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"
52 static void tce_build_pSeries(struct iommu_table *tbl, long index,
53 long npages, unsigned long uaddr,
54 enum dma_data_direction direction)
60 index <<= TCE_PAGE_FACTOR;
61 npages <<= TCE_PAGE_FACTOR;
63 proto_tce = TCE_PCI_READ; // Read allowed
65 if (direction != DMA_TO_DEVICE)
66 proto_tce |= TCE_PCI_WRITE;
68 tcep = ((u64 *)tbl->it_base) + index;
71 /* can't move this out since we might cross LMB boundary */
72 rpn = (virt_to_abs(uaddr)) >> TCE_SHIFT;
73 *tcep = proto_tce | (rpn & TCE_RPN_MASK) << TCE_RPN_SHIFT;
75 uaddr += TCE_PAGE_SIZE;
81 static void tce_free_pSeries(struct iommu_table *tbl, long index, long npages)
85 npages <<= TCE_PAGE_FACTOR;
86 index <<= TCE_PAGE_FACTOR;
88 tcep = ((u64 *)tbl->it_base) + index;
94 static unsigned long tce_get_pseries(struct iommu_table *tbl, long index)
98 index <<= TCE_PAGE_FACTOR;
99 tcep = ((u64 *)tbl->it_base) + index;
104 static void tce_build_pSeriesLP(struct iommu_table *tbl, long tcenum,
105 long npages, unsigned long uaddr,
106 enum dma_data_direction direction)
112 tcenum <<= TCE_PAGE_FACTOR;
113 npages <<= TCE_PAGE_FACTOR;
115 rpn = (virt_to_abs(uaddr)) >> TCE_SHIFT;
116 proto_tce = TCE_PCI_READ;
117 if (direction != DMA_TO_DEVICE)
118 proto_tce |= TCE_PCI_WRITE;
121 tce = proto_tce | (rpn & TCE_RPN_MASK) << TCE_RPN_SHIFT;
122 rc = plpar_tce_put((u64)tbl->it_index, (u64)tcenum << 12, tce);
124 if (rc && printk_ratelimit()) {
125 printk("tce_build_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc);
126 printk("\tindex = 0x%lx\n", (u64)tbl->it_index);
127 printk("\ttcenum = 0x%lx\n", (u64)tcenum);
128 printk("\ttce val = 0x%lx\n", tce );
129 show_stack(current, (unsigned long *)__get_SP());
137 static DEFINE_PER_CPU(u64 *, tce_page) = NULL;
139 static void tce_buildmulti_pSeriesLP(struct iommu_table *tbl, long tcenum,
140 long npages, unsigned long uaddr,
141 enum dma_data_direction direction)
149 if (TCE_PAGE_FACTOR == 0 && npages == 1)
150 return tce_build_pSeriesLP(tbl, tcenum, npages, uaddr,
153 tcep = __get_cpu_var(tce_page);
155 /* This is safe to do since interrupts are off when we're called
156 * from iommu_alloc{,_sg}()
159 tcep = (u64 *)__get_free_page(GFP_ATOMIC);
160 /* If allocation fails, fall back to the loop implementation */
162 return tce_build_pSeriesLP(tbl, tcenum, npages,
164 __get_cpu_var(tce_page) = tcep;
167 tcenum <<= TCE_PAGE_FACTOR;
168 npages <<= TCE_PAGE_FACTOR;
170 rpn = (virt_to_abs(uaddr)) >> TCE_SHIFT;
171 proto_tce = TCE_PCI_READ;
172 if (direction != DMA_TO_DEVICE)
173 proto_tce |= TCE_PCI_WRITE;
175 /* We can map max one pageful of TCEs at a time */
178 * Set up the page with TCE data, looping through and setting
181 limit = min_t(long, npages, 4096/TCE_ENTRY_SIZE);
183 for (l = 0; l < limit; l++) {
184 tcep[l] = proto_tce | (rpn & TCE_RPN_MASK) << TCE_RPN_SHIFT;
188 rc = plpar_tce_put_indirect((u64)tbl->it_index,
190 (u64)virt_to_abs(tcep),
195 } while (npages > 0 && !rc);
197 if (rc && printk_ratelimit()) {
198 printk("tce_buildmulti_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc);
199 printk("\tindex = 0x%lx\n", (u64)tbl->it_index);
200 printk("\tnpages = 0x%lx\n", (u64)npages);
201 printk("\ttce[0] val = 0x%lx\n", tcep[0]);
202 show_stack(current, (unsigned long *)__get_SP());
206 static void tce_free_pSeriesLP(struct iommu_table *tbl, long tcenum, long npages)
210 tcenum <<= TCE_PAGE_FACTOR;
211 npages <<= TCE_PAGE_FACTOR;
214 rc = plpar_tce_put((u64)tbl->it_index, (u64)tcenum << 12, 0);
