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 /*  ----------------------------------- Trace & Debug */
31 #include <dspbridge/dbc.h>
32 
33 /*  ----------------------------------- OS Adaptation Layer */
34 #include <dspbridge/drv.h>
35 #include <dspbridge/sync.h>
36 
37 /* ------------------------------------ Hardware Abstraction Layer */
38 #include <hw_defs.h>
39 #include <hw_mmu.h>
40 
41 /*  ----------------------------------- Link Driver */
42 #include <dspbridge/dspdefs.h>
43 #include <dspbridge/dspchnl.h>
44 #include <dspbridge/dspdeh.h>
45 #include <dspbridge/dspio.h>
46 #include <dspbridge/dspmsg.h>
47 #include <dspbridge/pwr.h>
48 #include <dspbridge/io_sm.h>
49 
50 /*  ----------------------------------- Platform Manager */
51 #include <dspbridge/dev.h>
52 #include <dspbridge/dspapi.h>
53 #include <dspbridge/dmm.h>
54 #include <dspbridge/wdt.h>
55 
56 /*  ----------------------------------- Local */
57 #include "_tiomap.h"
58 #include "_tiomap_pwr.h"
59 #include "tiomap_io.h"
60 
61 /* Offset in shared mem to write to in order to synchronize start with DSP */
62 #define SHMSYNCOFFSET 4		/* GPP byte offset */
63 
64 #define BUFFERSIZE 1024
65 
66 #define TIHELEN_ACKTIMEOUT  10000
67 
68 #define MMU_SECTION_ADDR_MASK    0xFFF00000
69 #define MMU_SSECTION_ADDR_MASK   0xFF000000
70 #define MMU_LARGE_PAGE_MASK      0xFFFF0000
71 #define MMU_SMALL_PAGE_MASK      0xFFFFF000
72 #define OMAP3_IVA2_BOOTADDR_MASK 0xFFFFFC00
73 #define PAGES_II_LVL_TABLE   512
74 #define PHYS_TO_PAGE(phys)      pfn_to_page((phys) >> PAGE_SHIFT)
75 
76 /*
77  * This is a totally ugly layer violation, but needed until
78  * omap_ctrl_set_dsp_boot*() are provided.
79  */
80 #define OMAP3_IVA2_BOOTMOD_IDLE 1
81 #define OMAP2_CONTROL_GENERAL 0x270
82 #define OMAP343X_CONTROL_IVA2_BOOTADDR (OMAP2_CONTROL_GENERAL + 0x0190)
83 #define OMAP343X_CONTROL_IVA2_BOOTMOD (OMAP2_CONTROL_GENERAL + 0x0194)
84 
85 #define OMAP343X_CTRL_REGADDR(reg) \
86 	OMAP2_L4_IO_ADDRESS(OMAP343X_CTRL_BASE + (reg))
87 
88 
89 /* Forward Declarations: */
90 static int bridge_brd_monitor(struct bridge_dev_context *dev_ctxt);
91 static int bridge_brd_read(struct bridge_dev_context *dev_ctxt,
92 				  u8 *host_buff,
93 				  u32 dsp_addr, u32 ul_num_bytes,
94 				  u32 mem_type);
95 static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
96 				   u32 dsp_addr);
97 static int bridge_brd_status(struct bridge_dev_context *dev_ctxt,
98 				    int *board_state);
99 static int bridge_brd_stop(struct bridge_dev_context *dev_ctxt);
100 static int bridge_brd_write(struct bridge_dev_context *dev_ctxt,
101 				   u8 *host_buff,
102 				   u32 dsp_addr, u32 ul_num_bytes,
103 				   u32 mem_type);
104 static int bridge_brd_set_state(struct bridge_dev_context *dev_ctxt,
105 				    u32 brd_state);
106 static int bridge_brd_mem_copy(struct bridge_dev_context *dev_ctxt,
107 				   u32 dsp_dest_addr, u32 dsp_src_addr,
108 				   u32 ul_num_bytes, u32 mem_type);
109 static int bridge_brd_mem_write(struct bridge_dev_context *dev_ctxt,
110 				    u8 *host_buff, u32 dsp_addr,
111 				    u32 ul_num_bytes, u32 mem_type);
112 static int bridge_brd_mem_map(struct bridge_dev_context *dev_ctxt,
113 				  u32 ul_mpu_addr, u32 virt_addr,
114 				  u32 ul_num_bytes, u32 ul_map_attr,
115 				  struct page **mapped_pages);
116 static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctxt,
117 				     u32 virt_addr, u32 ul_num_bytes);
118 static int bridge_dev_create(struct bridge_dev_context
119 					**dev_cntxt,
120 					struct dev_object *hdev_obj,
121 					struct cfg_hostres *config_param);
122 static int bridge_dev_ctrl(struct bridge_dev_context *dev_context,
123 				  u32 dw_cmd, void *pargs);
124 static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt);
125 static u32 user_va2_pa(struct mm_struct *mm, u32 address);
126 static int pte_update(struct bridge_dev_context *dev_ctxt, u32 pa,
127 			     u32 va, u32 size,
128 			     struct hw_mmu_map_attrs_t *map_attrs);
129 static int pte_set(struct pg_table_attrs *pt, u32 pa, u32 va,
130 			  u32 size, struct hw_mmu_map_attrs_t *attrs);
131 static int mem_map_vmalloc(struct bridge_dev_context *dev_context,
132 				  u32 ul_mpu_addr, u32 virt_addr,
133 				  u32 ul_num_bytes,
134 				  struct hw_mmu_map_attrs_t *hw_attrs);
135 
136 bool wait_for_start(struct bridge_dev_context *dev_context, u32 dw_sync_addr);
137 
138 /*  ----------------------------------- Globals */
139 
140 /* Attributes of L2 page tables for DSP MMU */
141 struct page_info {
142 	u32 num_entries;	/* Number of valid PTEs in the L2 PT */
143 };
144 
145 /* Attributes used to manage the DSP MMU page tables */
146 struct pg_table_attrs {
147 	spinlock_t pg_lock;	/* Critical section object handle */
148 
149 	u32 l1_base_pa;		/* Physical address of the L1 PT */
150 	u32 l1_base_va;		/* Virtual  address of the L1 PT */
151 	u32 l1_size;		/* Size of the L1 PT */
152 	u32 l1_tbl_alloc_pa;
153 	/* Physical address of Allocated mem for L1 table. May not be aligned */
154 	u32 l1_tbl_alloc_va;
155 	/* Virtual address of Allocated mem for L1 table. May not be aligned */
156 	u32 l1_tbl_alloc_sz;
157 	/* Size of consistent memory allocated for L1 table.
158 	 * May not be aligned */
159 
160 	u32 l2_base_pa;		/* Physical address of the L2 PT */
161 	u32 l2_base_va;		/* Virtual  address of the L2 PT */
162 	u32 l2_size;		/* Size of the L2 PT */
163 	u32 l2_tbl_alloc_pa;
164 	/* Physical address of Allocated mem for L2 table. May not be aligned */
165 	u32 l2_tbl_alloc_va;
166 	/* Virtual address of Allocated mem for L2 table. May not be aligned */
167 	u32 l2_tbl_alloc_sz;
168 	/* Size of consistent memory allocated for L2 table.
169 	 * May not be aligned */
170 
171 	u32 l2_num_pages;	/* Number of allocated L2 PT */
172 	/* Array [l2_num_pages] of L2 PT info structs */
173 	struct page_info *pg_info;
174 };
175 
176 /*
177  *  This Bridge driver's function interface table.
178  */
179 static struct bridge_drv_interface drv_interface_fxns = {
180 	/* Bridge API ver. for which this bridge driver is built. */
181 	BRD_API_MAJOR_VERSION,
182 	BRD_API_MINOR_VERSION,
183 	bridge_dev_create,
184 	bridge_dev_destroy,
185 	bridge_dev_ctrl,
186 	bridge_brd_monitor,
187 	bridge_brd_start,
188 	bridge_brd_stop,
189 	bridge_brd_status,
190 	bridge_brd_read,
191 	bridge_brd_write,
192 	bridge_brd_set_state,
193 	bridge_brd_mem_copy,
194 	bridge_brd_mem_write,
195 	bridge_brd_mem_map,
196 	bridge_brd_mem_un_map,
197 	/* The following CHNL functions are provided by chnl_io.lib: */
198 	bridge_chnl_create,
199 	bridge_chnl_destroy,
200 	bridge_chnl_open,
201 	bridge_chnl_close,
202 	bridge_chnl_add_io_req,
203 	bridge_chnl_get_ioc,
204 	bridge_chnl_cancel_io,
205 	bridge_chnl_flush_io,
206 	bridge_chnl_get_info,
207 	bridge_chnl_get_mgr_info,
208 	bridge_chnl_idle,
209 	bridge_chnl_register_notify,
210 	/* The following IO functions are provided by chnl_io.lib: */
211 	bridge_io_create,
212 	bridge_io_destroy,
213 	bridge_io_on_loaded,
214 	bridge_io_get_proc_load,
215 	/* The following msg_ctrl functions are provided by chnl_io.lib: */
216 	bridge_msg_create,
217 	bridge_msg_create_queue,
218 	bridge_msg_delete,
219 	bridge_msg_delete_queue,
220 	bridge_msg_get,
221 	bridge_msg_put,
222 	bridge_msg_register_notify,
223 	bridge_msg_set_queue_id,
224 };
225 
226 static struct notifier_block dsp_mbox_notifier = {
227 	.notifier_call = io_mbox_msg,
228 };
229 
flush_all(struct bridge_dev_context * dev_context)230 static inline void flush_all(struct bridge_dev_context *dev_context)
231 {
232 	if (dev_context->brd_state == BRD_DSP_HIBERNATION ||
233 	    dev_context->brd_state == BRD_HIBERNATION)
234 		wake_dsp(dev_context, NULL);
235 
236 	hw_mmu_tlb_flush_all(dev_context->dsp_mmu_base);
237 }
238 
bad_page_dump(u32 pa,struct page * pg)239 static void bad_page_dump(u32 pa, struct page *pg)
240 {
241 	pr_emerg("DSPBRIDGE: MAP function: COUNT 0 FOR PA 0x%x\n", pa);
242 	pr_emerg("Bad page state in process '%s'\n"
243 		 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
244 		 "Backtrace:\n",
245 		 current->comm, pg, (int)(2 * sizeof(unsigned long)),
246 		 (unsigned long)pg->flags, pg->mapping,
247 		 page_mapcount(pg), page_count(pg));
248 	dump_stack();
249 }
250 
251 /*
252  *  ======== bridge_drv_entry ========
253  *  purpose:
254  *      Bridge Driver entry point.
