2 * Copyright (C) 2001 Mike Corrigan & Dave Engebretsen, IBM Corporation
6 * Copyright (C) 2004 Olof Johansson <olof@lixom.net>, IBM Corporation
7 * Copyright (C) 2006 Olof Johansson <olof@lixom.net>
9 * Dynamic DMA mapping support, pSeries-specific parts, both SMP and LPAR.
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License as published by
14 * the Free Software Foundation; either version 2 of the License, or
15 * (at your option) any later version.
17 * This program is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 * GNU General Public License for more details.
22 * You should have received a copy of the GNU General Public License
23 * along with this program; if not, write to the Free Software
24 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
27 #include <linux/init.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
31 #include <linux/spinlock.h>
32 #include <linux/string.h>
33 #include <linux/pci.h>
34 #include <linux/dma-mapping.h>
38 #include <asm/iommu.h>
39 #include <asm/pci-bridge.h>
40 #include <asm/machdep.h>
41 #include <asm/abs_addr.h>
42 #include <asm/pSeries_reconfig.h>
43 #include <asm/firmware.h>
45 #include <asm/ppc-pci.h>
48 #include "plpar_wrappers.h"
51 static int tce_build_pSeries(struct iommu_table *tbl, long index,
52 long npages, unsigned long uaddr,
53 enum dma_data_direction direction,
54 struct dma_attrs *attrs)
60 proto_tce = TCE_PCI_READ; // Read allowed
62 if (direction != DMA_TO_DEVICE)
63 proto_tce |= TCE_PCI_WRITE;
65 tcep = ((u64 *)tbl->it_base) + index;
68 /* can't move this out since we might cross LMB boundary */
69 rpn = (virt_to_abs(uaddr)) >> TCE_SHIFT;
70 *tcep = proto_tce | (rpn & TCE_RPN_MASK) << TCE_RPN_SHIFT;
72 uaddr += TCE_PAGE_SIZE;
79 static void tce_free_pSeries(struct iommu_table *tbl, long index, long npages)
83 tcep = ((u64 *)tbl->it_base) + index;
89 static unsigned long tce_get_pseries(struct iommu_table *tbl, long index)
93 tcep = ((u64 *)tbl->it_base) + index;
98 static void tce_free_pSeriesLP(struct iommu_table*, long, long);
99 static void tce_freemulti_pSeriesLP(struct iommu_table*, long, long);
101 static int tce_build_pSeriesLP(struct iommu_table *tbl, long tcenum,
102 long npages, unsigned long uaddr,
103 enum dma_data_direction direction,
104 struct dma_attrs *attrs)
110 long tcenum_start = tcenum, npages_start = npages;
112 rpn = (virt_to_abs(uaddr)) >> TCE_SHIFT;
113 proto_tce = TCE_PCI_READ;
114 if (direction != DMA_TO_DEVICE)
115 proto_tce |= TCE_PCI_WRITE;
118 tce = proto_tce | (rpn & TCE_RPN_MASK) << TCE_RPN_SHIFT;
119 rc = plpar_tce_put((u64)tbl->it_index, (u64)tcenum << 12, tce);
121 if (unlikely(rc == H_NOT_ENOUGH_RESOURCES)) {
123 tce_free_pSeriesLP(tbl, tcenum_start,
124 (npages_start - (npages + 1)));
128 if (rc && printk_ratelimit()) {
129 printk("tce_build_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc);
130 printk("\tindex = 0x%lx\n", (u64)tbl->it_index);
