4 * DSP-BIOS Bridge driver support functions for TI OMAP processors.
6 * Processor Manager Driver for TI OMAP3430 EVM.
8 * Copyright (C) 2005-2006 Texas Instruments, Inc.
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.
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.
19 #include <linux/platform_data/dsp-omap.h>
21 #include <linux/types.h>
22 /* ----------------------------------- Host OS */
23 #include <dspbridge/host_os.h>
25 #include <linux/mmzone.h>
27 /* ----------------------------------- DSP/BIOS Bridge */
28 #include <dspbridge/dbdefs.h>
30 /* ----------------------------------- OS Adaptation Layer */
31 #include <dspbridge/drv.h>
32 #include <dspbridge/sync.h>
34 /* ------------------------------------ Hardware Abstraction Layer */
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>
47 /* ----------------------------------- Platform Manager */
48 #include <dspbridge/dev.h>
49 #include <dspbridge/dspapi.h>
50 #include <dspbridge/dmm.h>
51 #include <dspbridge/wdt.h>
53 /* ----------------------------------- Local */
55 #include "_tiomap_pwr.h"
56 #include "tiomap_io.h"
58 /* Offset in shared mem to write to in order to synchronize start with DSP */
59 #define SHMSYNCOFFSET 4 /* GPP byte offset */
61 #define BUFFERSIZE 1024
63 #define TIHELEN_ACKTIMEOUT 10000
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)
77 /* Forward Declarations: */
78 static int bridge_brd_monitor(struct bridge_dev_context *dev_ctxt);
79 static int bridge_brd_read(struct bridge_dev_context *dev_ctxt,
81 u32 dsp_addr, u32 ul_num_bytes,
83 static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
85 static int bridge_brd_status(struct bridge_dev_context *dev_ctxt,
87 static int bridge_brd_stop(struct bridge_dev_context *dev_ctxt);
88 static int bridge_brd_write(struct bridge_dev_context *dev_ctxt,
90 u32 dsp_addr, u32 ul_num_bytes,
92 static int bridge_brd_set_state(struct bridge_dev_context *dev_ctxt,
94 static int bridge_brd_mem_copy(struct bridge_dev_context *dev_ctxt,
95 u32 dsp_dest_addr, u32 dsp_src_addr,
96 u32 ul_num_bytes, u32 mem_type);
97 static int bridge_brd_mem_write(struct bridge_dev_context *dev_ctxt,
98 u8 *host_buff, u32 dsp_addr,
99 u32 ul_num_bytes, u32 mem_type);
100 static int bridge_brd_mem_map(struct bridge_dev_context *dev_ctxt,
101 u32 ul_mpu_addr, u32 virt_addr,
102 u32 ul_num_bytes, u32 ul_map_attr,
103 struct page **mapped_pages);
104 static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctxt,
105 u32 virt_addr, u32 ul_num_bytes);
106 static int bridge_dev_create(struct bridge_dev_context
108 struct dev_object *hdev_obj,
109 struct cfg_hostres *config_param);
110 static int bridge_dev_ctrl(struct bridge_dev_context *dev_context,
111 u32 dw_cmd, void *pargs);
112 static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt);
113 static u32 user_va2_pa(struct mm_struct *mm, u32 address);
114 static int pte_update(struct bridge_dev_context *dev_ctxt, u32 pa,
116 struct hw_mmu_map_attrs_t *map_attrs);
117 static int pte_set(struct pg_table_attrs *pt, u32 pa, u32 va,
118 u32 size, struct hw_mmu_map_attrs_t *attrs);
119 static int mem_map_vmalloc(struct bridge_dev_context *dev_context,
120 u32 ul_mpu_addr, u32 virt_addr,
122 struct hw_mmu_map_attrs_t *hw_attrs);
124 bool wait_for_start(struct bridge_dev_context *dev_context,
125 void __iomem *sync_addr);
127 /* ----------------------------------- Globals */
129 /* Attributes of L2 page tables for DSP MMU */
131 u32 num_entries; /* Number of valid PTEs in the L2 PT */
134 /* Attributes used to manage the DSP MMU page tables */
135 struct pg_table_attrs {
136 spinlock_t pg_lock; /* Critical section object handle */
138 u32 l1_base_pa; /* Physical address of the L1 PT */
139 u32 l1_base_va; /* Virtual address of the L1 PT */
140 u32 l1_size; /* Size of the L1 PT */
142 /* Physical address of Allocated mem for L1 table. May not be aligned */
144 /* Virtual address of Allocated mem for L1 table. May not be aligned */
146 /* Size of consistent memory allocated for L1 table.
147 * May not be aligned */
149 u32 l2_base_pa; /* Physical address of the L2 PT */
150 u32 l2_base_va; /* Virtual address of the L2 PT */
151 u32 l2_size; /* Size of the L2 PT */
153 /* Physical address of Allocated mem for L2 table. May not be aligned */
155 /* Virtual address of Allocated mem for L2 table. May not be aligned */
157 /* Size of consistent memory allocated for L2 table.
158 * May not be aligned */
160 u32 l2_num_pages; /* Number of allocated L2 PT */
161 /* Array [l2_num_pages] of L2 PT info structs */
162 struct page_info *pg_info;
166 * This Bridge driver's function interface table.
168 static struct bridge_drv_interface drv_interface_fxns = {
169 /* Bridge API ver. for which this bridge driver is built. */
170 BRD_API_MAJOR_VERSION,
171 BRD_API_MINOR_VERSION,
181 bridge_brd_set_state,
183 bridge_brd_mem_write,
185 bridge_brd_mem_un_map,
186 /* The following CHNL functions are provided by chnl_io.lib: */
191 bridge_chnl_add_io_req,
193 bridge_chnl_cancel_io,
194 bridge_chnl_flush_io,
195 bridge_chnl_get_info,
196 bridge_chnl_get_mgr_info,
198 bridge_chnl_register_notify,
199 /* The following IO functions are provided by chnl_io.lib: */
203 bridge_io_get_proc_load,
204 /* The following msg_ctrl functions are provided by chnl_io.lib: */
206 bridge_msg_create_queue,
208 bridge_msg_delete_queue,
211 bridge_msg_register_notify,
212 bridge_msg_set_queue_id,
215 static struct notifier_block dsp_mbox_notifier = {
216 .notifier_call = io_mbox_msg,
219 static inline void flush_all(struct bridge_dev_context *dev_context)
221 if (dev_context->brd_state == BRD_DSP_HIBERNATION ||
222 dev_context->brd_state == BRD_HIBERNATION)
223 wake_dsp(dev_context, NULL);
225 hw_mmu_tlb_flush_all(dev_context->dsp_mmu_base);
228 static void bad_page_dump(u32 pa, struct page *pg)
230 pr_emerg("DSPBRIDGE: MAP function: COUNT 0 FOR PA 0x%x\n", pa);
231 pr_emerg("Bad page state in process '%s'\n"
232 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
234 current->comm, pg, (int)(2 * sizeof(unsigned long)),
235 (unsigned long)pg->flags, pg->mapping,
236 page_mapcount(pg), page_count(pg));
241 * ======== bridge_drv_entry ========
243 * Bridge Driver entry point.
245 void bridge_drv_entry(struct bridge_drv_interface **drv_intf,
246 const char *driver_file_name)
248 if (strcmp(driver_file_name, "UMA") == 0)
249 *drv_intf = &drv_interface_fxns;
251 dev_dbg(bridge, "%s Unknown Bridge file name", __func__);
256 * ======== bridge_brd_monitor ========
258 * This bridge_brd_monitor puts DSP into a Loadable state.
259 * i.e Application can load and start the device.
262 * Device in 'OFF' state.