255  */
bridge_drv_entry(struct bridge_drv_interface ** drv_intf,const char * driver_file_name)256 void bridge_drv_entry(struct bridge_drv_interface **drv_intf,
257 		   const char *driver_file_name)
258 {
259 
260 	DBC_REQUIRE(driver_file_name != NULL);
261 
262 	if (strcmp(driver_file_name, "UMA") == 0)
263 		*drv_intf = &drv_interface_fxns;
264 	else
265 		dev_dbg(bridge, "%s Unknown Bridge file name", __func__);
266 
267 }
268 
269 /*
270  *  ======== bridge_brd_monitor ========
271  *  purpose:
272  *      This bridge_brd_monitor puts DSP into a Loadable state.
273  *      i.e Application can load and start the device.
274  *
275  *  Preconditions:
276  *      Device in 'OFF' state.
277  */
bridge_brd_monitor(struct bridge_dev_context * dev_ctxt)278 static int bridge_brd_monitor(struct bridge_dev_context *dev_ctxt)
279 {
280 	struct bridge_dev_context *dev_context = dev_ctxt;
281 	u32 temp;
282 	struct omap_dsp_platform_data *pdata =
283 		omap_dspbridge_dev->dev.platform_data;
284 
285 	temp = (*pdata->dsp_prm_read)(OMAP3430_IVA2_MOD, OMAP2_PM_PWSTST) &
286 					OMAP_POWERSTATEST_MASK;
287 	if (!(temp & 0x02)) {
288 		/* IVA2 is not in ON state */
289 		/* Read and set PM_PWSTCTRL_IVA2  to ON */
290 		(*pdata->dsp_prm_rmw_bits)(OMAP_POWERSTATEST_MASK,
291 			PWRDM_POWER_ON, OMAP3430_IVA2_MOD, OMAP2_PM_PWSTCTRL);
292 		/* Set the SW supervised state transition */
293 		(*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_FORCE_WAKEUP,
294 					OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
295 
296 		/* Wait until the state has moved to ON */
297 		while ((*pdata->dsp_prm_read)(OMAP3430_IVA2_MOD, OMAP2_PM_PWSTST) &
298 						OMAP_INTRANSITION_MASK)
299 			;
300 		/* Disable Automatic transition */
301 		(*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_DISABLE_AUTO,
302 					OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
303 	}
304 	(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0,
305 					OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
306 	dsp_clk_enable(DSP_CLK_IVA2);
307 
308 	/* set the device state to IDLE */
309 	dev_context->brd_state = BRD_IDLE;
310 
311 	return 0;
312 }
313 
314 /*
315  *  ======== bridge_brd_read ========
316  *  purpose:
317  *      Reads buffers for DSP memory.
318  */
bridge_brd_read(struct bridge_dev_context * dev_ctxt,u8 * host_buff,u32 dsp_addr,u32 ul_num_bytes,u32 mem_type)319 static int bridge_brd_read(struct bridge_dev_context *dev_ctxt,
320 				  u8 *host_buff, u32 dsp_addr,
321 				  u32 ul_num_bytes, u32 mem_type)
322 {
323 	int status = 0;
324 	struct bridge_dev_context *dev_context = dev_ctxt;
325 	u32 offset;
326 	u32 dsp_base_addr = dev_ctxt->dsp_base_addr;
327 
328 	if (dsp_addr < dev_context->dsp_start_add) {
329 		status = -EPERM;
330 		return status;
331 	}
332 	/* change here to account for the 3 bands of the DSP internal memory */
333 	if ((dsp_addr - dev_context->dsp_start_add) <
334 	    dev_context->internal_size) {
335 		offset = dsp_addr - dev_context->dsp_start_add;
336 	} else {
337 		status = read_ext_dsp_data(dev_context, host_buff, dsp_addr,
338 					   ul_num_bytes, mem_type);
339 		return status;
340 	}
341 	/* copy the data from  DSP memory, */
342 	memcpy(host_buff, (void *)(dsp_base_addr + offset), ul_num_bytes);
343 	return status;
344 }
345 
346 /*
347  *  ======== bridge_brd_set_state ========
348  *  purpose:
349  *      This routine updates the Board status.
350  */
bridge_brd_set_state(struct bridge_dev_context * dev_ctxt,u32 brd_state)351 static int bridge_brd_set_state(struct bridge_dev_context *dev_ctxt,
352 				    u32 brd_state)
353 {
354 	int status = 0;
355 	struct bridge_dev_context *dev_context = dev_ctxt;
356 
357 	dev_context->brd_state = brd_state;
358 	return status;
359 }
360 
361 /*
362  *  ======== bridge_brd_start ========
363  *  purpose:
364  *      Initializes DSP MMU and Starts DSP.
365  *
366  *  Preconditions:
367  *  a) DSP domain is 'ACTIVE'.
368  *  b) DSP_RST1 is asserted.
369  *  b) DSP_RST2 is released.
370  */
bridge_brd_start(struct bridge_dev_context * dev_ctxt,u32 dsp_addr)371 static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
372 				   u32 dsp_addr)
373 {
374 	int status = 0;
375 	struct bridge_dev_context *dev_context = dev_ctxt;
376 	u32 dw_sync_addr = 0;
377 	u32 ul_shm_base;	/* Gpp Phys SM base addr(byte) */
378 	u32 ul_shm_base_virt;	/* Dsp Virt SM base addr */
379 	u32 ul_tlb_base_virt;	/* Base of MMU TLB entry */
380 	/* Offset of shm_base_virt from tlb_base_virt */
381 	u32 ul_shm_offset_virt;
382 	s32 entry_ndx;
383 	s32 itmp_entry_ndx = 0;	/* DSP-MMU TLB entry base address */
384 	struct cfg_hostres *resources = NULL;
385 	u32 temp;
386 	u32 ul_dsp_clk_rate;
387 	u32 ul_dsp_clk_addr;
388 	u32 ul_bios_gp_timer;
389 	u32 clk_cmd;
390 	struct io_mgr *hio_mgr;
391 	u32 ul_load_monitor_timer;
392 	struct omap_dsp_platform_data *pdata =
393 		omap_dspbridge_dev->dev.platform_data;
394 
395 	/* The device context contains all the mmu setup info from when the
396 	 * last dsp base image was loaded. The first entry is always
397 	 * SHMMEM base. */
398 	/* Get SHM_BEG - convert to byte address */
399 	(void)dev_get_symbol(dev_context->dev_obj, SHMBASENAME,
400 			     &ul_shm_base_virt);
401 	ul_shm_base_virt *= DSPWORDSIZE;
402 	DBC_ASSERT(ul_shm_base_virt != 0);
403 	/* DSP Virtual address */
404 	ul_tlb_base_virt = dev_context->atlb_entry[0].dsp_va;
405 	DBC_ASSERT(ul_tlb_base_virt <= ul_shm_base_virt);
406 	ul_shm_offset_virt =
407 	    ul_shm_base_virt - (ul_tlb_base_virt * DSPWORDSIZE);
408 	/* Kernel logical address */
409 	ul_shm_base = dev_context->atlb_entry[0].gpp_va + ul_shm_offset_virt;
410 
411 	DBC_ASSERT(ul_shm_base != 0);
412 	/* 2nd wd is used as sync field */
413 	dw_sync_addr = ul_shm_base + SHMSYNCOFFSET;
414 	/* Write a signature into the shm base + offset; this will
415 	 * get cleared when the DSP program starts. */
416 	if ((ul_shm_base_virt == 0) || (ul_shm_base == 0)) {
417 		pr_err("%s: Illegal SM base\n", __func__);
418 		status = -EPERM;
419 	} else
420 		__raw_writel(0xffffffff, dw_sync_addr);
421 
422 	if (!status) {
423 		resources = dev_context->resources;
424 		if (!resources)
425 			status = -EPERM;
426 
427 		/* Assert RST1 i.e only the RST only for DSP megacell */
428 		if (!status) {
429 			(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST1_IVA2_MASK,
430 					OMAP3430_RST1_IVA2_MASK, OMAP3430_IVA2_MOD,
431 					OMAP2_RM_RSTCTRL);
432 			/* Mask address with 1K for compatibility */
433 			__raw_writel(dsp_addr & OMAP3_IVA2_BOOTADDR_MASK,
434 					OMAP343X_CTRL_REGADDR(
435 					OMAP343X_CONTROL_IVA2_BOOTADDR));
436 			/*
437 			 * Set bootmode to self loop if dsp_debug flag is true
438 			 */
439 			__raw_writel((dsp_debug) ? OMAP3_IVA2_BOOTMOD_IDLE : 0,
440 					OMAP343X_CTRL_REGADDR(
441 					OMAP343X_CONTROL_IVA2_BOOTMOD));
442 		}
443 	}
444 	if (!status) {
445 		/* Reset and Unreset the RST2, so that BOOTADDR is copied to
446 		 * IVA2 SYSC register */
447 		(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK,
448 			OMAP3430_RST2_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
449 		udelay(100);
450 		(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0,
451 					OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
452 		udelay(100);
453 
454 		/* Disbale the DSP MMU */
455 		hw_mmu_disable(resources->dmmu_base);
456 		/* Disable TWL */
457 		hw_mmu_twl_disable(resources->dmmu_base);
458 
459 		/* Only make TLB entry if both addresses are non-zero */
460 		for (entry_ndx = 0; entry_ndx < BRDIOCTL_NUMOFMMUTLB;
461 		     entry_ndx++) {
462 			struct bridge_ioctl_extproc *e = &dev_context->atlb_entry[entry_ndx];
463 			struct hw_mmu_map_attrs_t map_attrs = {
464 				.endianism = e->endianism,
465 				.element_size = e->elem_size,
466 				.mixed_size = e->mixed_mode,
467 			};
468 
469 			if (!e->gpp_pa || !e->dsp_va)
470 				continue;
471 
472 			dev_dbg(bridge,
473 					"MMU %d, pa: 0x%x, va: 0x%x, size: 0x%x",
474 					itmp_entry_ndx,
475 					e->gpp_pa,
476 					e->dsp_va,
