xref: /illumos-gate/usr/src/uts/sun4/io/px/px_mmu.c (revision 15e1afcd)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 /*
26  * PX mmu initialization and configuration
27  */
28 #include <sys/types.h>
29 #include <sys/kmem.h>
30 #include <sys/async.h>
31 #include <sys/sysmacros.h>
32 #include <sys/sunddi.h>
33 #include <sys/ddi_impldefs.h>
34 #include <sys/vmem.h>
35 #include <sys/machsystm.h>	/* lddphys() */
36 #include <sys/iommutsb.h>
37 #include "px_obj.h"
38 
39 int
40 px_mmu_attach(px_t *px_p)
41 {
42 	dev_info_t		*dip = px_p->px_dip;
43 	px_mmu_t			*mmu_p;
44 	uint32_t		tsb_i = 0;
45 	char			map_name[32];
46 	px_dvma_range_prop_t	*dvma_prop;
47 	int			dvma_prop_len;
48 	uint32_t		cache_size, tsb_entries;
49 
50 	/*
51 	 * Allocate mmu state structure and link it to the
52 	 * px state structure.
53 	 */
54 	mmu_p = kmem_zalloc(sizeof (px_mmu_t), KM_SLEEP);
55 	if (mmu_p == NULL)
56 		return (DDI_FAILURE);
57 
58 	px_p->px_mmu_p = mmu_p;
59 	mmu_p->mmu_px_p = px_p;
60 	mmu_p->mmu_inst = ddi_get_instance(dip);
61 
62 	/*
63 	 * Check for "virtual-dma" property that specifies
64 	 * the DVMA range.
65 	 */
66 	if (ddi_getlongprop(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
67 	    "virtual-dma", (caddr_t)&dvma_prop, &dvma_prop_len) !=
68 	    DDI_PROP_SUCCESS) {
69 
70 		DBG(DBG_ATTACH, dip, "Getting virtual-dma failed\n");
71 
72 		kmem_free(mmu_p, sizeof (px_mmu_t));
73 		px_p->px_mmu_p = NULL;
74 
75 		return (DDI_FAILURE);
76 	}
77 
78 	mmu_p->mmu_dvma_base = dvma_prop->dvma_base;
79 	mmu_p->mmu_dvma_end = dvma_prop->dvma_base +
80 	    (dvma_prop->dvma_len - 1);
81 	tsb_entries = MMU_BTOP(dvma_prop->dvma_len);
82 
83 	kmem_free(dvma_prop, dvma_prop_len);
84 
85 	/*
86 	 * Setup base and bounds for DVMA and bypass mappings.
87 	 */
88 	mmu_p->mmu_dvma_cache_locks =
89 	    kmem_zalloc(px_dvma_page_cache_entries, KM_SLEEP);
90 
91 	mmu_p->dvma_base_pg = MMU_BTOP(mmu_p->mmu_dvma_base);
92 	mmu_p->mmu_dvma_reserve = tsb_entries >> 1;
93 	mmu_p->dvma_end_pg = MMU_BTOP(mmu_p->mmu_dvma_end);
94 
95 	/*
96 	 * Create a virtual memory map for dvma address space.
97 	 * Reserve 'size' bytes of low dvma space for fast track cache.
98 	 */
99 	(void) snprintf(map_name, sizeof (map_name), "%s%d_dvma",
100 	    ddi_driver_name(dip), ddi_get_instance(dip));
101 
102 	cache_size = MMU_PTOB(px_dvma_page_cache_entries *
103 	    px_dvma_page_cache_clustsz);
104 	mmu_p->mmu_dvma_fast_end = mmu_p->mmu_dvma_base +
105 	    cache_size - 1;
106 
107 	mmu_p->mmu_dvma_map = vmem_create(map_name,
108 	    (void *)(mmu_p->mmu_dvma_fast_end + 1),
109 	    MMU_PTOB(tsb_entries) - cache_size, MMU_PAGE_SIZE,
110 	    NULL, NULL, NULL, MMU_PAGE_SIZE, VM_SLEEP);
111 
112 	mutex_init(&mmu_p->dvma_debug_lock, NULL, MUTEX_DRIVER, NULL);
113 
114 	for (tsb_i = 0; tsb_i < tsb_entries; tsb_i++) {
115 		r_addr_t ra = 0;
116 		io_attributes_t attr;
117 		caddr_t va;
118 
119 		if (px_lib_iommu_getmap(px_p->px_dip, PCI_TSBID(0, tsb_i),
120 		    &attr, &ra) != DDI_SUCCESS)
121 			continue;
122 
123 		va = (caddr_t)(MMU_PTOB(mmu_p->dvma_base_pg + tsb_i));
124 
125 		if (va <= (caddr_t)mmu_p->mmu_dvma_fast_end) {
126 			uint32_t cache_i;
127 
128 			/*
129 			 * the va is within the *fast* dvma range; therefore,
130 			 * lock its fast dvma page cache cluster in order to
131 			 * both preserve the TTE and prevent the use of this
132 			 * fast dvma page cache cluster by px_dvma_map_fast().
133 			 * the lock value 0xFF comes from ldstub().
