1 /* $NetBSD: xen_bus_dma.c,v 1.32 2020/05/06 19:50:26 bouyer Exp $ */ 2 /* NetBSD bus_dma.c,v 1.21 2005/04/16 07:53:35 yamt Exp */ 3 4 /*- 5 * Copyright (c) 1996, 1997, 1998 The NetBSD Foundation, Inc. 6 * All rights reserved. 7 * 8 * This code is derived from software contributed to The NetBSD Foundation 9 * by Charles M. Hannum and by Jason R. Thorpe of the Numerical Aerospace 10 * Simulation Facility, NASA Ames Research Center. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 23 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 24 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 25 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <sys/cdefs.h> 35 __KERNEL_RCSID(0, "$NetBSD: xen_bus_dma.c,v 1.32 2020/05/06 19:50:26 bouyer Exp $"); 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/kernel.h> 40 #include <sys/mbuf.h> 41 #include <sys/proc.h> 42 43 #include <sys/bus.h> 44 #include <machine/bus_private.h> 45 46 #include <uvm/uvm.h> 47 48 #include "opt_xen.h" 49 50 /* No special needs */ 51 struct x86_bus_dma_tag xenbus_bus_dma_tag = { 52 ._tag_needs_free = 0, 53 ._bounce_thresh = 0, 54 ._bounce_alloc_lo = 0, 55 ._bounce_alloc_hi = 0, 56 ._may_bounce = NULL, 57 }; 58 59 #ifdef XENPV 60 61 extern paddr_t avail_end; 62 63 /* Pure 2^n version of get_order */ 64 static inline int get_order(unsigned long size) 65 { 66 int order = -1; 67 size = (size - 1) >> (PAGE_SHIFT - 1); 68 do { 69 size >>= 1; 70 order++; 71 } while (size); 72 return order; 73 } 74 75 static int 76 _xen_alloc_contig(bus_size_t size, bus_size_t alignment, 77 struct pglist *mlistp, int flags, bus_addr_t low, bus_addr_t high) 78 { 79 int order, i; 80 unsigned long npagesreq, npages, mfn; 81 bus_addr_t pa; 82 struct vm_page *pg, *pgnext; 83 int s, error; 84 struct xen_memory_reservation res; 85 86 /* 87 * When requesting a contigous memory region, the hypervisor will 88 * return a memory range aligned on size. 89 * The only way to enforce alignment is to request a memory region 90 * of size max(alignment, size). 91 */ 92 order = uimax(get_order(size), get_order(alignment)); 93 npages = (1 << order); 94 npagesreq = (size >> PAGE_SHIFT); 95 KASSERT(npages >= npagesreq); 96 97 /* get npages from UVM, and give them back to the hypervisor */ 98 error = uvm_pglistalloc(((psize_t)npages) << PAGE_SHIFT, 99 0, avail_end, 0, 0, mlistp, npages, (flags & BUS_DMA_NOWAIT) == 0); 100 if (error) 101 return (error); 102 103 for (pg = mlistp->tqh_first; pg != NULL; pg = pg->pageq.queue.tqe_next) { 104 pa = VM_PAGE_TO_PHYS(pg); 105 mfn = xpmap_ptom(pa) >> PAGE_SHIFT; 106 xpmap_ptom_unmap(pa); 107 set_xen_guest_handle(res.extent_start, &mfn); 108 res.nr_extents = 1; 109 res.extent_order = 0; 110 res.mem_flags = 0; 111 res.domid = DOMID_SELF; 112 error = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &res); 113 if (error != 1) { 114 #ifdef DEBUG 115 printf("xen_alloc_contig: XENMEM_decrease_reservation " 116 "failed: err %d (pa %#" PRIxPADDR " mfn %#lx)\n", 117 error, pa, mfn); 118 #endif 119 xpmap_ptom_map(pa, ptoa(mfn)); 120 121 error = ENOMEM; 122 goto failed; 123 } 124 } 125 /* Get the new contiguous