477 					e->size);
478 
479 			hw_mmu_tlb_add(dev_context->dsp_mmu_base,
480 					e->gpp_pa,
481 					e->dsp_va,
482 					e->size,
483 					itmp_entry_ndx,
484 					&map_attrs, 1, 1);
485 
486 			itmp_entry_ndx++;
487 		}
488 	}
489 
490 	/* Lock the above TLB entries and get the BIOS and load monitor timer
491 	 * information */
492 	if (!status) {
493 		hw_mmu_num_locked_set(resources->dmmu_base, itmp_entry_ndx);
494 		hw_mmu_victim_num_set(resources->dmmu_base, itmp_entry_ndx);
495 		hw_mmu_ttb_set(resources->dmmu_base,
496 			       dev_context->pt_attrs->l1_base_pa);
497 		hw_mmu_twl_enable(resources->dmmu_base);
498 		/* Enable the SmartIdle and AutoIdle bit for MMU_SYSCONFIG */
499 
500 		temp = __raw_readl((resources->dmmu_base) + 0x10);
501 		temp = (temp & 0xFFFFFFEF) | 0x11;
502 		__raw_writel(temp, (resources->dmmu_base) + 0x10);
503 
504 		/* Let the DSP MMU run */
505 		hw_mmu_enable(resources->dmmu_base);
506 
507 		/* Enable the BIOS clock */
508 		(void)dev_get_symbol(dev_context->dev_obj,
509 				     BRIDGEINIT_BIOSGPTIMER, &ul_bios_gp_timer);
510 		(void)dev_get_symbol(dev_context->dev_obj,
511 				     BRIDGEINIT_LOADMON_GPTIMER,
512 				     &ul_load_monitor_timer);
513 	}
514 
515 	if (!status) {
516 		if (ul_load_monitor_timer != 0xFFFF) {
517 			clk_cmd = (BPWR_ENABLE_CLOCK << MBX_PM_CLK_CMDSHIFT) |
518 			    ul_load_monitor_timer;
519 			dsp_peripheral_clk_ctrl(dev_context, &clk_cmd);
520 		} else {
521 			dev_dbg(bridge, "Not able to get the symbol for Load "
522 				"Monitor Timer\n");
523 		}
524 	}
525 
526 	if (!status) {
527 		if (ul_bios_gp_timer != 0xFFFF) {
528 			clk_cmd = (BPWR_ENABLE_CLOCK << MBX_PM_CLK_CMDSHIFT) |
529 			    ul_bios_gp_timer;
530 			dsp_peripheral_clk_ctrl(dev_context, &clk_cmd);
531 		} else {
532 			dev_dbg(bridge,
533 				"Not able to get the symbol for BIOS Timer\n");
534 		}
535 	}
536 
537 	if (!status) {
538 		/* Set the DSP clock rate */
539 		(void)dev_get_symbol(dev_context->dev_obj,
540 				     "_BRIDGEINIT_DSP_FREQ", &ul_dsp_clk_addr);
541 		/*Set Autoidle Mode for IVA2 PLL */
542 		(*pdata->dsp_cm_write)(1 << OMAP3430_AUTO_IVA2_DPLL_SHIFT,
543 				OMAP3430_IVA2_MOD, OMAP3430_CM_AUTOIDLE_PLL);
544 
545 		if ((unsigned int *)ul_dsp_clk_addr != NULL) {
546 			/* Get the clock rate */
547 			ul_dsp_clk_rate = dsp_clk_get_iva2_rate();
548 			dev_dbg(bridge, "%s: DSP clock rate (KHZ): 0x%x \n",
549 				__func__, ul_dsp_clk_rate);
550 			(void)bridge_brd_write(dev_context,
551 					       (u8 *) &ul_dsp_clk_rate,
552 					       ul_dsp_clk_addr, sizeof(u32), 0);
553 		}
554 		/*
555 		 * Enable Mailbox events and also drain any pending
556 		 * stale messages.
557 		 */
558 		dev_context->mbox = omap_mbox_get("dsp", &dsp_mbox_notifier);
559 		if (IS_ERR(dev_context->mbox)) {
560 			dev_context->mbox = NULL;
561 			pr_err("%s: Failed to get dsp mailbox handle\n",
562 								__func__);
563 			status = -EPERM;
564 		}
565 
566 	}
567 	if (!status) {
568 /*PM_IVA2GRPSEL_PER = 0xC0;*/
569 		temp = readl(resources->per_pm_base + 0xA8);
570 		temp = (temp & 0xFFFFFF30) | 0xC0;
571 		writel(temp, resources->per_pm_base + 0xA8);
572 
573 /*PM_MPUGRPSEL_PER &= 0xFFFFFF3F; */
574 		temp = readl(resources->per_pm_base + 0xA4);
575 		temp = (temp & 0xFFFFFF3F);
576 		writel(temp, resources->per_pm_base + 0xA4);
577 /*CM_SLEEPDEP_PER |= 0x04; */
578 		temp = readl(resources->per_base + 0x44);
579 		temp = (temp & 0xFFFFFFFB) | 0x04;
580 		writel(temp, resources->per_base + 0x44);
581 
582 /*CM_CLKSTCTRL_IVA2 = 0x00000003 -To Allow automatic transitions */
583 		(*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_ENABLE_AUTO,
584 					OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
585 
586 		/* Let DSP go */
587 		dev_dbg(bridge, "%s Unreset\n", __func__);
588 		/* Enable DSP MMU Interrupts */
589 		hw_mmu_event_enable(resources->dmmu_base,
590 				    HW_MMU_ALL_INTERRUPTS);
591 		/* release the RST1, DSP starts executing now .. */
592 		(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST1_IVA2_MASK, 0,
593 					OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
594 
595 		dev_dbg(bridge, "Waiting for Sync @ 0x%x\n", dw_sync_addr);
596 		dev_dbg(bridge, "DSP c_int00 Address =  0x%x\n", dsp_addr);
597 		if (dsp_debug)
598 			while (__raw_readw(dw_sync_addr))
599 				;
600 
601 		/* Wait for DSP to clear word in shared memory */
602 		/* Read the Location */
603 		if (!wait_for_start(dev_context, dw_sync_addr))
604 			status = -ETIMEDOUT;
605 
606 		/* Start wdt */
607 		dsp_wdt_sm_set((void *)ul_shm_base);
608 		dsp_wdt_enable(true);
609 
610 		status = dev_get_io_mgr(dev_context->dev_obj, &hio_mgr);
611 		if (hio_mgr) {
612 			io_sh_msetting(hio_mgr, SHM_OPPINFO, NULL);
613 			/* Write the synchronization bit to indicate the
614 			 * completion of OPP table update to DSP
615 			 */
616 			__raw_writel(0XCAFECAFE, dw_sync_addr);
617 
618 			/* update board state */
619 			dev_context->brd_state = BRD_RUNNING;
620 			/* (void)chnlsm_enable_interrupt(dev_context); */
621 		} else {
622 			dev_context->brd_state = BRD_UNKNOWN;
623 		}
624 	}
625 	return status;
626 }
627 
628 /*
629  *  ======== bridge_brd_stop ========
630  *  purpose:
631  *      Puts DSP in self loop.
632  *
633  *  Preconditions :
634  *  a) None
635  */
bridge_brd_stop(struct bridge_dev_context * dev_ctxt)636 static int bridge_brd_stop(struct bridge_dev_context *dev_ctxt)
637 {
638 	int status = 0;
639 	struct bridge_dev_context *dev_context = dev_ctxt;
640 	struct pg_table_attrs *pt_attrs;
641 	u32 dsp_pwr_state;
642 	struct omap_dsp_platform_data *pdata =
643 		omap_dspbridge_dev->dev.platform_data;
644 
645 	if (dev_context->brd_state == BRD_STOPPED)
646 		return status;
647 
648 	/* as per TRM, it is advised to first drive the IVA2 to 'Standby' mode,
649 	 * before turning off the clocks.. This is to ensure that there are no
650 	 * pending L3 or other transactons from IVA2 */
651 	dsp_pwr_state = (*pdata->dsp_prm_read)(OMAP3430_IVA2_MOD, OMAP2_PM_PWSTST) &
652 					OMAP_POWERSTATEST_MASK;
653 	if (dsp_pwr_state != PWRDM_POWER_OFF) {
654 		(*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0,
655 					OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
656 		sm_interrupt_dsp(dev_context, MBX_PM_DSPIDLE);
657 		mdelay(10);
658 
659 		/* IVA2 is not in OFF state */
660 		/* Set PM_PWSTCTRL_IVA2  to OFF */
661 		(*pdata->dsp_prm_rmw_bits)(OMAP_POWERSTATEST_MASK,
662 			PWRDM_POWER_OFF, OMAP3430_IVA2_MOD, OMAP2_PM_PWSTCTRL);
663 		/* Set the SW supervised state transition for Sleep */
664 		(*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_FORCE_SLEEP,
665 					OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
666 	}
667 	udelay(10);
668 	/* Release the Ext Base virtual Address as the next DSP Program
669 	 * may have a different load address */
670 	if (dev_context->dsp_ext_base_addr)
671 		dev_context->dsp_ext_base_addr = 0;
672 
673 	dev_context->brd_state = BRD_STOPPED;	/* update board state */
674 
675 	dsp_wdt_enable(false);
676 
677 	/* This is a good place to clear the MMU page tables as well */
678 	if (dev_context->pt_attrs) {
679 		pt_attrs = dev_context->pt_attrs;
680 		memset((u8 *) pt_attrs->l1_base_va, 0x00, pt_attrs->l1_size);
681 		memset((u8 *) pt_attrs->l2_base_va, 0x00, pt_attrs->l2_size);
682 		memset((u8 *) pt_attrs->pg_info, 0x00,
683 		       (pt_attrs->l2_num_pages * sizeof(struct page_info)));
684 	}
685 	/* Disable the mailbox interrupts */
686 	if (dev_context->mbox) {
687 		omap_mbox_disable_irq(dev_context->mbox, IRQ_RX);
688 		omap_mbox_put(dev_context->mbox, &dsp_mbox_notifier);
689 		dev_context->mbox = NULL;
690 	}
691 	/* Reset IVA2 clocks*/
692 	(*pdata->dsp_prm_write)(OMAP3430_RST1_IVA2_MASK | OMAP3430_RST2_IVA2_MASK |
693 			OMAP3430_RST3_IVA2_MASK, OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
694 
695 	dsp_clock_disable_all(dev_context->dsp_per_clks);
696 	dsp_clk_disable(DSP_CLK_IVA2);
697 
698 	return status;
699 }
700 
701 /*
702  *  ======== bridge_brd_status ========
703  *      Returns the board status.