134 			 */
135 			cache_i = tsb_i / px_dvma_page_cache_clustsz;
136 			ASSERT(cache_i < px_dvma_page_cache_entries);
137 			mmu_p->mmu_dvma_cache_locks[cache_i] = 0xFF;
138 		} else {
139 			(void) vmem_xalloc(mmu_p->mmu_dvma_map, MMU_PAGE_SIZE,
140 			    MMU_PAGE_SIZE, 0, 0, va, va + MMU_PAGE_SIZE,
141 			    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
142 		}
143 	}
144 
145 	return (DDI_SUCCESS);
146 }
147 
148 void
149 px_mmu_detach(px_t *px_p)
150 {
151 	px_mmu_t *mmu_p = px_p->px_mmu_p;
152 
153 	(void) px_lib_iommu_detach(px_p);
154 
155 	/*
156 	 * Free the dvma resource map.
157 	 */
158 	vmem_destroy(mmu_p->mmu_dvma_map);
159 
160 	kmem_free(mmu_p->mmu_dvma_cache_locks,
161 	    px_dvma_page_cache_entries);
162 
163 	if (PX_DVMA_DBG_ON(mmu_p))
164 		px_dvma_debug_fini(mmu_p);
165 
166 	mutex_destroy(&mmu_p->dvma_debug_lock);
167 
168 	/*
169 	 * Free the mmu state structure.
170 	 */
171 	kmem_free(mmu_p, sizeof (px_mmu_t));
172 	px_p->px_mmu_p = NULL;
173 }
174 
175 int
176 px_mmu_map_pages(px_mmu_t *mmu_p, ddi_dma_impl_t *mp, px_dvma_addr_t dvma_pg,
177     size_t npages, size_t pfn_index)
178 {
179 	dev_info_t	*dip = mmu_p->mmu_px_p->px_dip;
180 	px_dvma_addr_t	pg_index = MMU_PAGE_INDEX(mmu_p, dvma_pg);
181 	io_attributes_t	attr = PX_GET_MP_TTE(mp->dmai_tte);
182 
183 	ASSERT(npages <= mp->dmai_ndvmapages);
184 	DBG(DBG_MAP_WIN, dip, "px_mmu_map_pages:%x+%x=%x "
185 	    "npages=0x%x pfn_index=0x%x\n", (uint_t)mmu_p->dvma_base_pg,
186 	    (uint_t)pg_index, dvma_pg, (uint_t)npages, (uint_t)pfn_index);
187 
188 	if (px_lib_iommu_map(dip, PCI_TSBID(0, pg_index), npages,
189 	    PX_ADD_ATTR_EXTNS(attr, mp->dmai_bdf), (void *)mp, pfn_index,
190 	    MMU_MAP_PFN) != DDI_SUCCESS) {
191 		DBG(DBG_MAP_WIN, dip, "px_mmu_map_pages: "
192 		    "px_lib_iommu_map failed\n");
193 
194 		return (DDI_FAILURE);
195 	}
196 
197 	if (!PX_MAP_BUFZONE(mp))
198 		goto done;
199 
200 	DBG(DBG_MAP_WIN, dip, "px_mmu_map_pages: redzone pg=%x\n",
201 	    pg_index + npages);
202 
203 	ASSERT(PX_HAS_REDZONE(mp));
204 
205 	if (px_lib_iommu_map(dip, PCI_TSBID(0, pg_index + npages), 1,
206 	    PX_ADD_ATTR_EXTNS(attr, mp->dmai_bdf), (void *)mp,
207 	    pfn_index + npages - 1, MMU_MAP_PFN) != DDI_SUCCESS) {
208 		DBG(DBG_MAP_WIN, dip, "px_mmu_map_pages: mapping "
209 		    "REDZONE page failed\n");
210 
211 		if (px_lib_iommu_demap(dip, PCI_TSBID(0, pg_index), npages)
212 		    != DDI_SUCCESS) {
213 			DBG(DBG_MAP_WIN, dip, "px_lib_iommu_demap: failed\n");
214 		}
215 		return (DDI_FAILURE);
216 	}
217 
218 done:
219 	if (PX_DVMA_DBG_ON(mmu_p))
220 		px_dvma_alloc_debug(mmu_p, (char *)mp->dmai_mapping,
221 		    mp->dmai_size, mp);
222 
223 	return (DDI_SUCCESS);
224 }
225 
226 void
227 px_mmu_unmap_pages(px_mmu_t *mmu_p, ddi_dma_impl_t *mp, px_dvma_addr_t dvma_pg,
228     uint_t npages)
229 {
230 	px_dvma_addr_t	pg_index = MMU_PAGE_INDEX(mmu_p, dvma_pg);
231 
232 	DBG(DBG_UNMAP_WIN, mmu_p->mmu_px_p->px_dip,
233 	    "px_mmu_unmap_pages:%x+%x=%x npages=0x%x\n",
234 	    (uint_t)mmu_p->dvma_base_pg, (uint_t)pg_index, dvma_pg,
235 	    (uint_t)npages);
236 
237 	if (px_lib_iommu_demap(mmu_p->mmu_px_p->px_dip,
238 	    PCI_TSBID(0, pg_index), npages) != DDI_SUCCESS) {
239 		DBG(DBG_UNMAP_WIN, mmu_p->mmu_px_p->px_dip,
240 		    "px_lib_iommu_demap: failed\n");
