed Page Table mode
308 	 *	0b01	Reserved
309 	 *	0b10	Radix on HPT
310 	 *	0b11	Radix on Radix
311 	 */
312 	val = 0ULL;
313 	val = SET_FIELD(VAS_XLATE_MODE, val, radix_enabled() ? 3 : 2);
314 	write_hvwc_reg(window, VREG(XLATE_CTL), val);
315 
316 	/*
317 	 * TODO: Can we mfspr(AMR) even for user windows?
318 	 */
319 	val = 0ULL;
320 	val = SET_FIELD(VAS_AMR, val, mfspr(SPRN_AMR));
321 	write_hvwc_reg(window, VREG(AMR), val);
322 
323 	val = 0ULL;
324 	val = SET_FIELD(VAS_SEIDR, val, 0);
325 	write_hvwc_reg(window, VREG(SEIDR), val);
326 }
327 
328 /*
329  * Initialize Reserved Send Buffer Count for the send window. It involves
330  * writing to the register, reading it back to confirm that the hardware
331  * has enough buffers to reserve. See section 1.3.1.2.1 of VAS workbook.
332  *
333  * Since we can only make a best-effort attempt to fulfill the request,
334  * we don't return any errors if we cannot.
335  *
336  * TODO: Reserved (aka dedicated) send buffers are not supported yet.
337  */
338 static void init_rsvd_tx_buf_count(struct vas_window *txwin,
339 				struct vas_winctx *winctx)
340 {
341 	write_hvwc_reg(txwin, VREG(TX_RSVD_BUF_COUNT), 0ULL);
342 }
343 
344 /*
345  * init_winctx_regs()
346  *	Initialize window context registers for a receive window.
347  *	Except for caching control and marking window open, the registers
348  *	are initialized in the order listed in Section 3.1.4 (Window Context
349  *	Cache Register Details) of the VAS workbook although they don't need
350  *	to be.
351  *
352  * Design note: For NX receive windows, NX allocates the FIFO buffer in OPAL
353  *	(so that it can get a large contiguous area) and passes that buffer
354  *	to kernel via device tree. We now write that buffer address to the
355  *	FIFO BAR. Would it make sense to do this all in OPAL? i.e have OPAL
356  *	write the per-chip RX FIFO addresses to the windows during boot-up
357  *	as a one-time task? That could work for NX but what about other
358  *	receivers?  Let the receivers tell us the rx-fifo buffers for now.
359  */
360 int init_winctx_regs(struct vas_window *window, struct vas_winctx *winctx)
361 {
362 	u64 val;
363 	int fifo_size;
364 
365 	reset_window_regs(window);
366 
367 	val = 0ULL;
368 	val = SET_FIELD(VAS_LPID, val, winctx->lpid);
369 	write_hvwc_reg(window, VREG(LPID), val);
370 
371 	val = 0ULL;
372 	val = SET_FIELD(VAS_PID_ID, val, winctx->pidr);
373 	write_hvwc_reg(window, VREG(PID), val);
374 
375 	init_xlate_regs(window, winctx->user_win);
376 
377 	val = 0ULL;
378 	val = SET_FIELD(VAS_FAULT_TX_WIN, val, winctx->fault_win_id);
379 	write_hvwc_reg(window, VREG(FAULT_TX_WIN), val);
380 
381 	/* In PowerNV, interrupts go to HV. */
382 	write_hvwc_reg(window, VREG(OSU_INTR_SRC_RA), 0ULL);
383 
384 	val = 0ULL;
385 	val = SET_FIELD(VAS_HV_INTR_SRC_RA, val, winctx->irq_port);
386 	write_hvwc_reg(window, VREG(HV_INTR_SRC_RA), val);
387 
388 	val = 0ULL;
389 	val = SET_FIELD(VAS_PSWID_EA_HANDLE, val, winctx->pswid);
390 	write_hvwc_reg(window, VREG(PSWID), val);
391 
392 	write_hvwc_reg(window, VREG(SPARE1), 0ULL);
393 	write_hvwc_reg(window, VREG(SPARE2), 0ULL);
394 	write_hvwc_reg(window, VREG(SPARE3), 0ULL);
395 
396 	/*
397 	 * NOTE: VAS expects the FIFO address to be copied into the LFIFO_BAR
398 	 *	 register as is - do NOT shift the address into VAS_LFIFO_BAR
399 	 *	 bit fields! Ok to set the page migration select fields -
400 	 *	 VAS ignores the lower 10+ bits in the address anyway, because
401 	 *	 the minimum FIFO size is 1K?
402 	 *
403 	 * See also: Design note in function header.
404 	 */
405 	val = __pa(winctx->rx_fifo);
406 	val = SET_FIELD(VAS_PAGE_MIGRATION_SELECT, val, 0);
407 	write_hvwc_reg(window, VREG(LFIFO_BAR), val);
408 
409 	val = 0ULL;
410 	val = SET_FIELD(VAS_LDATA_STAMP, val, winctx->data_stamp);
411 	write_hvwc_reg(window, VREG(LDATA_STAMP_CTL), val);
412 
413 	val = 0ULL;
414 	val = SET_FIELD(VAS_LDMA_TYPE, val, winctx->dma_type);
415 	val = SET_FIELD(VAS_LDMA_FIFO_DISABLE, val, winctx->fifo_disable);
416 	write_hvwc_reg(window, VREG(LDMA_CACHE_CTL), val);
417 
418 	write_hvwc_reg(window, VREG(LRFIFO_PUSH), 0ULL);
419 	write_hvwc_reg(window, VREG(CURR_MSG_COUNT), 0ULL);
420 	write_hvwc_reg(window, VREG(LNOTIFY_AFTER_COUNT), 0ULL);
421 
422 	val = 0ULL;
423 	val = SET_FIELD(VAS_LRX_WCRED, val, winctx->wcreds_max);
424 	write_hvwc_reg(window, VREG(LRX_WCRED), val);
425 
426 	val = 0ULL;
427 	val = SET_FIELD(VAS_TX_WCRED, val, winctx->wcreds_max);
428 	write_hvwc_reg(window, VREG(TX_WCRED), val);
429 
430 	write_hvwc_reg(window, VREG(LRX_WCRED_ADDER), 0ULL);
431 	write_hvwc_reg(window, VREG(TX_WCRED_ADDER), 0ULL);
432 
433 	fifo_size = winctx->rx_fifo_size / 1024;
434 
435 	val = 0ULL;
436 	val = SET_FIELD(VAS_LFIFO_SIZE, val, ilog2(fifo_size));
437 	write_hvwc_reg(window, VREG(LFIFO_SIZE), val);
438 
439 	/* Update window control and caching control registers last so
440 	 * we mark the window open only after fully initializing it and
441 	 * pushing context to cache.