264 static int bridge_brd_monitor(struct bridge_dev_context *dev_ctxt)
266 struct bridge_dev_context *dev_context = dev_ctxt;
268 struct omap_dsp_platform_data *pdata =
269 omap_dspbridge_dev->dev.platform_data;
271 temp = (*pdata->dsp_prm_read)(OMAP3430_IVA2_MOD, OMAP2_PM_PWSTST) &
272 OMAP_POWERSTATEST_MASK;
273 if (!(temp & 0x02)) {
274 /* IVA2 is not in ON state */
275 /* Read and set PM_PWSTCTRL_IVA2 to ON */
276 (*pdata->dsp_prm_rmw_bits)(OMAP_POWERSTATEST_MASK,
277 PWRDM_POWER_ON, OMAP3430_IVA2_MOD, OMAP2_PM_PWSTCTRL);
278 /* Set the SW supervised state transition */
279 (*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_FORCE_WAKEUP,
280 OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
282 /* Wait until the state has moved to ON */
283 while (*pdata->dsp_prm_read(OMAP3430_IVA2_MOD, OMAP2_PM_PWSTST)&
284 OMAP_INTRANSITION_MASK);
285 /* Disable Automatic transition */
286 (*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_DISABLE_AUTO,
287 OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
289 (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0,
290 OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
291 dsp_clk_enable(DSP_CLK_IVA2);
293 /* set the device state to IDLE */
294 dev_context->brd_state = BRD_IDLE;
300 * ======== bridge_brd_read ========
302 * Reads buffers for DSP memory.
304 static int bridge_brd_read(struct bridge_dev_context *dev_ctxt,
305 u8 *host_buff, u32 dsp_addr,
306 u32 ul_num_bytes, u32 mem_type)
309 struct bridge_dev_context *dev_context = dev_ctxt;
311 u32 dsp_base_addr = dev_ctxt->dsp_base_addr;
313 if (dsp_addr < dev_context->dsp_start_add) {
317 /* change here to account for the 3 bands of the DSP internal memory */
318 if ((dsp_addr - dev_context->dsp_start_add) <
319 dev_context->internal_size) {
320 offset = dsp_addr - dev_context->dsp_start_add;
322 status = read_ext_dsp_data(dev_context, host_buff, dsp_addr,
323 ul_num_bytes, mem_type);
326 /* copy the data from DSP memory */
327 memcpy(host_buff, (void *)(dsp_base_addr + offset), ul_num_bytes);
332 * ======== bridge_brd_set_state ========
334 * This routine updates the Board status.
336 static int bridge_brd_set_state(struct bridge_dev_context *dev_ctxt,
340 struct bridge_dev_context *dev_context = dev_ctxt;
342 dev_context->brd_state = brd_state;
347 * ======== bridge_brd_start ========
349 * Initializes DSP MMU and Starts DSP.
352 * a) DSP domain is 'ACTIVE'.
353 * b) DSP_RST1 is asserted.
354 * b) DSP_RST2 is released.
356 static int bridge_brd_start(struct bridge_dev_context *dev_ctxt,
360 struct bridge_dev_context *dev_context = dev_ctxt;
361 void __iomem *sync_addr;
362 u32 ul_shm_base; /* Gpp Phys SM base addr(byte) */
363 u32 ul_shm_base_virt; /* Dsp Virt SM base addr */
364 u32 ul_tlb_base_virt; /* Base of MMU TLB entry */
366 /* Offset of shm_base_virt from tlb_base_virt */
367 u32 ul_shm_offset_virt;
369 s32 itmp_entry_ndx = 0; /* DSP-MMU TLB entry base address */
370 struct cfg_hostres *resources = NULL;
374 u32 ul_bios_gp_timer;
376 struct io_mgr *hio_mgr;
377 u32 ul_load_monitor_timer;
379 struct omap_dsp_platform_data *pdata =
380 omap_dspbridge_dev->dev.platform_data;
382 /* The device context contains all the mmu setup info from when the
383 * last dsp base image was loaded. The first entry is always
385 /* Get SHM_BEG - convert to byte address */
386 (void)dev_get_symbol(dev_context->dev_obj, SHMBASENAME,
388 ul_shm_base_virt *= DSPWORDSIZE;
389 /* DSP Virtual address */
390 ul_tlb_base_virt = dev_context->atlb_entry[0].dsp_va;
392 ul_shm_base_virt - (ul_tlb_base_virt * DSPWORDSIZE);
393 /* Kernel logical address */
394 ul_shm_base = dev_context->atlb_entry[0].gpp_va + ul_shm_offset_virt;
396 /* SHM physical sync address */
397 shm_sync_pa = dev_context->atlb_entry[0].gpp_pa + ul_shm_offset_virt +
400 /* 2nd wd is used as sync field */
401 sync_addr = ioremap(shm_sync_pa, SZ_32);
405 /* Write a signature into the shm base + offset; this will
406 * get cleared when the DSP program starts. */
407 if ((ul_shm_base_virt == 0) || (ul_shm_base == 0)) {
408 pr_err("%s: Illegal SM base\n", __func__);
411 __raw_writel(0xffffffff, sync_addr);
414 resources = dev_context->resources;
418 /* Assert RST1 i.e only the RST only for DSP megacell */
420 (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST1_IVA2_MASK,
421 OMAP3430_RST1_IVA2_MASK,
425 /* Mask address with 1K for compatibility */
426 pdata->set_bootaddr(dsp_addr &
427 OMAP3_IVA2_BOOTADDR_MASK);
428 pdata->set_bootmode(dsp_debug ? IDLE : DIRECT);
432 /* Reset and Unreset the RST2, so that BOOTADDR is copied to
433 * IVA2 SYSC register */
434 (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK,
435 OMAP3430_RST2_IVA2_MASK, OMAP3430_IVA2_MOD,
438 (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0,
439 OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
442 /* Disbale the DSP MMU */
443 hw_mmu_disable(resources->dmmu_base);
445 hw_mmu_twl_disable(resources->dmmu_base);
447 /* Only make TLB entry if both addresses are non-zero */
448 for (entry_ndx = 0; entry_ndx < BRDIOCTL_NUMOFMMUTLB;
450 struct bridge_ioctl_extproc *e =
451 &dev_context->atlb_entry[entry_ndx];
452 struct hw_mmu_map_attrs_t map_attrs = {
453 .endianism = e->endianism,
454 .element_size = e->elem_size,
455 .mixed_size = e->mixed_mode,
458 if (!e->gpp_pa || !e->dsp_va)
462 "MMU %d, pa: 0x%x, va: 0x%x, size: 0x%x",
468 hw_mmu_tlb_add(dev_context->dsp_mmu_base,
479 /* Lock the above TLB entries and get the BIOS and load monitor timer
482 hw_mmu_num_locked_set(resources->dmmu_base, itmp_entry_ndx);
483 hw_mmu_victim_num_set(resources->dmmu_base, itmp_entry_ndx);
484 hw_mmu_ttb_set(resources->dmmu_base,
485 dev_context->pt_attrs->l1_base_pa);
486 hw_mmu_twl_enable(resources->dmmu_base);
487 /* Enable the SmartIdle and AutoIdle bit for MMU_SYSCONFIG */
489 temp = __raw_readl((resources->dmmu_base) + 0x10);
490 temp = (temp & 0xFFFFFFEF) | 0x11;
491 __raw_writel(temp, (resources->dmmu_base) + 0x10);