memory extent */ 126 set_xen_guest_handle(res.extent_start, &mfn); 127 res.nr_extents = 1; 128 res.extent_order = order; 129 res.mem_flags = XENMEMF_address_bits(get_order(high) + PAGE_SHIFT); 130 res.domid = DOMID_SELF; 131 error = HYPERVISOR_memory_op(XENMEM_increase_reservation, &res); 132 if (error != 1) { 133 #ifdef DEBUG 134 printf("xen_alloc_contig: XENMEM_increase_reservation " 135 "failed: %d (order %d mem_flags %d)\n", 136 error, order, res.mem_flags); 137 #endif 138 error = ENOMEM; 139 pg = NULL; 140 goto failed; 141 } 142 s = splvm(); /* XXXSMP */ 143 /* Map the new extent in place of the old pages */ 144 for (pg = mlistp->tqh_first, i = 0; pg != NULL; pg = pgnext, i++) { 145 pgnext = pg->pageq.queue.tqe_next; 146 pa = VM_PAGE_TO_PHYS(pg); 147 xpmap_ptom_map(pa, ptoa(mfn+i)); 148 xpq_queue_machphys_update(((paddr_t)(mfn+i)) << PAGE_SHIFT, pa); 149 } 150 /* Flush updates through and flush the TLB */ 151 xpq_queue_tlb_flush(); 152 splx(s); 153 /* now that ptom/mtop are valid, give the extra pages back to UVM */ 154 for (pg = mlistp->tqh_first, i = 0; pg != NULL; pg = pgnext, i++) { 155 pgnext = pg->pageq.queue.tqe_next; 156 if (i >= npagesreq) { 157 TAILQ_REMOVE(mlistp, pg, pageq.queue); 158 uvm_pagefree(pg); 159 } 160 } 161 return 0; 162 163 failed: 164 /* 165 * Attempt to recover from a failed decrease or increase reservation: 166 * if decrease_reservation failed, we don't have given all pages 167 * back to Xen; give them back to UVM, and get the missing pages 168 * from Xen. 169 * if increase_reservation failed, we expect pg to be NULL and we just 170 * get back the missing pages from Xen one by one. 171 */ 172 /* give back remaining pages to UVM */ 173 for (; pg != NULL; pg = pgnext) { 174 pgnext = pg->pageq.queue.tqe_next; 175 TAILQ_REMOVE(mlistp, pg, pageq.queue); 176 uvm_pagefree(pg); 177 } 178 /* remplace the pages that we already gave to Xen */ 179 s = splvm(); /* XXXSMP */ 180 for (pg = mlistp->tqh_first; pg != NULL; pg = pgnext) { 181 pgnext = pg->pageq.queue.tqe_next; 182 set_xen_guest_handle(res.extent_start, &mfn); 183 res.nr_extents = 1; 184 res.extent_order = 0; 185 res.mem_flags = XENMEMF_address_bits(32); 186 res.domid = DOMID_SELF; 187 if (HYPERVISOR_memory_op(XENMEM_increase_reservation, &res) 188 < 0) { 189 printf("xen_alloc_contig: recovery " 190 "XENMEM_increase_reservation failed!\n"); 191 break; 192 } 193 pa = VM_PAGE_TO_PHYS(pg); 194 xpmap_ptom_map(pa, ptoa(mfn)); 195 xpq_queue_machphys_update(((paddr_t)mfn) << PAGE_SHIFT, pa); 196 /* slow but we don't care */ 197 xpq_queue_tlb_flush(); 198 TAILQ_REMOVE(mlistp, pg, pageq.queue); 199 uvm_pagefree(pg); 200 } 201 splx(s); 202 return error; 203 } 204 205 206 /* 207 * Allocate physical memory from the given physical address range. 208 * Called by DMA-safe memory allocation methods. 209 * We need our own version to deal with physical vs machine addresses. 