131 printk("\ttcenum = 0x%lx\n", (u64)tcenum);
132 printk("\ttce val = 0x%lx\n", tce );
133 show_stack(current, (unsigned long *)__get_SP());
142 static DEFINE_PER_CPU(u64 *, tce_page) = NULL;
144 static int tce_buildmulti_pSeriesLP(struct iommu_table *tbl, long tcenum,
145 long npages, unsigned long uaddr,
146 enum dma_data_direction direction,
147 struct dma_attrs *attrs)
154 long tcenum_start = tcenum, npages_start = npages;
158 return tce_build_pSeriesLP(tbl, tcenum, npages, uaddr,
162 tcep = __get_cpu_var(tce_page);
164 /* This is safe to do since interrupts are off when we're called
165 * from iommu_alloc{,_sg}()
168 tcep = (u64 *)__get_free_page(GFP_ATOMIC);
169 /* If allocation fails, fall back to the loop implementation */
171 return tce_build_pSeriesLP(tbl, tcenum, npages, uaddr,
174 __get_cpu_var(tce_page) = tcep;
177 rpn = (virt_to_abs(uaddr)) >> TCE_SHIFT;
178 proto_tce = TCE_PCI_READ;
179 if (direction != DMA_TO_DEVICE)
180 proto_tce |= TCE_PCI_WRITE;
182 /* We can map max one pageful of TCEs at a time */
185 * Set up the page with TCE data, looping through and setting
188 limit = min_t(long, npages, 4096/TCE_ENTRY_SIZE);
190 for (l = 0; l < limit; l++) {
191 tcep[l] = proto_tce | (rpn & TCE_RPN_MASK) << TCE_RPN_SHIFT;
195 rc = plpar_tce_put_indirect((u64)tbl->it_index,
197 (u64)virt_to_abs(tcep),
202 } while (npages > 0 && !rc);
204 if (unlikely(rc == H_NOT_ENOUGH_RESOURCES)) {
206 tce_freemulti_pSeriesLP(tbl, tcenum_start,
207 (npages_start - (npages + limit)));
211 if (rc && printk_ratelimit()) {
212 printk("tce_buildmulti_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc);
213 printk("\tindex = 0x%lx\n", (u64)tbl->it_index);
214 printk("\tnpages = 0x%lx\n", (u64)npages);
215 printk("\ttce[0] val = 0x%lx\n", tcep[0]);
216 show_stack(current, (unsigned long *)__get_SP());
221 static void tce_free_pSeriesLP(struct iommu_table *tbl, long tcenum, long npages)
226 rc = plpar_tce_put((u64)tbl->it_index, (u64)tcenum << 12, 0);
228 if (rc && printk_ratelimit()) {
229 printk("tce_free_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc);
230 printk("\tindex = 0x%lx\n", (u64)tbl->it_index);
231 printk("\ttcenum = 0x%lx\n", (u64)tcenum);
232 show_stack(current, (unsigned long *)__get_SP());
240 static void tce_freemulti_pSeriesLP(struct iommu_table *tbl, long tcenum, long npages)
244 rc = plpar_tce_stuff((u64)tbl->it_index, (u64)tcenum << 12, 0, npages);
246 if (rc && printk_ratelimit()) {
247 printk("tce_freemulti_pSeriesLP: plpar_tce_stuff failed\n");
248 printk("\trc = %ld\n", rc);
249 printk("\tindex = 0x%lx\n", (u64)tbl->it_index);
250 printk("\tnpages = 0x%lx\n", (u64)npages);
251 show_stack(current, (unsigned long *)__get_SP());
255 static unsigned long tce_get_pSeriesLP(struct iommu_table *tbl, long tcenum)
258 unsigned long tce_ret;
260 rc = plpar_tce_get((u64)tbl->it_index, (u64)tcenum << 12, &tce_ret);
262 if (rc && printk_ratelimit()) {
