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 void tce_build_pSeries(struct iommu_table
*tbl
, long index
,
52 long npages
, unsigned long uaddr
,
53 enum dma_data_direction direction
)
59 proto_tce
= TCE_PCI_READ
; // Read allowed
61 if (direction
!= DMA_TO_DEVICE
)
62 proto_tce
|= TCE_PCI_WRITE
;
64 tcep
= ((u64
*)tbl
->it_base
) + index
;
67 /* can't move this out since we might cross LMB boundary */
68 rpn
= (virt_to_abs(uaddr
)) >> TCE_SHIFT
;
69 *tcep
= proto_tce
| (rpn
& TCE_RPN_MASK
) << TCE_RPN_SHIFT
;
71 uaddr
+= TCE_PAGE_SIZE
;
77 static void tce_free_pSeries(struct iommu_table
*tbl
, long index
, long npages
)
81 tcep
= ((u64
*)tbl
->it_base
) + index
;
87 static unsigned long tce_get_pseries(struct iommu_table
*tbl
, long index
)
91 tcep
= ((u64
*)tbl
->it_base
) + index
;
96 static void tce_build_pSeriesLP(struct iommu_table
*tbl
, long tcenum
,
97 long npages
, unsigned long uaddr
,
98 enum dma_data_direction direction
)
104 rpn
= (virt_to_abs(uaddr
)) >> TCE_SHIFT
;
105 proto_tce
= TCE_PCI_READ
;
106 if (direction
!= DMA_TO_DEVICE
)
107 proto_tce
|= TCE_PCI_WRITE
;
110 tce
= proto_tce
| (rpn
& TCE_RPN_MASK
) << TCE_RPN_SHIFT
;
111 rc
= plpar_tce_put((u64
)tbl
->it_index
, (u64
)tcenum
<< 12, tce
);
113 if (rc
&& printk_ratelimit()) {
114 printk("tce_build_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc
);
115 printk("\tindex = 0x%lx\n", (u64
)tbl
->it_index
);
116 printk("\ttcenum = 0x%lx\n", (u64
)tcenum
);
117 printk("\ttce val = 0x%lx\n", tce
);
118 show_stack(current
, (unsigned long *)__get_SP());
126 static DEFINE_PER_CPU(u64
*, tce_page
) = NULL
;
128 static void tce_buildmulti_pSeriesLP(struct iommu_table
*tbl
, long tcenum
,
129 long npages
, unsigned long uaddr
,
130 enum dma_data_direction direction
)
139 return tce_build_pSeriesLP(tbl
, tcenum
, npages
, uaddr
,
142 tcep
= __get_cpu_var(tce_page
);
144 /* This is safe to do since interrupts are off when we're called
145 * from iommu_alloc{,_sg}()
148 tcep
= (u64
*)__get_free_page(GFP_ATOMIC
);
149 /* If allocation fails, fall back to the loop implementation */
151 return tce_build_pSeriesLP(tbl
, tcenum
, npages
,
153 __get_cpu_var(tce_page
) = tcep
;
156 rpn
= (virt_to_abs(uaddr
)) >> TCE_SHIFT
;
157 proto_tce
= TCE_PCI_READ
;
158 if (direction
!= DMA_TO_DEVICE
)
159 proto_tce
|= TCE_PCI_WRITE
;
161 /* We can map max one pageful of TCEs at a time */
164 * Set up the page with TCE data, looping through and setting
167 limit
= min_t(long, npages
, 4096/TCE_ENTRY_SIZE
);
169 for (l
= 0; l
< limit
; l
++) {
170 tcep
[l
] = proto_tce
| (rpn
& TCE_RPN_MASK
) << TCE_RPN_SHIFT
;
174 rc
= plpar_tce_put_indirect((u64
)tbl
->it_index
,
176 (u64
)virt_to_abs(tcep
),
181 } while (npages
> 0 && !rc
);
183 if (rc
&& printk_ratelimit()) {
184 printk("tce_buildmulti_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc
);
185 printk("\tindex = 0x%lx\n", (u64