704  */
bridge_brd_status(struct bridge_dev_context * dev_ctxt,int * board_state)705 static int bridge_brd_status(struct bridge_dev_context *dev_ctxt,
706 				    int *board_state)
707 {
708 	struct bridge_dev_context *dev_context = dev_ctxt;
709 	*board_state = dev_context->brd_state;
710 	return 0;
711 }
712 
713 /*
714  *  ======== bridge_brd_write ========
715  *      Copies the buffers to DSP internal or external memory.
716  */
bridge_brd_write(struct bridge_dev_context * dev_ctxt,u8 * host_buff,u32 dsp_addr,u32 ul_num_bytes,u32 mem_type)717 static int bridge_brd_write(struct bridge_dev_context *dev_ctxt,
718 				   u8 *host_buff, u32 dsp_addr,
719 				   u32 ul_num_bytes, u32 mem_type)
720 {
721 	int status = 0;
722 	struct bridge_dev_context *dev_context = dev_ctxt;
723 
724 	if (dsp_addr < dev_context->dsp_start_add) {
725 		status = -EPERM;
726 		return status;
727 	}
728 	if ((dsp_addr - dev_context->dsp_start_add) <
729 	    dev_context->internal_size) {
730 		status = write_dsp_data(dev_ctxt, host_buff, dsp_addr,
731 					ul_num_bytes, mem_type);
732 	} else {
733 		status = write_ext_dsp_data(dev_context, host_buff, dsp_addr,
734 					    ul_num_bytes, mem_type, false);
735 	}
736 
737 	return status;
738 }
739 
740 /*
741  *  ======== bridge_dev_create ========
742  *      Creates a driver object. Puts DSP in self loop.
743  */
bridge_dev_create(struct bridge_dev_context ** dev_cntxt,struct dev_object * hdev_obj,struct cfg_hostres * config_param)744 static int bridge_dev_create(struct bridge_dev_context
745 					**dev_cntxt,
746 					struct dev_object *hdev_obj,
747 					struct cfg_hostres *config_param)
748 {
749 	int status = 0;
750 	struct bridge_dev_context *dev_context = NULL;
751 	s32 entry_ndx;
752 	struct cfg_hostres *resources = config_param;
753 	struct pg_table_attrs *pt_attrs;
754 	u32 pg_tbl_pa;
755 	u32 pg_tbl_va;
756 	u32 align_size;
757 	struct drv_data *drv_datap = dev_get_drvdata(bridge);
758 
759 	/* Allocate and initialize a data structure to contain the bridge driver
760 	 *  state, which becomes the context for later calls into this driver */
761 	dev_context = kzalloc(sizeof(struct bridge_dev_context), GFP_KERNEL);
762 	if (!dev_context) {
763 		status = -ENOMEM;
764 		goto func_end;
765 	}
766 
767 	dev_context->dsp_start_add = (u32) OMAP_GEM_BASE;
768 	dev_context->self_loop = (u32) NULL;
769 	dev_context->dsp_per_clks = 0;
770 	dev_context->internal_size = OMAP_DSP_SIZE;
771 	/*  Clear dev context MMU table entries.
772 	 *  These get set on bridge_io_on_loaded() call after program loaded. */
773 	for (entry_ndx = 0; entry_ndx < BRDIOCTL_NUMOFMMUTLB; entry_ndx++) {
774 		dev_context->atlb_entry[entry_ndx].gpp_pa =
775 		    dev_context->atlb_entry[entry_ndx].dsp_va = 0;
776 	}
777 	dev_context->dsp_base_addr = (u32) MEM_LINEAR_ADDRESS((void *)
778 								 (config_param->
779 								  mem_base
780 								  [3]),
781 								 config_param->
782 								 mem_length
783 								 [3]);
784 	if (!dev_context->dsp_base_addr)
785 		status = -EPERM;
786 
787 	pt_attrs = kzalloc(sizeof(struct pg_table_attrs), GFP_KERNEL);
788 	if (pt_attrs != NULL) {
789 		pt_attrs->l1_size = SZ_16K; /* 4096 entries of 32 bits */
790 		align_size = pt_attrs->l1_size;
791 		/* Align sizes are expected to be power of 2 */
792 		/* we like to get aligned on L1 table size */
793 		pg_tbl_va = (u32) mem_alloc_phys_mem(pt_attrs->l1_size,
794 						     align_size, &pg_tbl_pa);
795 
796 		/* Check if the PA is aligned for us */
797 		if ((pg_tbl_pa) & (align_size - 1)) {
798 			/* PA not aligned to page table size ,
799 			 * try with more allocation and align */
800 			mem_free_phys_mem((void *)pg_tbl_va, pg_tbl_pa,
801 					  pt_attrs->l1_size);
802 			/* we like to get aligned on L1 table size */
803 			pg_tbl_va =
804 			    (u32) mem_alloc_phys_mem((pt_attrs->l1_size) * 2,
805 						     align_size, &pg_tbl_pa);
806 			/* We should be able to get aligned table now */
807 			pt_attrs->l1_tbl_alloc_pa = pg_tbl_pa;
808 			pt_attrs->l1_tbl_alloc_va = pg_tbl_va;
809 			pt_attrs->l1_tbl_alloc_sz = pt_attrs->l1_size * 2;
810 			/* Align the PA to the next 'align'  boundary */
811 			pt_attrs->l1_base_pa =
812 			    ((pg_tbl_pa) +
813 			     (align_size - 1)) & (~(align_size - 1));
814 			pt_attrs->l1_base_va =
815 			    pg_tbl_va + (pt_attrs->l1_base_pa - pg_tbl_pa);
816 		} else {
817 			/* We got aligned PA, cool */
818 			pt_attrs->l1_tbl_alloc_pa = pg_tbl_pa;
819 			pt_attrs->l1_tbl_alloc_va = pg_tbl_va;
820 			pt_attrs->l1_tbl_alloc_sz = pt_attrs->l1_size;
821 			pt_attrs->l1_base_pa = pg_tbl_pa;
822 			pt_attrs->l1_base_va = pg_tbl_va;
823 		}
824 		if (pt_attrs->l1_base_va)
825 			memset((u8 *) pt_attrs->l1_base_va, 0x00,
826 			       pt_attrs->l1_size);
827 
828 		/* number of L2 page tables = DMM pool used + SHMMEM +EXTMEM +
829 		 * L4 pages */
830 		pt_attrs->l2_num_pages = ((DMMPOOLSIZE >> 20) + 6);
831 		pt_attrs->l2_size = HW_MMU_COARSE_PAGE_SIZE *
832 		    pt_attrs->l2_num_pages;
833 		align_size = 4;	/* Make it u32 aligned */
834 		/* we like to get aligned on L1 table size */
835 		pg_tbl_va = (u32) mem_alloc_phys_mem(pt_attrs->l2_size,
836 						     align_size, &pg_tbl_pa);
837 		pt_attrs->l2_tbl_alloc_pa = pg_tbl_pa;
838 		pt_attrs->l2_tbl_alloc_va = pg_tbl_va;
839 		pt_attrs->l2_tbl_alloc_sz = pt_attrs->l2_size;
840 		pt_attrs->l2_base_pa = pg_tbl_pa;
841 		pt_attrs->l2_base_va = pg_tbl_va;
842 
843 		if (pt_attrs->l2_base_va)
844 			memset((u8 *) pt_attrs->l2_base_va, 0x00,
845 			       pt_attrs->l2_size);
846 
847 		pt_attrs->pg_info = kzalloc(pt_attrs->l2_num_pages *
848 					sizeof(struct page_info), GFP_KERNEL);
849 		dev_dbg(bridge,
850 			"L1 pa %x, va %x, size %x\n L2 pa %x, va "
851 			"%x, size %x\n", pt_attrs->l1_base_pa,
852 			pt_attrs->l1_base_va, pt_attrs->l1_size,
853 			pt_attrs->l2_base_pa, pt_attrs->l2_base_va,
854 			pt_attrs->l2_size);
855 		dev_dbg(bridge, "pt_attrs %p L2 NumPages %x pg_info %p\n",
856 			pt_attrs, pt_attrs->l2_num_pages, pt_attrs->pg_info);
857 	}
858 	if ((pt_attrs != NULL) && (pt_attrs->l1_base_va != 0) &&
859 	    (pt_attrs->l2_base_va != 0) && (pt_attrs->pg_info != NULL))
860 		dev_context->pt_attrs = pt_attrs;
861 	else
862 		status = -ENOMEM;
863 
864 	if (!status) {
865 		spin_lock_init(&pt_attrs->pg_lock);
866 		dev_context->tc_word_swap_on = drv_datap->tc_wordswapon;
867 
868 		/* Set the Clock Divisor for the DSP module */
869 		udelay(5);
870 		/* MMU address is obtained from the host
871 		 * resources struct */
872 		dev_context->dsp_mmu_base = resources->dmmu_base;
873 	}
874 	if (!status) {
875 		dev_context->dev_obj = hdev_obj;
876 		/* Store current board state. */
877 		dev_context->brd_state = BRD_UNKNOWN;
878 		dev_context->resources = resources;
879 		dsp_clk_enable(DSP_CLK_IVA2);
880 		bridge_brd_stop(dev_context);
881 		/* Return ptr to our device state to the DSP API for storage */
882 		*dev_cntxt = dev_context;
883 	} else {
884 		if (pt_attrs != NULL) {
885 			kfree(pt_attrs->pg_info);
886 
887 			if (pt_attrs->l2_tbl_alloc_va) {
888 				mem_free_phys_mem((void *)
889 						  pt_attrs->l2_tbl_alloc_va,
890 						  pt_attrs->l2_tbl_alloc_pa,
891 						  pt_attrs->l2_tbl_alloc_sz);
892 			}
893 			if (pt_attrs->l1_tbl_alloc_va) {
894 				mem_free_phys_mem((void *)
895 						  pt_attrs->l1_tbl_alloc_va,
896 						  pt_attrs->l1_tbl_alloc_pa,
897 						  pt_attrs->l1_tbl_alloc_sz);
898 			}
899 		}
900 		kfree(pt_attrs);
901 		kfree(dev_context);
902 	}
903 func_end:
904 	return status;
905 }
906 
907 /*
908  *  ======== bridge_dev_ctrl ========
909  *      Receives device specific commands.