263 printk("tce_get_pSeriesLP: plpar_tce_get failed. rc=%ld\n",
265 printk("\tindex = 0x%lx\n", (u64)tbl->it_index);
266 printk("\ttcenum = 0x%lx\n", (u64)tcenum);
267 show_stack(current, (unsigned long *)__get_SP());
274 static void iommu_table_setparms(struct pci_controller *phb,
275 struct device_node *dn,
276 struct iommu_table *tbl)
278 struct device_node *node;
279 const unsigned long *basep;
284 basep = of_get_property(node, "linux,tce-base", NULL);
285 sizep = of_get_property(node, "linux,tce-size", NULL);
286 if (basep == NULL || sizep == NULL) {
287 printk(KERN_ERR "PCI_DMA: iommu_table_setparms: %s has "
288 "missing tce entries !\n", dn->full_name);
292 tbl->it_base = (unsigned long)__va(*basep);
294 #ifndef CONFIG_CRASH_DUMP
295 memset((void *)tbl->it_base, 0, *sizep);
298 tbl->it_busno = phb->bus->number;
300 /* Units of tce entries */
301 tbl->it_offset = phb->dma_window_base_cur >> IOMMU_PAGE_SHIFT;
303 /* Test if we are going over 2GB of DMA space */
304 if (phb->dma_window_base_cur + phb->dma_window_size > 0x80000000ul) {
305 udbg_printf("PCI_DMA: Unexpected number of IOAs under this PHB.\n");
306 panic("PCI_DMA: Unexpected number of IOAs under this PHB.\n");
309 phb->dma_window_base_cur += phb->dma_window_size;
311 /* Set the tce table size - measured in entries */
312 tbl->it_size = phb->dma_window_size >> IOMMU_PAGE_SHIFT;
315 tbl->it_blocksize = 16;
316 tbl->it_type = TCE_PCI;
320 * iommu_table_setparms_lpar
322 * Function: On pSeries LPAR systems, return TCE table info, given a pci bus.
324 static void iommu_table_setparms_lpar(struct pci_controller *phb,
325 struct device_node *dn,
326 struct iommu_table *tbl,
327 const void *dma_window,
330 unsigned long offset, size;
332 tbl->it_busno = bussubno;
333 of_parse_dma_window(dn, dma_window, &tbl->it_index, &offset, &size);
336 tbl->it_blocksize = 16;
337 tbl->it_type = TCE_PCI;
338 tbl->it_offset = offset >> IOMMU_PAGE_SHIFT;
339 tbl->it_size = size >> IOMMU_PAGE_SHIFT;
342 static void pci_dma_bus_setup_pSeries(struct pci_bus *bus)
344 struct device_node *dn;
345 struct iommu_table *tbl;
346 struct device_node *isa_dn, *isa_dn_orig;
347 struct device_node *tmp;
351 dn = pci_bus_to_OF_node(bus);
353 pr_debug("pci_dma_bus_setup_pSeries: setting up bus %s\n", dn->full_name);
356 /* This is not a root bus, any setup will be done for the
357 * device-side of the bridge in iommu_dev_setup_pSeries().
363 /* Check if the ISA bus on the system is under
366 isa_dn = isa_dn_orig = of_find_node_by_type(NULL, "isa");
368 while (isa_dn && isa_dn != dn)
369 isa_dn = isa_dn->parent;
372 of_node_put(isa_dn_orig);
374 /* Count number of direct PCI children of the PHB. */
375 for (children = 0, tmp = dn->child; tmp; tmp = tmp->sibling)
378 pr_debug("Children: %d\n", children);
380 /* Calculate amount of DMA window per slot. Each window must be
381 * a power of two (due to pci_alloc_consistent requirements).
383 * Keep 256MB aside for PHBs with ISA.