493 /* Let the DSP MMU run */
494 hw_mmu_enable(resources->dmmu_base);
496 /* Enable the BIOS clock */
497 (void)dev_get_symbol(dev_context->dev_obj,
498 BRIDGEINIT_BIOSGPTIMER, &ul_bios_gp_timer);
499 (void)dev_get_symbol(dev_context->dev_obj,
500 BRIDGEINIT_LOADMON_GPTIMER,
501 &ul_load_monitor_timer);
505 if (ul_load_monitor_timer != 0xFFFF) {
506 clk_cmd = (BPWR_ENABLE_CLOCK << MBX_PM_CLK_CMDSHIFT) |
507 ul_load_monitor_timer;
508 dsp_peripheral_clk_ctrl(dev_context, &clk_cmd);
510 dev_dbg(bridge, "Not able to get the symbol for Load "
516 if (ul_bios_gp_timer != 0xFFFF) {
517 clk_cmd = (BPWR_ENABLE_CLOCK << MBX_PM_CLK_CMDSHIFT) |
519 dsp_peripheral_clk_ctrl(dev_context, &clk_cmd);
522 "Not able to get the symbol for BIOS Timer\n");
527 /* Set the DSP clock rate */
528 (void)dev_get_symbol(dev_context->dev_obj,
529 "_BRIDGEINIT_DSP_FREQ", &ul_dsp_clk_addr);
530 /*Set Autoidle Mode for IVA2 PLL */
531 (*pdata->dsp_cm_write)(1 << OMAP3430_AUTO_IVA2_DPLL_SHIFT,
532 OMAP3430_IVA2_MOD, OMAP3430_CM_AUTOIDLE_PLL);
534 if ((unsigned int *)ul_dsp_clk_addr != NULL) {
535 /* Get the clock rate */
536 ul_dsp_clk_rate = dsp_clk_get_iva2_rate();
537 dev_dbg(bridge, "%s: DSP clock rate (KHZ): 0x%x \n",
538 __func__, ul_dsp_clk_rate);
539 (void)bridge_brd_write(dev_context,
540 (u8 *) &ul_dsp_clk_rate,
541 ul_dsp_clk_addr, sizeof(u32), 0);
544 * Enable Mailbox events and also drain any pending
547 dev_context->mbox = omap_mbox_get("dsp", &dsp_mbox_notifier);
548 if (IS_ERR(dev_context->mbox)) {
549 dev_context->mbox = NULL;
550 pr_err("%s: Failed to get dsp mailbox handle\n",
557 /*PM_IVA2GRPSEL_PER = 0xC0;*/
558 temp = readl(resources->per_pm_base + 0xA8);
559 temp = (temp & 0xFFFFFF30) | 0xC0;
560 writel(temp, resources->per_pm_base + 0xA8);
562 /*PM_MPUGRPSEL_PER &= 0xFFFFFF3F; */
563 temp = readl(resources->per_pm_base + 0xA4);
564 temp = (temp & 0xFFFFFF3F);
565 writel(temp, resources->per_pm_base + 0xA4);
566 /*CM_SLEEPDEP_PER |= 0x04; */
567 temp = readl(resources->per_base + 0x44);
568 temp = (temp & 0xFFFFFFFB) | 0x04;
569 writel(temp, resources->per_base + 0x44);
571 /*CM_CLKSTCTRL_IVA2 = 0x00000003 -To Allow automatic transitions */
572 (*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_ENABLE_AUTO,
573 OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
576 dev_dbg(bridge, "%s Unreset\n", __func__);
577 /* Enable DSP MMU Interrupts */
578 hw_mmu_event_enable(resources->dmmu_base,
579 HW_MMU_ALL_INTERRUPTS);
580 /* release the RST1, DSP starts executing now .. */
581 (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST1_IVA2_MASK, 0,
582 OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
584 dev_dbg(bridge, "Waiting for Sync @ 0x%x\n", *(u32 *)sync_addr);
585 dev_dbg(bridge, "DSP c_int00 Address = 0x%x\n", dsp_addr);
587 while (__raw_readw(sync_addr))
590 /* Wait for DSP to clear word in shared memory */
591 /* Read the Location */
592 if (!wait_for_start(dev_context, sync_addr))
595 dev_get_symbol(dev_context->dev_obj, "_WDT_enable", &wdt_en);
598 dsp_wdt_sm_set((void *)ul_shm_base);
599 dsp_wdt_enable(true);
602 status = dev_get_io_mgr(dev_context->dev_obj, &hio_mgr);
604 io_sh_msetting(hio_mgr, SHM_OPPINFO, NULL);
605 /* Write the synchronization bit to indicate the
606 * completion of OPP table update to DSP
608 __raw_writel(0XCAFECAFE, sync_addr);
610 /* update board state */
611 dev_context->brd_state = BRD_RUNNING;
612 /* (void)chnlsm_enable_interrupt(dev_context); */
614 dev_context->brd_state = BRD_UNKNOWN;
624 * ======== bridge_brd_stop ========
626 * Puts DSP in self loop.
631 static int bridge_brd_stop(struct bridge_dev_context *dev_ctxt)
634 struct bridge_dev_context *dev_context = dev_ctxt;
635 struct pg_table_attrs *pt_attrs;
637 struct omap_dsp_platform_data *pdata =
638 omap_dspbridge_dev->dev.platform_data;
640 if (dev_context->brd_state == BRD_STOPPED)
643 /* as per TRM, it is advised to first drive the IVA2 to 'Standby' mode,
644 * before turning off the clocks.. This is to ensure that there are no
645 * pending L3 or other transactons from IVA2 */
646 dsp_pwr_state = (*pdata->dsp_prm_read)
647 (OMAP3430_IVA2_MOD, OMAP2_PM_PWSTST) & OMAP_POWERSTATEST_MASK;
648 if (dsp_pwr_state != PWRDM_POWER_OFF) {
649 (*pdata->dsp_prm_rmw_bits)(OMAP3430_RST2_IVA2_MASK, 0,
650 OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
651 sm_interrupt_dsp(dev_context, MBX_PM_DSPIDLE);
654 /* IVA2 is not in OFF state */
655 /* Set PM_PWSTCTRL_IVA2 to OFF */
656 (*pdata->dsp_prm_rmw_bits)(OMAP_POWERSTATEST_MASK,
657 PWRDM_POWER_OFF, OMAP3430_IVA2_MOD, OMAP2_PM_PWSTCTRL);
658 /* Set the SW supervised state transition for Sleep */
659 (*pdata->dsp_cm_write)(OMAP34XX_CLKSTCTRL_FORCE_SLEEP,
660 OMAP3430_IVA2_MOD, OMAP2_CM_CLKSTCTRL);
663 /* Release the Ext Base virtual Address as the next DSP Program
664 * may have a different load address */
665 if (dev_context->dsp_ext_base_addr)
666 dev_context->dsp_ext_base_addr = 0;
668 dev_context->brd_state = BRD_STOPPED; /* update board state */
670 dsp_wdt_enable(false);
672 /* This is a good place to clear the MMU page tables as well */
673 if (dev_context->pt_attrs) {
674 pt_attrs = dev_context->pt_attrs;
675 memset((u8 *) pt_attrs->l1_base_va, 0x00, pt_attrs->l1_size);
676 memset((u8 *) pt_attrs->l2_base_va, 0x00, pt_attrs->l2_size);
677 memset((u8 *) pt_attrs->pg_info, 0x00,
678 (pt_attrs->l2_num_pages * sizeof(struct page_info)));
680 /* Disable the mailbox interrupts */
681 if (dev_context->mbox) {
682 omap_mbox_disable_irq(dev_context->mbox, IRQ_RX);
683 omap_mbox_put(dev_context->mbox, &dsp_mbox_notifier);
684 dev_context->mbox = NULL;
686 /* Reset IVA2 clocks*/
687 (*pdata->dsp_prm_write)(OMAP3430_RST1_IVA2_MASK |
688 OMAP3430_RST2_IVA2_MASK | OMAP3430_RST3_IVA2_MASK,
689 OMAP3430_IVA2_MOD, OMAP2_RM_RSTCTRL);
691 dsp_clock_disable_all(dev_context->dsp_per_clks);
692 dsp_clk_disable(DSP_CLK_IVA2);
698 * ======== bridge_brd_status ========
699 * Returns the board status.