241 	}
242 
243 	if (!PX_MAP_BUFZONE(mp))
244 		return;
245 
246 	DBG(DBG_UNMAP_WIN, mmu_p->mmu_px_p->px_dip, "px_mmu_unmap_pages: "
247 	    "redzone pg=%x\n", pg_index + npages);
248 
249 	ASSERT(PX_HAS_REDZONE(mp));
250 
251 	if (px_lib_iommu_demap(mmu_p->mmu_px_p->px_dip,
252 	    PCI_TSBID(0, pg_index + npages), 1) != DDI_SUCCESS) {
253 		DBG(DBG_UNMAP_WIN, mmu_p->mmu_px_p->px_dip,
254 		    "px_lib_iommu_demap: failed\n");
255 	}
256 }
257 
258 /*
259  * px_mmu_map_window - map a dvma window into the mmu
260  * used by: px_dma_win(), px_dma_ctlops() - DDI_DMA_MOVWIN, DDI_DMA_NEXTWIN
261  * return value: none
262  */
263 /*ARGSUSED*/
264 int
265 px_mmu_map_window(px_mmu_t *mmu_p, ddi_dma_impl_t *mp, px_window_t win_no)
266 {
267 	uint32_t obj_pg0_off = mp->dmai_roffset;
268 	uint32_t win_pg0_off = win_no ? 0 : obj_pg0_off;
269 	size_t win_size = mp->dmai_winsize;
270 	size_t pfn_index = win_size * win_no;			/* temp value */
271 	size_t obj_off = win_no ? pfn_index - obj_pg0_off : 0;	/* xferred sz */
272 	px_dvma_addr_t dvma_pg = MMU_BTOP(mp->dmai_mapping);
273 	size_t res_size = mp->dmai_object.dmao_size - obj_off + win_pg0_off;
274 	int ret = DDI_SUCCESS;
275 
276 	ASSERT(!(win_size & MMU_PAGE_OFFSET));
277 	if (win_no >= mp->dmai_nwin)
278 		return (ret);
279 	if (res_size < win_size)		/* last window */
280 		win_size = res_size;		/* mp->dmai_winsize unchanged */
281 
282 	mp->dmai_mapping = MMU_PTOB(dvma_pg) | win_pg0_off;
283 	mp->dmai_size = win_size - win_pg0_off;	/* cur win xferrable size */
284 	mp->dmai_offset = obj_off;		/* win offset into object */
285 	pfn_index = MMU_BTOP(pfn_index);	/* index into pfnlist */
286 	ret = px_mmu_map_pages(mmu_p, mp, dvma_pg, MMU_BTOPR(win_size),
287 	    pfn_index);
288 
289 	return (ret);
290 }
291 
292 /*
293  * px_mmu_unmap_window
294  * This routine is called to break down the mmu mappings to a dvma window.
295  * Non partial mappings are viewed as single window mapping.
296  * used by: px_dma_unbindhdl(), px_dma_window(),
297  *	and px_dma_ctlops() - DDI_DMA_FREE, DDI_DMA_MOVWIN, DDI_DMA_NEXTWIN
298  * return value: none
299  */
300 /*ARGSUSED*/
301 void
302 px_mmu_unmap_window(px_mmu_t *mmu_p, ddi_dma_impl_t *mp)
303 {
304 	px_dvma_addr_t dvma_pg = MMU_BTOP(mp->dmai_mapping);
305 	uint_t npages = MMU_BTOP(mp->dmai_winsize);
306 
307 	px_mmu_unmap_pages(mmu_p, mp, dvma_pg, npages);
308 
309 	if (PX_DVMA_DBG_ON(mmu_p))
310 		px_dvma_free_debug(mmu_p, (char *)mp->dmai_mapping,
311 		    mp->dmai_size, mp);
312 }
313 
314 
315 #if 0
316 /*
317  * The following table is for reference only. It denotes the
318  * the TSB table size measured in number of 8 byte entries.
319  * It is represented by bits 3:0 in the MMU TSB CTRL REG.
320  */
321 static int px_mmu_tsb_sizes[] = {
322 	0x0,		/* 1K */
323 	0x1,		/* 2K */
324 	0x2,		/* 4K */
325 	0x3,		/* 8K */
326 	0x4,		/* 16K */
327 	0x5,		/* 32K */
328 	0x6,		/* 64K */
329 	0x7,		/* 128K */
330 	0x8		/* 256K */
331 };
332 #endif
333 
334 static char *px_mmu_errsts[] = {
335 	"Protection Error", "Invalid Error", "Timeout", "ECC Error(UE)"
336 };
337 
338 /*ARGSUSED*/
339 static int
340 px_log_mmu_err(px_t *px_p)
341 {
342 	/*
343 	 * Place holder, the correct eror bits need tobe logged.
344 	 */
345 	return (0);
346 }
347