1 /*
2  * tiomap.c
3  *
4  * DSP-BIOS Bridge driver support functions for TI OMAP processors.
5  *
6  * Processor Manager Driver for TI OMAP3430 EVM.
7  *
8  * Copyright (C) 2005-2006 Texas Instruments, Inc.
9  *
10  * This package 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 PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
15  * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
16  * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.
17  */
18 
19 #include <plat/dsp.h>
20 
21 #include <linux/types.h>
22 /*  ----------------------------------- Host OS */
23 #include <dspbridge/host_os.h>
24 #include <linux/mm.h>
25 #include <linux/mmzone.h>
26 
27 /*  ----------------------------------- DSP/BIOS Bridge */
28 #include <dspbridge/dbdefs.h>
29 
30 /*  ----------------------------------- OS Adaptation Layer */
31 #include <dspbridge/drv.h>
32 #include <dspbridge/sync.h>
33 
34 /* ------------------------------------ Hardware Abstraction Layer */
35 #include <hw_defs.h>
36 #include <hw_mmu.h>
37 
38 /*  ----------------------------------- Link Driver */
39 #include <dspbridge/dspdefs.h>
40 #include <dspbridge/dspchnl.h>
41 #include <dspbridge/dspdeh.h>
42 #include <dspbridge/dspio.h>
43 #include <dspbridge/dspmsg.h>
44 #include <dspbridge/pwr.h>
45 #include <dspbridge/io_sm.h>
46 
47 /*  ----------------------------------- Platform Manager */
48 #include <dspbridge/dev.h>
49 #include <dspbridge/dspapi.h>
50 #include <dspbridge/dmm.h>
51 #include <dspbridge/wdt.h>
52 
53 /*  ----------------------------------- Local */
54 #include "_tiomap.h"
55 #include "_tiomap_pwr.h"
56 #include "tiomap_io.h"
57 
58 /* Offset in shared mem to write to in order to synchronize start with DSP */
59 #define SHMSYNCOFFSET 4		/* GPP byte offset */
60 
61 #define BUFFERSIZE 1024
62 
63 #define TIHELEN_ACKTIMEOUT  10000
64 
65 #define MMU_SECTION_ADDR_MASK    0xFFF00000
66 #define MMU_SSECTION_ADDR_MASK   0xFF000000
67 #define MMU_LARGE_PAGE_MASK      0xFFFF0000
68 #define MMU_SMALL_PAGE_MASK      0xFFFFF000
69 #define OMAP3_IVA2_BOOTADDR_MASK 0xFFFFFC00
70 #define PAGES_II_LVL_TABLE   512
71 #define PHYS_TO_PAGE(phys)      pfn_to_page((phys) >> PAGE_SHIFT)
72 
73 /*
74  * This is a totally ugly layer violation, but needed until
75  * omap_ctrl_set_dsp_boot*() are provided.
76  */
77 #define OMAP3_IVA2_BOOTMOD_IDLE 1
78 #define OMAP2_CONTROL_GENERAL 0x270
79 #define OMAP343X_CONTROL_IVA2_BOOTADDR (OMAP2_CONTROL_GENERAL + 0x0190)
80 #define OMAP343X_CONTROL_IVA2_BOOTMOD (OMAP2_CONTROL_GENERAL + 0x0194)
81 
82 /* Forward Declarations: */
83 static int bridge_brd_monitor(struct bridge_dev_context *dev_ctxt);
84 static int bridge_brd_read(struct bridge_dev_context *dev_ctxt,
85 				  u8 *host_buff,
86 				  u32 dsp_addr, u32 ul_num_bytes,
87 				  u32 mem_type);
88 static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
89 				   u32 dsp_addr);
90 static int bridge_brd_status(struct bridge_dev_context *dev_ctxt,
91 				    int *board_state);
92 static int bridge_brd_stop(struct bridge_dev_context *dev_ctxt);
93 static int bridge_brd_write(struct bridge_dev_context *dev_ctxt,
94 				   u8 *host_buff,
95 				   u32 dsp_addr, u32 ul_num_bytes,
96 				   u32 mem_type);
97 static int bridge_brd_set_state(struct bridge_dev_context *dev_ctxt,
98 				    u32 brd_state);
99 static int bridge_brd_mem_copy(struct bridge_dev_context *dev_ctxt,
100 				   u32 dsp_dest_addr, u32 dsp_src_addr,
101 				   u32 ul_num_bytes, u32 mem_type);
102 static int bridge_brd_mem_write(struct bridge_dev_context *dev_ctxt,
103 				    u8 *host_buff, u32 dsp_addr,
104 				    u32 ul_num_bytes, u32 mem_type);
105 static int bridge_brd_mem_map(struct bridge_dev_context *dev_ctxt,
106 				  u32 ul_mpu_addr, u32 virt_addr,
107 				  u32 ul_num_bytes, u32 ul_map_attr,
108 				  struct page **mapped_pages);
109 static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctxt,
110 				     u32 virt_addr, u32 ul_num_bytes);
111 static int bridge_dev_create(struct bridge_dev_context
112 					**dev_cntxt,
113 					struct dev_object *hdev_obj,
114 					struct cfg_hostres *config_param);
115 static int bridge_dev_ctrl(struct bridge_dev_context *dev_context,
116 				  u32 dw_cmd, void *pargs);
117 static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt);
118 static u32 user_va2_pa(struct mm_struct *mm, u32 address);
119 static int pte_update(struct bridge_dev_context *dev_ctxt, u32 pa,
120 			     u32 va, u32 size,
121 			     struct hw_mmu_map_attrs_t *map_attrs);
122 static int pte_set(struct pg_table_attrs *pt, u32 pa, u32 va,
123 			  u32 size, struct hw_mmu_map_attrs_t *attrs);
124 static int mem_map_vmalloc(struct bridge_dev_context *dev_context,
125 				  u32 ul_mpu_addr, u32 virt_addr,
126 				  u32 ul_num_bytes,
127 				  struct hw_mmu_map_attrs_t *hw_attrs);
128 
129 bool wait_for_start(struct bridge_dev_context *dev_context, u32 dw_sync_addr);
130 
131 /*  ----------------------------------- Globals */
132 
133 /* Attributes of L2 page tables for DSP MMU */
134 struct page_info {
135 	u32 num_entries;	/* Number of valid PTEs in the L2 PT */
136 };
137 
138 /* Attributes used to manage the DSP MMU page tables */
139 struct pg_table_attrs {
140 	spinlock_t pg_lock;	/* Critical section object handle */
141 
142 	u32 l1_base_pa;		/* Physical address of the L1 PT */
143 	u32 l1_base_va;		/* Virtual  address of the L1 PT */
144 	u32 l1_size;		/* Size of the L1 PT */
145 	u32 l1_tbl_alloc_pa;
146 	/* Physical address of Allocated mem for L1 table. May not be aligned */
147 	u32 l1_tbl_alloc_va;
148 	/* Virtual address of Allocated mem for L1 table. May not be aligned */
149 	u32 l1_tbl_alloc_sz;
150 	/* Size of consistent memory allocated for L1 table.
151 	 * May not be aligned */
152 
153 	u32 l2_base_pa;		/* Physical address of the L2 PT */
154 	u32 l2_base_va;		/* Virtual  address of the L2 PT */
155 	u32 l2_size;		/* Size of the L2 PT */
156 	u32 l2_tbl_alloc_pa;
157 	/* Physical address of Allocated mem for L2 table. May not be aligned */
158 	u32 l2_tbl_alloc_va;
159 	/* Virtual address of Allocated mem for L2 table. May not be aligned */
160 	u32 l2_tbl_alloc_sz;
161 	/* Size of consistent memory allocated for L2 table.
162 	 * May not be aligned */
163 
164 	u32 l2_num_pages;	/* Number of allocated L2 PT */
165 	/* Array [l2_num_pages] of L2 PT info structs */
166 	struct page_info *pg_info;
167 };
168 
169 /*
170  *  This Bridge driver's function interface table.
171  */
172 static struct bridge_drv_interface drv_interface_fxns = {
173 	/* Bridge API ver. for which this bridge driver is built. */
174 	BRD_API_MAJOR_VERSION,
175 	BRD_API_MINOR_VERSION,
176 	bridge_dev_create,
177 	bridge_dev_destroy,
178 	bridge_dev_ctrl,
179 	bridge_brd_monitor,
180 	bridge_brd_start,
181 	bridge_brd_stop,
182 	bridge_brd_status,
183 	bridge_brd_read,
184 	bridge_brd_write,
185 	bridge_brd_set_state,
186 	bridge_brd_mem_copy,
187 	bridge_brd_mem_write,
188 	bridge_brd_mem_map,
189 	bridge_brd_mem_un_map,
190 	/* The following CHNL functions are provided by chnl_io.lib: */
191 	bridge_chnl_create,
192 	bridge_chnl_destroy,
193 	bridge_chnl_open,
194 	bridge_chnl_close,
195 	bridge_chnl_add_io_req,
196 	bridge_chnl_get_ioc,
197 	bridge_chnl_cancel_io,
198 	bridge_chnl_flush_io,
199 	bridge_chnl_get_info,
200 	bridge_chnl_get_mgr_info,
201 	bridge_chnl_idle,
202 	bridge_chnl_register_notify,
203 	/* The following IO functions are provided by chnl_io.lib: */
204 	bridge_io_create,
205 	bridge_io_destroy,
206 	bridge_io_on_loaded,
207 	bridge_io_get_proc_load,
208 	/* The following msg_ctrl functions are provided by chnl_io.lib: */
209 	bridge_msg_create,
210 	bridge_msg_create_queue,
211 	bridge_msg_delete,
212 	bridge_msg_delete_queue,
213 	bridge_msg_get,
214 	bridge_msg_put,
215 	bridge_msg_register_notify,
216 	bridge_msg_set_queue_id,
217 };
218 
219 static struct notifier_block dsp_mbox_notifier = {
220 	.notifier_call = io_mbox_msg,
221 };
222 
flush_all(struct bridge_dev_context * dev_context)223 static inline void flush_all(struct bridge_dev_context *dev_context)
224 {
225 	if (dev_context->brd_state == BRD_DSP_HIBERNATION ||
226 	    dev_context->brd_state == BRD_HIBERNATION)
227 		wake_dsp(dev_context, NULL);
228 
229 	hw_mmu_tlb_flush_all(dev_context->dsp_mmu_base);
230 }
231 
bad_page_dump(u32 pa,struct page * pg)232 static void bad_page_dump(u32 pa, struct page *pg)
233 {
234 	pr_emerg("DSPBRIDGE: MAP function: COUNT 0 FOR PA 0x%x\n", pa);
235 	pr_emerg("Bad page state in process '%s'\n"
236 		 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
237 		 "Backtrace:\n",
238 		 current->comm, pg, (int)(2 * sizeof(unsigned long)),
239 		 (unsigned long)pg->flags, pg->mapping,
240 		 page_mapcount(pg), page_count(pg));
241 	dump_stack();
242 }
243 
244 /*
245  *  ======== bridge_drv_entry ========
246  *  purpose:
247  *      Bridge Driver entry point.