910  */
bridge_dev_ctrl(struct bridge_dev_context * dev_context,u32 dw_cmd,void * pargs)911 static int bridge_dev_ctrl(struct bridge_dev_context *dev_context,
912 				  u32 dw_cmd, void *pargs)
913 {
914 	int status = 0;
915 	struct bridge_ioctl_extproc *pa_ext_proc =
916 					(struct bridge_ioctl_extproc *)pargs;
917 	s32 ndx;
918 
919 	switch (dw_cmd) {
920 	case BRDIOCTL_CHNLREAD:
921 		break;
922 	case BRDIOCTL_CHNLWRITE:
923 		break;
924 	case BRDIOCTL_SETMMUCONFIG:
925 		/* store away dsp-mmu setup values for later use */
926 		for (ndx = 0; ndx < BRDIOCTL_NUMOFMMUTLB; ndx++, pa_ext_proc++)
927 			dev_context->atlb_entry[ndx] = *pa_ext_proc;
928 		break;
929 	case BRDIOCTL_DEEPSLEEP:
930 	case BRDIOCTL_EMERGENCYSLEEP:
931 		/* Currently only DSP Idle is supported Need to update for
932 		 * later releases */
933 		status = sleep_dsp(dev_context, PWR_DEEPSLEEP, pargs);
934 		break;
935 	case BRDIOCTL_WAKEUP:
936 		status = wake_dsp(dev_context, pargs);
937 		break;
938 	case BRDIOCTL_CLK_CTRL:
939 		status = 0;
940 		/* Looking For Baseport Fix for Clocks */
941 		status = dsp_peripheral_clk_ctrl(dev_context, pargs);
942 		break;
943 	case BRDIOCTL_PWR_HIBERNATE:
944 		status = handle_hibernation_from_dsp(dev_context);
945 		break;
946 	case BRDIOCTL_PRESCALE_NOTIFY:
947 		status = pre_scale_dsp(dev_context, pargs);
948 		break;
949 	case BRDIOCTL_POSTSCALE_NOTIFY:
950 		status = post_scale_dsp(dev_context, pargs);
951 		break;
952 	case BRDIOCTL_CONSTRAINT_REQUEST:
953 		status = handle_constraints_set(dev_context, pargs);
954 		break;
955 	default:
956 		status = -EPERM;
957 		break;
958 	}
959 	return status;
960 }
961 
962 /*
963  *  ======== bridge_dev_destroy ========
964  *      Destroys the driver object.
965  */
bridge_dev_destroy(struct bridge_dev_context * dev_ctxt)966 static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt)
967 {
968 	struct pg_table_attrs *pt_attrs;
969 	int status = 0;
970 	struct bridge_dev_context *dev_context = (struct bridge_dev_context *)
971 	    dev_ctxt;
972 	struct cfg_hostres *host_res;
973 	u32 shm_size;
974 	struct drv_data *drv_datap = dev_get_drvdata(bridge);
975 
976 	/* It should never happen */
977 	if (!dev_ctxt)
978 		return -EFAULT;
979 
980 	/* first put the device to stop state */
981 	bridge_brd_stop(dev_context);
982 	if (dev_context->pt_attrs) {
983 		pt_attrs = dev_context->pt_attrs;
984 		kfree(pt_attrs->pg_info);
985 
986 		if (pt_attrs->l2_tbl_alloc_va) {
987 			mem_free_phys_mem((void *)pt_attrs->l2_tbl_alloc_va,
988 					  pt_attrs->l2_tbl_alloc_pa,
989 					  pt_attrs->l2_tbl_alloc_sz);
990 		}
991 		if (pt_attrs->l1_tbl_alloc_va) {
992 			mem_free_phys_mem((void *)pt_attrs->l1_tbl_alloc_va,
993 					  pt_attrs->l1_tbl_alloc_pa,
994 					  pt_attrs->l1_tbl_alloc_sz);
995 		}
996 		kfree(pt_attrs);
997 
998 	}
999 
1000 	if (dev_context->resources) {
1001 		host_res = dev_context->resources;
1002 		shm_size = drv_datap->shm_size;
1003 		if (shm_size >= 0x10000) {
1004 			if ((host_res->mem_base[1]) &&
1005 			    (host_res->mem_phys[1])) {
1006 				mem_free_phys_mem((void *)
1007 						  host_res->mem_base
1008 						  [1],
1009 						  host_res->mem_phys
1010 						  [1], shm_size);
1011 			}
1012 		} else {
1013 			dev_dbg(bridge, "%s: Error getting shm size "
1014 				"from registry: %x. Not calling "
1015 				"mem_free_phys_mem\n", __func__,
1016 				status);
1017 		}
1018 		host_res->mem_base[1] = 0;
1019 		host_res->mem_phys[1] = 0;
1020 
1021 		if (host_res->mem_base[0])
1022 			iounmap((void *)host_res->mem_base[0]);
1023 		if (host_res->mem_base[2])
1024 			iounmap((void *)host_res->mem_base[2]);
1025 		if (host_res->mem_base[3])
1026 			iounmap((void *)host_res->mem_base[3]);
1027 		if (host_res->mem_base[4])
1028 			iounmap((void *)host_res->mem_base[4]);
1029 		if (host_res->dmmu_base)
1030 			iounmap(host_res->dmmu_base);
1031 		if (host_res->per_base)
1032 			iounmap(host_res->per_base);
1033 		if (host_res->per_pm_base)
1034 			iounmap((void *)host_res->per_pm_base);
1035 		if (host_res->core_pm_base)
1036 			iounmap((void *)host_res->core_pm_base);
1037 
1038 		host_res->mem_base[0] = (u32) NULL;
1039 		host_res->mem_base[2] = (u32) NULL;
1040 		host_res->mem_base[3] = (u32) NULL;
1041 		host_res->mem_base[4] = (u32) NULL;
1042 		host_res->dmmu_base = NULL;
1043 
1044 		kfree(host_res);
1045 	}
1046 
1047 	/* Free the driver's device context: */
1048 	kfree(drv_datap->base_img);
1049 	kfree((void *)dev_ctxt);
1050 	return status;
1051 }
1052 
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)1053 static int bridge_brd_mem_copy(struct bridge_dev_context *dev_ctxt,
1054 				   u32 dsp_dest_addr, u32 dsp_src_addr,
1055 				   u32 ul_num_bytes, u32 mem_type)
1056 {
1057 	int status = 0;
1058 	u32 src_addr = dsp_src_addr;
1059 	u32 dest_addr = dsp_dest_addr;
1060 	u32 copy_bytes = 0;
1061 	u32 total_bytes = ul_num_bytes;
1062 	u8 host_buf[BUFFERSIZE];
1063 	struct bridge_dev_context *dev_context = dev_ctxt;
1064 	while (total_bytes > 0 && !status) {
1065 		copy_bytes =
1066 		    total_bytes > BUFFERSIZE ? BUFFERSIZE : total_bytes;
1067 		/* Read from External memory */
1068 		status = read_ext_dsp_data(dev_ctxt, host_buf, src_addr,
1069 					   copy_bytes, mem_type);
1070 		if (!status) {
1071 			if (dest_addr < (dev_context->dsp_start_add +
1072 					 dev_context->internal_size)) {
1073 				/* Write to Internal memory */
1074 				status = write_dsp_data(dev_ctxt, host_buf,
1075 							dest_addr, copy_bytes,
1076 							mem_type);
1077 			} else {
1078 				/* Write to External memory */
1079 				status =
1080 				    write_ext_dsp_data(dev_ctxt, host_buf,
1081 						       dest_addr, copy_bytes,
1082 						       mem_type, false);
1083 			}
1084 		}
1085 		total_bytes -= copy_bytes;
1086 		src_addr += copy_bytes;
1087 		dest_addr += copy_bytes;
1088 	}
1089 	return status;
1090 }
1091 
1092 /* 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)1093 static int bridge_brd_mem_write(struct bridge_dev_context *dev_ctxt,
1094 				    u8 *host_buff, u32 dsp_addr,
1095 				    u32 ul_num_bytes, u32 mem_type)
1096 {
1097 	int status = 0;
1098 	struct bridge_dev_context *dev_context = dev_ctxt;
1099 	u32 ul_remain_bytes = 0;
1100 	u32 ul_bytes = 0;
1101 	ul_remain_bytes = ul_num_bytes;
1102 	while (ul_remain_bytes > 0 && !status) {
1103 		ul_bytes =
1104 		    ul_remain_bytes > BUFFERSIZE ? BUFFERSIZE : ul_remain_bytes;
1105 		if (dsp_addr < (dev_context->dsp_start_add +
1106 				 dev_context->internal_size)) {
1107 			status =
1108 			    write_dsp_data(dev_ctxt, host_buff, dsp_addr,
1109 					   ul_bytes, mem_type);
1110 		} else {
1111 			status = write_ext_dsp_data(dev_ctxt, host_buff,
1112 						    dsp_addr, ul_bytes,
1113 						    mem_type, true);
1114 		}
1115 		ul_remain_bytes -= ul_bytes;
1116 		dsp_addr += ul_bytes;
1117 		host_buff = host_buff + ul_bytes;
1118 	}
1119 	return status;
1120 }
1121 
1122 /*
1123  *  ======== bridge_brd_mem_map ========
1124  *      This function maps MPU buffer to the DSP address space. It performs
1125  *  linear to physical address translation if required. It translates each
1126  *  page since linear addresses can be physically non-contiguous
1127  *  All address & size arguments are assumed to be page aligned (in proc.c)