216 if (rc && printk_ratelimit()) {
217 printk("tce_free_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc);
218 printk("\tindex = 0x%lx\n", (u64)tbl->it_index);
219 printk("\ttcenum = 0x%lx\n", (u64)tcenum);
220 show_stack(current, (unsigned long *)__get_SP());
228 static void tce_freemulti_pSeriesLP(struct iommu_table *tbl, long tcenum, long npages)
232 tcenum <<= TCE_PAGE_FACTOR;
233 npages <<= TCE_PAGE_FACTOR;
235 rc = plpar_tce_stuff((u64)tbl->it_index, (u64)tcenum << 12, 0, npages);
237 if (rc && printk_ratelimit()) {
238 printk("tce_freemulti_pSeriesLP: plpar_tce_stuff failed\n");
239 printk("\trc = %ld\n", rc);
240 printk("\tindex = 0x%lx\n", (u64)tbl->it_index);
241 printk("\tnpages = 0x%lx\n", (u64)npages);
242 show_stack(current, (unsigned long *)__get_SP());
246 static unsigned long tce_get_pSeriesLP(struct iommu_table *tbl, long tcenum)
249 unsigned long tce_ret;
251 tcenum <<= TCE_PAGE_FACTOR;
252 rc = plpar_tce_get((u64)tbl->it_index, (u64)tcenum << 12, &tce_ret);
254 if (rc && printk_ratelimit()) {
255 printk("tce_get_pSeriesLP: plpar_tce_get failed. rc=%ld\n",
257 printk("\tindex = 0x%lx\n", (u64)tbl->it_index);
258 printk("\ttcenum = 0x%lx\n", (u64)tcenum);
259 show_stack(current, (unsigned long *)__get_SP());
265 static void iommu_table_setparms(struct pci_controller *phb,
266 struct device_node *dn,
267 struct iommu_table *tbl)
269 struct device_node *node;
270 unsigned long *basep;
273 node = (struct device_node *)phb->arch_data;
275 basep = (unsigned long *)get_property(node, "linux,tce-base", NULL);
276 sizep = (unsigned int *)get_property(node, "linux,tce-size", NULL);
277 if (basep == NULL || sizep == NULL) {
278 printk(KERN_ERR "PCI_DMA: iommu_table_setparms: %s has "
279 "missing tce entries !\n", dn->full_name);
283 tbl->it_base = (unsigned long)__va(*basep);
285 #ifndef CONFIG_CRASH_DUMP
286 memset((void *)tbl->it_base, 0, *sizep);
289 tbl->it_busno = phb->bus->number;
291 /* Units of tce entries */
292 tbl->it_offset = phb->dma_window_base_cur >> PAGE_SHIFT;
294 /* Test if we are going over 2GB of DMA space */
295 if (phb->dma_window_base_cur + phb->dma_window_size > 0x80000000ul) {
296 udbg_printf("PCI_DMA: Unexpected number of IOAs under this PHB.\n");
297 panic("PCI_DMA: Unexpected number of IOAs under this PHB.\n");
300 phb->dma_window_base_cur += phb->dma_window_size;
302 /* Set the tce table size - measured in entries */
303 tbl->it_size = phb->dma_window_size >> PAGE_SHIFT;
306 tbl->it_blocksize = 16;
307 tbl->it_type = TCE_PCI;
311 * iommu_table_setparms_lpar
313 * Function: On pSeries LPAR systems, return TCE table info, given a pci bus.
315 static void iommu_table_setparms_lpar(struct pci_controller *phb,
316 struct device_node *dn,
317 struct iommu_table *tbl,
318 unsigned char *dma_window)
320 unsigned long offset, size;
322 tbl->it_busno = PCI_DN(dn)->bussubno;
323 of_parse_dma_window(dn, dma_window, &tbl->it_index, &offset, &size);
326 tbl->it_blocksize = 16;
327 tbl->it_type = TCE_PCI;
328 tbl->it_offset = offset >> PAGE_SHIFT;
329 tbl->it_size = size >> PAGE_SHIFT;
332 static void iommu_bus_setup_pSeries(struct pci_bus *bus)
334 struct device_node *dn;
335 struct iommu_table *tbl;
336 struct device_node *isa_dn, *isa_dn_orig;
337 struct device_node *tmp;
341 DBG("iommu_bus_setup_pSeries, bus %p, bus->self %p\n", bus, bus->self);
343 dn = pci_bus_to_OF_node(bus);
347 /* This is not a root bus, any setup will be done for the
348 * device-side of the bridge in iommu_dev_setup_pSeries().