248  */
bridge_drv_entry(struct bridge_drv_interface ** drv_intf,const char * driver_file_name)249 void bridge_drv_entry(struct bridge_drv_interface **drv_intf,
250 		   const char *driver_file_name)
251 {
252 	if (strcmp(driver_file_name, "UMA") == 0)
253 		*drv_intf = &drv_interface_fxns;
254 	else
255 		dev_dbg(bridge, "%s Unknown Bridge file name", __func__);
256 
257 }
258 
259 /*
260  *  ======== bridge_brd_monitor ========
261  *  purpose:
262  *      This bridge_brd_monitor puts DSP into a Loadable state.
263  *      i.e Application can load and start the device.
264  *
265  *  Preconditions:
266  *      Device in 'OFF' state.
267  */
bridge_brd_monitor(struct bridge_dev_context * dev_ctxt)268 static int bridge_brd_monitor(struct bridge_dev_context *dev_ctxt)
269 {
270 	struct bridge_dev_context *dev_context = dev_ctxt;
271 	u32 temp;
272 	struct omap_dsp_platform_data *pdata =
273 		omap_dspbridge_dev->dev.platform_data;
274 
275 	temp = (*pdata->dsp_prm_read)(OMAP3430_IVA2_MOD, OMAP2_PM_PWSTST) &
276 					OMAP_POWERSTATEST_MASK;
277 	if (!(temp & 0x02)) {
278 		/* IVA2 is not in ON state */
279 		/* Read and set PM_PWSTCTRL_IVA2  to ON */
280 		(*pdata->dsp_prm_rmw_bits)(OMAP_POWERSTATEST_MASK,
281 			PWRDM_POWER_ON, OMAP3430_IVA2_MOD, OMAP2_PM_PWSTCTRL);
282 		/* Set the SW supervised state transition */
283 		(*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_FORCE_WAKEUP,
284 					OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
285 
286 		/* Wait until the state has moved to ON */
287 		while ((*pdata->dsp_prm_read)(OMAP3430_IVA2_MOD, OMAP2_PM_PWSTST) &
288 						OMAP_INTRANSITION_MASK)
289 			;
290 		/* Disable Automatic transition */
291 		(*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_DISABLE_AUTO,
292 					OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
293 	}
294 	(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0,
295 					OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
296 	dsp_clk_enable(DSP_CLK_IVA2);
297 
298 	/* set the device state to IDLE */
299 	dev_context->brd_state = BRD_IDLE;
300 
301 	return 0;
302 }
303 
304 /*
305  *  ======== bridge_brd_read ========
306  *  purpose:
307  *      Reads buffers for DSP memory.
308  */
bridge_brd_read(struct bridge_dev_context * dev_ctxt,u8 * host_buff,u32 dsp_addr,u32 ul_num_bytes,u32 mem_type)309 static int bridge_brd_read(struct bridge_dev_context *dev_ctxt,
310 				  u8 *host_buff, u32 dsp_addr,
311 				  u32 ul_num_bytes, u32 mem_type)
312 {
313 	int status = 0;
314 	struct bridge_dev_context *dev_context = dev_ctxt;
315 	u32 offset;
316 	u32 dsp_base_addr = dev_ctxt->dsp_base_addr;
317 
318 	if (dsp_addr < dev_context->dsp_start_add) {
319 		status = -EPERM;
320 		return status;
321 	}
322 	/* change here to account for the 3 bands of the DSP internal memory */
323 	if ((dsp_addr - dev_context->dsp_start_add) <
324 	    dev_context->internal_size) {
325 		offset = dsp_addr - dev_context->dsp_start_add;
326 	} else {
327 		status = read_ext_dsp_data(dev_context, host_buff, dsp_addr,
328 					   ul_num_bytes, mem_type);
329 		return status;
330 	}
331 	/* copy the data from  DSP memory, */
332 	memcpy(host_buff, (void *)(dsp_base_addr + offset), ul_num_bytes);
333 	return status;
334 }
335 
336 /*
337  *  ======== bridge_brd_set_state ========
338  *  purpose:
339  *      This routine updates the Board status.
340  */
bridge_brd_set_state(struct bridge_dev_context * dev_ctxt,u32 brd_state)341 static int bridge_brd_set_state(struct bridge_dev_context *dev_ctxt,
342 				    u32 brd_state)
343 {
344 	int status = 0;
345 	struct bridge_dev_context *dev_context = dev_ctxt;
346 
347 	dev_context->brd_state = brd_state;
348 	return status;
349 }
350 
351 /*
352  *  ======== bridge_brd_start ========
353  *  purpose:
354  *      Initializes DSP MMU and Starts DSP.
355  *
356  *  Preconditions:
357  *  a) DSP domain is 'ACTIVE'.
358  *  b) DSP_RST1 is asserted.
359  *  b) DSP_RST2 is released.
360  */
bridge_brd_start(struct bridge_dev_context * dev_ctxt,u32 dsp_addr)361 static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
362 				   u32 dsp_addr)
363 {
364 	int status = 0;
365 	struct bridge_dev_context *dev_context = dev_ctxt;
366 	u32 dw_sync_addr = 0;
367 	u32 ul_shm_base;	/* Gpp Phys SM base addr(byte) */
368 	u32 ul_shm_base_virt;	/* Dsp Virt SM base addr */
369 	u32 ul_tlb_base_virt;	/* Base of MMU TLB entry */
370 	/* Offset of shm_base_virt from tlb_base_virt */
371 	u32 ul_shm_offset_virt;
372 	s32 entry_ndx;
373 	s32 itmp_entry_ndx = 0;	/* DSP-MMU TLB entry base address */
374 	struct cfg_hostres *resources = NULL;
375 	u32 temp;
376 	u32 ul_dsp_clk_rate;
377 	u32 ul_dsp_clk_addr;
378 	u32 ul_bios_gp_timer;
379 	u32 clk_cmd;
380 	struct io_mgr *hio_mgr;
381 	u32 ul_load_monitor_timer;
382 	u32 wdt_en = 0;
383 	struct omap_dsp_platform_data *pdata =
384 		omap_dspbridge_dev->dev.platform_data;
385 
386 	/* The device context contains all the mmu setup info from when the
387 	 * last dsp base image was loaded. The first entry is always
388 	 * SHMMEM base. */
389 	/* Get SHM_BEG - convert to byte address */
390 	(void)dev_get_symbol(dev_context->dev_obj, SHMBASENAME,
391 			     &ul_shm_base_virt);
392 	ul_shm_base_virt *= DSPWORDSIZE;
393 	/* DSP Virtual address */
394 	ul_tlb_base_virt = dev_context->atlb_entry[0].dsp_va;
395 	ul_shm_offset_virt =
396 	    ul_shm_base_virt - (ul_tlb_base_virt * DSPWORDSIZE);
397 	/* Kernel logical address */
398 	ul_shm_base = dev_context->atlb_entry[0].gpp_va + ul_shm_offset_virt;
399 
400 	/* 2nd wd is used as sync field */
401 	dw_sync_addr = ul_shm_base + SHMSYNCOFFSET;
402 	/* Write a signature into the shm base + offset; this will
403 	 * get cleared when the DSP program starts. */
404 	if ((ul_shm_base_virt == 0) || (ul_shm_base == 0)) {
405 		pr_err("%s: Illegal SM base\n", __func__);
406 		status = -EPERM;
407 	} else
408 		__raw_writel(0xffffffff, dw_sync_addr);
409 
410 	if (!status) {
411 		resources = dev_context->resources;
412 		if (!resources)
413 			status = -EPERM;
414 
415 		/* Assert RST1 i.e only the RST only for DSP megacell */
416 		if (!status) {
417 			/*
418 			 * XXX: ioremapping  MUST be removed once ctrl
419 			 * function is made available.