442 	 */
443 
444 	write_hvwc_reg(window, VREG(WIN_STATUS), 0ULL);
445 
446 	init_rsvd_tx_buf_count(window, winctx);
447 
448 	/* for a send window, point to the matching receive window */
449 	val = 0ULL;
450 	val = SET_FIELD(VAS_LRX_WIN_ID, val, winctx->rx_win_id);
451 	write_hvwc_reg(window, VREG(LRFIFO_WIN_PTR), val);
452 
453 	write_hvwc_reg(window, VREG(SPARE4), 0ULL);
454 
455 	val = 0ULL;
456 	val = SET_FIELD(VAS_NOTIFY_DISABLE, val, winctx->notify_disable);
457 	val = SET_FIELD(VAS_INTR_DISABLE, val, winctx->intr_disable);
458 	val = SET_FIELD(VAS_NOTIFY_EARLY, val, winctx->notify_early);
459 	val = SET_FIELD(VAS_NOTIFY_OSU_INTR, val, winctx->notify_os_intr_reg);
460 	write_hvwc_reg(window, VREG(LNOTIFY_CTL), val);
461 
462 	val = 0ULL;
463 	val = SET_FIELD(VAS_LNOTIFY_PID, val, winctx->lnotify_pid);
464 	write_hvwc_reg(window, VREG(LNOTIFY_PID), val);
465 
466 	val = 0ULL;
467 	val = SET_FIELD(VAS_LNOTIFY_LPID, val, winctx->lnotify_lpid);
468 	write_hvwc_reg(window, VREG(LNOTIFY_LPID), val);
469 
470 	val = 0ULL;
471 	val = SET_FIELD(VAS_LNOTIFY_TID, val, winctx->lnotify_tid);
472 	write_hvwc_reg(window, VREG(LNOTIFY_TID), val);
473 
474 	val = 0ULL;
475 	val = SET_FIELD(VAS_LNOTIFY_MIN_SCOPE, val, winctx->min_scope);
476 	val = SET_FIELD(VAS_LNOTIFY_MAX_SCOPE, val, winctx->max_scope);
477 	write_hvwc_reg(window, VREG(LNOTIFY_SCOPE), val);
478 
479 	/* Skip read-only registers NX_UTIL and NX_UTIL_SE */
480 
481 	write_hvwc_reg(window, VREG(SPARE5), 0ULL);
482 	write_hvwc_reg(window, VREG(NX_UTIL_ADDER), 0ULL);
483 	write_hvwc_reg(window, VREG(SPARE6), 0ULL);
484 
485 	/* Finally, push window context to memory and... */
486 	val = 0ULL;
487 	val = SET_FIELD(VAS_PUSH_TO_MEM, val, 1);
488 	write_hvwc_reg(window, VREG(WIN_CTX_CACHING_CTL), val);
489 
490 	/* ... mark the window open for business */
491 	val = 0ULL;
492 	val = SET_FIELD(VAS_WINCTL_REJ_NO_CREDIT, val, winctx->rej_no_credit);
493 	val = SET_FIELD(VAS_WINCTL_PIN, val, winctx->pin_win);
494 	val = SET_FIELD(VAS_WINCTL_TX_WCRED_MODE, val, winctx->tx_wcred_mode);
495 	val = SET_FIELD(VAS_WINCTL_RX_WCRED_MODE, val, winctx->rx_wcred_mode);
496 	val = SET_FIELD(VAS_WINCTL_TX_WORD_MODE, val, winctx->tx_word_mode);
497 	val = SET_FIELD(VAS_WINCTL_RX_WORD_MODE, val, winctx->rx_word_mode);
498 	val = SET_FIELD(VAS_WINCTL_FAULT_WIN, val, winctx->fault_win);
499 	val = SET_FIELD(VAS_WINCTL_NX_WIN, val, winctx->nx_win);
500 	val = SET_FIELD(VAS_WINCTL_OPEN, val, 1);
501 	write_hvwc_reg(window, VREG(WINCTL), val);
502 
503 	return 0;
504 }
505 
506 static void vas_release_window_id(struct ida *ida, int winid)
507 {
508 	ida_free(ida, winid);
509 }
510 
511 static int vas_assign_window_id(struct ida *ida)
512 {
513 	int winid = ida_alloc_max(ida, VAS_WINDOWS_PER_CHIP - 1, GFP_KERNEL);
514 
515 	if (winid == -ENOSPC) {
516 		pr_err("Too many (%d) open windows\n", VAS_WINDOWS_PER_CHIP);
517 		return -EAGAIN;
518 	}
519 
520 	return winid;
521 }
522 
523 static void vas_window_free(struct vas_window *window)
524 {
525 	int winid = window->winid;
526 	struct vas_instance *vinst = window->vinst;
527 
528 	unmap_winctx_mmio_bars(window);
529 
530 	vas_window_free_dbgdir(window);
531 
532 	kfree(window);
533 
534 	vas_release_window_id(&vinst->ida, winid);
535 }
536 
537 static struct vas_window *vas_window_alloc(struct vas_instance *vinst)
538 {
539 	int winid;
540 	struct vas_window *window;
541 
542 	winid = vas_assign_window_id(&vinst->ida);
543 	if (winid < 0)
544 		return ERR_PTR(winid);
545 
546 	window = kzalloc(sizeof(*window), GFP_KERNEL);
547 	if (!window)
548 		goto out_free;
549 
550 	window->vinst = vinst;
551 	window->winid = winid;
552 
553 	if (map_winctx_mmio_bars(window))
554 		goto out_free;
555 
556 	vas_window_init_dbgdir(window);
557 
558 	return window;
559 
560 out_free:
561 	kfree(window);
562 	vas_release_window_id(&vinst->ida, winid);
563 	return ERR_PTR(-ENOMEM);
564 }
565 
566 static void put_rx_win(struct vas_window *rxwin)
567 {
568 	/* Better not be a send window! */
569 	WARN_ON_ONCE(rxwin->tx_win);
570 
571 	atomic_dec(&rxwin->num_txwins);
572 }
573 
574 /*
575  * Find the user space receive window given the @pswid.
