2 * arch/ppc64/kernel/pSeries_iommu.c
4 * Copyright (C) 2001 Mike Corrigan & Dave Engebretsen, IBM Corporation
8 * Copyright (C) 2004 Olof Johansson <olof@austin.ibm.com>, IBM Corporation
10 * Dynamic DMA mapping support, pSeries-specific parts, both SMP and LPAR.
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
23 * You should have received a copy of the GNU General Public License
24 * along with this program; if not, write to the Free Software
25 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
28 #include <linux/config.h>
29 #include <linux/init.h>
30 #include <linux/types.h>
31 #include <linux/slab.h>
33 #include <linux/spinlock.h>
34 #include <linux/string.h>
35 #include <linux/pci.h>
36 #include <linux/dma-mapping.h>
40 #include <asm/ppcdebug.h>
41 #include <asm/iommu.h>
42 #include <asm/pci-bridge.h>
43 #include <asm/machdep.h>
44 #include <asm/abs_addr.h>
45 #include <asm/plpar_wrappers.h>
49 static void tce_build_pSeries(struct iommu_table
*tbl
, long index
,
50 long npages
, unsigned long uaddr
,
51 enum dma_data_direction direction
)
57 t
.te_rdwr
= 1; // Read allowed
59 if (direction
!= DMA_TO_DEVICE
)
62 tp
= ((union tce_entry
*)tbl
->it_base
) + index
;
65 /* can't move this out since we might cross LMB boundary */
66 t
.te_rpn
= (virt_to_abs(uaddr
)) >> PAGE_SHIFT
;
68 tp
->te_word
= t
.te_word
;
76 static void tce_free_pSeries(struct iommu_table
*tbl
, long index
, long npages
)
82 tp
= ((union tce_entry
*)tbl
->it_base
) + index
;
85 tp
->te_word
= t
.te_word
;
92 static void tce_build_pSeriesLP(struct iommu_table
*tbl
, long tcenum
,
93 long npages
, unsigned long uaddr
,
94 enum dma_data_direction direction
)
100 tce
.te_rpn
= (virt_to_abs(uaddr
)) >> PAGE_SHIFT
;
102 if (direction
!= DMA_TO_DEVICE
)
106 rc
= plpar_tce_put((u64
)tbl
->it_index
,
110 if (rc
&& printk_ratelimit()) {
111 printk("tce_build_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc
);
112 printk("\tindex = 0x%lx\n", (u64
)tbl
->it_index
);
113 printk("\ttcenum = 0x%lx\n", (u64
)tcenum
);
114 printk("\ttce val = 0x%lx\n", tce
.te_word
);
115 show_stack(current
, (unsigned long *)__get_SP());
123 DEFINE_PER_CPU(void *, tce_page
) = NULL
;
125 static void tce_buildmulti_pSeriesLP(struct iommu_table
*tbl
, long tcenum
,
126 long npages
, unsigned long uaddr
,
127 enum dma_data_direction direction
)
130 union tce_entry tce
, *tcep
;
134 return tce_build_pSeriesLP(tbl
, tcenum
, npages
, uaddr
,
137 tcep
= __get_cpu_var(tce_page
);
139 /* This is safe to do since interrupts are off when we're called
140 * from iommu_alloc{,_sg}()
143 tcep
= (void *)__get_free_page(GFP_ATOMIC
);
144 /* If allocation fails, fall back to the loop implementation */
146 return tce_build_pSeriesLP(tbl
, tcenum
, npages
,
148 __get_cpu_var(tce_page
) = tcep
;
152 tce
.te_rpn
= (virt_to_abs(uaddr
)) >> PAGE_SHIFT
;
154 if (direction
!= DMA_TO_DEVICE
)
157 /* We can map max one pageful of TCEs at a time */
160 * Set up the page with TCE data, looping through and setting
163 limit
= min_t(long, npages
, PAGE_SIZE
/sizeof(union tce_entry
));
165 for (l
= 0; l
< limit
; l
++) {
170 rc
= plpar_tce_put_indirect((u64
)tbl
->it_index
,
172 (u64
)virt_to_abs(tcep
),
177 } while (npages
> 0 && !rc
);
179 if (rc
&& printk_ratelimit()) {
180 printk("tce_buildmulti_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc
);
181 printk("\tindex = 0x%lx\n", (u64
)tbl
->it_index
);
182 printk("\tnpages = 0x%lx\n", (u64
)npages
);
183 printk("\ttce[0] val = 0x%lx\n", tcep
[0].te_word
);
184 show_stack(current
, (unsigned long *)__get_SP());
188 static void tce_free_pSeriesLP(struct iommu_table
*tbl
, long tcenum
, long npages
)
196 rc
= plpar_tce_put((u64
)tbl
->it_index
,
200 if (rc
&& printk_ratelimit()) {