420 			 */
421 			void __iomem *ctrl = ioremap(OMAP343X_CTRL_BASE, SZ_4K);
422 			if (!ctrl)
423 				return -ENOMEM;
424 
425 			(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST1_IVA2_MASK,
426 					OMAP3430_RST1_IVA2_MASK, OMAP3430_IVA2_MOD,
427 					OMAP2_RM_RSTCTRL);
428 			/* Mask address with 1K for compatibility */
429 			__raw_writel(dsp_addr & OMAP3_IVA2_BOOTADDR_MASK,
430 					ctrl + OMAP343X_CONTROL_IVA2_BOOTADDR);
431 			/*
432 			 * Set bootmode to self loop if dsp_debug flag is true
433 			 */
434 			__raw_writel((dsp_debug) ? OMAP3_IVA2_BOOTMOD_IDLE : 0,
435 					ctrl + OMAP343X_CONTROL_IVA2_BOOTMOD);
436 
437 			iounmap(ctrl);
438 		}
439 	}
440 	if (!status) {
441 		/* Reset and Unreset the RST2, so that BOOTADDR is copied to
442 		 * IVA2 SYSC register */
443 		(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK,
444 			OMAP3430_RST2_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
445 		udelay(100);
446 		(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0,
447 					OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
448 		udelay(100);
449 
450 		/* Disbale the DSP MMU */
451 		hw_mmu_disable(resources->dmmu_base);
452 		/* Disable TWL */
453 		hw_mmu_twl_disable(resources->dmmu_base);
454 
455 		/* Only make TLB entry if both addresses are non-zero */
456 		for (entry_ndx = 0; entry_ndx < BRDIOCTL_NUMOFMMUTLB;
457 		     entry_ndx++) {
458 			struct bridge_ioctl_extproc *e = &dev_context->atlb_entry[entry_ndx];
459 			struct hw_mmu_map_attrs_t map_attrs = {
460 				.endianism = e->endianism,
461 				.element_size = e->elem_size,
462 				.mixed_size = e->mixed_mode,
463 			};
464 
465 			if (!e->gpp_pa || !e->dsp_va)
466 				continue;
467 
468 			dev_dbg(bridge,
469 					"MMU %d, pa: 0x%x, va: 0x%x, size: 0x%x",
470 					itmp_entry_ndx,
471 					e->gpp_pa,
472 					e->dsp_va,
473 					e->size);
474 
475 			hw_mmu_tlb_add(dev_context->dsp_mmu_base,
476 					e->gpp_pa,
477 					e->dsp_va,
478 					e->size,
479 					itmp_entry_ndx,
480 					&map_attrs, 1, 1);
481 
482 			itmp_entry_ndx++;
483 		}
484 	}
485 
486 	/* Lock the above TLB entries and get the BIOS and load monitor timer
487 	 * information */
488 	if (!status) {
489 		hw_mmu_num_locked_set(resources->dmmu_base, itmp_entry_ndx);
490 		hw_mmu_victim_num_set(resources->dmmu_base, itmp_entry_ndx);
491 		hw_mmu_ttb_set(resources->dmmu_base,
492 			       dev_context->pt_attrs->l1_base_pa);
493 		hw_mmu_twl_enable(resources->dmmu_base);
494 		/* Enable the SmartIdle and AutoIdle bit for MMU_SYSCONFIG */
495 
496 		temp = __raw_readl((resources->dmmu_base) + 0x10);
497 		temp = (temp & 0xFFFFFFEF) | 0x11;
498 		__raw_writel(temp, (resources->dmmu_base) + 0x10);
499 
500 		/* Let the DSP MMU run */
501 		hw_mmu_enable(resources->dmmu_base);
502 
503 		/* Enable the BIOS clock */
504 		(void)dev_get_symbol(dev_context->dev_obj,
505 				     BRIDGEINIT_BIOSGPTIMER, &ul_bios_gp_timer);
506 		(void)dev_get_symbol(dev_context->dev_obj,
507 				     BRIDGEINIT_LOADMON_GPTIMER,
508 				     &ul_load_monitor_timer);
509 	}
510 
511 	if (!status) {
512 		if (ul_load_monitor_timer != 0xFFFF) {
513 			clk_cmd = (BPWR_ENABLE_CLOCK << MBX_PM_CLK_CMDSHIFT) |
514 			    ul_load_monitor_timer;
515 			dsp_peripheral_clk_ctrl(dev_context, &clk_cmd);
516 		} else {
517 			dev_dbg(bridge, "Not able to get the symbol for Load "
518 				"Monitor Timer\n");
519 		}
520 	}
521 
522 	if (!status) {
523 		if (ul_bios_gp_timer != 0xFFFF) {
524 			clk_cmd = (BPWR_ENABLE_CLOCK << MBX_PM_CLK_CMDSHIFT) |
525 			    ul_bios_gp_timer;
526 			dsp_peripheral_clk_ctrl(dev_context, &clk_cmd);
527 		} else {
528 			dev_dbg(bridge,
529 				"Not able to get the symbol for BIOS Timer\n");
530 		}
531 	}
532 
533 	if (!status) {
534 		/* Set the DSP clock rate */
535 		(void)dev_get_symbol(dev_context->dev_obj,
536 				     "_BRIDGEINIT_DSP_FREQ", &ul_dsp_clk_addr);
537 		/*Set Autoidle Mode for IVA2 PLL */
538 		(*pdata->dsp_cm_write)(1 << OMAP3430_AUTO_IVA2_DPLL_SHIFT,
539 				OMAP3430_IVA2_MOD, OMAP3430_CM_AUTOIDLE_PLL);
540 
541 		if ((unsigned int *)ul_dsp_clk_addr != NULL) {
542 			/* Get the clock rate */
543 			ul_dsp_clk_rate = dsp_clk_get_iva2_rate();
544 			dev_dbg(bridge, "%s: DSP clock rate (KHZ): 0x%x \n",
545 				__func__, ul_dsp_clk_rate);
546 			(void)bridge_brd_write(dev_context,
547 					       (u8 *) &ul_dsp_clk_rate,
548 					       ul_dsp_clk_addr, sizeof(u32), 0);
549 		}
550 		/*
551 		 * Enable Mailbox events and also drain any pending
552 		 * stale messages.
553 		 */
554 		dev_context->mbox = omap_mbox_get("dsp", &dsp_mbox_notifier);
555 		if (IS_ERR(dev_context->mbox)) {
556 			dev_context->mbox = NULL;
557 			pr_err("%s: Failed to get dsp mailbox handle\n",
558 								__func__);
559 			status = -EPERM;
560 		}
561 
562 	}
563 	if (!status) {
564 /*PM_IVA2GRPSEL_PER = 0xC0;*/
565 		temp = readl(resources->per_pm_base + 0xA8);
566 		temp = (temp & 0xFFFFFF30) | 0xC0;
567 		writel(temp, resources->per_pm_base + 0xA8);
568 
569 /*PM_MPUGRPSEL_PER &= 0xFFFFFF3F; */
570 		temp = readl(resources->per_pm_base + 0xA4);
571 		temp = (temp & 0xFFFFFF3F);
572 		writel(temp, resources->per_pm_base + 0xA4);
573 /*CM_SLEEPDEP_PER |= 0x04; */
574 		temp = readl(resources->per_base + 0x44);
575 		temp = (temp & 0xFFFFFFFB) | 0x04;
576 		writel(temp, resources->per_base + 0x44);
577 
578 /*CM_CLKSTCTRL_IVA2 = 0x00000003 -To Allow automatic transitions */
579 		(*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_ENABLE_AUTO,
580 					OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
581 
582 		/* Let DSP go */
583 		dev_dbg(bridge, "%s Unreset\n", __func__);
584 		/* Enable DSP MMU Interrupts */
585 		hw_mmu_event_enable(resources->dmmu_base,
586 				    HW_MMU_ALL_INTERRUPTS);
587 		/* release the RST1, DSP starts executing now .. */
588 		(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST1_IVA2_MASK, 0,
589 					OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
590 
591 		dev_dbg(bridge, "Waiting for Sync @ 0x%x\n", dw_sync_addr);
592 		dev_dbg(bridge, "DSP c_int00 Address =  0x%x\n", dsp_addr);
593 		if (dsp_debug)
594 			while (__raw_readw(dw_sync_addr))
595 				;
596 
597 		/* Wait for DSP to clear word in shared memory */
598 		/* Read the Location */
599 		if (!wait_for_start(dev_context, dw_sync_addr))
600 			status = -ETIMEDOUT;
601 
602 		dev_get_symbol(dev_context->dev_obj, "_WDT_enable", &wdt_en);
603 		if (wdt_en) {
604 			/* Start wdt */
605 			dsp_wdt_sm_set((void *)ul_shm_base);
606 			dsp_wdt_enable(true);
607 		}
608 
609 		status = dev_get_io_mgr(dev_context->dev_obj, &hio_mgr);
610 		if (hio_mgr) {
611 			io_sh_msetting(hio_mgr, SHM_OPPINFO, NULL);
612 			/* Write the synchronization bit to indicate the
613 			 * completion of OPP table update to DSP
614 			 */
615 			__raw_writel(0XCAFECAFE, dw_sync_addr);
616 
617 			/* update board state */
618 			dev_context->brd_state = BRD_RUNNING;
619 			/* (void)chnlsm_enable_interrupt(dev_context); */
620 		} else {
621 			dev_context->brd_state = BRD_UNKNOWN;
622 		}
623 	}
624 	return status;
625 }
626 
627 /*
628  *  ======== bridge_brd_stop ========
629  *  purpose:
630  *      Puts DSP in self loop.
631  *
632  *  Preconditions :
633  *  a) None
634  */
bridge_brd_stop(struct bridge_dev_context * dev_ctxt)635 static int bridge_brd_stop(struct bridge_dev_context *dev_ctxt)
636 {
637 	int status = 0;
638 	struct bridge_dev_context *dev_context = dev_ctxt;
639 	struct pg_table_attrs *pt_attrs;
640 	u32 dsp_pwr_state;
641 	struct omap_dsp_platform_data *pdata =
642 		omap_dspbridge_dev->dev.platform_data;
643 
644 	if (dev_context->brd_state == BRD_STOPPED)
645 		return status;
646 
647 	/* as per TRM, it is advised to first drive the IVA2 to 'Standby' mode,
648 	 * before turning off the clocks.. This is to ensure that there are no
649 	 * pending L3 or other transactons from IVA2 */
650 	dsp_pwr_state = (*pdata->dsp_prm_read)(OMAP3430_IVA2_MOD, OMAP2_PM_PWSTST) &
651 					OMAP_POWERSTATEST_MASK;
652 	if (dsp_pwr_state != PWRDM_POWER_OFF) {
653 		(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0,
654 					OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
655 		sm_interrupt_dsp(dev_context, MBX_PM_DSPIDLE);
656 		mdelay(10);
657 
658 		/* IVA2 is not in OFF state */
659 		/* Set PM_PWSTCTRL_IVA2  to OFF */
660 		(*pdata->dsp_prm_rmw_bits)(OMAP_POWERSTATEST_MASK,
661 			PWRDM_POWER_OFF, OMAP3430_IVA2_MOD, OMAP2_PM_PWSTCTRL);
662 		/* Set the SW supervised state transition for Sleep */
663 		(*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_FORCE_SLEEP,
664 					OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
665 	}
666 	udelay(10);
667 	/* Release the Ext Base virtual Address as the next DSP Program
668 	 * may have a different load address */
669 	if (dev_context->dsp_ext_base_addr)
670 		dev_context->dsp_ext_base_addr = 0;
671 
672 	dev_context->brd_state = BRD_STOPPED;	/* update board state */
673 
674 	dsp_wdt_enable(false);
675 
676 	/* This is a good place to clear the MMU page tables as well */
677 	if (dev_context->pt_attrs) {
678 		pt_attrs = dev_context->pt_attrs;
679 		memset((u8 *) pt_attrs->l1_base_va, 0x00, pt_attrs->l1_size);
680 		memset((u8 *) pt_attrs->l2_base_va, 0x00, pt_attrs->l2_size);
681 		memset((u8 *) pt_attrs->pg_info, 0x00,
682 		       (pt_attrs->l2_num_pages * sizeof(struct page_info)));
683 	}
684 	/* Disable the mailbox interrupts */
685 	if (dev_context->mbox) {
686 		omap_mbox_disable_irq(dev_context->mbox, IRQ_RX);
687 		omap_mbox_put(dev_context->mbox, &dsp_mbox_notifier);
688 		dev_context->mbox = NULL;
689 	}
690 	/* Reset IVA2 clocks*/
691 	(*pdata->dsp_prm_write)(OMAP3430_RST1_IVA2_MASK | OMAP3430_RST2_IVA2_MASK |
692 			OMAP3430_RST3_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
693 
694 	dsp_clock_disable_all(dev_context->dsp_per_clks);
695 	dsp_clk_disable(DSP_CLK_IVA2);
696 
697 	return status;
698 }
699 
700 /*
701  *  ======== bridge_brd_status ========
702  *      Returns the board status.