576  *      - We must have a valid vasid and it must belong to this instance.
577  *        (so both send and receive windows are on the same VAS instance)
578  *      - The window must refer to an OPEN, FTW, RECEIVE window.
579  *
580  * NOTE: We access ->windows[] table and assume that vinst->mutex is held.
581  */
582 static struct vas_window *get_user_rxwin(struct vas_instance *vinst, u32 pswid)
583 {
584 	int vasid, winid;
585 	struct vas_window *rxwin;
586 
587 	decode_pswid(pswid, &vasid, &winid);
588 
589 	if (vinst->vas_id != vasid)
590 		return ERR_PTR(-EINVAL);
591 
592 	rxwin = vinst->windows[winid];
593 
594 	if (!rxwin || rxwin->tx_win || rxwin->cop != VAS_COP_TYPE_FTW)
595 		return ERR_PTR(-EINVAL);
596 
597 	return rxwin;
598 }
599 
600 /*
601  * Get the VAS receive window associated with NX engine identified
602  * by @cop and if applicable, @pswid.
603  *
604  * See also function header of set_vinst_win().
605  */
606 static struct vas_window *get_vinst_rxwin(struct vas_instance *vinst,
607 			enum vas_cop_type cop, u32 pswid)
608 {
609 	struct vas_window *rxwin;
610 
611 	mutex_lock(&vinst->mutex);
612 
613 	if (cop == VAS_COP_TYPE_FTW)
614 		rxwin = get_user_rxwin(vinst, pswid);
615 	else
616 		rxwin = vinst->rxwin[cop] ?: ERR_PTR(-EINVAL);
617 
618 	if (!IS_ERR(rxwin))
619 		atomic_inc(&rxwin->num_txwins);
620 
621 	mutex_unlock(&vinst->mutex);
622 
623 	return rxwin;
624 }
625 
626 /*
627  * We have two tables of windows in a VAS instance. The first one,
628  * ->windows[], contains all the windows in the instance and allows
629  * looking up a window by its id. It is used to look up send windows
630  * during fault handling and receive windows when pairing user space
631  * send/receive windows.
632  *
633  * The second table, ->rxwin[], contains receive windows that are
634  * associated with NX engines. This table has VAS_COP_TYPE_MAX
635  * entries and is used to look up a receive window by its
636  * coprocessor type.
637  *
638  * Here, we save @window in the ->windows[] table. If it is a receive
639  * window, we also save the window in the ->rxwin[] table.
640  */
641 static void set_vinst_win(struct vas_instance *vinst,
642 			struct vas_window *window)
643 {
644 	int id = window->winid;
645 
646 	mutex_lock(&vinst->mutex);
647 
648 	/*
649 	 * There should only be one receive window for a coprocessor type
650 	 * unless its a user (FTW) window.
651 	 */
652 	if (!window->user_win && !window->tx_win) {
653 		WARN_ON_ONCE(vinst->rxwin[window->cop]);
654 		vinst->rxwin[window->cop] = window;
655 	}
656 
657 	WARN_ON_ONCE(vinst->windows[id] != NULL);
658 	vinst->windows[id] = window;
659 
660 	mutex_unlock(&vinst->mutex);
661 }
662 
663 /*
664  * Clear this window from the table(s) of windows for this VAS instance.
665  * See also function header of set_vinst_win().
666  */
667 static void clear_vinst_win(struct vas_window *window)
668 {
669 	int id = window->winid;
670 	struct vas_instance *vinst = window->vinst;
671 
672 	mutex_lock(&vinst->mutex);
673 
674 	if (!window->user_win && !window->tx_win) {
675 		WARN_ON_ONCE(!vinst->rxwin[window->cop]);
676 		vinst->rxwin[window->cop] = NULL;
677 	}
678 
679 	WARN_ON_ONCE(vinst->windows[id] != window);
680 	vinst->windows[id] = NULL;
681 
682 	mutex_unlock(&vinst->mutex);
683 }
684 
685 static void init_winctx_for_rxwin(struct vas_window *rxwin,
686 			struct vas_rx_win_attr *rxattr,
687 			struct vas_winctx *winctx)
688 {
689 	/*
690 	 * We first zero (memset()) all fields and only set non-zero fields.