701 static int bridge_brd_status(struct bridge_dev_context *dev_ctxt,
704 struct bridge_dev_context *dev_context = dev_ctxt;
705 *board_state = dev_context->brd_state;
710 * ======== bridge_brd_write ========
711 * Copies the buffers to DSP internal or external memory.
713 static int bridge_brd_write(struct bridge_dev_context *dev_ctxt,
714 u8 *host_buff, u32 dsp_addr,
715 u32 ul_num_bytes, u32 mem_type)
718 struct bridge_dev_context *dev_context = dev_ctxt;
720 if (dsp_addr < dev_context->dsp_start_add) {
724 if ((dsp_addr - dev_context->dsp_start_add) <
725 dev_context->internal_size) {
726 status = write_dsp_data(dev_ctxt, host_buff, dsp_addr,
727 ul_num_bytes, mem_type);
729 status = write_ext_dsp_data(dev_context, host_buff, dsp_addr,
730 ul_num_bytes, mem_type, false);
737 * ======== bridge_dev_create ========
738 * Creates a driver object. Puts DSP in self loop.
740 static int bridge_dev_create(struct bridge_dev_context
742 struct dev_object *hdev_obj,
743 struct cfg_hostres *config_param)
746 struct bridge_dev_context *dev_context = NULL;
748 struct cfg_hostres *resources = config_param;
749 struct pg_table_attrs *pt_attrs;
753 struct drv_data *drv_datap = dev_get_drvdata(bridge);
755 /* Allocate and initialize a data structure to contain the bridge driver
756 * state, which becomes the context for later calls into this driver */
757 dev_context = kzalloc(sizeof(struct bridge_dev_context), GFP_KERNEL);
763 dev_context->dsp_start_add = (u32) OMAP_GEM_BASE;
764 dev_context->self_loop = (u32) NULL;
765 dev_context->dsp_per_clks = 0;
766 dev_context->internal_size = OMAP_DSP_SIZE;
767 /* Clear dev context MMU table entries.
768 * These get set on bridge_io_on_loaded() call after program loaded. */
769 for (entry_ndx = 0; entry_ndx < BRDIOCTL_NUMOFMMUTLB; entry_ndx++) {
770 dev_context->atlb_entry[entry_ndx].gpp_pa =
771 dev_context->atlb_entry[entry_ndx].dsp_va = 0;
773 dev_context->dsp_base_addr = (u32) MEM_LINEAR_ADDRESS((void *)
780 if (!dev_context->dsp_base_addr)
783 pt_attrs = kzalloc(sizeof(struct pg_table_attrs), GFP_KERNEL);
784 if (pt_attrs != NULL) {
785 pt_attrs->l1_size = SZ_16K; /* 4096 entries of 32 bits */
786 align_size = pt_attrs->l1_size;
787 /* Align sizes are expected to be power of 2 */
788 /* we like to get aligned on L1 table size */
789 pg_tbl_va = (u32) mem_alloc_phys_mem(pt_attrs->l1_size,
790 align_size, &pg_tbl_pa);
792 /* Check if the PA is aligned for us */
793 if ((pg_tbl_pa) & (align_size - 1)) {
794 /* PA not aligned to page table size ,
795 * try with more allocation and align */
796 mem_free_phys_mem((void *)pg_tbl_va, pg_tbl_pa,
798 /* we like to get aligned on L1 table size */
800 (u32) mem_alloc_phys_mem((pt_attrs->l1_size) * 2,
801 align_size, &pg_tbl_pa);
802 /* We should be able to get aligned table now */
803 pt_attrs->l1_tbl_alloc_pa = pg_tbl_pa;
804 pt_attrs->l1_tbl_alloc_va = pg_tbl_va;
805 pt_attrs->l1_tbl_alloc_sz = pt_attrs->l1_size * 2;
806 /* Align the PA to the next 'align' boundary */
807 pt_attrs->l1_base_pa =
809 (align_size - 1)) & (~(align_size - 1));
810 pt_attrs->l1_base_va =
811 pg_tbl_va + (pt_attrs->l1_base_pa - pg_tbl_pa);
813 /* We got aligned PA, cool */
814 pt_attrs->l1_tbl_alloc_pa = pg_tbl_pa;
815 pt_attrs->l1_tbl_alloc_va = pg_tbl_va;
816 pt_attrs->l1_tbl_alloc_sz = pt_attrs->l1_size;
817 pt_attrs->l1_base_pa = pg_tbl_pa;
818 pt_attrs->l1_base_va = pg_tbl_va;
820 if (pt_attrs->l1_base_va)
821 memset((u8 *) pt_attrs->l1_base_va, 0x00,
824 /* number of L2 page tables = DMM pool used + SHMMEM +EXTMEM +
826 pt_attrs->l2_num_pages = ((DMMPOOLSIZE >> 20) + 6);
827 pt_attrs->l2_size = HW_MMU_COARSE_PAGE_SIZE *
828 pt_attrs->l2_num_pages;
829 align_size = 4; /* Make it u32 aligned */
830 /* we like to get aligned on L1 table size */
831 pg_tbl_va = (u32) mem_alloc_phys_mem(pt_attrs->l2_size,
832 align_size, &pg_tbl_pa);
833 pt_attrs->l2_tbl_alloc_pa = pg_tbl_pa;
834 pt_attrs->l2_tbl_alloc_va = pg_tbl_va;
835 pt_attrs->l2_tbl_alloc_sz = pt_attrs->l2_size;
836 pt_attrs->l2_base_pa = pg_tbl_pa;
837 pt_attrs->l2_base_va = pg_tbl_va;
839 if (pt_attrs->l2_base_va)
840 memset((u8 *) pt_attrs->l2_base_va, 0x00,
843 pt_attrs->pg_info = kzalloc(pt_attrs->l2_num_pages *
844 sizeof(struct page_info), GFP_KERNEL);
846 "L1 pa %x, va %x, size %x\n L2 pa %x, va "
847 "%x, size %x\n", pt_attrs->l1_base_pa,
848 pt_attrs->l1_base_va, pt_attrs->l1_size,
849 pt_attrs->l2_base_pa, pt_attrs->l2_base_va,
851 dev_dbg(bridge, "pt_attrs %p L2 NumPages %x pg_info %p\n",
852 pt_attrs, pt_attrs->l2_num_pages, pt_attrs->pg_info);
854 if ((pt_attrs != NULL) && (pt_attrs->l1_base_va != 0) &&
855 (pt_attrs->l2_base_va != 0) && (pt_attrs->pg_info != NULL))
856 dev_context->pt_attrs = pt_attrs;
861 spin_lock_init(&pt_attrs->pg_lock);
862 dev_context->tc_word_swap_on = drv_datap->tc_wordswapon;
864 /* Set the Clock Divisor for the DSP module */
866 /* MMU address is obtained from the host
867 * resources struct */
868 dev_context->dsp_mmu_base = resources->dmmu_base;
871 dev_context->dev_obj = hdev_obj;
872 /* Store current board state. */
873 dev_context->brd_state = BRD_UNKNOWN;
874 dev_context->resources = resources;
875 dsp_clk_enable(DSP_CLK_IVA2);
876 bridge_brd_stop(dev_context);
877 /* Return ptr to our device state to the DSP API for storage */
878 *dev_cntxt = dev_context;
880 if (pt_attrs != NULL) {
881 kfree(pt_attrs->pg_info);
883 if (pt_attrs->l2_tbl_alloc_va) {
884 mem_free_phys_mem((void *)
885 pt_attrs->l2_tbl_alloc_va,
886 pt_attrs->l2_tbl_alloc_pa,
887 pt_attrs->l2_tbl_alloc_sz);
889 if (pt_attrs->l1_tbl_alloc_va) {
890 mem_free_phys_mem((void *)
891 pt_attrs->l1_tbl_alloc_va,
892 pt_attrs->l1_tbl_alloc_pa,
893 pt_attrs->l1_tbl_alloc_sz);
904 * ======== bridge_dev_ctrl ========
905 * Receives device specific commands.