703  */
bridge_brd_status(struct bridge_dev_context * dev_ctxt,int * board_state)704 static int bridge_brd_status(struct bridge_dev_context *dev_ctxt,
705 				    int *board_state)
706 {
707 	struct bridge_dev_context *dev_context = dev_ctxt;
708 	*board_state = dev_context->brd_state;
709 	return 0;
710 }
711 
712 /*
713  *  ======== bridge_brd_write ========
714  *      Copies the buffers to DSP internal or external memory.
715  */
bridge_brd_write(struct bridge_dev_context * dev_ctxt,u8 * host_buff,u32 dsp_addr,u32 ul_num_bytes,u32 mem_type)716 static int bridge_brd_write(struct bridge_dev_context *dev_ctxt,
717 				   u8 *host_buff, u32 dsp_addr,
718 				   u32 ul_num_bytes, u32 mem_type)
719 {
720 	int status = 0;
721 	struct bridge_dev_context *dev_context = dev_ctxt;
722 
723 	if (dsp_addr < dev_context->dsp_start_add) {
724 		status = -EPERM;
725 		return status;
726 	}
727 	if ((dsp_addr - dev_context->dsp_start_add) <
728 	    dev_context->internal_size) {
729 		status = write_dsp_data(dev_ctxt, host_buff, dsp_addr,
730 					ul_num_bytes, mem_type);
731 	} else {
732 		status = write_ext_dsp_data(dev_context, host_buff, dsp_addr,
733 					    ul_num_bytes, mem_type, false);
734 	}
735 
736 	return status;
737 }
738 
739 /*
740  *  ======== bridge_dev_create ========
741  *      Creates a driver object. Puts DSP in self loop.
742  */
bridge_dev_create(struct bridge_dev_context ** dev_cntxt,struct dev_object * hdev_obj,struct cfg_hostres * config_param)743 static int bridge_dev_create(struct bridge_dev_context
744 					**dev_cntxt,
745 					struct dev_object *hdev_obj,
746 					struct cfg_hostres *config_param)
747 {
748 	int status = 0;
749 	struct bridge_dev_context *dev_context = NULL;
750 	s32 entry_ndx;
751 	struct cfg_hostres *resources = config_param;
752 	struct pg_table_attrs *pt_attrs;
753 	u32 pg_tbl_pa;
754 	u32 pg_tbl_va;
755 	u32 align_size;
756 	struct drv_data *drv_datap = dev_get_drvdata(bridge);
757 
758 	/* Allocate and initialize a data structure to contain the bridge driver
759 	 *  state, which becomes the context for later calls into this driver */
760 	dev_context = kzalloc(sizeof(struct bridge_dev_context), GFP_KERNEL);
761 	if (!dev_context) {
762 		status = -ENOMEM;
763 		goto func_end;
764 	}
765 
766 	dev_context->dsp_start_add = (u32) OMAP_GEM_BASE;
767 	dev_context->self_loop = (u32) NULL;
768 	dev_context->dsp_per_clks = 0;
769 	dev_context->internal_size = OMAP_DSP_SIZE;
770 	/*  Clear dev context MMU table entries.
771 	 *  These get set on bridge_io_on_loaded() call after program loaded. */
772 	for (entry_ndx = 0; entry_ndx < BRDIOCTL_NUMOFMMUTLB; entry_ndx++) {
773 		dev_context->atlb_entry[entry_ndx].gpp_pa =
774 		    dev_context->atlb_entry[entry_ndx].dsp_va = 0;
775 	}
776 	dev_context->dsp_base_addr = (u32) MEM_LINEAR_ADDRESS((void *)
777 								 (config_param->
778 								  mem_base
779 								  [3]),
780 								 config_param->
781 								 mem_length
782 								 [3]);
783 	if (!dev_context->dsp_base_addr)
784 		status = -EPERM;
785 
786 	pt_attrs = kzalloc(sizeof(struct pg_table_attrs), GFP_KERNEL);
787 	if (pt_attrs != NULL) {
788 		pt_attrs->l1_size = SZ_16K; /* 4096 entries of 32 bits */
789 		align_size = pt_attrs->l1_size;
790 		/* Align sizes are expected to be power of 2 */
791 		/* we like to get aligned on L1 table size */
792 		pg_tbl_va = (u32) mem_alloc_phys_mem(pt_attrs->l1_size,
793 						     align_size, &pg_tbl_pa);
794 
795 		/* Check if the PA is aligned for us */
796 		if ((pg_tbl_pa) & (align_size - 1)) {
797 			/* PA not aligned to page table size ,
798 			 * try with more allocation and align */
799 			mem_free_phys_mem((void *)pg_tbl_va, pg_tbl_pa,
800 					  pt_attrs->l1_size);
801 			/* we like to get aligned on L1 table size */
802 			pg_tbl_va =
803 			    (u32) mem_alloc_phys_mem((pt_attrs->l1_size) * 2,
804 						     align_size, &pg_tbl_pa);
805 			/* We should be able to get aligned table now */
806 			pt_attrs->l1_tbl_alloc_pa = pg_tbl_pa;
807 			pt_attrs->l1_tbl_alloc_va = pg_tbl_va;
808 			pt_attrs->l1_tbl_alloc_sz = pt_attrs->l1_size * 2;
809 			/* Align the PA to the next 'align'  boundary */
810 			pt_attrs->l1_base_pa =
811 			    ((pg_tbl_pa) +
812 			     (align_size - 1)) & (~(align_size - 1));
813 			pt_attrs->l1_base_va =
814 			    pg_tbl_va + (pt_attrs->l1_base_pa - pg_tbl_pa);
815 		} else {
816 			/* We got aligned PA, cool */
817 			pt_attrs->l1_tbl_alloc_pa = pg_tbl_pa;
818 			pt_attrs->l1_tbl_alloc_va = pg_tbl_va;
819 			pt_attrs->l1_tbl_alloc_sz = pt_attrs->l1_size;
820 			pt_attrs->l1_base_pa = pg_tbl_pa;
821 			pt_attrs->l1_base_va = pg_tbl_va;
822 		}
823 		if (pt_attrs->l1_base_va)
824 			memset((u8 *) pt_attrs->l1_base_va, 0x00,
825 			       pt_attrs->l1_size);
826 
827 		/* number of L2 page tables = DMM pool used + SHMMEM +EXTMEM +
828 		 * L4 pages */
829 		pt_attrs->l2_num_pages = ((DMMPOOLSIZE >> 20) + 6);
830 		pt_attrs->l2_size = HW_MMU_COARSE_PAGE_SIZE *
831 		    pt_attrs->l2_num_pages;
832 		align_size = 4;	/* Make it u32 aligned */
833 		/* we like to get aligned on L1 table size */
834 		pg_tbl_va = (u32) mem_alloc_phys_mem(pt_attrs->l2_size,
835 						     align_size, &pg_tbl_pa);
836 		pt_attrs->l2_tbl_alloc_pa = pg_tbl_pa;
837 		pt_attrs->l2_tbl_alloc_va = pg_tbl_va;
838 		pt_attrs->l2_tbl_alloc_sz = pt_attrs->l2_size;
839 		pt_attrs->l2_base_pa = pg_tbl_pa;
840 		pt_attrs->l2_base_va = pg_tbl_va;
841 
842 		if (pt_attrs->l2_base_va)
843 			memset((u8 *) pt_attrs->l2_base_va, 0x00,
844 			       pt_attrs->l2_size);
845 
846 		pt_attrs->pg_info = kzalloc(pt_attrs->l2_num_pages *
847 					sizeof(struct page_info), GFP_KERNEL);
848 		dev_dbg(bridge,
849 			"L1 pa %x, va %x, size %x\n L2 pa %x, va "
850 			"%x, size %x\n", pt_attrs->l1_base_pa,
851 			pt_attrs->l1_base_va, pt_attrs->l1_size,
852 			pt_attrs->l2_base_pa, pt_attrs->l2_base_va,
853 			pt_attrs->l2_size);
854 		dev_dbg(bridge, "pt_attrs %p L2 NumPages %x pg_info %p\n",
855 			pt_attrs, pt_attrs->l2_num_pages, pt_attrs->pg_info);
856 	}
857 	if ((pt_attrs != NULL) && (pt_attrs->l1_base_va != 0) &&
858 	    (pt_attrs->l2_base_va != 0) && (pt_attrs->pg_info != NULL))
859 		dev_context->pt_attrs = pt_attrs;
860 	else
861 		status = -ENOMEM;
862 
863 	if (!status) {
864 		spin_lock_init(&pt_attrs->pg_lock);
865 		dev_context->tc_word_swap_on = drv_datap->tc_wordswapon;
866 
867 		/* Set the Clock Divisor for the DSP module */
868 		udelay(5);
869 		/* MMU address is obtained from the host
870 		 * resources struct */
871 		dev_context->dsp_mmu_base = resources->dmmu_base;
872 	}
873 	if (!status) {
874 		dev_context->dev_obj = hdev_obj;
875 		/* Store current board state. */
876 		dev_context->brd_state = BRD_UNKNOWN;
877 		dev_context->resources = resources;
878 		dsp_clk_enable(DSP_CLK_IVA2);
879 		bridge_brd_stop(dev_context);
880 		/* Return ptr to our device state to the DSP API for storage */
881 		*dev_cntxt = dev_context;
882 	} else {
883 		if (pt_attrs != NULL) {
884 			kfree(pt_attrs->pg_info);
885 
886 			if (pt_attrs->l2_tbl_alloc_va) {
887 				mem_free_phys_mem((void *)
888 						  pt_attrs->l2_tbl_alloc_va,
889 						  pt_attrs->l2_tbl_alloc_pa,
890 						  pt_attrs->l2_tbl_alloc_sz);
891 			}
892 			if (pt_attrs->l1_tbl_alloc_va) {
893 				mem_free_phys_mem((void *)
894 						  pt_attrs->l1_tbl_alloc_va,
895 						  pt_attrs->l1_tbl_alloc_pa,
896 						  pt_attrs->l1_tbl_alloc_sz);
897 			}
898 		}
899 		kfree(pt_attrs);
900 		kfree(dev_context);
901 	}
902 func_end:
903 	return status;
904 }
905 
906 /*
907  *  ======== bridge_dev_ctrl ========
908  *      Receives device specific commands.