691 	 * Following fields are 0/false but maybe deserve a comment:
692 	 *
693 	 *	->notify_os_intr_reg	In powerNV, send intrs to HV
694 	 *	->notify_disable	False for NX windows
695 	 *	->intr_disable		False for Fault Windows
696 	 *	->xtra_write		False for NX windows
697 	 *	->notify_early		NA for NX windows
698 	 *	->rsvd_txbuf_count	NA for Rx windows
699 	 *	->lpid, ->pid, ->tid	NA for Rx windows
700 	 */
701 
702 	memset(winctx, 0, sizeof(struct vas_winctx));
703 
704 	winctx->rx_fifo = rxattr->rx_fifo;
705 	winctx->rx_fifo_size = rxattr->rx_fifo_size;
706 	winctx->wcreds_max = rxwin->wcreds_max;
707 	winctx->pin_win = rxattr->pin_win;
708 
709 	winctx->nx_win = rxattr->nx_win;
710 	winctx->fault_win = rxattr->fault_win;
711 	winctx->user_win = rxattr->user_win;
712 	winctx->rej_no_credit = rxattr->rej_no_credit;
713 	winctx->rx_word_mode = rxattr->rx_win_ord_mode;
714 	winctx->tx_word_mode = rxattr->tx_win_ord_mode;
715 	winctx->rx_wcred_mode = rxattr->rx_wcred_mode;
716 	winctx->tx_wcred_mode = rxattr->tx_wcred_mode;
717 	winctx->notify_early = rxattr->notify_early;
718 
719 	if (winctx->nx_win) {
720 		winctx->data_stamp = true;
721 		winctx->intr_disable = true;
722 		winctx->pin_win = true;
723 
724 		WARN_ON_ONCE(winctx->fault_win);
725 		WARN_ON_ONCE(!winctx->rx_word_mode);
726 		WARN_ON_ONCE(!winctx->tx_word_mode);
727 		WARN_ON_ONCE(winctx->notify_after_count);
728 	} else if (winctx->fault_win) {
729 		winctx->notify_disable = true;
730 	} else if (winctx->user_win) {
731 		/*
732 		 * Section 1.8.1 Low Latency Core-Core Wake up of
733 		 * the VAS workbook:
734 		 *
735 		 *      - disable credit checks ([tr]x_wcred_mode = false)
736 		 *      - disable FIFO writes
737 		 *      - enable ASB_Notify, disable interrupt
738 		 */
739 		winctx->fifo_disable = true;
740 		winctx->intr_disable = true;
741 		winctx->rx_fifo = NULL;
742 	}
743 
744 	winctx->lnotify_lpid = rxattr->lnotify_lpid;
745 	winctx->lnotify_pid = rxattr->lnotify_pid;
746 	winctx->lnotify_tid = rxattr->lnotify_tid;
747 	winctx->pswid = rxattr->pswid;
748 	winctx->dma_type = VAS_DMA_TYPE_INJECT;
749 	winctx->tc_mode = rxattr->tc_mode;
750 
751 	winctx->min_scope = VAS_SCOPE_LOCAL;
752 	winctx->max_scope = VAS_SCOPE_VECTORED_GROUP;
753 	if (rxwin->vinst->virq)
754 		winctx->irq_port = rxwin->vinst->irq_port;
755 }
756 
757 static bool rx_win_args_valid(enum vas_cop_type cop,
758 			struct vas_rx_win_attr *attr)
759 {
760 	pr_debug("Rxattr: fault %d, notify %d, intr %d, early %d, fifo %d\n",
761 			attr->fault_win, attr->notify_disable,
762 			attr->intr_disable, attr->notify_early,
763 			attr->rx_fifo_size);
764 
765 	if (cop >= VAS_COP_TYPE_MAX)
766 		return false;
767 
768 	if (cop != VAS_COP_TYPE_FTW &&
769 				attr->rx_fifo_size < VAS_RX_FIFO_SIZE_MIN)
770 		return false;
771 
772 	if (attr->rx_fifo_size > VAS_RX_FIFO_SIZE_MAX)
773 		return false;
774 
775 	if (!attr->wcreds_max)
776 		return false;
777 
778 	if (attr->nx_win) {
779 		/* cannot be fault or user window if it is nx */
780 		if (attr->fault_win || attr->user_win)
781 			return false;
782 		/*
783 		 * Section 3.1.4.32: NX Windows must not disable notification,
784 		 *	and must not enable interrupts or early notification.
785 		 */
786 		if (attr->notify_disable || !attr->intr_disable ||
787 				attr->notify_early)
788 			return false;
789 	} else if (attr->fault_win) {
790 		/* cannot be both fault and user window */
791 		if (attr->user_win)
792 			return false;
793 
794 		/*
795 		 * Section 3.1.4.32: Fault windows must disable notification
796 		 *	but not interrupts.
797 		 */
798 		if (!attr->notify_disable || attr->intr_disable)
799 			return false;
800 
801 	} else if (attr->user_win) {
802 		/*
803 		 * User receive windows are only for fast-thread-wakeup
804 		 * (FTW). They don't need a FIFO and must disable interrupts
805 		 */
806 		if (attr->rx_fifo || attr->rx_fifo_size || !attr->intr_disable)
807 			return false;
808 	} else {
809 		/* Rx window must be one of NX or Fault or User window. */
810 		return false;
811 	}
812 
813 	return true;
814 }
815 
816 void vas_init_rx_win_attr(struct vas_rx_win_attr *rxattr, enum vas_cop_type cop)
817 {
818 	memset(rxattr, 0, sizeof(*rxattr));
819 
820 	if (cop == VAS_COP_TYPE_842 || cop == VAS_COP_TYPE_842_HIPRI ||
821 		cop == VAS_COP_TYPE_GZIP || cop == VAS_COP_TYPE_GZIP_HIPRI) {
822 		rxattr->pin_win = true;
823 		rxattr->nx_win = true;
824 		rxattr->fault_win = false;
825 		rxattr->intr_disable = true;
826 		rxattr->rx_wcred_mode = true;
827 		rxattr->tx_wcred_mode = true;
828 		rxattr->rx_win_ord_mode = true;
829 		rxattr->tx_win_ord_mode = true;
830 	} else if (cop == VAS_COP_TYPE_FAULT) {
831 		rxattr->pin_win = true;
832 		rxattr->fault_win = true;
833 		rxattr->notify_disable = true;
834 		rxattr->rx_wcred_mode = true;
835 		rxattr->rx_win_ord_mode = true;
836 		rxattr->rej_no_credit = true;
837 		rxattr->tc_mode = VAS_THRESH_DISABLED;
838 	} else if (cop == VAS_COP_TYPE_FTW) {
839 		rxattr->user_win = true;
840 		rxattr->intr_disable = true;
841 
842 		/*
843 		 * As noted in the VAS Workbook we disable credit checks.
844 		 * If we enable credit checks in the future, we must also
845 		 * implement a mechanism to return the user credits or new
846 		 * paste operations will fail.