387 /* No ISA/IDE - just set window size and return */
388 pci->phb->dma_window_size = 0x80000000ul; /* To be divided */
390 while (pci->phb->dma_window_size * children > 0x80000000ul)
391 pci->phb->dma_window_size >>= 1;
392 pr_debug("No ISA/IDE, window size is 0x%lx\n",
393 pci->phb->dma_window_size);
394 pci->phb->dma_window_base_cur = 0;
399 /* If we have ISA, then we probably have an IDE
400 * controller too. Allocate a 128MB table but
401 * skip the first 128MB to avoid stepping on ISA
404 pci->phb->dma_window_size = 0x8000000ul;
405 pci->phb->dma_window_base_cur = 0x8000000ul;
407 tbl = kmalloc_node(sizeof(struct iommu_table), GFP_KERNEL,
410 iommu_table_setparms(pci->phb, dn, tbl);
411 pci->iommu_table = iommu_init_table(tbl, pci->phb->node);
413 /* Divide the rest (1.75GB) among the children */
414 pci->phb->dma_window_size = 0x80000000ul;
415 while (pci->phb->dma_window_size * children > 0x70000000ul)
416 pci->phb->dma_window_size >>= 1;
418 pr_debug("ISA/IDE, window size is 0x%lx\n", pci->phb->dma_window_size);
422 static void pci_dma_bus_setup_pSeriesLP(struct pci_bus *bus)
424 struct iommu_table *tbl;
425 struct device_node *dn, *pdn;
427 const void *dma_window = NULL;
429 dn = pci_bus_to_OF_node(bus);
431 pr_debug("pci_dma_bus_setup_pSeriesLP: setting up bus %s\n",
434 /* Find nearest ibm,dma-window, walking up the device tree */
435 for (pdn = dn; pdn != NULL; pdn = pdn->parent) {
436 dma_window = of_get_property(pdn, "ibm,dma-window", NULL);
437 if (dma_window != NULL)
441 if (dma_window == NULL) {
442 pr_debug(" no ibm,dma-window property !\n");
448 pr_debug(" parent is %s, iommu_table: 0x%p\n",
449 pdn->full_name, ppci->iommu_table);
451 if (!ppci->iommu_table) {
452 tbl = kmalloc_node(sizeof(struct iommu_table), GFP_KERNEL,
454 iommu_table_setparms_lpar(ppci->phb, pdn, tbl, dma_window,
456 ppci->iommu_table = iommu_init_table(tbl, ppci->phb->node);
457 pr_debug(" created table: %p\n", ppci->iommu_table);
461 PCI_DN(dn)->iommu_table = ppci->iommu_table;
465 static void pci_dma_dev_setup_pSeries(struct pci_dev *dev)
467 struct device_node *dn;
468 struct iommu_table *tbl;
470 pr_debug("pci_dma_dev_setup_pSeries: %s\n", pci_name(dev));
472 dn = dev->dev.archdata.of_node;
474 /* If we're the direct child of a root bus, then we need to allocate
475 * an iommu table ourselves. The bus setup code should have setup
476 * the window sizes already.
478 if (!dev->bus->self) {
479 struct pci_controller *phb = PCI_DN(dn)->phb;
481 pr_debug(" --> first child, no bridge. Allocating iommu table.\n");
482 tbl = kmalloc_node(sizeof(struct iommu_table), GFP_KERNEL,
484 iommu_table_setparms(phb, dn, tbl);
485 PCI_DN(dn)->iommu_table = iommu_init_table(tbl, phb->node);
486 dev->dev.archdata.dma_data = PCI_DN(dn)->iommu_table;
490 /* If this device is further down the bus tree, search upwards until
491 * an already allocated iommu table is found and use that.
494 while (dn && PCI_DN(dn) && PCI_DN(dn)->iommu_table == NULL)
497 if (dn && PCI_DN(dn))
498 dev->dev.archdata.dma_data = PCI_DN(dn)->iommu_table;
500 printk(KERN_WARNING "iommu: Device %s has no iommu table\n",
504 static void pci_dma_dev_setup_pSeriesLP(struct pci_dev *dev)
506 struct device_node *pdn, *dn;
507 struct iommu_table *tbl;
508 const void *dma_window = NULL;
511 pr_debug("pci_dma_dev_setup_pSeriesLP: %s\n", pci_name(dev));
513 /* dev setup for LPAR is a little tricky, since the device tree might
514 * contain the dma-window properties per-device and not neccesarily
515 * for the bus. So we need to search upwards in the tree until we
516 * either hit a dma-window property, OR find a parent with a table
519 dn = pci_device_to_OF_node(dev);
520 pr_debug(" node is %s\n", dn->full_name);
522 for (pdn = dn; pdn && PCI_DN(pdn) && !PCI_DN(pdn)->iommu_table;
524 dma_window = of_get_property(pdn, "ibm,dma-window", NULL);
529 if (!pdn || !PCI_DN(pdn)) {
530 printk(KERN_WARNING "pci_dma_dev_setup_pSeriesLP: "
531 "no DMA window found for pci dev=%s dn=%s\n",
532 pci_name(dev), dn? dn->full_name : "<null>");
535 pr_debug(" parent is %s\n", pdn->full_name);
537 /* Check for parent == NULL so we don't try to setup the empty EADS
538 * slots on POWER4 machines.