909  */
bridge_dev_ctrl(struct bridge_dev_context * dev_context,u32 dw_cmd,void * pargs)910 static int bridge_dev_ctrl(struct bridge_dev_context *dev_context,
911 				  u32 dw_cmd, void *pargs)
912 {
913 	int status = 0;
914 	struct bridge_ioctl_extproc *pa_ext_proc =
915 					(struct bridge_ioctl_extproc *)pargs;
916 	s32 ndx;
917 
918 	switch (dw_cmd) {
919 	case BRDIOCTL_CHNLREAD:
920 		break;
921 	case BRDIOCTL_CHNLWRITE:
922 		break;
923 	case BRDIOCTL_SETMMUCONFIG:
924 		/* store away dsp-mmu setup values for later use */
925 		for (ndx = 0; ndx < BRDIOCTL_NUMOFMMUTLB; ndx++, pa_ext_proc++)
926 			dev_context->atlb_entry[ndx] = *pa_ext_proc;
927 		break;
928 	case BRDIOCTL_DEEPSLEEP:
929 	case BRDIOCTL_EMERGENCYSLEEP:
930 		/* Currently only DSP Idle is supported Need to update for
931 		 * later releases */
932 		status = sleep_dsp(dev_context, PWR_DEEPSLEEP, pargs);
933 		break;
934 	case BRDIOCTL_WAKEUP:
935 		status = wake_dsp(dev_context, pargs);
936 		break;
937 	case BRDIOCTL_CLK_CTRL:
938 		status = 0;
939 		/* Looking For Baseport Fix for Clocks */
940 		status = dsp_peripheral_clk_ctrl(dev_context, pargs);
941 		break;
942 	case BRDIOCTL_PWR_HIBERNATE:
943 		status = handle_hibernation_from_dsp(dev_context);
944 		break;
945 	case BRDIOCTL_PRESCALE_NOTIFY:
946 		status = pre_scale_dsp(dev_context, pargs);
947 		break;
948 	case BRDIOCTL_POSTSCALE_NOTIFY:
949 		status = post_scale_dsp(dev_context, pargs);
950 		break;
951 	case BRDIOCTL_CONSTRAINT_REQUEST:
952 		status = handle_constraints_set(dev_context, pargs);
953 		break;
954 	default:
955 		status = -EPERM;
956 		break;
957 	}
958 	return status;
959 }
960 
961 /*
962  *  ======== bridge_dev_destroy ========
963  *      Destroys the driver object.
964  */
bridge_dev_destroy(struct bridge_dev_context * dev_ctxt)965 static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt)
966 {
967 	struct pg_table_attrs *pt_attrs;
968 	int status = 0;
969 	struct bridge_dev_context *dev_context = (struct bridge_dev_context *)
970 	    dev_ctxt;
971 	struct cfg_hostres *host_res;
972 	u32 shm_size;
973 	struct drv_data *drv_datap = dev_get_drvdata(bridge);
974 
975 	/* It should never happen */
976 	if (!dev_ctxt)
977 		return -EFAULT;
978 
979 	/* first put the device to stop state */
980 	bridge_brd_stop(dev_context);
981 	if (dev_context->pt_attrs) {
982 		pt_attrs = dev_context->pt_attrs;
983 		kfree(pt_attrs->pg_info);
984 
985 		if (pt_attrs->l2_tbl_alloc_va) {
986 			mem_free_phys_mem((void *)pt_attrs->l2_tbl_alloc_va,
987 					  pt_attrs->l2_tbl_alloc_pa,
988 					  pt_attrs->l2_tbl_alloc_sz);
989 		}
990 		if (pt_attrs->l1_tbl_alloc_va) {
991 			mem_free_phys_mem((void *)pt_attrs->l1_tbl_alloc_va,
992 					  pt_attrs->l1_tbl_alloc_pa,
993 					  pt_attrs->l1_tbl_alloc_sz);
994 		}
995 		kfree(pt_attrs);
996 
997 	}
998 
999 	if (dev_context->resources) {
1000 		host_res = dev_context->resources;
1001 		shm_size = drv_datap->shm_size;
1002 		if (shm_size >= 0x10000) {
1003 			if ((host_res->mem_base[1]) &&
1004 			    (host_res->mem_phys[1])) {
1005 				mem_free_phys_mem((void *)
1006 						  host_res->mem_base
1007 						  [1],
1008 						  host_res->mem_phys
1009 						  [1], shm_size);
1010 			}
1011 		} else {
1012 			dev_dbg(bridge, "%s: Error getting shm size "
1013 				"from registry: %x. Not calling "
1014 				"mem_free_phys_mem\n", __func__,
1015 				status);
1016 		}
1017 		host_res->mem_base[1] = 0;
1018 		host_res->mem_phys[1] = 0;
1019 
1020 		if (host_res->mem_base[0])
1021 			iounmap((void *)host_res->mem_base[0]);
1022 		if (host_res->mem_base[2])
1023 			iounmap((void *)host_res->mem_base[2]);
1024 		if (host_res->mem_base[3])
1025 			iounmap((void *)host_res->mem_base[3]);
1026 		if (host_res->mem_base[4])
1027 			iounmap((void *)host_res->mem_base[4]);
1028 		if (host_res->dmmu_base)
1029 			iounmap(host_res->dmmu_base);
1030 		if (host_res->per_base)
1031 			iounmap(host_res->per_base);
1032 		if (host_res->per_pm_base)
1033 			iounmap((void *)host_res->per_pm_base);
1034 		if (host_res->core_pm_base)
1035 			iounmap((void *)host_res->core_pm_base);
1036 
1037 		host_res->mem_base[0] = (u32) NULL;
1038 		host_res->mem_base[2] = (u32) NULL;
1039 		host_res->mem_base[3] = (u32) NULL;
1040 		host_res->mem_base[4] = (u32) NULL;
1041 		host_res->dmmu_base = NULL;
1042 
1043 		kfree(host_res);
1044 	}
1045 
1046 	/* Free the driver's device context: */
1047 	kfree(drv_datap->base_img);
1048 	kfree((void *)dev_ctxt);
1049 	return status;
1050 }
1051 
bridge_brd_mem_copy(struct bridge_dev_context * dev_ctxt,u32 dsp_dest_addr,u32 dsp_src_addr,u32 ul_num_bytes,u32 mem_type)1052 static int bridge_brd_mem_copy(struct bridge_dev_context *dev_ctxt,
1053 				   u32 dsp_dest_addr, u32 dsp_src_addr,
1054 				   u32 ul_num_bytes, u32 mem_type)
1055 {
1056 	int status = 0;
1057 	u32 src_addr = dsp_src_addr;
1058 	u32 dest_addr = dsp_dest_addr;
1059 	u32 copy_bytes = 0;
1060 	u32 total_bytes = ul_num_bytes;
1061 	u8 host_buf[BUFFERSIZE];
1062 	struct bridge_dev_context *dev_context = dev_ctxt;
1063 	while (total_bytes > 0 && !status) {
1064 		copy_bytes =
1065 		    total_bytes > BUFFERSIZE ? BUFFERSIZE : total_bytes;
1066 		/* Read from External memory */
1067 		status = read_ext_dsp_data(dev_ctxt, host_buf, src_addr,
1068 					   copy_bytes, mem_type);
1069 		if (!status) {
1070 			if (dest_addr < (dev_context->dsp_start_add +
1071 					 dev_context->internal_size)) {
1072 				/* Write to Internal memory */
1073 				status = write_dsp_data(dev_ctxt, host_buf,
1074 							dest_addr, copy_bytes,
1075 							mem_type);
1076 			} else {
1077 				/* Write to External memory */
1078 				status =
1079 				    write_ext_dsp_data(dev_ctxt, host_buf,
1080 						       dest_addr, copy_bytes,
1081 						       mem_type, false);
1082 			}
1083 		}
1084 		total_bytes -= copy_bytes;
1085 		src_addr += copy_bytes;
1086 		dest_addr += copy_bytes;
1087 	}
1088 	return status;
1089 }
1090 
1091 /* Mem Write does not halt the DSP to write unlike bridge_brd_write */
bridge_brd_mem_write(struct bridge_dev_context * dev_ctxt,u8 * host_buff,u32 dsp_addr,u32 ul_num_bytes,u32 mem_type)1092 static int bridge_brd_mem_write(struct bridge_dev_context *dev_ctxt,
1093 				    u8 *host_buff, u32 dsp_addr,
1094 				    u32 ul_num_bytes, u32 mem_type)
1095 {
1096 	int status = 0;
1097 	struct bridge_dev_context *dev_context = dev_ctxt;
1098 	u32 ul_remain_bytes = 0;
1099 	u32 ul_bytes = 0;
1100 	ul_remain_bytes = ul_num_bytes;
1101 	while (ul_remain_bytes > 0 && !status) {
1102 		ul_bytes =
1103 		    ul_remain_bytes > BUFFERSIZE ? BUFFERSIZE : ul_remain_bytes;
1104 		if (dsp_addr < (dev_context->dsp_start_add +
1105 				 dev_context->internal_size)) {
1106 			status =
1107 			    write_dsp_data(dev_ctxt, host_buff, dsp_addr,
1108 					   ul_bytes, mem_type);
1109 		} else {
1110 			status = write_ext_dsp_data(dev_ctxt, host_buff,
1111 						    dsp_addr, ul_bytes,
1112 						    mem_type, true);
1113 		}
1114 		ul_remain_bytes -= ul_bytes;
1115 		dsp_addr += ul_bytes;
1116 		host_buff = host_buff + ul_bytes;
1117 	}
1118 	return status;
1119 }
1120 
1121 /*
1122  *  ======== bridge_brd_mem_map ========
1123  *      This function maps MPU buffer to the DSP address space. It performs
1124  *  linear to physical address translation if required. It translates each
1125  *  page since linear addresses can be physically non-contiguous