907 static int bridge_dev_ctrl(struct bridge_dev_context *dev_context,
908 u32 dw_cmd, void *pargs)
911 struct bridge_ioctl_extproc *pa_ext_proc =
912 (struct bridge_ioctl_extproc *)pargs;
916 case BRDIOCTL_CHNLREAD:
918 case BRDIOCTL_CHNLWRITE:
920 case BRDIOCTL_SETMMUCONFIG:
921 /* store away dsp-mmu setup values for later use */
922 for (ndx = 0; ndx < BRDIOCTL_NUMOFMMUTLB; ndx++, pa_ext_proc++)
923 dev_context->atlb_entry[ndx] = *pa_ext_proc;
925 case BRDIOCTL_DEEPSLEEP:
926 case BRDIOCTL_EMERGENCYSLEEP:
927 /* Currently only DSP Idle is supported Need to update for
929 status = sleep_dsp(dev_context, PWR_DEEPSLEEP, pargs);
931 case BRDIOCTL_WAKEUP:
932 status = wake_dsp(dev_context, pargs);
934 case BRDIOCTL_CLK_CTRL:
936 /* Looking For Baseport Fix for Clocks */
937 status = dsp_peripheral_clk_ctrl(dev_context, pargs);
939 case BRDIOCTL_PWR_HIBERNATE:
940 status = handle_hibernation_from_dsp(dev_context);
942 case BRDIOCTL_PRESCALE_NOTIFY:
943 status = pre_scale_dsp(dev_context, pargs);
945 case BRDIOCTL_POSTSCALE_NOTIFY:
946 status = post_scale_dsp(dev_context, pargs);
948 case BRDIOCTL_CONSTRAINT_REQUEST:
949 status = handle_constraints_set(dev_context, pargs);
959 * ======== bridge_dev_destroy ========
960 * Destroys the driver object.
962 static int bridge_dev_destroy(struct bridge_dev_context *dev_ctxt)
964 struct pg_table_attrs *pt_attrs;
966 struct bridge_dev_context *dev_context = (struct bridge_dev_context *)
968 struct cfg_hostres *host_res;
970 struct drv_data *drv_datap = dev_get_drvdata(bridge);
972 /* It should never happen */
976 /* first put the device to stop state */
977 bridge_brd_stop(dev_context);
978 if (dev_context->pt_attrs) {
979 pt_attrs = dev_context->pt_attrs;
980 kfree(pt_attrs->pg_info);
982 if (pt_attrs->l2_tbl_alloc_va) {
983 mem_free_phys_mem((void *)pt_attrs->l2_tbl_alloc_va,
984 pt_attrs->l2_tbl_alloc_pa,
985 pt_attrs->l2_tbl_alloc_sz);
987 if (pt_attrs->l1_tbl_alloc_va) {
988 mem_free_phys_mem((void *)pt_attrs->l1_tbl_alloc_va,
989 pt_attrs->l1_tbl_alloc_pa,
990 pt_attrs->l1_tbl_alloc_sz);
996 if (dev_context->resources) {
997 host_res = dev_context->resources;
998 shm_size = drv_datap->shm_size;
999 if (shm_size >= 0x10000) {
1000 if ((host_res->mem_base[1]) &&
1001 (host_res->mem_phys[1])) {
1002 mem_free_phys_mem((void *)
1009 dev_dbg(bridge, "%s: Error getting shm size "
1010 "from registry: %x. Not calling "
1011 "mem_free_phys_mem\n", __func__,
1014 host_res->mem_base[1] = 0;
1015 host_res->mem_phys[1] = 0;
1017 if (host_res->mem_base[0])
1018 iounmap((void *)host_res->mem_base[0]);
1019 if (host_res->mem_base[2])
1020 iounmap((void *)host_res->mem_base[2]);
1021 if (host_res->mem_base[3])
1022 iounmap((void *)host_res->mem_base[3]);
1023 if (host_res->mem_base[4])
1024 iounmap((void *)host_res->mem_base[4]);
1025 if (host_res->dmmu_base)
1026 iounmap(host_res->dmmu_base);
1027 if (host_res->per_base)
1028 iounmap(host_res->per_base);
1029 if (host_res->per_pm_base)
1030 iounmap((void *)host_res->per_pm_base);
1031 if (host_res->core_pm_base)
1032 iounmap((void *)host_res->core_pm_base);
1034 host_res->mem_base[0] = (u32) NULL;
1035 host_res->mem_base[2] = (u32) NULL;
1036 host_res->mem_base[3] = (u32) NULL;
1037 host_res->mem_base[4] = (u32) NULL;
1038 host_res->dmmu_base = NULL;
1043 /* Free the driver's device context: */
1044 kfree(drv_datap->base_img);
1045 kfree((void *)dev_ctxt);
1049 static int bridge_brd_mem_copy(struct bridge_dev_context *dev_ctxt,
1050 u32 dsp_dest_addr, u32 dsp_src_addr,
1051 u32 ul_num_bytes, u32 mem_type)
1054 u32 src_addr = dsp_src_addr;
1055 u32 dest_addr = dsp_dest_addr;
1057 u32 total_bytes = ul_num_bytes;
1058 u8 host_buf[BUFFERSIZE];
1059 struct bridge_dev_context *dev_context = dev_ctxt;
1060 while (total_bytes > 0 && !status) {
1062 total_bytes > BUFFERSIZE ? BUFFERSIZE : total_bytes;
1063 /* Read from External memory */
1064 status = read_ext_dsp_data(dev_ctxt, host_buf, src_addr,
1065 copy_bytes, mem_type);
1067 if (dest_addr < (dev_context->dsp_start_add +
1068 dev_context->internal_size)) {
1069 /* Write to Internal memory */
1070 status = write_dsp_data(dev_ctxt, host_buf,
1071 dest_addr, copy_bytes,
1074 /* Write to External memory */
1076 write_ext_dsp_data(dev_ctxt, host_buf,
1077 dest_addr, copy_bytes,
1081 total_bytes -= copy_bytes;
1082 src_addr += copy_bytes;
1083 dest_addr += copy_bytes;
1088 /* Mem Write does not halt the DSP to write unlike bridge_brd_write */
1089 static int bridge_brd_mem_write(struct bridge_dev_context *dev_ctxt,
1090 u8 *host_buff, u32 dsp_addr,
1091 u32 ul_num_bytes, u32 mem_type)
1094 struct bridge_dev_context *dev_context = dev_ctxt;
1095 u32 ul_remain_bytes = 0;
1097 ul_remain_bytes = ul_num_bytes;
1098 while (ul_remain_bytes > 0 && !status) {
1100 ul_remain_bytes > BUFFERSIZE ? BUFFERSIZE : ul_remain_bytes;
1101 if (dsp_addr < (dev_context->dsp_start_add +
1102 dev_context->internal_size)) {
1104 write_dsp_data(dev_ctxt, host_buff, dsp_addr,
1105 ul_bytes, mem_type);
1107 status = write_ext_dsp_data(dev_ctxt, host_buff,
1111 ul_remain_bytes -= ul_bytes;
1112 dsp_addr += ul_bytes;
1113 host_buff = host_buff + ul_bytes;
1119 * ======== bridge_brd_mem_map ========
1120 * This function maps MPU buffer to the DSP address space. It performs
1121 * linear to physical address translation if required. It translates each
1122 * page since linear addresses can be physically non-contiguous