847 		 */
848 	}
849 }
850 EXPORT_SYMBOL_GPL(vas_init_rx_win_attr);
851 
852 struct vas_window *vas_rx_win_open(int vasid, enum vas_cop_type cop,
853 			struct vas_rx_win_attr *rxattr)
854 {
855 	struct vas_window *rxwin;
856 	struct vas_winctx winctx;
857 	struct vas_instance *vinst;
858 
859 	trace_vas_rx_win_open(current, vasid, cop, rxattr);
860 
861 	if (!rx_win_args_valid(cop, rxattr))
862 		return ERR_PTR(-EINVAL);
863 
864 	vinst = find_vas_instance(vasid);
865 	if (!vinst) {
866 		pr_devel("vasid %d not found!\n", vasid);
867 		return ERR_PTR(-EINVAL);
868 	}
869 	pr_devel("Found instance %d\n", vasid);
870 
871 	rxwin = vas_window_alloc(vinst);
872 	if (IS_ERR(rxwin)) {
873 		pr_devel("Unable to allocate memory for Rx window\n");
874 		return rxwin;
875 	}
876 
877 	rxwin->tx_win = false;
878 	rxwin->nx_win = rxattr->nx_win;
879 	rxwin->user_win = rxattr->user_win;
880 	rxwin->cop = cop;
881 	rxwin->wcreds_max = rxattr->wcreds_max;
882 
883 	init_winctx_for_rxwin(rxwin, rxattr, &winctx);
884 	init_winctx_regs(rxwin, &winctx);
885 
886 	set_vinst_win(vinst, rxwin);
887 
888 	return rxwin;
889 }
890 EXPORT_SYMBOL_GPL(vas_rx_win_open);
891 
892 void vas_init_tx_win_attr(struct vas_tx_win_attr *txattr, enum vas_cop_type cop)
893 {
894 	memset(txattr, 0, sizeof(*txattr));
895 
896 	if (cop == VAS_COP_TYPE_842 || cop == VAS_COP_TYPE_842_HIPRI ||
897 		cop == VAS_COP_TYPE_GZIP || cop == VAS_COP_TYPE_GZIP_HIPRI) {
898 		txattr->rej_no_credit = false;
899 		txattr->rx_wcred_mode = true;
900 		txattr->tx_wcred_mode = true;
901 		txattr->rx_win_ord_mode = true;
902 		txattr->tx_win_ord_mode = true;
903 	} else if (cop == VAS_COP_TYPE_FTW) {
904 		txattr->user_win = true;
905 	}
906 }
907 EXPORT_SYMBOL_GPL(vas_init_tx_win_attr);
908 
909 static void init_winctx_for_txwin(struct vas_window *txwin,
910 			struct vas_tx_win_attr *txattr,
911 			struct vas_winctx *winctx)
912 {
913 	/*
914 	 * We first zero all fields and only set non-zero ones. Following
915 	 * are some fields set to 0/false for the stated reason:
916 	 *
917 	 *	->notify_os_intr_reg	In powernv, send intrs to HV
918 	 *	->rsvd_txbuf_count	Not supported yet.
919 	 *	->notify_disable	False for NX windows
920 	 *	->xtra_write		False for NX windows
921 	 *	->notify_early		NA for NX windows
922 	 *	->lnotify_lpid		NA for Tx windows
923 	 *	->lnotify_pid		NA for Tx windows
924 	 *	->lnotify_tid		NA for Tx windows
925 	 *	->tx_win_cred_mode	Ignore for now for NX windows
926 	 *	->rx_win_cred_mode	Ignore for now for NX windows
927 	 */
928 	memset(winctx, 0, sizeof(struct vas_winctx));
929 
930 	winctx->wcreds_max = txwin->wcreds_max;
931 
932 	winctx->user_win = txattr->user_win;
933 	winctx->nx_win = txwin->rxwin->nx_win;
934 	winctx->pin_win = txattr->pin_win;
935 	winctx->rej_no_credit = txattr->rej_no_credit;
936 	winctx->rsvd_txbuf_enable = txattr->rsvd_txbuf_enable;
937 
938 	winctx->rx_wcred_mode = txattr->rx_wcred_mode;
939 	winctx->tx_wcred_mode = txattr->tx_wcred_mode;
940 	winctx->rx_word_mode = txattr->rx_win_ord_mode;
941 	winctx->tx_word_mode = txattr->tx_win_ord_mode;
942 	winctx->rsvd_txbuf_count = txattr->rsvd_txbuf_count;
943 
944 	winctx->intr_disable = true;
945 	if (winctx->nx_win)
946 		winctx->data_stamp = true;
947 
948 	winctx->lpid = txattr->lpid;
949 	winctx->pidr = txattr->pidr;
950 	winctx->rx_win_id = txwin->rxwin->winid;
951 	/*
952 	 * IRQ and fault window setup is successful. Set fault window
953 	 * for the send window so that ready to handle faults.
954 	 */
955 	if (txwin->vinst->virq)
956 		winctx->fault_win_id = txwin->vinst->fault_win->winid;
957 
958 	winctx->dma_type = VAS_DMA_TYPE_INJECT;
959 	winctx->tc_mode = txattr->tc_mode;
960 	winctx->min_scope = VAS_SCOPE_LOCAL;
961 	winctx->max_scope = VAS_SCOPE_VECTORED_GROUP;
962 	if (txwin->vinst->virq)
963 		winctx->irq_port = txwin->vinst->irq_port;
964 
965 	winctx->pswid = txattr->pswid ? txattr->pswid :
966 			encode_pswid(txwin->vinst->vas_id, txwin->winid);
967 }
968 
969 static bool tx_win_args_valid(enum vas_cop_type cop,
970 			struct vas_tx_win_attr *attr)
971 {
972 	if (attr->tc_mode != VAS_THRESH_DISABLED)
973 		return false;
974 
975 	if (cop > VAS_COP_TYPE_MAX)
976 		return false;
977 
978 	if (attr->wcreds_max > VAS_TX_WCREDS_MAX)
979 		return false;
980 
981 	if (attr->user_win) {
982 		if (attr->rsvd_txbuf_count)
983 			return false;
984 
985 		if (cop != VAS_COP_TYPE_FTW && cop != VAS_COP_TYPE_GZIP &&
986 			cop != VAS_COP_TYPE_GZIP_HIPRI)
987 			return false;
988 	}
989 
990 	return true;
991 }
992 
993 struct vas_window *vas_tx_win_open(int vasid, enum vas_cop_type cop,
994 			struct vas_tx_win_attr *attr)
995 {
996 	int rc;
997 	struct vas_window *txwin;
998 	struct vas_window *rxwin;
999 	struct vas_winctx winctx;
1000 	struct vas_instance *vinst;
1001 
1002 	trace_vas_tx_win_open(current, vasid, cop, attr);
1003 
1004 	if (!tx_win_args_valid(cop, attr))
1005 		return ERR_PTR(-EINVAL);
1006 
1007 	/*
1008 	 * If caller did not specify a vasid but specified the PSWID of a
1009 	 * receive window (applicable only to FTW windows), use the vasid
1010 	 * from that receive window.