1126  *  All address & size arguments are assumed to be page aligned (in proc.c)
1127  *
1128  *  TODO: Disable MMU while updating the page tables (but that'll stall DSP)
1129  */
bridge_brd_mem_map(struct bridge_dev_context * dev_ctxt,u32 ul_mpu_addr,u32 virt_addr,u32 ul_num_bytes,u32 ul_map_attr,struct page ** mapped_pages)1130 static int bridge_brd_mem_map(struct bridge_dev_context *dev_ctxt,
1131 				  u32 ul_mpu_addr, u32 virt_addr,
1132 				  u32 ul_num_bytes, u32 ul_map_attr,
1133 				  struct page **mapped_pages)
1134 {
1135 	u32 attrs;
1136 	int status = 0;
1137 	struct bridge_dev_context *dev_context = dev_ctxt;
1138 	struct hw_mmu_map_attrs_t hw_attrs;
1139 	struct vm_area_struct *vma;
1140 	struct mm_struct *mm = current->mm;
1141 	u32 write = 0;
1142 	u32 num_usr_pgs = 0;
1143 	struct page *mapped_page, *pg;
1144 	s32 pg_num;
1145 	u32 va = virt_addr;
1146 	struct task_struct *curr_task = current;
1147 	u32 pg_i = 0;
1148 	u32 mpu_addr, pa;
1149 
1150 	dev_dbg(bridge,
1151 		"%s hDevCtxt %p, pa %x, va %x, size %x, ul_map_attr %x\n",
1152 		__func__, dev_ctxt, ul_mpu_addr, virt_addr, ul_num_bytes,
1153 		ul_map_attr);
1154 	if (ul_num_bytes == 0)
1155 		return -EINVAL;
1156 
1157 	if (ul_map_attr & DSP_MAP_DIR_MASK) {
1158 		attrs = ul_map_attr;
1159 	} else {
1160 		/* Assign default attributes */
1161 		attrs = ul_map_attr | (DSP_MAPVIRTUALADDR | DSP_MAPELEMSIZE16);
1162 	}
1163 	/* Take mapping properties */
1164 	if (attrs & DSP_MAPBIGENDIAN)
1165 		hw_attrs.endianism = HW_BIG_ENDIAN;
1166 	else
1167 		hw_attrs.endianism = HW_LITTLE_ENDIAN;
1168 
1169 	hw_attrs.mixed_size = (enum hw_mmu_mixed_size_t)
1170 	    ((attrs & DSP_MAPMIXEDELEMSIZE) >> 2);
1171 	/* Ignore element_size if mixed_size is enabled */
1172 	if (hw_attrs.mixed_size == 0) {
1173 		if (attrs & DSP_MAPELEMSIZE8) {
1174 			/* Size is 8 bit */
1175 			hw_attrs.element_size = HW_ELEM_SIZE8BIT;
1176 		} else if (attrs & DSP_MAPELEMSIZE16) {
1177 			/* Size is 16 bit */
1178 			hw_attrs.element_size = HW_ELEM_SIZE16BIT;
1179 		} else if (attrs & DSP_MAPELEMSIZE32) {
1180 			/* Size is 32 bit */
1181 			hw_attrs.element_size = HW_ELEM_SIZE32BIT;
1182 		} else if (attrs & DSP_MAPELEMSIZE64) {
1183 			/* Size is 64 bit */
1184 			hw_attrs.element_size = HW_ELEM_SIZE64BIT;
1185 		} else {
1186 			/*
1187 			 * Mixedsize isn't enabled, so size can't be
1188 			 * zero here
1189 			 */
1190 			return -EINVAL;
1191 		}
1192 	}
1193 	if (attrs & DSP_MAPDONOTLOCK)
1194 		hw_attrs.donotlockmpupage = 1;
1195 	else
1196 		hw_attrs.donotlockmpupage = 0;
1197 
1198 	if (attrs & DSP_MAPVMALLOCADDR) {
1199 		return mem_map_vmalloc(dev_ctxt, ul_mpu_addr, virt_addr,
1200 				       ul_num_bytes, &hw_attrs);
1201 	}
1202 	/*
1203 	 * Do OS-specific user-va to pa translation.
1204 	 * Combine physically contiguous regions to reduce TLBs.
1205 	 * Pass the translated pa to pte_update.
1206 	 */
1207 	if ((attrs & DSP_MAPPHYSICALADDR)) {
1208 		status = pte_update(dev_context, ul_mpu_addr, virt_addr,
1209 				    ul_num_bytes, &hw_attrs);
1210 		goto func_cont;
1211 	}
1212 
1213 	/*
1214 	 * Important Note: ul_mpu_addr is mapped from user application process
1215 	 * to current process - it must lie completely within the current
1216 	 * virtual memory address space in order to be of use to us here!
1217 	 */
1218 	down_read(&mm->mmap_sem);
1219 	vma = find_vma(mm, ul_mpu_addr);
1220 	if (vma)
1221 		dev_dbg(bridge,
1222 			"VMAfor UserBuf: ul_mpu_addr=%x, ul_num_bytes=%x, "
1223 			"vm_start=%lx, vm_end=%lx, vm_flags=%lx\n", ul_mpu_addr,
1224 			ul_num_bytes, vma->vm_start, vma->vm_end,
1225 			vma->vm_flags);
1226 
1227 	/*
1228 	 * It is observed that under some circumstances, the user buffer is
1229 	 * spread across several VMAs. So loop through and check if the entire
1230 	 * user buffer is covered
1231 	 */
1232 	while ((vma) && (ul_mpu_addr + ul_num_bytes > vma->vm_end)) {
1233 		/* jump to the next VMA region */
1234 		vma = find_vma(mm, vma->vm_end + 1);
1235 		dev_dbg(bridge,
1236 			"VMA for UserBuf ul_mpu_addr=%x ul_num_bytes=%x, "
1237 			"vm_start=%lx, vm_end=%lx, vm_flags=%lx\n", ul_mpu_addr,
1238 			ul_num_bytes, vma->vm_start, vma->vm_end,
1239 			vma->vm_flags);
1240 	}
1241 	if (!vma) {
1242 		pr_err("%s: Failed to get VMA region for 0x%x (%d)\n",
1243 		       __func__, ul_mpu_addr, ul_num_bytes);
1244 		status = -EINVAL;
1245 		up_read(&mm->mmap_sem);
1246 		goto func_cont;
1247 	}
1248 
1249 	if (vma->vm_flags & VM_IO) {
1250 		num_usr_pgs = ul_num_bytes / PG_SIZE4K;
1251 		mpu_addr = ul_mpu_addr;
1252 
1253 		/* Get the physical addresses for user buffer */
1254 		for (pg_i = 0; pg_i < num_usr_pgs; pg_i++) {
1255 			pa = user_va2_pa(mm, mpu_addr);
1256 			if (!pa) {
1257 				status = -EPERM;
1258 				pr_err("DSPBRIDGE: VM_IO mapping physical"
1259 				       "address is invalid\n");
1260 				break;
1261 			}
1262 			if (pfn_valid(__phys_to_pfn(pa))) {
1263 				pg = PHYS_TO_PAGE(pa);
1264 				get_page(pg);
1265 				if (page_count(pg) < 1) {
1266 					pr_err("Bad page in VM_IO buffer\n");
1267 					bad_page_dump(pa, pg);
1268 				}
1269 			}
1270 			status = pte_set(dev_context->pt_attrs, pa,
1271 					 va, HW_PAGE_SIZE4KB, &hw_attrs);
1272 			if (status)
1273 				break;
1274 
1275 			va += HW_PAGE_SIZE4KB;
1276 			mpu_addr += HW_PAGE_SIZE4KB;
1277 			pa += HW_PAGE_SIZE4KB;
1278 		}
1279 	} else {
1280 		num_usr_pgs = ul_num_bytes / PG_SIZE4K;
1281 		if (vma->vm_flags & (VM_WRITE | VM_MAYWRITE))
1282 			write = 1;
1283 
1284 		for (pg_i = 0; pg_i < num_usr_pgs; pg_i++) {
1285 			pg_num = get_user_pages(curr_task, mm, ul_mpu_addr, 1,
1286 						write, 1, &mapped_page, NULL);
1287 			if (pg_num > 0) {
1288 				if (page_count(mapped_page) < 1) {
1289 					pr_err("Bad page count after doing"
1290 					       "get_user_pages on"
1291 					       "user buffer\n");
1292 					bad_page_dump(page_to_phys(mapped_page),
1293 						      mapped_page);
1294 				}
1295 				status = pte_set(dev_context->pt_attrs,
1296 						 page_to_phys(mapped_page), va,
1297 						 HW_PAGE_SIZE4KB, &hw_attrs);
1298 				if (status)
1299 					break;
1300 
1301 				if (mapped_pages)
1302 					mapped_pages[pg_i] = mapped_page;
1303 
1304 				va += HW_PAGE_SIZE4KB;
1305 				ul_mpu_addr += HW_PAGE_SIZE4KB;
1306 			} else {
1307 				pr_err("DSPBRIDGE: get_user_pages FAILED,"
1308 				       "MPU addr = 0x%x,"
1309 				       "vma->vm_flags = 0x%lx,"
1310 				       "get_user_pages Err"
1311 				       "Value = %d, Buffer"
1312 				       "size=0x%x\n", ul_mpu_addr,
1313 				       vma->vm_flags, pg_num, ul_num_bytes);
1314 				status = -EPERM;
1315 				break;
1316 			}
1317 		}
1318 	}
1319 	up_read(&mm->mmap_sem);
1320 func_cont:
1321 	if (status) {
1322 		/*
1323 		 * Roll out the mapped pages incase it failed in middle of
1324 		 * mapping
1325 		 */
1326 		if (pg_i) {
1327 			bridge_brd_mem_un_map(dev_context, virt_addr,
1328 					   (pg_i * PG_SIZE4K));
1329 		}
1330 		status = -EPERM;
1331 	}
1332 	/*
1333 	 * In any case, flush the TLB
1334 	 * This is called from here instead from pte_update to avoid unnecessary
1335 	 * repetition while mapping non-contiguous physical regions of a virtual
1336 	 * region
1337 	 */
1338 	flush_all(dev_context);
1339 	dev_dbg(bridge, "%s status %x\n", __func__, status);
1340 	return status;
1341 }
1342 
1343 /*
1344  *  ======== bridge_brd_mem_un_map ========
1345  *      Invalidate the PTEs for the DSP VA block to be unmapped.