1123 * All address & size arguments are assumed to be page aligned (in proc.c)
1125 * TODO: Disable MMU while updating the page tables (but that'll stall DSP)
1127 static int bridge_brd_mem_map(struct bridge_dev_context *dev_ctxt,
1128 u32 ul_mpu_addr, u32 virt_addr,
1129 u32 ul_num_bytes, u32 ul_map_attr,
1130 struct page **mapped_pages)
1134 struct bridge_dev_context *dev_context = dev_ctxt;
1135 struct hw_mmu_map_attrs_t hw_attrs;
1136 struct vm_area_struct *vma;
1137 struct mm_struct *mm = current->mm;
1139 u32 num_usr_pgs = 0;
1140 struct page *mapped_page, *pg;
1143 struct task_struct *curr_task = current;
1148 "%s hDevCtxt %p, pa %x, va %x, size %x, ul_map_attr %x\n",
1149 __func__, dev_ctxt, ul_mpu_addr, virt_addr, ul_num_bytes,
1151 if (ul_num_bytes == 0)
1154 if (ul_map_attr & DSP_MAP_DIR_MASK) {
1155 attrs = ul_map_attr;
1157 /* Assign default attributes */
1158 attrs = ul_map_attr | (DSP_MAPVIRTUALADDR | DSP_MAPELEMSIZE16);
1160 /* Take mapping properties */
1161 if (attrs & DSP_MAPBIGENDIAN)
1162 hw_attrs.endianism = HW_BIG_ENDIAN;
1164 hw_attrs.endianism = HW_LITTLE_ENDIAN;
1166 hw_attrs.mixed_size = (enum hw_mmu_mixed_size_t)
1167 ((attrs & DSP_MAPMIXEDELEMSIZE) >> 2);
1168 /* Ignore element_size if mixed_size is enabled */
1169 if (hw_attrs.mixed_size == 0) {
1170 if (attrs & DSP_MAPELEMSIZE8) {
1172 hw_attrs.element_size = HW_ELEM_SIZE8BIT;
1173 } else if (attrs & DSP_MAPELEMSIZE16) {
1174 /* Size is 16 bit */
1175 hw_attrs.element_size = HW_ELEM_SIZE16BIT;
1176 } else if (attrs & DSP_MAPELEMSIZE32) {
1177 /* Size is 32 bit */
1178 hw_attrs.element_size = HW_ELEM_SIZE32BIT;
1179 } else if (attrs & DSP_MAPELEMSIZE64) {
1180 /* Size is 64 bit */
1181 hw_attrs.element_size = HW_ELEM_SIZE64BIT;
1184 * Mixedsize isn't enabled, so size can't be
1190 if (attrs & DSP_MAPDONOTLOCK)
1191 hw_attrs.donotlockmpupage = 1;
1193 hw_attrs.donotlockmpupage = 0;
1195 if (attrs & DSP_MAPVMALLOCADDR) {
1196 return mem_map_vmalloc(dev_ctxt, ul_mpu_addr, virt_addr,
1197 ul_num_bytes, &hw_attrs);
1200 * Do OS-specific user-va to pa translation.
1201 * Combine physically contiguous regions to reduce TLBs.
1202 * Pass the translated pa to pte_update.
1204 if ((attrs & DSP_MAPPHYSICALADDR)) {
1205 status = pte_update(dev_context, ul_mpu_addr, virt_addr,
1206 ul_num_bytes, &hw_attrs);
1211 * Important Note: ul_mpu_addr is mapped from user application process
1212 * to current process - it must lie completely within the current
1213 * virtual memory address space in order to be of use to us here!
1215 down_read(&mm->mmap_sem);
1216 vma = find_vma(mm, ul_mpu_addr);
1219 "VMAfor UserBuf: ul_mpu_addr=%x, ul_num_bytes=%x, "
1220 "vm_start=%lx, vm_end=%lx, vm_flags=%lx\n", ul_mpu_addr,
1221 ul_num_bytes, vma->vm_start, vma->vm_end,
1225 * It is observed that under some circumstances, the user buffer is
1226 * spread across several VMAs. So loop through and check if the entire
1227 * user buffer is covered
1229 while ((vma) && (ul_mpu_addr + ul_num_bytes > vma->vm_end)) {
1230 /* jump to the next VMA region */
1231 vma = find_vma(mm, vma->vm_end + 1);
1233 "VMA for UserBuf ul_mpu_addr=%x ul_num_bytes=%x, "
1234 "vm_start=%lx, vm_end=%lx, vm_flags=%lx\n", ul_mpu_addr,
1235 ul_num_bytes, vma->vm_start, vma->vm_end,
1239 pr_err("%s: Failed to get VMA region for 0x%x (%d)\n",
1240 __func__, ul_mpu_addr, ul_num_bytes);
1242 up_read(&mm->mmap_sem);
1246 if (vma->vm_flags & VM_IO) {
1247 num_usr_pgs = ul_num_bytes / PG_SIZE4K;
1248 mpu_addr = ul_mpu_addr;
1250 /* Get the physical addresses for user buffer */
1251 for (pg_i = 0; pg_i < num_usr_pgs; pg_i++) {
1252 pa = user_va2_pa(mm, mpu_addr);
1255 pr_err("DSPBRIDGE: VM_IO mapping physical"
1256 "address is invalid\n");
1259 if (pfn_valid(__phys_to_pfn(pa))) {
1260 pg = PHYS_TO_PAGE(pa);
1262 if (page_count(pg) < 1) {
1263 pr_err("Bad page in VM_IO buffer\n");
1264 bad_page_dump(pa, pg);
1267 status = pte_set(dev_context->pt_attrs, pa,
1268 va, HW_PAGE_SIZE4KB, &hw_attrs);
1272 va += HW_PAGE_SIZE4KB;
1273 mpu_addr += HW_PAGE_SIZE4KB;
1274 pa += HW_PAGE_SIZE4KB;
1277 num_usr_pgs = ul_num_bytes / PG_SIZE4K;
1278 if (vma->vm_flags & (VM_WRITE | VM_MAYWRITE))
1281 for (pg_i = 0; pg_i < num_usr_pgs; pg_i++) {
1282 pg_num = get_user_pages(curr_task, mm, ul_mpu_addr, 1,
1283 write, 1, &mapped_page, NULL);
1285 if (page_count(mapped_page) < 1) {
1286 pr_err("Bad page count after doing"
1289 bad_page_dump(page_to_phys(mapped_page),
1292 status = pte_set(dev_context->pt_attrs,
1293 page_to_phys(mapped_page), va,
1294 HW_PAGE_SIZE4KB, &hw_attrs);
1299 mapped_pages[pg_i] = mapped_page;
1301 va += HW_PAGE_SIZE4KB;
1302 ul_mpu_addr += HW_PAGE_SIZE4KB;
1304 pr_err("DSPBRIDGE: get_user_pages FAILED,"
1306 "vma->vm_flags = 0x%lx,"
1307 "get_user_pages Err"
1308 "Value = %d, Buffer"
1309 "size=0x%x\n", ul_mpu_addr,
1310 vma->vm_flags, pg_num, ul_num_bytes);
1316 up_read(&mm->mmap_sem);
1320 * Roll out the mapped pages incase it failed in middle of
1324 bridge_brd_mem_un_map(dev_context, virt_addr,
1325 (pg_i * PG_SIZE4K));
1330 * In any case, flush the TLB
1331 * This is called from here instead from pte_update to avoid unnecessary
1332 * repetition while mapping non-contiguous physical regions of a virtual
1335 flush_all(dev_context);
1336 dev_dbg(bridge, "%s status %x\n", __func__, status);
1341 * ======== bridge_brd_mem_un_map ========
1342 * Invalidate the PTEs for the DSP VA block to be unmapped.