1128  *
1129  *  TODO: Disable MMU while updating the page tables (but that'll stall DSP)
1130  */
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)1131 static int bridge_brd_mem_map(struct bridge_dev_context *dev_ctxt,
1132 				  u32 ul_mpu_addr, u32 virt_addr,
1133 				  u32 ul_num_bytes, u32 ul_map_attr,
1134 				  struct page **mapped_pages)
1135 {
1136 	u32 attrs;
1137 	int status = 0;
1138 	struct bridge_dev_context *dev_context = dev_ctxt;
1139 	struct hw_mmu_map_attrs_t hw_attrs;
1140 	struct vm_area_struct *vma;
1141 	struct mm_struct *mm = current->mm;
1142 	u32 write = 0;
1143 	u32 num_usr_pgs = 0;
1144 	struct page *mapped_page, *pg;
1145 	s32 pg_num;
1146 	u32 va = virt_addr;
1147 	struct task_struct *curr_task = current;
1148 	u32 pg_i = 0;
1149 	u32 mpu_addr, pa;
1150 
1151 	dev_dbg(bridge,
1152 		"%s hDevCtxt %p, pa %x, va %x, size %x, ul_map_attr %x\n",
1153 		__func__, dev_ctxt, ul_mpu_addr, virt_addr, ul_num_bytes,
1154 		ul_map_attr);
1155 	if (ul_num_bytes == 0)
1156 		return -EINVAL;
1157 
1158 	if (ul_map_attr & DSP_MAP_DIR_MASK) {
1159 		attrs = ul_map_attr;
1160 	} else {
1161 		/* Assign default attributes */
1162 		attrs = ul_map_attr | (DSP_MAPVIRTUALADDR | DSP_MAPELEMSIZE16);
1163 	}
1164 	/* Take mapping properties */
1165 	if (attrs & DSP_MAPBIGENDIAN)
1166 		hw_attrs.endianism = HW_BIG_ENDIAN;
1167 	else
1168 		hw_attrs.endianism = HW_LITTLE_ENDIAN;
1169 
1170 	hw_attrs.mixed_size = (enum hw_mmu_mixed_size_t)
1171 	    ((attrs & DSP_MAPMIXEDELEMSIZE) >> 2);
1172 	/* Ignore element_size if mixed_size is enabled */
1173 	if (hw_attrs.mixed_size == 0) {
1174 		if (attrs & DSP_MAPELEMSIZE8) {
1175 			/* Size is 8 bit */
1176 			hw_attrs.element_size = HW_ELEM_SIZE8BIT;
1177 		} else if (attrs & DSP_MAPELEMSIZE16) {
1178 			/* Size is 16 bit */
1179 			hw_attrs.element_size = HW_ELEM_SIZE16BIT;
1180 		} else if (attrs & DSP_MAPELEMSIZE32) {
1181 			/* Size is 32 bit */
1182 			hw_attrs.element_size = HW_ELEM_SIZE32BIT;
1183 		} else if (attrs & DSP_MAPELEMSIZE64) {
1184 			/* Size is 64 bit */
1185 			hw_attrs.element_size = HW_ELEM_SIZE64BIT;
1186 		} else {
1187 			/*
1188 			 * Mixedsize isn't enabled, so size can't be
1189 			 * zero here
1190 			 */
1191 			return -EINVAL;
1192 		}
1193 	}
1194 	if (attrs & DSP_MAPDONOTLOCK)
1195 		hw_attrs.donotlockmpupage = 1;
1196 	else
1197 		hw_attrs.donotlockmpupage = 0;
1198 
1199 	if (attrs & DSP_MAPVMALLOCADDR) {
1200 		return mem_map_vmalloc(dev_ctxt, ul_mpu_addr, virt_addr,
1201 				       ul_num_bytes, &hw_attrs);
1202 	}
1203 	/*
1204 	 * Do OS-specific user-va to pa translation.
1205 	 * Combine physically contiguous regions to reduce TLBs.
1206 	 * Pass the translated pa to pte_update.
1207 	 */
1208 	if ((attrs & DSP_MAPPHYSICALADDR)) {
1209 		status = pte_update(dev_context, ul_mpu_addr, virt_addr,
1210 				    ul_num_bytes, &hw_attrs);
1211 		goto func_cont;
1212 	}
1213 
1214 	/*
1215 	 * Important Note: ul_mpu_addr is mapped from user application process
1216 	 * to current process - it must lie completely within the current
1217 	 * virtual memory address space in order to be of use to us here!
1218 	 */
1219 	down_read(&mm->mmap_sem);
1220 	vma = find_vma(mm, ul_mpu_addr);
1221 	if (vma)
1222 		dev_dbg(bridge,
1223 			"VMAfor UserBuf: ul_mpu_addr=%x, ul_num_bytes=%x, "
1224 			"vm_start=%lx, vm_end=%lx, vm_flags=%lx\n", ul_mpu_addr,
1225 			ul_num_bytes, vma->vm_start, vma->vm_end,
1226 			vma->vm_flags);
1227 
1228 	/*
1229 	 * It is observed that under some circumstances, the user buffer is
1230 	 * spread across several VMAs. So loop through and check if the entire
1231 	 * user buffer is covered
1232 	 */
1233 	while ((vma) && (ul_mpu_addr + ul_num_bytes > vma->vm_end)) {
1234 		/* jump to the next VMA region */
1235 		vma = find_vma(mm, vma->vm_end + 1);
1236 		dev_dbg(bridge,
1237 			"VMA for UserBuf ul_mpu_addr=%x ul_num_bytes=%x, "
1238 			"vm_start=%lx, vm_end=%lx, vm_flags=%lx\n", ul_mpu_addr,
1239 			ul_num_bytes, vma->vm_start, vma->vm_end,
1240 			vma->vm_flags);
1241 	}
1242 	if (!vma) {
1243 		pr_err("%s: Failed to get VMA region for 0x%x (%d)\n",
1244 		       __func__, ul_mpu_addr, ul_num_bytes);
1245 		status = -EINVAL;
1246 		up_read(&mm->mmap_sem);
1247 		goto func_cont;
1248 	}
1249 
1250 	if (vma->vm_flags & VM_IO) {
1251 		num_usr_pgs = ul_num_bytes / PG_SIZE4K;
1252 		mpu_addr = ul_mpu_addr;
1253 
1254 		/* Get the physical addresses for user buffer */
1255 		for (pg_i = 0; pg_i < num_usr_pgs; pg_i++) {
1256 			pa = user_va2_pa(mm, mpu_addr);
1257 			if (!pa) {
1258 				status = -EPERM;
1259 				pr_err("DSPBRIDGE: VM_IO mapping physical"
1260 				       "address is invalid\n");
1261 				break;
1262 			}
1263 			if (pfn_valid(__phys_to_pfn(pa))) {
1264 				pg = PHYS_TO_PAGE(pa);
1265 				get_page(pg);
1266 				if (page_count(pg) < 1) {
1267 					pr_err("Bad page in VM_IO buffer\n");
1268 					bad_page_dump(pa, pg);
1269 				}
1270 			}
1271 			status = pte_set(dev_context->pt_attrs, pa,
1272 					 va, HW_PAGE_SIZE4KB, &hw_attrs);
1273 			if (status)
1274 				break;
1275 
1276 			va += HW_PAGE_SIZE4KB;
1277 			mpu_addr += HW_PAGE_SIZE4KB;
1278 			pa += HW_PAGE_SIZE4KB;
1279 		}
1280 	} else {
1281 		num_usr_pgs = ul_num_bytes / PG_SIZE4K;
1282 		if (vma->vm_flags & (VM_WRITE | VM_MAYWRITE))
1283 			write = 1;
1284 
1285 		for (pg_i = 0; pg_i < num_usr_pgs; pg_i++) {
1286 			pg_num = get_user_pages(curr_task, mm, ul_mpu_addr, 1,
1287 						write, 1, &mapped_page, NULL);
1288 			if (pg_num > 0) {
1289 				if (page_count(mapped_page) < 1) {
1290 					pr_err("Bad page count after doing"
1291 					       "get_user_pages on"
1292 					       "user buffer\n");
1293 					bad_page_dump(page_to_phys(mapped_page),
1294 						      mapped_page);
1295 				}
1296 				status = pte_set(dev_context->pt_attrs,
1297 						 page_to_phys(mapped_page), va,
1298 						 HW_PAGE_SIZE4KB, &hw_attrs);
1299 				if (status)
1300 					break;
1301 
1302 				if (mapped_pages)
1303 					mapped_pages[pg_i] = mapped_page;
1304 
1305 				va += HW_PAGE_SIZE4KB;
1306 				ul_mpu_addr += HW_PAGE_SIZE4KB;
1307 			} else {
1308 				pr_err("DSPBRIDGE: get_user_pages FAILED,"
1309 				       "MPU addr = 0x%x,"
1310 				       "vma->vm_flags = 0x%lx,"
1311 				       "get_user_pages Err"
1312 				       "Value = %d, Buffer"
1313 				       "size=0x%x\n", ul_mpu_addr,
1314 				       vma->vm_flags, pg_num, ul_num_bytes);
1315 				status = -EPERM;
1316 				break;
1317 			}
1318 		}
1319 	}
1320 	up_read(&mm->mmap_sem);
1321 func_cont:
1322 	if (status) {
1323 		/*
1324 		 * Roll out the mapped pages incase it failed in middle of
1325 		 * mapping
1326 		 */
1327 		if (pg_i) {
1328 			bridge_brd_mem_un_map(dev_context, virt_addr,
1329 					   (pg_i * PG_SIZE4K));
1330 		}
1331 		status = -EPERM;
1332 	}
1333 	/*
1334 	 * In any case, flush the TLB
1335 	 * This is called from here instead from pte_update to avoid unnecessary
1336 	 * repetition while mapping non-contiguous physical regions of a virtual
1337 	 * region
1338 	 */
1339 	flush_all(dev_context);
1340 	dev_dbg(bridge, "%s status %x\n", __func__, status);
1341 	return status;
1342 }
1343 
1344 /*
1345  *  ======== bridge_brd_mem_un_map ========
1346  *      Invalidate the PTEs for the DSP VA block to be unmapped.