)tbl
->it_index
);
186 printk("\tnpages = 0x%lx\n", (u64
)npages
);
187 printk("\ttce[0] val = 0x%lx\n", tcep
[0]);
188 show_stack(current
, (unsigned long *)__get_SP());
192 static void tce_free_pSeriesLP(struct iommu_table
*tbl
, long tcenum
, long npages
)
197 rc
= plpar_tce_put((u64
)tbl
->it_index
, (u64
)tcenum
<< 12, 0);
199 if (rc
&& printk_ratelimit()) {
200 printk("tce_free_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc
);
201 printk("\tindex = 0x%lx\n", (u64
)tbl
->it_index
);
202 printk("\ttcenum = 0x%lx\n", (u64
)tcenum
);
203 show_stack(current
, (unsigned long *)__get_SP());
211 static void tce_freemulti_pSeriesLP(struct iommu_table
*tbl
, long tcenum
, long npages
)
215 rc
= plpar_tce_stuff((u64
)tbl
->it_index
, (u64
)tcenum
<< 12, 0, npages
);
217 if (rc
&& printk_ratelimit()) {
218 printk("tce_freemulti_pSeriesLP: plpar_tce_stuff failed\n");
219 printk("\trc = %ld\n", rc
);
220 printk("\tindex = 0x%lx\n", (u64
)tbl
->it_index
);
221 printk("\tnpages = 0x%lx\n", (u64
)npages
);
222 show_stack(current
, (unsigned long *)__get_SP());
226 static unsigned long tce_get_pSeriesLP(struct iommu_table
*tbl
, long tcenum
)
229 unsigned long tce_ret
;
231 rc
= plpar_tce_get((u64
)tbl
->it_index
, (u64
)tcenum
<< 12, &tce_ret
);
233 if (rc
&& printk_ratelimit()) {
234 printk("tce_get_pSeriesLP: plpar_tce_get failed. rc=%ld\n",
236 printk("\tindex = 0x%lx\n", (u64
)tbl
->it_index
);
237 printk("\ttcenum = 0x%lx\n", (u64
)tcenum
);
238 show_stack(current
, (unsigned long *)__get_SP());
245 static void iommu_table_setparms(struct pci_controller
*phb
,
246 struct device_node
*dn
,
247 struct iommu_table
*tbl
)
249 struct device_node
*node
;
250 const unsigned long *basep
;
255 basep
= of_get_property(node
, "linux,tce-base", NULL
);
256 sizep
= of_get_property(node
, "linux,tce-size", NULL
);
257 if (basep
== NULL
|| sizep
== NULL
) {
258 printk(KERN_ERR
"PCI_DMA: iommu_table_setparms: %s has "
259 "missing tce entries !\n", dn
->full_name
);
263 tbl
->it_base
= (unsigned long)__va(*basep
);
265 #ifndef CONFIG_CRASH_DUMP
266 memset((void *)tbl
->it_base
, 0, *sizep
);
269 tbl
->it_busno
= phb
->bus
->number
;
271 /* Units of tce entries */
272 tbl
->it_offset
= phb
->dma_window_base_cur
>> IOMMU_PAGE_SHIFT
;
274 /* Test if we are going over 2GB of DMA space */
275 if (phb
->dma_window_base_cur
+ phb
->dma_window_size
> 0x80000000ul
) {
276 udbg_printf("PCI_DMA: Unexpected number of IOAs under this PHB.\n");
277 panic("PCI_DMA: Unexpected number of IOAs under this PHB.\n");
280 phb
->dma_window_base_cur
+= phb
->dma_window_size
;
282 /* Set the tce table size - measured in entries */
283 tbl
->it_size
= phb
->dma_window_size
>> IOMMU_PAGE_SHIFT
;
286 tbl
->it_blocksize
= 16;
287 tbl
->it_type
= TCE_PCI
;
291 * iommu_table_setparms_lpar
293 * Function: On pSeries LPAR systems, return TCE table info, given a pci bus.