210 */ 211 int 212 _xen_bus_dmamem_alloc_range(bus_dma_tag_t t, bus_size_t size, 213 bus_size_t alignment, bus_size_t boundary, bus_dma_segment_t *segs, 214 int nsegs, int *rsegs, int flags, bus_addr_t low, bus_addr_t high) 215 { 216 bus_addr_t curaddr, lastaddr; 217 struct vm_page *m; 218 struct pglist mlist; 219 int curseg, error; 220 int doingrealloc = 0; 221 bus_size_t uboundary; 222 223 /* Always round the size. */ 224 size = round_page(size); 225 226 KASSERT((alignment & (alignment - 1)) == 0); 227 KASSERT((boundary & (boundary - 1)) == 0); 228 KASSERT(boundary >= PAGE_SIZE || boundary == 0); 229 230 if (alignment < PAGE_SIZE) 231 alignment = PAGE_SIZE; 232 233 /* 234 * Allocate pages from the VM system. 235 * We accept boundaries < size, splitting in multiple segments 236 * if needed. uvm_pglistalloc does not, so compute an appropriate 237 * boundary: next power of 2 >= size 238 */ 239 if (boundary == 0) 240 uboundary = 0; 241 else { 242 uboundary = boundary; 243 while (uboundary < size) 244 uboundary = uboundary << 1; 245 } 246 error = uvm_pglistalloc(size, 0, avail_end, alignment, uboundary, 247 &mlist, nsegs, (flags & BUS_DMA_NOWAIT) == 0); 248 if (error) 249 return (error); 250 again: 251 252 /* 253 * Compute the location, size, and number of segments actually 254 * returned by the VM code. 255 */ 256 m = mlist.tqh_first; 257 curseg = 0; 258 curaddr = lastaddr = segs[curseg].ds_addr = _BUS_VM_PAGE_TO_BUS(m); 259 if (curaddr < low || curaddr >= high) 260 goto badaddr; 261 segs[curseg].ds_len = PAGE_SIZE; 262 m = m->pageq.queue.tqe_next; 263 if ((segs[curseg].ds_addr & (alignment - 1)) != 0) 264 goto dorealloc; 265 266 for (; m != NULL; m = m->pageq.queue.tqe_next) { 267 curaddr = _BUS_VM_PAGE_TO_BUS(m); 268 if (curaddr < low || curaddr >= high) 269 goto badaddr; 270 if (curaddr == (lastaddr + PAGE_SIZE) && 271 (lastaddr & boundary) == (curaddr & boundary)) { 272 segs[curseg].ds_len += PAGE_SIZE; 273 } else { 274 curseg++; 275 if (curseg >= nsegs || 276 (curaddr & (alignment - 1)) != 0) { 277 if (doingrealloc) 278 return EFBIG; 279 else 280 goto dorealloc; 281 } 282 segs[curseg].ds_addr = curaddr; 283 segs[curseg].ds_len = PAGE_SIZE; 284 } 285 lastaddr = curaddr; 286 } 287 288 *rsegs = curseg + 1; 289 return (0); 290 291 badaddr: 292 if (doingrealloc == 0) 293 goto dorealloc; 294 if (curaddr < low) { 295 /* no way to enforce this */ 296 printf("_xen_bus_dmamem_alloc_range: no way to " 297 "enforce address range (0x%" PRIx64 " - 0x%" PRIx64 ")\n", 298 (uint64_t)low, (uint64_t)high); 299 uvm_pglistfree(&mlist); 300 return EINVAL; 301 } 302 printf("xen_bus_dmamem_alloc_range: " 303 "curraddr=0x%lx > high=0x%lx\n", 304 (u_long)curaddr, (u_long)high); 305 panic("xen_bus_dmamem_alloc_range 1"); 306 dorealloc: 307 if (doingrealloc == 1) 308 panic("_xen_bus_dmamem_alloc_range: " 309 "xen_alloc_contig returned " 310 "too much segments"); 311 doingrealloc = 1; 312 /* 313 * Too much segments, or memory doesn't fit 314 * constraints. Free this memory and 315 * get a contigous segment from the hypervisor. 316 */ 317 uvm_pglistfree(&mlist); 318 for (curseg = 0; curseg < nsegs; curseg++) { 319 segs[curseg].ds_addr = 0; 320 segs[curseg].ds_len = 0; 321 } 322 error = _xen_alloc_contig(size, alignment, 323 &mlist, flags, low, high); 324 if (error) 325 return error; 326 goto again; 327 } 328 #endif /* XENPV */ 329