1011 	 */
1012 	if (vasid == -1 && attr->pswid)
1013 		decode_pswid(attr->pswid, &vasid, NULL);
1014 
1015 	vinst = find_vas_instance(vasid);
1016 	if (!vinst) {
1017 		pr_devel("vasid %d not found!\n", vasid);
1018 		return ERR_PTR(-EINVAL);
1019 	}
1020 
1021 	rxwin = get_vinst_rxwin(vinst, cop, attr->pswid);
1022 	if (IS_ERR(rxwin)) {
1023 		pr_devel("No RxWin for vasid %d, cop %d\n", vasid, cop);
1024 		return rxwin;
1025 	}
1026 
1027 	txwin = vas_window_alloc(vinst);
1028 	if (IS_ERR(txwin)) {
1029 		rc = PTR_ERR(txwin);
1030 		goto put_rxwin;
1031 	}
1032 
1033 	txwin->cop = cop;
1034 	txwin->tx_win = 1;
1035 	txwin->rxwin = rxwin;
1036 	txwin->nx_win = txwin->rxwin->nx_win;
1037 	txwin->user_win = attr->user_win;
1038 	txwin->wcreds_max = attr->wcreds_max ?: VAS_WCREDS_DEFAULT;
1039 
1040 	init_winctx_for_txwin(txwin, attr, &winctx);
1041 
1042 	init_winctx_regs(txwin, &winctx);
1043 
1044 	/*
1045 	 * If its a kernel send window, map the window address into the
1046 	 * kernel's address space. For user windows, user must issue an
1047 	 * mmap() to map the window into their address space.
1048 	 *
1049 	 * NOTE: If kernel ever resubmits a user CRB after handling a page
1050 	 *	 fault, we will need to map this into kernel as well.
1051 	 */
1052 	if (!txwin->user_win) {
1053 		txwin->paste_kaddr = map_paste_region(txwin);
1054 		if (IS_ERR(txwin->paste_kaddr)) {
1055 			rc = PTR_ERR(txwin->paste_kaddr);
1056 			goto free_window;
1057 		}
1058 	} else {
1059 		/*
1060 		 * Interrupt hanlder or fault window setup failed. Means
1061 		 * NX can not generate fault for page fault. So not
1062 		 * opening for user space tx window.
1063 		 */
1064 		if (!vinst->virq) {
1065 			rc = -ENODEV;
1066 			goto free_window;
1067 		}
1068 
1069 		/*
1070 		 * Window opened by a child thread may not be closed when
1071 		 * it exits. So take reference to its pid and release it
1072 		 * when the window is free by parent thread.
1073 		 * Acquire a reference to the task's pid to make sure
1074 		 * pid will not be re-used - needed only for multithread
1075 		 * applications.
1076 		 */
1077 		txwin->pid = get_task_pid(current, PIDTYPE_PID);
1078 		/*
1079 		 * Acquire a reference to the task's mm.
1080 		 */
1081 		txwin->mm = get_task_mm(current);
1082 
1083 		if (!txwin->mm) {
1084 			put_pid(txwin->pid);
1085 			pr_err("VAS: pid(%d): mm_struct is not found\n",
1086 					current->pid);
1087 			rc = -EPERM;
1088 			goto free_window;
1089 		}
1090 
1091 		mmgrab(txwin->mm);
1092 		mmput(txwin->mm);
1093 		mm_context_add_vas_window(txwin->mm);
1094 		/*
1095 		 * Process closes window during exit. In the case of
1096 		 * multithread application, the child thread can open
1097 		 * window and can exit without closing it. Expects parent
1098 		 * thread to use and close the window. So do not need
1099 		 * to take pid reference for parent thread.
1100 		 */
1101 		txwin->tgid = find_get_pid(task_tgid_vnr(current));
1102 		/*
1103 		 * Even a process that has no foreign real address mapping can
1104 		 * use an unpaired COPY instruction (to no real effect). Issue
1105 		 * CP_ABORT to clear any pending COPY and prevent a covert
1106 		 * channel.
1107 		 *
1108 		 * __switch_to() will issue CP_ABORT on future context switches
1109 		 * if process / thread has any open VAS window (Use
1110 		 * current->mm->context.vas_windows).
1111 		 */
1112 		asm volatile(PPC_CP_ABORT);
1113 	}
1114 
1115 	set_vinst_win(vinst, txwin);
1116 
1117 	return txwin;
1118 
1119 free_window:
1120 	vas_window_free(txwin);
1121 
1122 put_rxwin:
1123 	put_rx_win(rxwin);
1124 	return ERR_PTR(rc);
1125 
1126 }
1127 EXPORT_SYMBOL_GPL(vas_tx_win_open);
1128 
1129 int vas_copy_crb(void *crb, int offset)
1130 {
1131 	return vas_copy(crb, offset);
1132 }
1133 EXPORT_SYMBOL_GPL(vas_copy_crb);
1134 
1135 #define RMA_LSMP_REPORT_ENABLE PPC_BIT(53)
1136 int vas_paste_crb(struct vas_window *txwin, int offset, bool re)
1137 {
1138 	int rc;
1139 	void *addr;
1140 	uint64_t val;
1141 
1142 	trace_vas_paste_crb(current, txwin);
1143 
1144 	/*
1145 	 * Only NX windows are supported for now and hardware assumes
1146 	 * report-enable flag is set for NX windows. Ensure software
1147 	 * complies too.