1347  *
1348  *      PTEs of a mapped memory block are contiguous in any page table
1349  *      So, instead of looking up the PTE address for every 4K block,
1350  *      we clear consecutive PTEs until we unmap all the bytes
1351  */
bridge_brd_mem_un_map(struct bridge_dev_context * dev_ctxt,u32 virt_addr,u32 ul_num_bytes)1352 static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctxt,
1353 				     u32 virt_addr, u32 ul_num_bytes)
1354 {
1355 	u32 l1_base_va;
1356 	u32 l2_base_va;
1357 	u32 l2_base_pa;
1358 	u32 l2_page_num;
1359 	u32 pte_val;
1360 	u32 pte_size;
1361 	u32 pte_count;
1362 	u32 pte_addr_l1;
1363 	u32 pte_addr_l2 = 0;
1364 	u32 rem_bytes;
1365 	u32 rem_bytes_l2;
1366 	u32 va_curr;
1367 	struct page *pg = NULL;
1368 	int status = 0;
1369 	struct bridge_dev_context *dev_context = dev_ctxt;
1370 	struct pg_table_attrs *pt = dev_context->pt_attrs;
1371 	u32 temp;
1372 	u32 paddr;
1373 	u32 numof4k_pages = 0;
1374 
1375 	va_curr = virt_addr;
1376 	rem_bytes = ul_num_bytes;
1377 	rem_bytes_l2 = 0;
1378 	l1_base_va = pt->l1_base_va;
1379 	pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va_curr);
1380 	dev_dbg(bridge, "%s dev_ctxt %p, va %x, NumBytes %x l1_base_va %x, "
1381 		"pte_addr_l1 %x\n", __func__, dev_ctxt, virt_addr,
1382 		ul_num_bytes, l1_base_va, pte_addr_l1);
1383 
1384 	while (rem_bytes && !status) {
1385 		u32 va_curr_orig = va_curr;
1386 		/* Find whether the L1 PTE points to a valid L2 PT */
1387 		pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va_curr);
1388 		pte_val = *(u32 *) pte_addr_l1;
1389 		pte_size = hw_mmu_pte_size_l1(pte_val);
1390 
1391 		if (pte_size != HW_MMU_COARSE_PAGE_SIZE)
1392 			goto skip_coarse_page;
1393 
1394 		/*
1395 		 * Get the L2 PA from the L1 PTE, and find
1396 		 * corresponding L2 VA
1397 		 */
1398 		l2_base_pa = hw_mmu_pte_coarse_l1(pte_val);
1399 		l2_base_va = l2_base_pa - pt->l2_base_pa + pt->l2_base_va;
1400 		l2_page_num =
1401 		    (l2_base_pa - pt->l2_base_pa) / HW_MMU_COARSE_PAGE_SIZE;
1402 		/*
1403 		 * Find the L2 PTE address from which we will start
1404 		 * clearing, the number of PTEs to be cleared on this
1405 		 * page, and the size of VA space that needs to be
1406 		 * cleared on this L2 page
1407 		 */
1408 		pte_addr_l2 = hw_mmu_pte_addr_l2(l2_base_va, va_curr);
1409 		pte_count = pte_addr_l2 & (HW_MMU_COARSE_PAGE_SIZE - 1);
1410 		pte_count = (HW_MMU_COARSE_PAGE_SIZE - pte_count) / sizeof(u32);
1411 		if (rem_bytes < (pte_count * PG_SIZE4K))
1412 			pte_count = rem_bytes / PG_SIZE4K;
1413 		rem_bytes_l2 = pte_count * PG_SIZE4K;
1414 
1415 		/*
1416 		 * Unmap the VA space on this L2 PT. A quicker way
1417 		 * would be to clear pte_count entries starting from
1418 		 * pte_addr_l2. However, below code checks that we don't
1419 		 * clear invalid entries or less than 64KB for a 64KB
1420 		 * entry. Similar checking is done for L1 PTEs too
1421 		 * below
1422 		 */
1423 		while (rem_bytes_l2 && !status) {
1424 			pte_val = *(u32 *) pte_addr_l2;
1425 			pte_size = hw_mmu_pte_size_l2(pte_val);
1426 			/* va_curr aligned to pte_size? */
1427 			if (pte_size == 0 || rem_bytes_l2 < pte_size ||
1428 			    va_curr & (pte_size - 1)) {
1429 				status = -EPERM;
1430 				break;
1431 			}
1432 
1433 			/* Collect Physical addresses from VA */
1434 			paddr = (pte_val & ~(pte_size - 1));
1435 			if (pte_size == HW_PAGE_SIZE64KB)
1436 				numof4k_pages = 16;
1437 			else
1438 				numof4k_pages = 1;
1439 			temp = 0;
1440 			while (temp++ < numof4k_pages) {
1441 				if (!pfn_valid(__phys_to_pfn(paddr))) {
1442 					paddr += HW_PAGE_SIZE4KB;
1443 					continue;
1444 				}
1445 				pg = PHYS_TO_PAGE(paddr);
1446 				if (page_count(pg) < 1) {
1447 					pr_info("DSPBRIDGE: UNMAP function: "
1448 						"COUNT 0 FOR PA 0x%x, size = "
1449 						"0x%x\n", paddr, ul_num_bytes);
1450 					bad_page_dump(paddr, pg);
1451 				} else {
1452 					set_page_dirty(pg);
1453 					page_cache_release(pg);
1454 				}
1455 				paddr += HW_PAGE_SIZE4KB;
1456 			}
1457 			if (hw_mmu_pte_clear(pte_addr_l2, va_curr, pte_size)) {
1458 				status = -EPERM;
1459 				goto EXIT_LOOP;
1460 			}
1461 
1462 			status = 0;
1463 			rem_bytes_l2 -= pte_size;
1464 			va_curr += pte_size;
1465 			pte_addr_l2 += (pte_size >> 12) * sizeof(u32);
1466 		}
1467 		spin_lock(&pt->pg_lock);
1468 		if (rem_bytes_l2 == 0) {
1469 			pt->pg_info[l2_page_num].num_entries -= pte_count;
1470 			if (pt->pg_info[l2_page_num].num_entries == 0) {
1471 				/*
1472 				 * Clear the L1 PTE pointing to the L2 PT
1473 				 */
1474 				if (!hw_mmu_pte_clear(l1_base_va, va_curr_orig,
1475 						     HW_MMU_COARSE_PAGE_SIZE))
1476 					status = 0;
1477 				else {
1478 					status = -EPERM;
1479 					spin_unlock(&pt->pg_lock);
1480 					goto EXIT_LOOP;
1481 				}
1482 			}
1483 			rem_bytes -= pte_count * PG_SIZE4K;
1484 		} else
1485 			status = -EPERM;
1486 
1487 		spin_unlock(&pt->pg_lock);
1488 		continue;
1489 skip_coarse_page:
1490 		/* va_curr aligned to pte_size? */
1491 		/* pte_size = 1 MB or 16 MB */
1492 		if (pte_size == 0 || rem_bytes < pte_size ||
1493 		    va_curr & (pte_size - 1)) {
1494 			status = -EPERM;
1495 			break;
1496 		}
1497 
1498 		if (pte_size == HW_PAGE_SIZE1MB)
1499 			numof4k_pages = 256;
1500 		else
1501 			numof4k_pages = 4096;
1502 		temp = 0;
1503 		/* Collect Physical addresses from VA */
1504 		paddr = (pte_val & ~(pte_size - 1));
1505 		while (temp++ < numof4k_pages) {
1506 			if (pfn_valid(__phys_to_pfn(paddr))) {
1507 				pg = PHYS_TO_PAGE(paddr);
1508 				if (page_count(pg) < 1) {
1509 					pr_info("DSPBRIDGE: UNMAP function: "
1510 						"COUNT 0 FOR PA 0x%x, size = "
1511 						"0x%x\n", paddr, ul_num_bytes);
1512 					bad_page_dump(paddr, pg);
1513 				} else {
1514 					set_page_dirty(pg);
1515 					page_cache_release(pg);
1516 				}
1517 			}
1518 			paddr += HW_PAGE_SIZE4KB;
1519 		}
1520 		if (!hw_mmu_pte_clear(l1_base_va, va_curr, pte_size)) {
1521 			status = 0;
1522 			rem_bytes -= pte_size;
1523 			va_curr += pte_size;
1524 		} else {
1525 			status = -EPERM;
1526 			goto EXIT_LOOP;
1527 		}
1528 	}
1529 	/*
1530 	 * It is better to flush the TLB here, so that any stale old entries
1531 	 * get flushed
1532 	 */
1533 EXIT_LOOP:
1534 	flush_all(dev_context);
1535 	dev_dbg(bridge,
1536 		"%s: va_curr %x, pte_addr_l1 %x pte_addr_l2 %x rem_bytes %x,"
1537 		" rem_bytes_l2 %x status %x\n", __func__, va_curr, pte_addr_l1,
1538 		pte_addr_l2, rem_bytes, rem_bytes_l2, status);
1539 	return status;
1540 }
1541 
1542 /*
1543  *  ======== user_va2_pa ========
1544  *  Purpose:
1545  *      This function walks through the page tables to convert a userland
1546  *      virtual address to physical address
1547  */
user_va2_pa(struct mm_struct * mm,u32 address)1548 static u32 user_va2_pa(struct mm_struct *mm, u32 address)
1549 {
1550 	pgd_t *pgd;
1551 	pmd_t *pmd;
1552 	pte_t *ptep, pte;
1553 
1554 	pgd = pgd_offset(mm, address);
1555 	if (!(pgd_none(*pgd) || pgd_bad(*pgd))) {
1556 		pmd = pmd_offset(pgd, address);
1557 		if (!(pmd_none(*pmd) || pmd_bad(*pmd))) {
1558 			ptep = pte_offset_map(pmd, address);
1559 			if (ptep) {
1560 				pte = *ptep;
1561 				if (pte_present(pte))
1562 					return pte & PAGE_MASK;
1563 			}
1564 		}
1565 	}
1566 
1567 	return 0;
1568 }
1569 
1570 /*
1571  *  ======== pte_update ========
1572  *      This function calculates the optimum page-aligned addresses and sizes
1573  *      Caller must pass page-aligned values
1574  */
pte_update(struct bridge_dev_context * dev_ctxt,u32 pa,u32 va,u32 size,struct hw_mmu_map_attrs_t * map_attrs)1575 static int pte_update(struct bridge_dev_context *dev_ctxt, u32 pa,