1344 * PTEs of a mapped memory block are contiguous in any page table
1345 * So, instead of looking up the PTE address for every 4K block,
1346 * we clear consecutive PTEs until we unmap all the bytes
1348 static int bridge_brd_mem_un_map(struct bridge_dev_context *dev_ctxt,
1349 u32 virt_addr, u32 ul_num_bytes)
1359 u32 pte_addr_l2 = 0;
1363 struct page *pg = NULL;
1365 struct bridge_dev_context *dev_context = dev_ctxt;
1366 struct pg_table_attrs *pt = dev_context->pt_attrs;
1369 u32 numof4k_pages = 0;
1371 va_curr = virt_addr;
1372 rem_bytes = ul_num_bytes;
1374 l1_base_va = pt->l1_base_va;
1375 pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va_curr);
1376 dev_dbg(bridge, "%s dev_ctxt %p, va %x, NumBytes %x l1_base_va %x, "
1377 "pte_addr_l1 %x\n", __func__, dev_ctxt, virt_addr,
1378 ul_num_bytes, l1_base_va, pte_addr_l1);
1380 while (rem_bytes && !status) {
1381 u32 va_curr_orig = va_curr;
1382 /* Find whether the L1 PTE points to a valid L2 PT */
1383 pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va_curr);
1384 pte_val = *(u32 *) pte_addr_l1;
1385 pte_size = hw_mmu_pte_size_l1(pte_val);
1387 if (pte_size != HW_MMU_COARSE_PAGE_SIZE)
1388 goto skip_coarse_page;
1391 * Get the L2 PA from the L1 PTE, and find
1392 * corresponding L2 VA
1394 l2_base_pa = hw_mmu_pte_coarse_l1(pte_val);
1395 l2_base_va = l2_base_pa - pt->l2_base_pa + pt->l2_base_va;
1397 (l2_base_pa - pt->l2_base_pa) / HW_MMU_COARSE_PAGE_SIZE;
1399 * Find the L2 PTE address from which we will start
1400 * clearing, the number of PTEs to be cleared on this
1401 * page, and the size of VA space that needs to be
1402 * cleared on this L2 page
1404 pte_addr_l2 = hw_mmu_pte_addr_l2(l2_base_va, va_curr);
1405 pte_count = pte_addr_l2 & (HW_MMU_COARSE_PAGE_SIZE - 1);
1406 pte_count = (HW_MMU_COARSE_PAGE_SIZE - pte_count) / sizeof(u32);
1407 if (rem_bytes < (pte_count * PG_SIZE4K))
1408 pte_count = rem_bytes / PG_SIZE4K;
1409 rem_bytes_l2 = pte_count * PG_SIZE4K;
1412 * Unmap the VA space on this L2 PT. A quicker way
1413 * would be to clear pte_count entries starting from
1414 * pte_addr_l2. However, below code checks that we don't
1415 * clear invalid entries or less than 64KB for a 64KB
1416 * entry. Similar checking is done for L1 PTEs too
1419 while (rem_bytes_l2 && !status) {
1420 pte_val = *(u32 *) pte_addr_l2;
1421 pte_size = hw_mmu_pte_size_l2(pte_val);
1422 /* va_curr aligned to pte_size? */
1423 if (pte_size == 0 || rem_bytes_l2 < pte_size ||
1424 va_curr & (pte_size - 1)) {
1429 /* Collect Physical addresses from VA */
1430 paddr = (pte_val & ~(pte_size - 1));
1431 if (pte_size == HW_PAGE_SIZE64KB)
1436 while (temp++ < numof4k_pages) {
1437 if (!pfn_valid(__phys_to_pfn(paddr))) {
1438 paddr += HW_PAGE_SIZE4KB;
1441 pg = PHYS_TO_PAGE(paddr);
1442 if (page_count(pg) < 1) {
1443 pr_info("DSPBRIDGE: UNMAP function: "
1444 "COUNT 0 FOR PA 0x%x, size = "
1445 "0x%x\n", paddr, ul_num_bytes);
1446 bad_page_dump(paddr, pg);
1449 page_cache_release(pg);
1451 paddr += HW_PAGE_SIZE4KB;
1453 if (hw_mmu_pte_clear(pte_addr_l2, va_curr, pte_size)) {
1459 rem_bytes_l2 -= pte_size;
1460 va_curr += pte_size;
1461 pte_addr_l2 += (pte_size >> 12) * sizeof(u32);
1463 spin_lock(&pt->pg_lock);
1464 if (rem_bytes_l2 == 0) {
1465 pt->pg_info[l2_page_num].num_entries -= pte_count;
1466 if (pt->pg_info[l2_page_num].num_entries == 0) {
1468 * Clear the L1 PTE pointing to the L2 PT
1470 if (!hw_mmu_pte_clear(l1_base_va, va_curr_orig,
1471 HW_MMU_COARSE_PAGE_SIZE))
1475 spin_unlock(&pt->pg_lock);
1479 rem_bytes -= pte_count * PG_SIZE4K;
1483 spin_unlock(&pt->pg_lock);
1486 /* va_curr aligned to pte_size? */
1487 /* pte_size = 1 MB or 16 MB */
1488 if (pte_size == 0 || rem_bytes < pte_size ||
1489 va_curr & (pte_size - 1)) {
1494 if (pte_size == HW_PAGE_SIZE1MB)
1495 numof4k_pages = 256;
1497 numof4k_pages = 4096;
1499 /* Collect Physical addresses from VA */
1500 paddr = (pte_val & ~(pte_size - 1));
1501 while (temp++ < numof4k_pages) {
1502 if (pfn_valid(__phys_to_pfn(paddr))) {
1503 pg = PHYS_TO_PAGE(paddr);
1504 if (page_count(pg) < 1) {
1505 pr_info("DSPBRIDGE: UNMAP function: "
1506 "COUNT 0 FOR PA 0x%x, size = "
1507 "0x%x\n", paddr, ul_num_bytes);
1508 bad_page_dump(paddr, pg);
1511 page_cache_release(pg);
1514 paddr += HW_PAGE_SIZE4KB;
1516 if (!hw_mmu_pte_clear(l1_base_va, va_curr, pte_size)) {
1518 rem_bytes -= pte_size;
1519 va_curr += pte_size;
1526 * It is better to flush the TLB here, so that any stale old entries
1530 flush_all(dev_context);
1532 "%s: va_curr %x, pte_addr_l1 %x pte_addr_l2 %x rem_bytes %x,"
1533 " rem_bytes_l2 %x status %x\n", __func__, va_curr, pte_addr_l1,
1534 pte_addr_l2, rem_bytes, rem_bytes_l2, status);
1539 * ======== user_va2_pa ========
1541 * This function walks through the page tables to convert a userland
1542 * virtual address to physical address
1544 static u32 user_va2_pa(struct mm_struct *mm, u32 address)
1551 pgd = pgd_offset(mm, address);
1552 if (pgd_none(*pgd) || pgd_bad(*pgd))
1555 pud = pud_offset(pgd, address);
1556 if (pud_none(*pud) || pud_bad(*pud))
1559 pmd = pmd_offset(pud, address);
1560 if (pmd_none(*pmd) || pmd_bad(*pmd))
1563 ptep = pte_offset_map(pmd, address);
1566 if (pte_present(pte))
1567 return pte & PAGE_MASK;
1574 * ======== pte_update ========
1575 * This function calculates the optimum page-aligned addresses and sizes
1576 * Caller must pass page-aligned values