1148 	 */
1149 	WARN_ON_ONCE(txwin->nx_win && !re);
1150 
1151 	addr = txwin->paste_kaddr;
1152 	if (re) {
1153 		/*
1154 		 * Set the REPORT_ENABLE bit (equivalent to writing
1155 		 * to 1K offset of the paste address)
1156 		 */
1157 		val = SET_FIELD(RMA_LSMP_REPORT_ENABLE, 0ULL, 1);
1158 		addr += val;
1159 	}
1160 
1161 	/*
1162 	 * Map the raw CR value from vas_paste() to an error code (there
1163 	 * is just pass or fail for now though).
1164 	 */
1165 	rc = vas_paste(addr, offset);
1166 	if (rc == 2)
1167 		rc = 0;
1168 	else
1169 		rc = -EINVAL;
1170 
1171 	pr_debug("Txwin #%d: Msg count %llu\n", txwin->winid,
1172 			read_hvwc_reg(txwin, VREG(LRFIFO_PUSH)));
1173 
1174 	return rc;
1175 }
1176 EXPORT_SYMBOL_GPL(vas_paste_crb);
1177 
1178 /*
1179  * If credit checking is enabled for this window, poll for the return
1180  * of window credits (i.e for NX engines to process any outstanding CRBs).
1181  * Since NX-842 waits for the CRBs to be processed before closing the
1182  * window, we should not have to wait for too long.
1183  *
1184  * TODO: We retry in 10ms intervals now. We could/should probably peek at
1185  *	the VAS_LRFIFO_PUSH_OFFSET register to get an estimate of pending
1186  *	CRBs on the FIFO and compute the delay dynamically on each retry.
1187  *	But that is not really needed until we support NX-GZIP access from
1188  *	user space. (NX-842 driver waits for CSB and Fast thread-wakeup
1189  *	doesn't use credit checking).
1190  */
1191 static void poll_window_credits(struct vas_window *window)
1192 {
1193 	u64 val;
1194 	int creds, mode;
1195 	int count = 0;
1196 
1197 	val = read_hvwc_reg(window, VREG(WINCTL));
1198 	if (window->tx_win)
1199 		mode = GET_FIELD(VAS_WINCTL_TX_WCRED_MODE, val);
1200 	else
1201 		mode = GET_FIELD(VAS_WINCTL_RX_WCRED_MODE, val);
1202 
1203 	if (!mode)
1204 		return;
1205 retry:
1206 	if (window->tx_win) {
1207 		val = read_hvwc_reg(window, VREG(TX_WCRED));
1208 		creds = GET_FIELD(VAS_TX_WCRED, val);
1209 	} else {
1210 		val = read_hvwc_reg(window, VREG(LRX_WCRED));
1211 		creds = GET_FIELD(VAS_LRX_WCRED, val);
1212 	}
1213 
1214 	/*
1215 	 * Takes around few milliseconds to complete all pending requests
1216 	 * and return credits.
1217 	 * TODO: Scan fault FIFO and invalidate CRBs points to this window
1218 	 *       and issue CRB Kill to stop all pending requests. Need only
1219 	 *       if there is a bug in NX or fault handling in kernel.
1220 	 */
1221 	if (creds < window->wcreds_max) {
1222 		val = 0;
1223 		set_current_state(TASK_UNINTERRUPTIBLE);
1224 		schedule_timeout(msecs_to_jiffies(10));
1225 		count++;
1226 		/*
1227 		 * Process can not close send window until all credits are
1228 		 * returned.
1229 		 */
1230 		if (!(count % 1000))
1231 			pr_warn_ratelimited("VAS: pid %d stuck. Waiting for credits returned for Window(%d). creds %d, Retries %d\n",
1232 				vas_window_pid(window), window->winid,
1233 				creds, count);
1234 
1235 		goto retry;
1236 	}
1237 }
1238 
1239 /*
1240  * Wait for the window to go to "not-busy" state. It should only take a
1241  * short time to queue a CRB, so window should not be busy for too long.
1242  * Trying 5ms intervals.
1243  */
1244 static void poll_window_busy_state(struct vas_window *window)
1245 {
1246 	int busy;
1247 	u64 val;
1248 	int count = 0;
1249 
1250 retry:
1251 	val = read_hvwc_reg(window, VREG(WIN_STATUS));
1252 	busy = GET_FIELD(VAS_WIN_BUSY, val);
1253 	if (busy) {
1254 		val = 0;
1255 		set_current_state(TASK_UNINTERRUPTIBLE);
1256 		schedule_timeout(msecs_to_jiffies(10));
1257 		count++;
1258 		/*
1259 		 * Takes around few milliseconds to process all pending
1260 		 * requests.
1261 		 */
1262 		if (!(count % 1000))
1263 			pr_warn_ratelimited("VAS: pid %d stuck. Window (ID=%d) is in busy state. Retries %d\n",
1264 				vas_window_pid(window), window->winid, count);
1265 
1266 		goto retry;
1267 	}
1268 }
1269 
1270 /*
1271  * Have the hardware cast a window out of cache and wait for it to
1272  * be completed.
1273  *
1274  * NOTE: It can take a relatively long time to cast the window context
1275  *	out of the cache. It is not strictly necessary to cast out if:
1276  *
1277  *	- we clear the "Pin Window" bit (so hardware is free to evict)
1278  *
1279  *	- we re-initialize the window context when it is reassigned.
1280  *
1281  *	We do the former in vas_win_close() and latter in vas_win_open().
1282  *	So, ignoring the cast-out for now. We can add it as needed. If
1283  *	casting out becomes necessary we should consider offloading the
1284  *	job to a worker thread, so the window close can proceed quickly.