201 printk("tce_free_pSeriesLP: plpar_tce_put failed. rc=%ld\n", rc
);
202 printk("\tindex = 0x%lx\n", (u64
)tbl
->it_index
);
203 printk("\ttcenum = 0x%lx\n", (u64
)tcenum
);
204 printk("\ttce val = 0x%lx\n", tce
.te_word
);
205 show_stack(current
, (unsigned long *)__get_SP());
213 static void tce_freemulti_pSeriesLP(struct iommu_table
*tbl
, long tcenum
, long npages
)
220 rc
= plpar_tce_stuff((u64
)tbl
->it_index
,
225 if (rc
&& printk_ratelimit()) {
226 printk("tce_freemulti_pSeriesLP: plpar_tce_stuff failed\n");
227 printk("\trc = %ld\n", rc
);
228 printk("\tindex = 0x%lx\n", (u64
)tbl
->it_index
);
229 printk("\tnpages = 0x%lx\n", (u64
)npages
);
230 printk("\ttce val = 0x%lx\n", tce
.te_word
);
231 show_stack(current
, (unsigned long *)__get_SP());
236 static void iommu_buses_init(void)
238 struct pci_controller
*phb
, *tmp
;
239 struct device_node
*dn
, *first_dn
;
240 int num_slots
, num_slots_ilog2
;
242 unsigned long tcetable_ilog2
;
245 * We default to a TCE table that maps 2GB (4MB table, 22 bits),
246 * however some machines have a 3GB IO hole and for these we
247 * create a table that maps 1GB (2MB table, 21 bits)
249 if (io_hole_start
< 0x80000000UL
)
254 /* XXX Should we be using pci_root_buses instead? -ojn
257 list_for_each_entry_safe(phb
, tmp
, &hose_list
, list_node
) {
258 first_dn
= ((struct device_node
*)phb
->arch_data
)->child
;
260 /* Carve 2GB into the largest dma_window_size possible */
261 for (dn
= first_dn
, num_slots
= 0; dn
!= NULL
; dn
= dn
->sibling
)
263 num_slots_ilog2
= __ilog2(num_slots
);
265 if ((1<<num_slots_ilog2
) != num_slots
)
268 phb
->dma_window_size
= 1 << (tcetable_ilog2
- num_slots_ilog2
);
270 /* Reserve 16MB of DMA space on the first PHB.
271 * We should probably be more careful and use firmware props.
272 * In reality this space is remapped, not lost. But we don't
273 * want to get that smart to handle it -- too much work.
275 phb
->dma_window_base_cur
= first_phb
? (1 << 12) : 0;
278 for (dn
= first_dn
; dn
!= NULL
; dn
= dn
->sibling
)
279 iommu_devnode_init(dn
);
284 static void iommu_buses_init_lpar(struct list_head
*bus_list
)
286 struct list_head
*ln
;
288 struct device_node
*busdn
;
289 unsigned int *dma_window
;
291 for (ln
=bus_list
->next
; ln
!= bus_list
; ln
=ln
->next
) {
293 busdn
= PCI_GET_DN(bus
);
295 dma_window
= (unsigned int *)get_property(busdn
, "ibm,dma-window", NULL
);
297 /* Bussubno hasn't been copied yet.
298 * Do it now because iommu_table_setparms_lpar needs it.
300 busdn
->bussubno
= bus
->number
;
301 iommu_devnode_init(busdn
);
304 /* look for a window on a bridge even if the PHB had one */
305 iommu_buses_init_lpar(&bus
->children
);
310 static void iommu_table_setparms(struct pci_controller
*phb
,
311 struct device_node
*dn
,
312 struct iommu_table
*tbl
)
314 struct device_node
*node
;
315 unsigned long *basep
;
318 node
= (struct device_node
*)phb
->arch_data
;
320 if (get_property(node
, "linux,has-tce-table", NULL
) == NULL
) {
321 printk(KERN_ERR
"PCI_DMA: iommu_table_setparms: %s has no tce table !\n",
325 basep
= (unsigned long *)get_property(node
, "linux,tce-base", NULL
);
326 sizep
= (unsigned int *)get_property(node
, "linux,tce-size", NULL
);
327 if (basep
== NULL
|| sizep
== NULL
) {
328 printk(KERN_ERR
"PCI_DMA: iommu_table_setparms: %s has missing tce"
329 " entries !\n", dn
->full_name
);
332 memset((void *)(*basep
), 0, *sizep
);
334 tbl
->it_busno
= phb
->bus
->number
;
336 /* Units of tce entries */
337 tbl
->it_offset
= phb
->dma_window_base_cur
;
339 /* Adjust the current table offset to the next
340 * region. Measured in TCE entries. Force an
341 * alignment to the size allotted per IOA. This
342 * makes it easier to remove the 1st 16MB.
