1 // SPDX-License-Identifier: GPL-2.0
3 * channel program interfaces
5 * Copyright IBM Corp. 2017
7 * Author(s): Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com>
8 * Xiao Feng Ren <renxiaof@linux.vnet.ibm.com>
12 #include <linux/slab.h>
13 #include <linux/iommu.h>
14 #include <linux/vfio.h>
15 #include <asm/idals.h>
17 #include "vfio_ccw_cp.h"
20 /* Starting guest physical I/O address. */
21 unsigned long pa_iova
;
22 /* Array that stores PFNs of the pages need to pin. */
23 unsigned long *pa_iova_pfn
;
24 /* Array that receives PFNs of the pages pinned. */
25 unsigned long *pa_pfn
;
26 /* Number of pages pinned from @pa_iova. */
31 struct list_head next
;
33 /* Guest physical address of the current chain. */
35 /* Count of the valid ccws in chain. */
37 /* Pinned PAGEs for the original data. */
38 struct pfn_array
*ch_pa
;
42 * pfn_array_alloc() - alloc memory for PFNs
43 * @pa: pfn_array on which to perform the operation
44 * @iova: target guest physical address
45 * @len: number of bytes that should be pinned from @iova
47 * Attempt to allocate memory for PFNs.
50 * We expect (pa_nr == 0) and (pa_iova_pfn == NULL), any field in
51 * this structure will be filled in by this function.
54 * 0 if PFNs are allocated
55 * -EINVAL if pa->pa_nr is not initially zero, or pa->pa_iova_pfn is not NULL
56 * -ENOMEM if alloc failed
58 static int pfn_array_alloc(struct pfn_array
*pa
, u64 iova
, unsigned int len
)
62 if (pa
->pa_nr
|| pa
->pa_iova_pfn
)
67 pa
->pa_nr
= ((iova
& ~PAGE_MASK
) + len
+ (PAGE_SIZE
- 1)) >> PAGE_SHIFT
;
71 pa
->pa_iova_pfn
= kcalloc(pa
->pa_nr
,
72 sizeof(*pa
->pa_iova_pfn
) +
75 if (unlikely(!pa
->pa_iova_pfn
)) {
79 pa
->pa_pfn
= pa
->pa_iova_pfn
+ pa
->pa_nr
;
81 pa
->pa_iova_pfn
[0] = pa
->pa_iova
>> PAGE_SHIFT
;
82 pa
->pa_pfn
[0] = -1ULL;
83 for (i
= 1; i
< pa
->pa_nr
; i
++) {
84 pa
->pa_iova_pfn
[i
] = pa
->pa_iova_pfn
[i
- 1] + 1;
85 pa
->pa_pfn
[i
] = -1ULL;
92 * pfn_array_pin() - Pin user pages in memory
93 * @pa: pfn_array on which to perform the operation
94 * @mdev: the mediated device to perform pin operations
96 * Returns number of pages pinned upon success.
97 * If the pin request partially succeeds, or fails completely,
98 * all pages are left unpinned and a negative error value is returned.
100 static int pfn_array_pin(struct pfn_array
*pa
, struct device
*mdev
)
104 ret
= vfio_pin_pages(mdev
, pa
->pa_iova_pfn
, pa
->pa_nr
,
105 IOMMU_READ
| IOMMU_WRITE
, pa
->pa_pfn
);
109 } else if (ret
> 0 && ret
!= pa
->pa_nr
) {
110 vfio_unpin_pages(mdev
, pa
->pa_iova_pfn
, ret
);
123 /* Unpin the pages before releasing the memory. */
124 static void pfn_array_unpin_free(struct pfn_array
*pa
, struct device
*mdev
)
126 /* Only unpin if any pages were pinned to begin with */
128 vfio_unpin_pages(mdev
, pa
->pa_iova_pfn
, pa
->pa_nr
);
130 kfree(pa
->pa_iova_pfn
);
133 static bool pfn_array_iova_pinned(struct pfn_array
*pa
, unsigned long iova
)
135 unsigned long iova_pfn
= iova
>> PAGE_SHIFT
;
138 for (i
= 0; i
< pa
->pa_nr
; i
++)
139 if (pa
->pa_iova_pfn
[i
] == iova_pfn
)
144 /* Create the list of IDAL words for a pfn_array. */
145 static inline void pfn_array_idal_create_words(
146 struct pfn_array
*pa
,
147 unsigned long *idaws
)
152 * Idal words (execept the first one) rely on the memory being 4k
153 * aligned. If a user virtual address is 4K aligned, then it's
154 * corresponding kernel physical address will also be 4K aligned. Thus
155 * there will be no problem here to simply use the phys to create an
159 for (i
= 0; i
< pa
->pa_nr
; i
++)
160 idaws
[i
] = pa
->pa_pfn
[i
] << PAGE_SHIFT
;
162 /* Adjust the first IDAW, since it may not start on a page boundary */
163 idaws
[0] += pa
->pa_iova
& (PAGE_SIZE
- 1);
166 static void convert_ccw0_to_ccw1(struct ccw1
*source
, unsigned long len
)
169 struct ccw1
*pccw1
= source
;
172 for (i
= 0; i
< len
; i
++) {
173 ccw0
= *(struct ccw0
*)pccw1
;
174 if ((pccw1
->cmd_code
& 0x0f) == CCW_CMD_TIC
) {
175 pccw1
->cmd_code
= CCW_CMD_TIC
;
179 pccw1
->cmd_code
= ccw0
.cmd_code
;
180 pccw1
->flags
= ccw0
.flags
;
181 pccw1
->count
= ccw0
.count
;
183 pccw1
->cda
= ccw0
.cda
;
189 * Within the domain (@mdev), copy @n bytes from a guest physical
190 * address (@iova) to a host physical address (@to).