1346  *
1347  *      PTEs of a mapped memory block are contiguous in any page table
1348  *      So, instead of looking up the PTE address for every 4K block,
1349  *      we clear consecutive PTEs until we unmap all the bytes
1350  */
bridge_brd_mem_un_map(struct bridge_dev_context * dev_ctxt,u32 virt_addr,u32 ul_num_bytes)1351 static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctxt,
1352 				     u32 virt_addr, u32 ul_num_bytes)
1353 {
1354 	u32 l1_base_va;
1355 	u32 l2_base_va;
1356 	u32 l2_base_pa;
1357 	u32 l2_page_num;
1358 	u32 pte_val;
1359 	u32 pte_size;
1360 	u32 pte_count;
1361 	u32 pte_addr_l1;
1362 	u32 pte_addr_l2 = 0;
1363 	u32 rem_bytes;
1364 	u32 rem_bytes_l2;
1365 	u32 va_curr;
1366 	struct page *pg = NULL;
1367 	int status = 0;
1368 	struct bridge_dev_context *dev_context = dev_ctxt;
1369 	struct pg_table_attrs *pt = dev_context->pt_attrs;
1370 	u32 temp;
1371 	u32 paddr;
1372 	u32 numof4k_pages = 0;
1373 
1374 	va_curr = virt_addr;
1375 	rem_bytes = ul_num_bytes;
1376 	rem_bytes_l2 = 0;
1377 	l1_base_va = pt->l1_base_va;
1378 	pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va_curr);
1379 	dev_dbg(bridge, "%s dev_ctxt %p, va %x, NumBytes %x l1_base_va %x, "
1380 		"pte_addr_l1 %x\n", __func__, dev_ctxt, virt_addr,
1381 		ul_num_bytes, l1_base_va, pte_addr_l1);
1382 
1383 	while (rem_bytes && !status) {
1384 		u32 va_curr_orig = va_curr;
1385 		/* Find whether the L1 PTE points to a valid L2 PT */
1386 		pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va_curr);
1387 		pte_val = *(u32 *) pte_addr_l1;
1388 		pte_size = hw_mmu_pte_size_l1(pte_val);
1389 
1390 		if (pte_size != HW_MMU_COARSE_PAGE_SIZE)
1391 			goto skip_coarse_page;
1392 
1393 		/*
1394 		 * Get the L2 PA from the L1 PTE, and find
1395 		 * corresponding L2 VA
1396 		 */
1397 		l2_base_pa = hw_mmu_pte_coarse_l1(pte_val);
1398 		l2_base_va = l2_base_pa - pt->l2_base_pa + pt->l2_base_va;
1399 		l2_page_num =
1400 		    (l2_base_pa - pt->l2_base_pa) / HW_MMU_COARSE_PAGE_SIZE;
1401 		/*
1402 		 * Find the L2 PTE address from which we will start
1403 		 * clearing, the number of PTEs to be cleared on this
1404 		 * page, and the size of VA space that needs to be
1405 		 * cleared on this L2 page
1406 		 */
1407 		pte_addr_l2 = hw_mmu_pte_addr_l2(l2_base_va, va_curr);
1408 		pte_count = pte_addr_l2 & (HW_MMU_COARSE_PAGE_SIZE - 1);
1409 		pte_count = (HW_MMU_COARSE_PAGE_SIZE - pte_count) / sizeof(u32);
1410 		if (rem_bytes < (pte_count * PG_SIZE4K))
1411 			pte_count = rem_bytes / PG_SIZE4K;
1412 		rem_bytes_l2 = pte_count * PG_SIZE4K;
1413 
1414 		/*
1415 		 * Unmap the VA space on this L2 PT. A quicker way
1416 		 * would be to clear pte_count entries starting from
1417 		 * pte_addr_l2. However, below code checks that we don't
1418 		 * clear invalid entries or less than 64KB for a 64KB
1419 		 * entry. Similar checking is done for L1 PTEs too
1420 		 * below
1421 		 */
1422 		while (rem_bytes_l2 && !status) {
1423 			pte_val = *(u32 *) pte_addr_l2;
1424 			pte_size = hw_mmu_pte_size_l2(pte_val);
1425 			/* va_curr aligned to pte_size? */
1426 			if (pte_size == 0 || rem_bytes_l2 < pte_size ||
1427 			    va_curr & (pte_size - 1)) {
1428 				status = -EPERM;
1429 				break;
1430 			}
1431 
1432 			/* Collect Physical addresses from VA */
1433 			paddr = (pte_val & ~(pte_size - 1));
1434 			if (pte_size == HW_PAGE_SIZE64KB)
1435 				numof4k_pages = 16;
1436 			else
1437 				numof4k_pages = 1;
1438 			temp = 0;
1439 			while (temp++ < numof4k_pages) {
1440 				if (!pfn_valid(__phys_to_pfn(paddr))) {
1441 					paddr += HW_PAGE_SIZE4KB;
1442 					continue;
1443 				}
1444 				pg = PHYS_TO_PAGE(paddr);
1445 				if (page_count(pg) < 1) {
1446 					pr_info("DSPBRIDGE: UNMAP function: "
1447 						"COUNT 0 FOR PA 0x%x, size = "
1448 						"0x%x\n", paddr, ul_num_bytes);
1449 					bad_page_dump(paddr, pg);
1450 				} else {
1451 					set_page_dirty(pg);
1452 					page_cache_release(pg);
1453 				}
1454 				paddr += HW_PAGE_SIZE4KB;
1455 			}
1456 			if (hw_mmu_pte_clear(pte_addr_l2, va_curr, pte_size)) {
1457 				status = -EPERM;
1458 				goto EXIT_LOOP;
1459 			}
1460 
1461 			status = 0;
1462 			rem_bytes_l2 -= pte_size;
1463 			va_curr += pte_size;
1464 			pte_addr_l2 += (pte_size >> 12) * sizeof(u32);
1465 		}
1466 		spin_lock(&pt->pg_lock);
1467 		if (rem_bytes_l2 == 0) {
1468 			pt->pg_info[l2_page_num].num_entries -= pte_count;
1469 			if (pt->pg_info[l2_page_num].num_entries == 0) {
1470 				/*
1471 				 * Clear the L1 PTE pointing to the L2 PT
1472 				 */
1473 				if (!hw_mmu_pte_clear(l1_base_va, va_curr_orig,
1474 						     HW_MMU_COARSE_PAGE_SIZE))
1475 					status = 0;
1476 				else {
1477 					status = -EPERM;
1478 					spin_unlock(&pt->pg_lock);
1479 					goto EXIT_LOOP;
1480 				}
1481 			}
1482 			rem_bytes -= pte_count * PG_SIZE4K;
1483 		} else
1484 			status = -EPERM;
1485 
1486 		spin_unlock(&pt->pg_lock);
1487 		continue;
1488 skip_coarse_page:
1489 		/* va_curr aligned to pte_size? */
1490 		/* pte_size = 1 MB or 16 MB */
1491 		if (pte_size == 0 || rem_bytes < pte_size ||
1492 		    va_curr & (pte_size - 1)) {
1493 			status = -EPERM;
1494 			break;
1495 		}
1496 
1497 		if (pte_size == HW_PAGE_SIZE1MB)
1498 			numof4k_pages = 256;
1499 		else
1500 			numof4k_pages = 4096;
1501 		temp = 0;
1502 		/* Collect Physical addresses from VA */
1503 		paddr = (pte_val & ~(pte_size - 1));
1504 		while (temp++ < numof4k_pages) {
1505 			if (pfn_valid(__phys_to_pfn(paddr))) {
1506 				pg = PHYS_TO_PAGE(paddr);
1507 				if (page_count(pg) < 1) {
1508 					pr_info("DSPBRIDGE: UNMAP function: "
1509 						"COUNT 0 FOR PA 0x%x, size = "
1510 						"0x%x\n", paddr, ul_num_bytes);
1511 					bad_page_dump(paddr, pg);
1512 				} else {
1513 					set_page_dirty(pg);
1514 					page_cache_release(pg);
1515 				}
1516 			}
1517 			paddr += HW_PAGE_SIZE4KB;
1518 		}
1519 		if (!hw_mmu_pte_clear(l1_base_va, va_curr, pte_size)) {
1520 			status = 0;
1521 			rem_bytes -= pte_size;
1522 			va_curr += pte_size;
1523 		} else {
1524 			status = -EPERM;
1525 			goto EXIT_LOOP;
1526 		}
1527 	}
1528 	/*
1529 	 * It is better to flush the TLB here, so that any stale old entries
1530 	 * get flushed
1531 	 */
1532 EXIT_LOOP:
1533 	flush_all(dev_context);
1534 	dev_dbg(bridge,
1535 		"%s: va_curr %x, pte_addr_l1 %x pte_addr_l2 %x rem_bytes %x,"
1536 		" rem_bytes_l2 %x status %x\n", __func__, va_curr, pte_addr_l1,
1537 		pte_addr_l2, rem_bytes, rem_bytes_l2, status);
1538 	return status;
1539 }
1540 
1541 /*
1542  *  ======== user_va2_pa ========
1543  *  Purpose:
1544  *      This function walks through the page tables to convert a userland
1545  *      virtual address to physical address
1546  */
user_va2_pa(struct mm_struct * mm,u32 address)1547 static u32 user_va2_pa(struct mm_struct *mm, u32 address)
1548 {
1549 	pgd_t *pgd;
1550 	pmd_t *pmd;
1551 	pte_t *ptep, pte;
1552 
1553 	pgd = pgd_offset(mm, address);
1554 	if (!(pgd_none(*pgd) || pgd_bad(*pgd))) {
1555 		pmd = pmd_offset(pgd, address);
1556 		if (!(pmd_none(*pmd) || pmd_bad(*pmd))) {
1557 			ptep = pte_offset_map(pmd, address);
1558 			if (ptep) {
1559 				pte = *ptep;
1560 				if (pte_present(pte))
1561 					return pte & PAGE_MASK;
1562 			}
1563 		}
1564 	}
1565 
1566 	return 0;
1567 }
1568 
1569 /*
1570  *  ======== pte_update ========
1571  *      This function calculates the optimum page-aligned addresses and sizes
1572  *      Caller must pass page-aligned values
1573  */
pte_update(struct bridge_dev_context * dev_ctxt,u32 pa,u32 va,u32 size,struct hw_mmu_map_attrs_t * map_attrs)1574 static int pte_update(struct bridge_dev_context *dev_ctxt, u32 pa,
1575 			     u32 va, u32 size,