353 /* Check if the ISA bus on the system is under
356 isa_dn = isa_dn_orig = of_find_node_by_type(NULL, "isa");
358 while (isa_dn && isa_dn != dn)
359 isa_dn = isa_dn->parent;
362 of_node_put(isa_dn_orig);
364 /* Count number of direct PCI children of the PHB. */
365 for (children = 0, tmp = dn->child; tmp; tmp = tmp->sibling)
368 DBG("Children: %d\n", children);
370 /* Calculate amount of DMA window per slot. Each window must be
371 * a power of two (due to pci_alloc_consistent requirements).
373 * Keep 256MB aside for PHBs with ISA.
377 /* No ISA/IDE - just set window size and return */
378 pci->phb->dma_window_size = 0x80000000ul; /* To be divided */
380 while (pci->phb->dma_window_size * children > 0x80000000ul)
381 pci->phb->dma_window_size >>= 1;
382 DBG("No ISA/IDE, window size is 0x%lx\n",
383 pci->phb->dma_window_size);
384 pci->phb->dma_window_base_cur = 0;
389 /* If we have ISA, then we probably have an IDE
390 * controller too. Allocate a 128MB table but
391 * skip the first 128MB to avoid stepping on ISA
394 pci->phb->dma_window_size = 0x8000000ul;
395 pci->phb->dma_window_base_cur = 0x8000000ul;
397 tbl = kmalloc_node(sizeof(struct iommu_table), GFP_KERNEL,
400 iommu_table_setparms(pci->phb, dn, tbl);
401 pci->iommu_table = iommu_init_table(tbl, pci->phb->node);
403 /* Divide the rest (1.75GB) among the children */
404 pci->phb->dma_window_size = 0x80000000ul;
405 while (pci->phb->dma_window_size * children > 0x70000000ul)
406 pci->phb->dma_window_size >>= 1;
408 DBG("ISA/IDE, window size is 0x%lx\n", pci->phb->dma_window_size);
413 static void iommu_bus_setup_pSeriesLP(struct pci_bus *bus)
415 struct iommu_table *tbl;
416 struct device_node *dn, *pdn;
418 unsigned char *dma_window = NULL;
420 DBG("iommu_bus_setup_pSeriesLP, bus %p, bus->self %p\n", bus, bus->self);
422 dn = pci_bus_to_OF_node(bus);
424 /* Find nearest ibm,dma-window, walking up the device tree */
425 for (pdn = dn; pdn != NULL; pdn = pdn->parent) {
426 dma_window = get_property(pdn, "ibm,dma-window", NULL);
427 if (dma_window != NULL)
431 if (dma_window == NULL) {
432 DBG("iommu_bus_setup_pSeriesLP: bus %s seems to have no ibm,dma-window property\n", dn->full_name);
437 if (!ppci->iommu_table) {
438 /* Bussubno hasn't been copied yet.
439 * Do it now because iommu_table_setparms_lpar needs it.
442 ppci->bussubno = bus->number;
444 tbl = kmalloc_node(sizeof(struct iommu_table), GFP_KERNEL,
447 iommu_table_setparms_lpar(ppci->phb, pdn, tbl, dma_window);
449 ppci->iommu_table = iommu_init_table(tbl, ppci->phb->node);
453 PCI_DN(dn)->iommu_table = ppci->iommu_table;
457 static void iommu_dev_setup_pSeries(struct pci_dev *dev)
459 struct device_node *dn, *mydn;
460 struct iommu_table *tbl;
462 DBG("iommu_dev_setup_pSeries, dev %p (%s)\n", dev, pci_name(dev));
464 mydn = dn = pci_device_to_OF_node(dev);
466 /* If we're the direct child of a root bus, then we need to allocate
467 * an iommu table ourselves. The bus setup code should have setup
468 * the window sizes already.