192 static long copy_from_iova(struct device
*mdev
,
196 struct pfn_array pa
= {0};
201 ret
= pfn_array_alloc(&pa
, iova
, n
);
205 ret
= pfn_array_pin(&pa
, mdev
);
207 pfn_array_unpin_free(&pa
, mdev
);
212 for (i
= 0; i
< pa
.pa_nr
; i
++) {
213 from
= pa
.pa_pfn
[i
] << PAGE_SHIFT
;
216 from
+= iova
& (PAGE_SIZE
- 1);
217 m
-= iova
& (PAGE_SIZE
- 1);
221 memcpy(to
+ (n
- l
), (void *)from
, m
);
228 pfn_array_unpin_free(&pa
, mdev
);
234 * Helpers to operate ccwchain.
236 #define ccw_is_read(_ccw) (((_ccw)->cmd_code & 0x03) == 0x02)
237 #define ccw_is_read_backward(_ccw) (((_ccw)->cmd_code & 0x0F) == 0x0C)
238 #define ccw_is_sense(_ccw) (((_ccw)->cmd_code & 0x0F) == CCW_CMD_BASIC_SENSE)
240 #define ccw_is_noop(_ccw) ((_ccw)->cmd_code == CCW_CMD_NOOP)
242 #define ccw_is_tic(_ccw) ((_ccw)->cmd_code == CCW_CMD_TIC)
244 #define ccw_is_idal(_ccw) ((_ccw)->flags & CCW_FLAG_IDA)
245 #define ccw_is_skip(_ccw) ((_ccw)->flags & CCW_FLAG_SKIP)
247 #define ccw_is_chain(_ccw) ((_ccw)->flags & (CCW_FLAG_CC | CCW_FLAG_DC))
250 * ccw_does_data_transfer()
252 * Determine whether a CCW will move any data, such that the guest pages
253 * would need to be pinned before performing the I/O.
255 * Returns 1 if yes, 0 if no.
257 static inline int ccw_does_data_transfer(struct ccw1
*ccw
)
259 /* If the count field is zero, then no data will be transferred */
263 /* If the command is a NOP, then no data will be transferred */
264 if (ccw_is_noop(ccw
))
267 /* If the skip flag is off, then data will be transferred */
268 if (!ccw_is_skip(ccw
))
272 * If the skip flag is on, it is only meaningful if the command
273 * code is a read, read backward, sense, or sense ID. In those
274 * cases, no data will be transferred.
276 if (ccw_is_read(ccw
) || ccw_is_read_backward(ccw
))
279 if (ccw_is_sense(ccw
))
282 /* The skip flag is on, but it is ignored for this command code. */
287 * is_cpa_within_range()
289 * @cpa: channel program address being questioned
290 * @head: address of the beginning of a CCW chain
291 * @len: number of CCWs within the chain
293 * Determine whether the address of a CCW (whether a new chain,
294 * or the target of a TIC) falls within a range (including the end points).
296 * Returns 1 if yes, 0 if no.