1576 			     struct hw_mmu_map_attrs_t *map_attrs)
1577 {
1578 	u32 i;
1579 	u32 all_bits;
1580 	u32 pa_curr = pa;
1581 	u32 va_curr = va;
1582 	u32 num_bytes = size;
1583 	struct bridge_dev_context *dev_context = dev_ctxt;
1584 	int status = 0;
1585 	u32 page_size[] = { HW_PAGE_SIZE16MB, HW_PAGE_SIZE1MB,
1586 		HW_PAGE_SIZE64KB, HW_PAGE_SIZE4KB
1587 	};
1588 
1589 	while (num_bytes && !status) {
1590 		/* To find the max. page size with which both PA & VA are
1591 		 * aligned */
1592 		all_bits = pa_curr | va_curr;
1593 
1594 		for (i = 0; i < 4; i++) {
1595 			if ((num_bytes >= page_size[i]) && ((all_bits &
1596 							     (page_size[i] -
1597 							      1)) == 0)) {
1598 				status =
1599 				    pte_set(dev_context->pt_attrs, pa_curr,
1600 					    va_curr, page_size[i], map_attrs);
1601 				pa_curr += page_size[i];
1602 				va_curr += page_size[i];
1603 				num_bytes -= page_size[i];
1604 				/* Don't try smaller sizes. Hopefully we have
1605 				 * reached an address aligned to a bigger page
1606 				 * size */
1607 				break;
1608 			}
1609 		}
1610 	}
1611 
1612 	return status;
1613 }
1614 
1615 /*
1616  *  ======== pte_set ========
1617  *      This function calculates PTE address (MPU virtual) to be updated
1618  *      It also manages the L2 page tables
1619  */
pte_set(struct pg_table_attrs * pt,u32 pa,u32 va,u32 size,struct hw_mmu_map_attrs_t * attrs)1620 static int pte_set(struct pg_table_attrs *pt, u32 pa, u32 va,
1621 			  u32 size, struct hw_mmu_map_attrs_t *attrs)
1622 {
1623 	u32 i;
1624 	u32 pte_val;
1625 	u32 pte_addr_l1;
1626 	u32 pte_size;
1627 	/* Base address of the PT that will be updated */
1628 	u32 pg_tbl_va;
1629 	u32 l1_base_va;
1630 	/* Compiler warns that the next three variables might be used
1631 	 * uninitialized in this function. Doesn't seem so. Working around,
1632 	 * anyways. */
1633 	u32 l2_base_va = 0;
1634 	u32 l2_base_pa = 0;
1635 	u32 l2_page_num = 0;
1636 	int status = 0;
1637 
1638 	l1_base_va = pt->l1_base_va;
1639 	pg_tbl_va = l1_base_va;
1640 	if ((size == HW_PAGE_SIZE64KB) || (size == HW_PAGE_SIZE4KB)) {
1641 		/* Find whether the L1 PTE points to a valid L2 PT */
1642 		pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va);
1643 		if (pte_addr_l1 <= (pt->l1_base_va + pt->l1_size)) {
1644 			pte_val = *(u32 *) pte_addr_l1;
1645 			pte_size = hw_mmu_pte_size_l1(pte_val);
1646 		} else {
1647 			return -EPERM;
1648 		}
1649 		spin_lock(&pt->pg_lock);
1650 		if (pte_size == HW_MMU_COARSE_PAGE_SIZE) {
1651 			/* Get the L2 PA from the L1 PTE, and find
1652 			 * corresponding L2 VA */
1653 			l2_base_pa = hw_mmu_pte_coarse_l1(pte_val);
1654 			l2_base_va =
1655 			    l2_base_pa - pt->l2_base_pa + pt->l2_base_va;
1656 			l2_page_num =
1657 			    (l2_base_pa -
1658 			     pt->l2_base_pa) / HW_MMU_COARSE_PAGE_SIZE;
1659 		} else if (pte_size == 0) {
1660 			/* L1 PTE is invalid. Allocate a L2 PT and
1661 			 * point the L1 PTE to it */
1662 			/* Find a free L2 PT. */
1663 			for (i = 0; (i < pt->l2_num_pages) &&
1664 			     (pt->pg_info[i].num_entries != 0); i++)
1665 				;
1666 			if (i < pt->l2_num_pages) {
1667 				l2_page_num = i;
1668 				l2_base_pa = pt->l2_base_pa + (l2_page_num *
1669 						HW_MMU_COARSE_PAGE_SIZE);
1670 				l2_base_va = pt->l2_base_va + (l2_page_num *
1671 						HW_MMU_COARSE_PAGE_SIZE);
1672 				/* Endianness attributes are ignored for
1673 				 * HW_MMU_COARSE_PAGE_SIZE */
1674 				status =
1675 				    hw_mmu_pte_set(l1_base_va, l2_base_pa, va,
1676 						   HW_MMU_COARSE_PAGE_SIZE,
1677 						   attrs);
1678 			} else {
1679 				status = -ENOMEM;
1680 			}
1681 		} else {
1682 			/* Found valid L1 PTE of another size.
1683 			 * Should not overwrite it. */
1684 			status = -EPERM;
1685 		}
1686 		if (!status) {
1687 			pg_tbl_va = l2_base_va;
1688 			if (size == HW_PAGE_SIZE64KB)
1689 				pt->pg_info[l2_page_num].num_entries += 16;
1690 			else
1691 				pt->pg_info[l2_page_num].num_entries++;
1692 			dev_dbg(bridge, "PTE: L2 BaseVa %x, BasePa %x, PageNum "
1693 				"%x, num_entries %x\n", l2_base_va,
1694 				l2_base_pa, l2_page_num,
1695 				pt->pg_info[l2_page_num].num_entries);
1696 		}
1697 		spin_unlock(&pt->pg_lock);
1698 	}
1699 	if (!status) {
1700 		dev_dbg(bridge, "PTE: pg_tbl_va %x, pa %x, va %x, size %x\n",
1701 			pg_tbl_va, pa, va, size);
1702 		dev_dbg(bridge, "PTE: endianism %x, element_size %x, "
1703 			"mixed_size %x\n", attrs->endianism,
1704 			attrs->element_size, attrs->mixed_size);
1705 		status = hw_mmu_pte_set(pg_tbl_va, pa, va, size, attrs);
1706 	}
1707 
1708 	return status;
1709 }
1710 
1711 /* Memory map kernel VA -- memory allocated with vmalloc */
mem_map_vmalloc(struct bridge_dev_context * dev_context,u32 ul_mpu_addr,u32 virt_addr,u32 ul_num_bytes,struct hw_mmu_map_attrs_t * hw_attrs)1712 static int mem_map_vmalloc(struct bridge_dev_context *dev_context,
1713 				  u32 ul_mpu_addr, u32 virt_addr,
1714 				  u32 ul_num_bytes,
1715 				  struct hw_mmu_map_attrs_t *hw_attrs)
1716 {
1717 	int status = 0;
1718 	struct page *page[1];
1719 	u32 i;
1720 	u32 pa_curr;
1721 	u32 pa_next;
1722 	u32 va_curr;
1723 	u32 size_curr;
1724 	u32 num_pages;
1725 	u32 pa;
1726 	u32 num_of4k_pages;
1727 	u32 temp = 0;
1728 
1729 	/*
1730 	 * Do Kernel va to pa translation.
1731 	 * Combine physically contiguous regions to reduce TLBs.
1732 	 * Pass the translated pa to pte_update.
1733 	 */
1734 	num_pages = ul_num_bytes / PAGE_SIZE;	/* PAGE_SIZE = OS page size */
1735 	i = 0;
1736 	va_curr = ul_mpu_addr;
1737 	page[0] = vmalloc_to_page((void *)va_curr);
1738 	pa_next = page_to_phys(page[0]);
1739 	while (!status && (i < num_pages)) {
1740 		/*
1741 		 * Reuse pa_next from the previous iteraion to avoid
1742 		 * an extra va2pa call
1743 		 */
1744 		pa_curr = pa_next;
1745 		size_curr = PAGE_SIZE;
1746 		/*
1747 		 * If the next page is physically contiguous,
1748 		 * map it with the current one by increasing
1749 		 * the size of the region to be mapped
1750 		 */
1751 		while (++i < num_pages) {
1752 			page[0] =
1753 			    vmalloc_to_page((void *)(va_curr + size_curr));
1754 			pa_next = page_to_phys(page[0]);
1755 
1756 			if (pa_next == (pa_curr + size_curr))
1757 				size_curr += PAGE_SIZE;
1758 			else
1759 				break;
1760 
1761 		}
1762 		if (pa_next == 0) {
1763 			status = -ENOMEM;
1764 			break;
1765 		}
1766 		pa = pa_curr;
1767 		num_of4k_pages = size_curr / HW_PAGE_SIZE4KB;
1768 		while (temp++ < num_of4k_pages) {
1769 			get_page(PHYS_TO_PAGE(pa));
1770 			pa += HW_PAGE_SIZE4KB;
1771 		}
1772 		status = pte_update(dev_context, pa_curr, virt_addr +
1773 				    (va_curr - ul_mpu_addr), size_curr,
1774 				    hw_attrs);
1775 		va_curr += size_curr;
1776 	}
1777 	/*
1778 	 * In any case, flush the TLB
1779 	 * This is called from here instead from pte_update to avoid unnecessary
1780 	 * repetition while mapping non-contiguous physical regions of a virtual
1781 	 * region
1782 	 */
1783 	flush_all(dev_context);
1784 	dev_dbg(bridge, "%s status %x\n", __func__, status);
1785 	return status;
1786 }
1787 
1788 /*
1789  *  ======== wait_for_start ========
1790  *      Wait for the singal from DSP that it has started, or time out.
1791  */
wait_for_start(struct bridge_dev_context * dev_context,u32 dw_sync_addr)1792 bool wait_for_start(struct bridge_dev_context *dev_context, u32 dw_sync_addr)
1793 {
1794 	u16 timeout = TIHELEN_ACKTIMEOUT;
1795 
1796 	/*  Wait for response from board */
1797 	while (__raw_readw(dw_sync_addr) && --timeout)
1798 		udelay(10);
1799 
1800 	/*  If timed out: return false */
1801 	if (!timeout) {
1802 		pr_err("%s: Timed out waiting DSP to Start\n", __func__);
1803 		return false;
1804 	}
1805 	return true;
1806 }
1807