470 if (!dev->bus->self) {
471 DBG(" --> first child, no bridge. Allocating iommu table.\n");
472 tbl = kmalloc_node(sizeof(struct iommu_table), GFP_KERNEL,
473 PCI_DN(dn)->phb->node);
474 iommu_table_setparms(PCI_DN(dn)->phb, dn, tbl);
475 PCI_DN(dn)->iommu_table = iommu_init_table(tbl,
476 PCI_DN(dn)->phb->node);
481 /* If this device is further down the bus tree, search upwards until
482 * an already allocated iommu table is found and use that.
485 while (dn && PCI_DN(dn) && PCI_DN(dn)->iommu_table == NULL)
488 if (dn && PCI_DN(dn)) {
489 PCI_DN(mydn)->iommu_table = PCI_DN(dn)->iommu_table;
491 DBG("iommu_dev_setup_pSeries, dev %p (%s) has no iommu table\n", dev, pci_name(dev));
495 static int iommu_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *node)
498 struct device_node *np = node;
499 struct pci_dn *pci = PCI_DN(np);
502 case PSERIES_RECONFIG_REMOVE:
503 if (pci && pci->iommu_table &&
504 get_property(np, "ibm,dma-window", NULL))
505 iommu_free_table(np);
514 static struct notifier_block iommu_reconfig_nb = {
515 .notifier_call = iommu_reconfig_notifier,
518 static void iommu_dev_setup_pSeriesLP(struct pci_dev *dev)
520 struct device_node *pdn, *dn;
521 struct iommu_table *tbl;
522 unsigned char *dma_window = NULL;
525 DBG("iommu_dev_setup_pSeriesLP, dev %p (%s)\n", dev, pci_name(dev));
527 /* dev setup for LPAR is a little tricky, since the device tree might
528 * contain the dma-window properties per-device and not neccesarily
529 * for the bus. So we need to search upwards in the tree until we
530 * either hit a dma-window property, OR find a parent with a table
533 dn = pci_device_to_OF_node(dev);
535 for (pdn = dn; pdn && PCI_DN(pdn) && !PCI_DN(pdn)->iommu_table;
537 dma_window = get_property(pdn, "ibm,dma-window", NULL);
542 /* Check for parent == NULL so we don't try to setup the empty EADS
543 * slots on POWER4 machines.
545 if (dma_window == NULL || pdn->parent == NULL) {
546 DBG("No dma window for device, linking to parent\n");
547 PCI_DN(dn)->iommu_table = PCI_DN(pdn)->iommu_table;
550 DBG("Found DMA window, allocating table\n");
554 if (!pci->iommu_table) {
555 /* iommu_table_setparms_lpar needs bussubno. */
556 pci->bussubno = pci->phb->bus->number;
558 tbl = kmalloc_node(sizeof(struct iommu_table), GFP_KERNEL,
561 iommu_table_setparms_lpar(pci->phb, pdn, tbl, dma_window);
563 pci->iommu_table = iommu_init_table(tbl, pci->phb->node);
567 PCI_DN(dn)->iommu_table = pci->iommu_table;
570 static void iommu_bus_setup_null(struct pci_bus *b) { }
571 static void iommu_dev_setup_null(struct pci_dev *d) { }
573 /* These are called very early. */
574 void iommu_init_early_pSeries(void)
576 if (of_chosen && get_property(of_chosen, "linux,iommu-off", NULL)) {
577 /* Direct I/O, IOMMU off */
578 ppc_md.iommu_dev_setup = iommu_dev_setup_null;
579 ppc_md.iommu_bus_setup = iommu_bus_setup_null;
580 pci_direct_iommu_init();
585 if (firmware_has_feature(FW_FEATURE_LPAR)) {
586 if (firmware_has_feature(FW_FEATURE_MULTITCE)) {
587 ppc_md.tce_build = tce_buildmulti_pSeriesLP;
588 ppc_md.tce_free = tce_freemulti_pSeriesLP;
590 ppc_md.tce_build = tce_build_pSeriesLP;
591 ppc_md.tce_free = tce_free_pSeriesLP;
593 ppc_md.tce_get = tce_get_pSeriesLP;
594 ppc_md.iommu_bus_setup = iommu_bus_setup_pSeriesLP;
595 ppc_md.iommu_dev_setup = iommu_dev_setup_pSeriesLP;
597 ppc_md.tce_build = tce_build_pSeries;
598 ppc_md.tce_free = tce_free_pSeries;
599 ppc_md.tce_get = tce_get_pseries;
600 ppc_md.iommu_bus_setup = iommu_bus_setup_pSeries;
601 ppc_md.iommu_dev_setup = iommu_dev_setup_pSeries;
605 pSeries_reconfig_notifier_register(&iommu_reconfig_nb);