295 static void iommu_table_setparms_lpar(struct pci_controller
*phb
,
296 struct device_node
*dn
,
297 struct iommu_table
*tbl
,
298 const void *dma_window
,
301 unsigned long offset
, size
;
303 tbl
->it_busno
= bussubno
;
304 of_parse_dma_window(dn
, dma_window
, &tbl
->it_index
, &offset
, &size
);
307 tbl
->it_blocksize
= 16;
308 tbl
->it_type
= TCE_PCI
;
309 tbl
->it_offset
= offset
>> IOMMU_PAGE_SHIFT
;
310 tbl
->it_size
= size
>> IOMMU_PAGE_SHIFT
;
313 static void pci_dma_bus_setup_pSeries(struct pci_bus
*bus
)
315 struct device_node
*dn
;
316 struct iommu_table
*tbl
;
317 struct device_node
*isa_dn
, *isa_dn_orig
;
318 struct device_node
*tmp
;
322 dn
= pci_bus_to_OF_node(bus
);
324 pr_debug("pci_dma_bus_setup_pSeries: setting up bus %s\n", dn
->full_name
);
327 /* This is not a root bus, any setup will be done for the
328 * device-side of the bridge in iommu_dev_setup_pSeries().
334 /* Check if the ISA bus on the system is under
337 isa_dn
= isa_dn_orig
= of_find_node_by_type(NULL
, "isa");
339 while (isa_dn
&& isa_dn
!= dn
)
340 isa_dn
= isa_dn
->parent
;
343 of_node_put(isa_dn_orig
);
345 /* Count number of direct PCI children of the PHB. */
346 for (children
= 0, tmp
= dn
->child
; tmp
; tmp
= tmp
->sibling
)
349 pr_debug("Children: %d\n", children
);
351 /* Calculate amount of DMA window per slot. Each window must be
352 * a power of two (due to pci_alloc_consistent requirements).
354 * Keep 256MB aside for PHBs with ISA.
358 /* No ISA/IDE - just set window size and return */
359 pci
->phb
->dma_window_size
= 0x80000000ul
; /* To be divided */
361 while (pci
->phb
->dma_window_size
* children
> 0x80000000ul
)
362 pci
->phb
->dma_window_size
>>= 1;
363 pr_debug("No ISA/IDE, window size is 0x%lx\n",
364 pci
->phb
->dma_window_size
);
365 pci
->phb
->dma_window_base_cur
= 0;
370 /* If we have ISA, then we probably have an IDE
371 * controller too. Allocate a 128MB table but
372 * skip the first 128MB to avoid stepping on ISA
375 pci
->phb
->dma_window_size
= 0x8000000ul
;
376 pci
->phb
->dma_window_base_cur
= 0x8000000ul
;
378 tbl
= kmalloc_node(sizeof(struct iommu_table
), GFP_KERNEL
,
381 iommu_table_setparms(pci
->phb
, dn
, tbl
);
382 pci
->iommu_table
= iommu_init_table(tbl
, pci
->phb
->node
);
384 /* Divide the rest (1.75GB) among the children */
385 pci
->phb
->dma_window_size
= 0x80000000ul
;
386 while (pci
->phb
->dma_window_size
* children
> 0x70000000ul
)
387 pci
->phb
->dma_window_size
>>= 1;
389 pr_debug("ISA/IDE, window size is 0x%lx\n", pci
->phb
->dma_window_size
);
393 static void pci_dma_bus_setup_pSeriesLP(struct pci_bus
*bus
)
395 struct iommu_table
*tbl
;
396 struct device_node
*dn
, *pdn
;
398 const void *dma_window
= NULL
;
400 dn
= pci_bus_to_OF_node(bus
);
402 pr_debug("pci_dma_bus_setup_pSeriesLP: setting up bus %s\n",
405 /* Find nearest ibm,dma-window, walking up the device tree */
406 for (pdn
= dn