540 if (dma_window == NULL || pdn->parent == NULL) {
541 pr_debug(" no dma window for device, linking to parent\n");
542 dev->dev.archdata.dma_data = PCI_DN(pdn)->iommu_table;
547 if (!pci->iommu_table) {
548 tbl = kmalloc_node(sizeof(struct iommu_table), GFP_KERNEL,
550 iommu_table_setparms_lpar(pci->phb, pdn, tbl, dma_window,
551 pci->phb->bus->number);
552 pci->iommu_table = iommu_init_table(tbl, pci->phb->node);
553 pr_debug(" created table: %p\n", pci->iommu_table);
555 pr_debug(" found DMA window, table: %p\n", pci->iommu_table);
558 dev->dev.archdata.dma_data = pci->iommu_table;
560 #else /* CONFIG_PCI */
561 #define pci_dma_bus_setup_pSeries NULL
562 #define pci_dma_dev_setup_pSeries NULL
563 #define pci_dma_bus_setup_pSeriesLP NULL
564 #define pci_dma_dev_setup_pSeriesLP NULL
565 #endif /* !CONFIG_PCI */
567 static int iommu_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *node)
570 struct device_node *np = node;
571 struct pci_dn *pci = PCI_DN(np);
574 case PSERIES_RECONFIG_REMOVE:
575 if (pci && pci->iommu_table &&
576 of_get_property(np, "ibm,dma-window", NULL))
577 iommu_free_table(pci->iommu_table, np->full_name);
586 static struct notifier_block iommu_reconfig_nb = {
587 .notifier_call = iommu_reconfig_notifier,
590 /* These are called very early. */
591 void iommu_init_early_pSeries(void)
593 if (of_chosen && of_get_property(of_chosen, "linux,iommu-off", NULL)) {
594 /* Direct I/O, IOMMU off */
595 ppc_md.pci_dma_dev_setup = NULL;
596 ppc_md.pci_dma_bus_setup = NULL;
597 set_pci_dma_ops(&dma_direct_ops);
601 if (firmware_has_feature(FW_FEATURE_LPAR)) {
602 if (firmware_has_feature(FW_FEATURE_MULTITCE)) {
603 ppc_md.tce_build = tce_buildmulti_pSeriesLP;
604 ppc_md.tce_free = tce_freemulti_pSeriesLP;
606 ppc_md.tce_build = tce_build_pSeriesLP;
607 ppc_md.tce_free = tce_free_pSeriesLP;
609 ppc_md.tce_get = tce_get_pSeriesLP;
610 ppc_md.pci_dma_bus_setup = pci_dma_bus_setup_pSeriesLP;
611 ppc_md.pci_dma_dev_setup = pci_dma_dev_setup_pSeriesLP;
613 ppc_md.tce_build = tce_build_pSeries;
614 ppc_md.tce_free = tce_free_pSeries;
615 ppc_md.tce_get = tce_get_pseries;
616 ppc_md.pci_dma_bus_setup = pci_dma_bus_setup_pSeries;
617 ppc_md.pci_dma_dev_setup = pci_dma_dev_setup_pSeries;
621 pSeries_reconfig_notifier_register(&iommu_reconfig_nb);
623 set_pci_dma_ops(&dma_iommu_ops);