344 phb
->dma_window_base_cur
+= (phb
->dma_window_size
>>3);
345 phb
->dma_window_base_cur
&=
346 ~((phb
->dma_window_size
>>3)-1);
348 /* Set the tce table size - measured in pages */
349 tbl
->it_size
= ((phb
->dma_window_base_cur
-
350 tbl
->it_offset
) << 3) >> PAGE_SHIFT
;
352 /* Test if we are going over 2GB of DMA space */
353 if (phb
->dma_window_base_cur
> (1 << 19))
354 panic("PCI_DMA: Unexpected number of IOAs under this PHB.\n");
356 tbl
->it_base
= *basep
;
358 tbl
->it_entrysize
= sizeof(union tce_entry
);
359 tbl
->it_blocksize
= 16;
360 tbl
->it_type
= TCE_PCI
;
364 * iommu_table_setparms_lpar
366 * Function: On pSeries LPAR systems, return TCE table info, given a pci bus.
368 * ToDo: properly interpret the ibm,dma-window property. The definition is:
369 * logical-bus-number (1 word)
370 * phys-address (#address-cells words)
371 * size (#cell-size words)
373 * Currently we hard code these sizes (more or less).
375 static void iommu_table_setparms_lpar(struct pci_controller
*phb
,
376 struct device_node
*dn
,
377 struct iommu_table
*tbl
)
379 unsigned int *dma_window
;
381 dma_window
= (unsigned int *)get_property(dn
, "ibm,dma-window", NULL
);
384 panic("iommu_table_setparms_lpar: device %s has no"
385 " ibm,dma-window property!\n", dn
->full_name
);
387 tbl
->it_busno
= dn
->bussubno
;
388 tbl
->it_size
= (((((unsigned long)dma_window
[4] << 32) |
389 (unsigned long)dma_window
[5]) >> PAGE_SHIFT
) << 3) >> PAGE_SHIFT
;
390 tbl
->it_offset
= ((((unsigned long)dma_window
[2] << 32) |
391 (unsigned long)dma_window
[3]) >> 12);
393 tbl
->it_index
= dma_window
[0];
394 tbl
->it_entrysize
= sizeof(union tce_entry
);
395 tbl
->it_blocksize
= 16;
396 tbl
->it_type
= TCE_PCI
;
400 void iommu_devnode_init(struct device_node
*dn
)
402 struct iommu_table
*tbl
;
404 tbl
= (struct iommu_table
*)kmalloc(sizeof(struct iommu_table
),
407 if (systemcfg
->platform
== PLATFORM_PSERIES_LPAR
)
408 iommu_table_setparms_lpar(dn
->phb
, dn
, tbl
);
410 iommu_table_setparms(dn
->phb
, dn
, tbl
);
412 dn
->iommu_table
= iommu_init_table(tbl
);
416 void iommu_setup_pSeries(void)
418 struct pci_dev
*dev
= NULL
;
419 struct device_node
*dn
, *mydn
;
421 if (systemcfg
->platform
== PLATFORM_PSERIES_LPAR
)
422 iommu_buses_init_lpar(&pci_root_buses
);
426 /* Now copy the iommu_table ptr from the bus devices down to every
427 * pci device_node. This means get_iommu_table() won't need to search
428 * up the device tree to find it.
430 while ((dev
= pci_find_device(PCI_ANY_ID
, PCI_ANY_ID
, dev
)) != NULL
) {
431 mydn
= dn
= PCI_GET_DN(dev
);
433 while (dn
&& dn
->iommu_table
== NULL
)
436 mydn
->iommu_table
= dn
->iommu_table
;
441 /* These are called very early. */
442 void tce_init_pSeries(void)
444 if (!(systemcfg
->platform
& PLATFORM_LPAR
)) {
445 ppc_md
.tce_build
= tce_build_pSeries
;
446 ppc_md
.tce_free
= tce_free_pSeries
;
447 } else if (cur_cpu_spec
->firmware_features
& FW_FEATURE_MULTITCE
) {
448 ppc_md
.tce_build
= tce_buildmulti_pSeriesLP
;
449 ppc_md
.tce_free
= tce_freemulti_pSeriesLP
;
451 ppc_md
.tce_build
= tce_build_pSeriesLP
;
452 ppc_md
.tce_free
= tce_free_pSeriesLP
;