; pdn
!= NULL
; pdn
= pdn
->parent
) {
407 dma_window
= of_get_property(pdn
, "ibm,dma-window", NULL
);
408 if (dma_window
!= NULL
)
412 if (dma_window
== NULL
) {
413 pr_debug(" no ibm,dma-window property !\n");
419 pr_debug(" parent is %s, iommu_table: 0x%p\n",
420 pdn
->full_name
, ppci
->iommu_table
);
422 if (!ppci
->iommu_table
) {
423 tbl
= kmalloc_node(sizeof(struct iommu_table
), GFP_KERNEL
,
425 iommu_table_setparms_lpar(ppci
->phb
, pdn
, tbl
, dma_window
,
427 ppci
->iommu_table
= iommu_init_table(tbl
, ppci
->phb
->node
);
428 pr_debug(" created table: %p\n", ppci
->iommu_table
);
432 PCI_DN(dn
)->iommu_table
= ppci
->iommu_table
;
436 static void pci_dma_dev_setup_pSeries(struct pci_dev
*dev
)
438 struct device_node
*dn
;
439 struct iommu_table
*tbl
;
441 pr_debug("pci_dma_dev_setup_pSeries: %s\n", pci_name(dev
));
443 dn
= dev
->dev
.archdata
.of_node
;
445 /* If we're the direct child of a root bus, then we need to allocate
446 * an iommu table ourselves. The bus setup code should have setup
447 * the window sizes already.
449 if (!dev
->bus
->self
) {
450 struct pci_controller
*phb
= PCI_DN(dn
)->phb
;
452 pr_debug(" --> first child, no bridge. Allocating iommu table.\n");
453 tbl
= kmalloc_node(sizeof(struct iommu_table
), GFP_KERNEL
,
455 iommu_table_setparms(phb
, dn
, tbl
);
456 PCI_DN(dn
)->iommu_table
= iommu_init_table(tbl
, phb
->node
);
457 dev
->dev
.archdata
.dma_data
= PCI_DN(dn
)->iommu_table
;
461 /* If this device is further down the bus tree, search upwards until
462 * an already allocated iommu table is found and use that.
465 while (dn
&& PCI_DN(dn
) && PCI_DN(dn
)->iommu_table
== NULL
)
468 if (dn
&& PCI_DN(dn
))
469 dev
->dev
.archdata
.dma_data
= PCI_DN(dn
)->iommu_table
;
471 printk(KERN_WARNING
"iommu: Device %s has no iommu table\n",
475 static void pci_dma_dev_setup_pSeriesLP(struct pci_dev
*dev
)
477 struct device_node
*pdn
, *dn
;
478 struct iommu_table
*tbl
;
479 const void *dma_window
= NULL
;
482 pr_debug("pci_dma_dev_setup_pSeriesLP: %s\n", pci_name(dev
));
484 /* dev setup for LPAR is a little tricky, since the device tree might
485 * contain the dma-window properties per-device and not neccesarily
486 * for the bus. So we need to search upwards in the tree until we
487 * either hit a dma-window property, OR find a parent with a table
490 dn
= pci_device_to_OF_node(dev
);
491 pr_debug(" node is %s\n", dn
->full_name
);
493 for (pdn
= dn
; pdn
&& PCI_DN(pdn
) && !PCI_DN(pdn
)->iommu_table
;
495 dma_window
= of_get_property(pdn
, "ibm,dma-window", NULL
);
500 if (!pdn
|| !PCI_DN(pdn
)) {
501 printk(KERN_WARNING
"pci_dma_dev_setup_pSeriesLP: "
502 "no DMA window found for pci dev=%s dn=%s\n",
503 pci_name(dev
), dn
? dn
->full_name
: "<null>");
506 pr_debug(" parent is %s\n", pdn
->full_name
);
508 /* Check for parent == NULL so we don't try to setup the empty EADS
509 * slots on POWER4 machines.