1285  */
1286 static void poll_window_castout(struct vas_window *window)
1287 {
1288 	/* stub for now */
1289 }
1290 
1291 /*
1292  * Unpin and close a window so no new requests are accepted and the
1293  * hardware can evict this window from cache if necessary.
1294  */
1295 static void unpin_close_window(struct vas_window *window)
1296 {
1297 	u64 val;
1298 
1299 	val = read_hvwc_reg(window, VREG(WINCTL));
1300 	val = SET_FIELD(VAS_WINCTL_PIN, val, 0);
1301 	val = SET_FIELD(VAS_WINCTL_OPEN, val, 0);
1302 	write_hvwc_reg(window, VREG(WINCTL), val);
1303 }
1304 
1305 /*
1306  * Close a window.
1307  *
1308  * See Section 1.12.1 of VAS workbook v1.05 for details on closing window:
1309  *	- Disable new paste operations (unmap paste address)
1310  *	- Poll for the "Window Busy" bit to be cleared
1311  *	- Clear the Open/Enable bit for the Window.
1312  *	- Poll for return of window Credits (implies FIFO empty for Rx win?)
1313  *	- Unpin and cast window context out of cache
1314  *
1315  * Besides the hardware, kernel has some bookkeeping of course.
1316  */
1317 int vas_win_close(struct vas_window *window)
1318 {
1319 	if (!window)
1320 		return 0;
1321 
1322 	if (!window->tx_win && atomic_read(&window->num_txwins) != 0) {
1323 		pr_devel("Attempting to close an active Rx window!\n");
1324 		WARN_ON_ONCE(1);
1325 		return -EBUSY;
1326 	}
1327 
1328 	unmap_paste_region(window);
1329 
1330 	poll_window_busy_state(window);
1331 
1332 	unpin_close_window(window);
1333 
1334 	poll_window_credits(window);
1335 
1336 	clear_vinst_win(window);
1337 
1338 	poll_window_castout(window);
1339 
1340 	/* if send window, drop reference to matching receive window */
1341 	if (window->tx_win) {
1342 		if (window->user_win) {
1343 			/* Drop references to pid and mm */
1344 			put_pid(window->pid);
1345 			if (window->mm) {
1346 				mm_context_remove_vas_window(window->mm);
1347 				mmdrop(window->mm);
1348 			}
1349 		}
1350 		put_rx_win(window->rxwin);
1351 	}
1352 
1353 	vas_window_free(window);
1354 
1355 	return 0;
1356 }
1357 EXPORT_SYMBOL_GPL(vas_win_close);
1358 
1359 /*
1360  * Return credit for the given window.
1361  * Send windows and fault window uses credit mechanism as follows:
1362  *
1363  * Send windows:
1364  * - The default number of credits available for each send window is
1365  *   1024. It means 1024 requests can be issued asynchronously at the
1366  *   same time. If the credit is not available, that request will be
1367  *   returned with RMA_Busy.
1368  * - One credit is taken when NX request is issued.
1369  * - This credit is returned after NX processed that request.
1370  * - If NX encounters translation error, kernel will return the
1371  *   credit on the specific send window after processing the fault CRB.
1372  *
1373  * Fault window:
1374  * - The total number credits available is FIFO_SIZE/CRB_SIZE.
1375  *   Means 4MB/128 in the current implementation. If credit is not
1376  *   available, RMA_Reject is returned.
1377  * - A credit is taken when NX pastes CRB in fault FIFO.
1378  * - The kernel with return credit on fault window after reading entry
1379  *   from fault FIFO.
1380  */
1381 void vas_return_credit(struct vas_window *window, bool tx)
1382 {
1383 	uint64_t val;
1384 
1385 	val = 0ULL;
1386 	if (tx) { /* send window */
1387 		val = SET_FIELD(VAS_TX_WCRED, val, 1);
1388 		write_hvwc_reg(window, VREG(TX_WCRED_ADDER), val);
1389 	} else {
1390 		val = SET_FIELD(VAS_LRX_WCRED, val, 1);
1391 		write_hvwc_reg(window, VREG(LRX_WCRED_ADDER), val);
1392 	}
1393 }
1394 
1395 struct vas_window *vas_pswid_to_window(struct vas_instance *vinst,
1396 		uint32_t pswid)
1397 {
1398 	struct vas_window *window;
1399 	int winid;
1400 
1401 	if (!pswid) {
1402 		pr_devel("%s: called for pswid 0!\n", __func__);
1403 		return ERR_PTR(-ESRCH);
1404 	}
1405 
1406 	decode_pswid(pswid, NULL, &winid);
1407 
1408 	if (winid >= VAS_WINDOWS_PER_CHIP)
1409 		return ERR_PTR(-ESRCH);
1410 
1411 	/*
1412 	 * If application closes the window before the hardware
1413 	 * returns the fault CRB, we should wait in vas_win_close()
1414 	 * for the pending requests. so the window must be active
1415 	 * and the process alive.
1416 	 *
1417 	 * If its a kernel process, we should not get any faults and
1418 	 * should not get here.
1419 	 */
1420 	window = vinst->windows[winid];
1421 
1422 	if (!window) {
1423 		pr_err("PSWID decode: Could not find window for winid %d pswid %d vinst 0x%p\n",
1424 			winid, pswid, vinst);
1425 		return NULL;
1426 	}
1427 
1428 	/*
1429 	 * Do some sanity checks on the decoded window.  Window should be
1430 	 * NX GZIP user send window. FTW windows should not incur faults
1431 	 * since their CRBs are ignored (not queued on FIFO or processed
1432 	 * by NX).
1433 	 */
1434 	if (!window->tx_win || !window->user_win || !window->nx_win ||
1435 			window->cop == VAS_COP_TYPE_FAULT ||
1436 			window->cop == VAS_COP_TYPE_FTW) {
1437 		pr_err("PSWID decode: id %d, tx %d, user %d, nx %d, cop %d\n",
1438 			winid, window->tx_win, window->user_win,
1439 			window->nx_win, window->cop);
1440 		WARN_ON(1);
1441 	}
1442 
1443 	return window;
1444 }
1445