1576 			     u32 va, u32 size,
1577 			     struct hw_mmu_map_attrs_t *map_attrs)
1578 {
1579 	u32 i;
1580 	u32 all_bits;
1581 	u32 pa_curr = pa;
1582 	u32 va_curr = va;
1583 	u32 num_bytes = size;
1584 	struct bridge_dev_context *dev_context = dev_ctxt;
1585 	int status = 0;
1586 	u32 page_size[] = { HW_PAGE_SIZE16MB, HW_PAGE_SIZE1MB,
1587 		HW_PAGE_SIZE64KB, HW_PAGE_SIZE4KB
1588 	};
1589 
1590 	while (num_bytes && !status) {
1591 		/* To find the max. page size with which both PA & VA are
1592 		 * aligned */
1593 		all_bits = pa_curr | va_curr;
1594 
1595 		for (i = 0; i < 4; i++) {
1596 			if ((num_bytes >= page_size[i]) && ((all_bits &
1597 							     (page_size[i] -
1598 							      1)) == 0)) {
1599 				status =
1600 				    pte_set(dev_context->pt_attrs, pa_curr,
1601 					    va_curr, page_size[i], map_attrs);
1602 				pa_curr += page_size[i];
1603 				va_curr += page_size[i];
1604 				num_bytes -= page_size[i];
1605 				/* Don't try smaller sizes. Hopefully we have
1606 				 * reached an address aligned to a bigger page
1607 				 * size */
1608 				break;
1609 			}
1610 		}
1611 	}
1612 
1613 	return status;
1614 }
1615 
1616 /*
1617  *  ======== pte_set ========
1618  *      This function calculates PTE address (MPU virtual) to be updated
1619  *      It also manages the L2 page tables
1620  */
pte_set(struct pg_table_attrs * pt,u32 pa,u32 va,u32 size,struct hw_mmu_map_attrs_t * attrs)1621 static int pte_set(struct pg_table_attrs *pt, u32 pa, u32 va,
1622 			  u32 size, struct hw_mmu_map_attrs_t *attrs)
1623 {
1624 	u32 i;
1625 	u32 pte_val;
1626 	u32 pte_addr_l1;
1627 	u32 pte_size;
1628 	/* Base address of the PT that will be updated */
1629 	u32 pg_tbl_va;
1630 	u32 l1_base_va;
1631 	/* Compiler warns that the next three variables might be used
1632 	 * uninitialized in this function. Doesn't seem so. Working around,
1633 	 * anyways. */
1634 	u32 l2_base_va = 0;
1635 	u32 l2_base_pa = 0;
1636 	u32 l2_page_num = 0;
1637 	int status = 0;
1638 
1639 	l1_base_va = pt->l1_base_va;
1640 	pg_tbl_va = l1_base_va;
1641 	if ((size == HW_PAGE_SIZE64KB) || (size == HW_PAGE_SIZE4KB)) {
1642 		/* Find whether the L1 PTE points to a valid L2 PT */
1643 		pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va);
1644 		if (pte_addr_l1 <= (pt->l1_base_va + pt->l1_size)) {
1645 			pte_val = *(u32 *) pte_addr_l1;
1646 			pte_size = hw_mmu_pte_size_l1(pte_val);
1647 		} else {
1648 			return -EPERM;
1649 		}
1650 		spin_lock(&pt->pg_lock);
1651 		if (pte_size == HW_MMU_COARSE_PAGE_SIZE) {
1652 			/* Get the L2 PA from the L1 PTE, and find
1653 			 * corresponding L2 VA */
1654 			l2_base_pa = hw_mmu_pte_coarse_l1(pte_val);
1655 			l2_base_va =
1656 			    l2_base_pa - pt->l2_base_pa + pt->l2_base_va;
1657 			l2_page_num =
1658 			    (l2_base_pa -
1659 			     pt->l2_base_pa) / HW_MMU_COARSE_PAGE_SIZE;
1660 		} else if (pte_size == 0) {
1661 			/* L1 PTE is invalid. Allocate a L2 PT and
1662 			 * point the L1 PTE to it */
1663 			/* Find a free L2 PT. */
1664 			for (i = 0; (i < pt->l2_num_pages) &&
1665 			     (pt->pg_info[i].num_entries != 0); i++)
1666 				;
1667 			if (i < pt->l2_num_pages) {
1668 				l2_page_num = i;
1669 				l2_base_pa = pt->l2_base_pa + (l2_page_num *
1670 						HW_MMU_COARSE_PAGE_SIZE);
1671 				l2_base_va = pt->l2_base_va + (l2_page_num *
1672 						HW_MMU_COARSE_PAGE_SIZE);
1673 				/* Endianness attributes are ignored for
1674 				 * HW_MMU_COARSE_PAGE_SIZE */
1675 				status =
1676 				    hw_mmu_pte_set(l1_base_va, l2_base_pa, va,
1677 						   HW_MMU_COARSE_PAGE_SIZE,
1678 						   attrs);
1679 			} else {
1680 				status = -ENOMEM;
1681 			}
1682 		} else {
1683 			/* Found valid L1 PTE of another size.
1684 			 * Should not overwrite it. */
1685 			status = -EPERM;
1686 		}
1687 		if (!status) {
1688 			pg_tbl_va = l2_base_va;
1689 			if (size == HW_PAGE_SIZE64KB)
1690 				pt->pg_info[l2_page_num].num_entries += 16;
1691 			else
1692 				pt->pg_info[l2_page_num].num_entries++;
1693 			dev_dbg(bridge, "PTE: L2 BaseVa %x, BasePa %x, PageNum "
1694 				"%x, num_entries %x\n", l2_base_va,
1695 				l2_base_pa, l2_page_num,
1696 				pt->pg_info[l2_page_num].num_entries);
1697 		}
1698 		spin_unlock(&pt->pg_lock);
1699 	}
1700 	if (!status) {
1701 		dev_dbg(bridge, "PTE: pg_tbl_va %x, pa %x, va %x, size %x\n",
1702 			pg_tbl_va, pa, va, size);
1703 		dev_dbg(bridge, "PTE: endianism %x, element_size %x, "
1704 			"mixed_size %x\n", attrs->endianism,
1705 			attrs->element_size, attrs->mixed_size);
1706 		status = hw_mmu_pte_set(pg_tbl_va, pa, va, size, attrs);
1707 	}
1708 
1709 	return status;
1710 }
1711 
1712 /* 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)1713 static int mem_map_vmalloc(struct bridge_dev_context *dev_context,
1714 				  u32 ul_mpu_addr, u32 virt_addr,
1715 				  u32 ul_num_bytes,
1716 				  struct hw_mmu_map_attrs_t *hw_attrs)
1717 {
1718 	int status = 0;
1719 	struct page *page[1];
1720 	u32 i;
1721 	u32 pa_curr;
1722 	u32 pa_next;
1723 	u32 va_curr;
1724 	u32 size_curr;
1725 	u32 num_pages;
1726 	u32 pa;
1727 	u32 num_of4k_pages;
1728 	u32 temp = 0;
1729 
1730 	/*
1731 	 * Do Kernel va to pa translation.
1732 	 * Combine physically contiguous regions to reduce TLBs.
1733 	 * Pass the translated pa to pte_update.
1734 	 */
1735 	num_pages = ul_num_bytes / PAGE_SIZE;	/* PAGE_SIZE = OS page size */
1736 	i = 0;
1737 	va_curr = ul_mpu_addr;
1738 	page[0] = vmalloc_to_page((void *)va_curr);
1739 	pa_next = page_to_phys(page[0]);
1740 	while (!status && (i < num_pages)) {
1741 		/*
1742 		 * Reuse pa_next from the previous iteraion to avoid
1743 		 * an extra va2pa call
1744 		 */
1745 		pa_curr = pa_next;
1746 		size_curr = PAGE_SIZE;
1747 		/*
1748 		 * If the next page is physically contiguous,
1749 		 * map it with the current one by increasing
1750 		 * the size of the region to be mapped
1751 		 */
1752 		while (++i < num_pages) {
1753 			page[0] =
1754 			    vmalloc_to_page((void *)(va_curr + size_curr));
1755 			pa_next = page_to_phys(page[0]);
1756 
1757 			if (pa_next == (pa_curr + size_curr))
1758 				size_curr += PAGE_SIZE;
1759 			else
1760 				break;
1761 
1762 		}
1763 		if (pa_next == 0) {
1764 			status = -ENOMEM;
1765 			break;
1766 		}
1767 		pa = pa_curr;
1768 		num_of4k_pages = size_curr / HW_PAGE_SIZE4KB;
1769 		while (temp++ < num_of4k_pages) {
1770 			get_page(PHYS_TO_PAGE(pa));
1771 			pa += HW_PAGE_SIZE4KB;
1772 		}
1773 		status = pte_update(dev_context, pa_curr, virt_addr +
1774 				    (va_curr - ul_mpu_addr), size_curr,
1775 				    hw_attrs);
1776 		va_curr += size_curr;
1777 	}
1778 	/*
1779 	 * In any case, flush the TLB
1780 	 * This is called from here instead from pte_update to avoid unnecessary
1781 	 * repetition while mapping non-contiguous physical regions of a virtual
1782 	 * region
1783 	 */
1784 	flush_all(dev_context);
1785 	dev_dbg(bridge, "%s status %x\n", __func__, status);
1786 	return status;
1787 }
1788 
1789 /*
1790  *  ======== wait_for_start ========
1791  *      Wait for the singal from DSP that it has started, or time out.
1792  */
wait_for_start(struct bridge_dev_context * dev_context,u32 dw_sync_addr)1793 bool wait_for_start(struct bridge_dev_context *dev_context, u32 dw_sync_addr)
1794 {
1795 	u16 timeout = TIHELEN_ACKTIMEOUT;
1796 
1797 	/*  Wait for response from board */
1798 	while (__raw_readw(dw_sync_addr) && --timeout)
1799 		udelay(10);
1800 
1801 	/*  If timed out: return false */
1802 	if (!timeout) {
1803 		pr_err("%s: Timed out waiting DSP to Start\n", __func__);
1804 		return false;
1805 	}
1806 	return true;
1807 }
1808