511 if (dma_window
== NULL
|| pdn
->parent
== NULL
) {
512 pr_debug(" no dma window for device, linking to parent\n");
513 dev
->dev
.archdata
.dma_data
= PCI_DN(pdn
)->iommu_table
;
518 if (!pci
->iommu_table
) {
519 tbl
= kmalloc_node(sizeof(struct iommu_table
), GFP_KERNEL
,
521 iommu_table_setparms_lpar(pci
->phb
, pdn
, tbl
, dma_window
,
522 pci
->phb
->bus
->number
);
523 pci
->iommu_table
= iommu_init_table(tbl
, pci
->phb
->node
);
524 pr_debug(" created table: %p\n", pci
->iommu_table
);
526 pr_debug(" found DMA window, table: %p\n", pci
->iommu_table
);
529 dev
->dev
.archdata
.dma_data
= pci
->iommu_table
;
531 #else /* CONFIG_PCI */
532 #define pci_dma_bus_setup_pSeries NULL
533 #define pci_dma_dev_setup_pSeries NULL
534 #define pci_dma_bus_setup_pSeriesLP NULL
535 #define pci_dma_dev_setup_pSeriesLP NULL
536 #endif /* !CONFIG_PCI */
538 static int iommu_reconfig_notifier(struct notifier_block
*nb
, unsigned long action
, void *node
)
541 struct device_node
*np
= node
;
542 struct pci_dn
*pci
= PCI_DN(np
);
545 case PSERIES_RECONFIG_REMOVE
:
546 if (pci
&& pci
->iommu_table
&&
547 of_get_property(np
, "ibm,dma-window", NULL
))
548 iommu_free_table(pci
->iommu_table
, np
->full_name
);
557 static struct notifier_block iommu_reconfig_nb
= {
558 .notifier_call
= iommu_reconfig_notifier
,
561 /* These are called very early. */
562 void iommu_init_early_pSeries(void)
564 if (of_chosen
&& of_get_property(of_chosen
, "linux,iommu-off", NULL
)) {
565 /* Direct I/O, IOMMU off */
566 ppc_md
.pci_dma_dev_setup
= NULL
;
567 ppc_md
.pci_dma_bus_setup
= NULL
;
568 set_pci_dma_ops(&dma_direct_ops
);
572 if (firmware_has_feature(FW_FEATURE_LPAR
)) {
573 if (firmware_has_feature(FW_FEATURE_MULTITCE
)) {
574 ppc_md
.tce_build
= tce_buildmulti_pSeriesLP
;
575 ppc_md
.tce_free
= tce_freemulti_pSeriesLP
;
577 ppc_md
.tce_build
= tce_build_pSeriesLP
;
578 ppc_md
.tce_free
= tce_free_pSeriesLP
;
580 ppc_md
.tce_get
= tce_get_pSeriesLP
;
581 ppc_md
.pci_dma_bus_setup
= pci_dma_bus_setup_pSeriesLP
;
582 ppc_md
.pci_dma_dev_setup
= pci_dma_dev_setup_pSeriesLP
;
584 ppc_md
.tce_build
= tce_build_pSeries
;
585 ppc_md
.tce_free
= tce_free_pSeries
;
586 ppc_md
.tce_get
= tce_get_pseries
;
587 ppc_md
.pci_dma_bus_setup
= pci_dma_bus_setup_pSeries
;
588 ppc_md
.pci_dma_dev_setup
= pci_dma_dev_setup_pSeries
;
592 pSeries_reconfig_notifier_register(&iommu_reconfig_nb
);
594 set_pci_dma_ops(&dma_iommu_ops
);