298 static inline int is_cpa_within_range(u32 cpa
, u32 head
, int len
)
300 u32 tail
= head
+ (len
- 1) * sizeof(struct ccw1
);
302 return (head
<= cpa
&& cpa
<= tail
);
305 static inline int is_tic_within_range(struct ccw1
*ccw
, u32 head
, int len
)
307 if (!ccw_is_tic(ccw
))
310 return is_cpa_within_range(ccw
->cda
, head
, len
);
313 static struct ccwchain
*ccwchain_alloc(struct channel_program
*cp
, int len
)
315 struct ccwchain
*chain
;
319 /* Make ccw address aligned to 8. */
320 size
= ((sizeof(*chain
) + 7L) & -8L) +
321 sizeof(*chain
->ch_ccw
) * len
+
322 sizeof(*chain
->ch_pa
) * len
;
323 chain
= kzalloc(size
, GFP_DMA
| GFP_KERNEL
);
327 data
= (u8
*)chain
+ ((sizeof(*chain
) + 7L) & -8L);
328 chain
->ch_ccw
= (struct ccw1
*)data
;
330 data
= (u8
*)(chain
->ch_ccw
) + sizeof(*chain
->ch_ccw
) * len
;
331 chain
->ch_pa
= (struct pfn_array
*)data
;
335 list_add_tail(&chain
->next
, &cp
->ccwchain_list
);
340 static void ccwchain_free(struct ccwchain
*chain
)
342 list_del(&chain
->next
);
346 /* Free resource for a ccw that allocated memory for its cda. */
347 static void ccwchain_cda_free(struct ccwchain
*chain
, int idx
)
349 struct ccw1
*ccw
= chain
->ch_ccw
+ idx
;
354 kfree((void *)(u64
)ccw
->cda
);
358 * ccwchain_calc_length - calculate the length of the ccw chain.
359 * @iova: guest physical address of the target ccw chain
360 * @cp: channel_program on which to perform the operation
362 * This is the chain length not considering any TICs.
363 * You need to do a new round for each TIC target.
365 * The program is also validated for absence of not yet supported
366 * indirect data addressing scenarios.
368 * Returns: the length of the ccw chain or -errno.
370 static int ccwchain_calc_length(u64 iova
, struct channel_program
*cp
)
372 struct ccw1
*ccw
= cp
->guest_cp
;
379 * As we don't want to fail direct addressing even if the
380 * orb specified one of the unsupported formats, we defer
381 * checking for IDAWs in unsupported formats to here.
383 if ((!cp
->orb
.cmd
.c64
|| cp
->orb
.cmd
.i2k
) && ccw_is_idal(ccw
))
387 * We want to keep counting if the current CCW has the
388 * command-chaining flag enabled, or if it is a TIC CCW
389 * that loops back into the current chain. The latter
390 * is used for device orientation, where the CCW PRIOR to
391 * the TIC can either jump to the TIC or a CCW immediately
392 * after the TIC, depending on the results of its operation.
394 if (!ccw_is_chain(ccw
) && !is_tic_within_range(ccw
, iova
, cnt
))
398 } while (cnt
< CCWCHAIN_LEN_MAX
+ 1);
400 if (cnt
== CCWCHAIN_LEN_MAX
+ 1)
406 static int tic_target_chain_exists(struct ccw1
*tic
, struct channel_program
*cp
)
408 struct ccwchain
*chain
;
411 list_for_each_entry(chain
, &cp
->ccwchain_list
, next
) {
412 ccw_head
= chain
->ch_iova
;
413 if (is_cpa_within_range(tic
->cda
, ccw_head
, chain
->ch_len
))
420 static int ccwchain_loop_tic(struct ccwchain
*chain
,
421 struct channel_program
*cp
);
423 static int ccwchain_handle_ccw(u32 cda
, struct channel_program
*cp
)
425 struct ccwchain
*chain
;
428 /* Copy 2K (the most we support today) of possible CCWs */
429 len
= copy_from_iova(cp
->mdev
, cp
->guest_cp
, cda
,
430 CCWCHAIN_LEN_MAX
* sizeof(struct ccw1
));
434 /* Convert any Format-0 CCWs to Format-1 */
435 if (!cp
->orb
.cmd
.fmt
)
436 convert_ccw0_to_ccw1(cp
->guest_cp
, CCWCHAIN_LEN_MAX
);
438 /* Count the CCWs in the current chain */
439 len