1578 static int pte_update(struct bridge_dev_context *dev_ctxt, u32 pa,
1580 struct hw_mmu_map_attrs_t *map_attrs)
1586 u32 num_bytes = size;
1587 struct bridge_dev_context *dev_context = dev_ctxt;
1589 u32 page_size[] = { HW_PAGE_SIZE16MB, HW_PAGE_SIZE1MB,
1590 HW_PAGE_SIZE64KB, HW_PAGE_SIZE4KB
1593 while (num_bytes && !status) {
1594 /* To find the max. page size with which both PA & VA are
1596 all_bits = pa_curr | va_curr;
1598 for (i = 0; i < 4; i++) {
1599 if ((num_bytes >= page_size[i]) && ((all_bits &
1603 pte_set(dev_context->pt_attrs, pa_curr,
1604 va_curr, page_size[i], map_attrs);
1605 pa_curr += page_size[i];
1606 va_curr += page_size[i];
1607 num_bytes -= page_size[i];
1608 /* Don't try smaller sizes. Hopefully we have
1609 * reached an address aligned to a bigger page
1620 * ======== pte_set ========
1621 * This function calculates PTE address (MPU virtual) to be updated
1622 * It also manages the L2 page tables
1624 static int pte_set(struct pg_table_attrs *pt, u32 pa, u32 va,
1625 u32 size, struct hw_mmu_map_attrs_t *attrs)
1631 /* Base address of the PT that will be updated */
1634 /* Compiler warns that the next three variables might be used
1635 * uninitialized in this function. Doesn't seem so. Working around,
1639 u32 l2_page_num = 0;
1642 l1_base_va = pt->l1_base_va;
1643 pg_tbl_va = l1_base_va;
1644 if ((size == HW_PAGE_SIZE64KB) || (size == HW_PAGE_SIZE4KB)) {
1645 /* Find whether the L1 PTE points to a valid L2 PT */
1646 pte_addr_l1 = hw_mmu_pte_addr_l1(l1_base_va, va);
1647 if (pte_addr_l1 <= (pt->l1_base_va + pt->l1_size)) {
1648 pte_val = *(u32 *) pte_addr_l1;
1649 pte_size = hw_mmu_pte_size_l1(pte_val);
1653 spin_lock(&pt->pg_lock);
1654 if (pte_size == HW_MMU_COARSE_PAGE_SIZE) {
1655 /* Get the L2 PA from the L1 PTE, and find
1656 * corresponding L2 VA */
1657 l2_base_pa = hw_mmu_pte_coarse_l1(pte_val);
1659 l2_base_pa - pt->l2_base_pa + pt->l2_base_va;
1662 pt->l2_base_pa) / HW_MMU_COARSE_PAGE_SIZE;
1663 } else if (pte_size == 0) {
1664 /* L1 PTE is invalid. Allocate a L2 PT and
1665 * point the L1 PTE to it */
1666 /* Find a free L2 PT. */
1667 for (i = 0; (i < pt->l2_num_pages) &&
1668 (pt->pg_info[i].num_entries != 0); i++)
1670 if (i < pt->l2_num_pages) {
1672 l2_base_pa = pt->l2_base_pa + (l2_page_num *
1673 HW_MMU_COARSE_PAGE_SIZE);
1674 l2_base_va = pt->l2_base_va + (l2_page_num *
1675 HW_MMU_COARSE_PAGE_SIZE);
1676 /* Endianness attributes are ignored for
1677 * HW_MMU_COARSE_PAGE_SIZE */
1679 hw_mmu_pte_set(l1_base_va, l2_base_pa, va,
1680 HW_MMU_COARSE_PAGE_SIZE,
1686 /* Found valid L1 PTE of another size.
1687 * Should not overwrite it. */
1691 pg_tbl_va = l2_base_va;
1692 if (size == HW_PAGE_SIZE64KB)
1693 pt->pg_info[l2_page_num].num_entries += 16;
1695 pt->pg_info[l2_page_num].num_entries++;
1696 dev_dbg(bridge, "PTE: L2 BaseVa %x, BasePa %x, PageNum "
1697 "%x, num_entries %x\n", l2_base_va,
1698 l2_base_pa, l2_page_num,
1699 pt->pg_info[l2_page_num].num_entries);
1701 spin_unlock(&pt->pg_lock);
1704 dev_dbg(bridge, "PTE: pg_tbl_va %x, pa %x, va %x, size %x\n",
1705 pg_tbl_va, pa, va, size);
1706 dev_dbg(bridge, "PTE: endianism %x, element_size %x, "
1707 "mixed_size %x\n", attrs->endianism,
1708 attrs->element_size, attrs->mixed_size);
1709 status = hw_mmu_pte_set(pg_tbl_va, pa, va, size, attrs);
1715 /* Memory map kernel VA -- memory allocated with vmalloc */
1716 static int mem_map_vmalloc(struct bridge_dev_context *dev_context,
1717 u32 ul_mpu_addr, u32 virt_addr,
1719 struct hw_mmu_map_attrs_t *hw_attrs)
1722 struct page *page[1];
1734 * Do Kernel va to pa translation.
1735 * Combine physically contiguous regions to reduce TLBs.
1736 * Pass the translated pa to pte_update.
1738 num_pages = ul_num_bytes / PAGE_SIZE; /* PAGE_SIZE = OS page size */
1740 va_curr = ul_mpu_addr;
1741 page[0] = vmalloc_to_page((void *)va_curr);
1742 pa_next = page_to_phys(page[0]);
1743 while (!status && (i < num_pages)) {
1745 * Reuse pa_next from the previous iteration to avoid
1746 * an extra va2pa call
1749 size_curr = PAGE_SIZE;
1751 * If the next page is physically contiguous,
1752 * map it with the current one by increasing
1753 * the size of the region to be mapped
1755 while (++i < num_pages) {
1757 vmalloc_to_page((void *)(va_curr + size_curr));
1758 pa_next = page_to_phys(page[0]);
1760 if (pa_next == (pa_curr + size_curr))
1761 size_curr += PAGE_SIZE;
1771 num_of4k_pages = size_curr / HW_PAGE_SIZE4KB;
1772 while (temp++ < num_of4k_pages) {
1773 get_page(PHYS_TO_PAGE(pa));
1774 pa += HW_PAGE_SIZE4KB;
1776 status = pte_update(dev_context, pa_curr, virt_addr +
1777 (va_curr - ul_mpu_addr), size_curr,
1779 va_curr += size_curr;
1782 * In any case, flush the TLB
1783 * This is called from here instead from pte_update to avoid unnecessary
1784 * repetition while mapping non-contiguous physical regions of a virtual
1787 flush_all(dev_context);
1788 dev_dbg(bridge, "%s status %x\n", __func__, status);
1793 * ======== wait_for_start ========
1794 * Wait for the singal from DSP that it has started, or time out.
1796 bool wait_for_start(struct bridge_dev_context *dev_context,
1797 void __iomem *sync_addr)
1799 u16 timeout = TIHELEN_ACKTIMEOUT;
1801 /* Wait for response from board */
1802 while (__raw_readw(sync_addr) && --timeout)
1805 /* If timed out: return false */
1807 pr_err("%s: Timed out waiting DSP to Start\n", __func__);