= ccwchain_calc_length(cda
, cp
);
443 /* Need alloc a new chain for this one. */
444 chain
= ccwchain_alloc(cp
, len
);
447 chain
->ch_iova
= cda
;
449 /* Copy the actual CCWs into the new chain */
450 memcpy(chain
->ch_ccw
, cp
->guest_cp
, len
* sizeof(struct ccw1
));
452 /* Loop for tics on this new chain. */
453 ret
= ccwchain_loop_tic(chain
, cp
);
456 ccwchain_free(chain
);
462 static int ccwchain_loop_tic(struct ccwchain
*chain
, struct channel_program
*cp
)
467 for (i
= 0; i
< chain
->ch_len
; i
++) {
468 tic
= chain
->ch_ccw
+ i
;
470 if (!ccw_is_tic(tic
))
473 /* May transfer to an existing chain. */
474 if (tic_target_chain_exists(tic
, cp
))
477 /* Build a ccwchain for the next segment */
478 ret
= ccwchain_handle_ccw(tic
->cda
, cp
);
486 static int ccwchain_fetch_tic(struct ccwchain
*chain
,
488 struct channel_program
*cp
)
490 struct ccw1
*ccw
= chain
->ch_ccw
+ idx
;
491 struct ccwchain
*iter
;
494 list_for_each_entry(iter
, &cp
->ccwchain_list
, next
) {
495 ccw_head
= iter
->ch_iova
;
496 if (is_cpa_within_range(ccw
->cda
, ccw_head
, iter
->ch_len
)) {
497 ccw
->cda
= (__u32
) (addr_t
) (((char *)iter
->ch_ccw
) +
498 (ccw
->cda
- ccw_head
));
506 static int ccwchain_fetch_direct(struct ccwchain
*chain
,
508 struct channel_program
*cp
)
511 struct pfn_array
*pa
;
513 unsigned long *idaws
;
516 int idaw_nr
, idal_len
;
519 ccw
= chain
->ch_ccw
+ idx
;
524 /* Calculate size of IDAL */
525 if (ccw_is_idal(ccw
)) {
526 /* Read first IDAW to see if it's 4K-aligned or not. */
527 /* All subsequent IDAws will be 4K-aligned. */
528 ret
= copy_from_iova(cp
->mdev
, &iova
, ccw
->cda
, sizeof(iova
));
534 idaw_nr
= idal_nr_words((void *)iova
, bytes
);
535 idal_len
= idaw_nr
* sizeof(*idaws
);
537 /* Allocate an IDAL from host storage */
538 idaws
= kcalloc(idaw_nr
, sizeof(*idaws
), GFP_DMA
| GFP_KERNEL
);
545 * Allocate an array of pfn's for pages to pin/translate.
546 * The number of pages is actually the count of the idaws
547 * required for the data transfer, since we only only support
550 pa
= chain
->ch_pa
+ idx
;
551 ret
= pfn_array_alloc(pa
, iova
, bytes
);
555 if (ccw_is_idal(ccw
)) {
556 /* Copy guest IDAL into host IDAL */
557 ret
= copy_from_iova(cp
->mdev
, idaws
, ccw
->cda
, idal_len
);
562 * Copy guest IDAWs into pfn_array, in case the memory they
563 * occupy is not contiguous.
565 for (i
= 0; i
< idaw_nr
; i
++)
566 pa
->pa_iova_pfn
[i
] = idaws
[i
] >> PAGE_SHIFT
;
569 * No action is required here; the iova addresses in pfn_array
570 * were initialized sequentially in pfn_array_alloc() beginning
571 * with the contents of ccw->cda.
575 if (ccw_does_data_transfer(ccw
)) {
576 ret
= pfn_array_pin(pa
, cp
->mdev
);
583 ccw
->cda
= (__u32
) virt_to_phys(idaws
);
584 ccw
->flags
|= CCW_FLAG_IDA
;
586 /* Populate the IDAL with pinned/translated addresses from pfn */
587 pfn_array_idal_create_words(pa
, idaws
);
592 pfn_array_unpin_free(pa
, cp
->mdev
);
602 * To reduce memory copy, we'll pin the cda page in memory,
603 * and to get rid of the cda 2G limitiaion of ccw1, we'll translate
604 * direct ccws to idal ccws.
606 static int ccwchain_fetch_one(struct ccwchain
*chain
,
608 struct channel_program
*cp
)
610 struct ccw1
*ccw
= chain
->ch_ccw
+ idx
;
613 return ccwchain_fetch_tic(chain
, idx
, cp
);
615 return ccwchain_fetch_direct(chain
, idx
, cp
);
619 * cp_init() - allocate ccwchains for a channel program.
620 * @cp: channel_program on which to perform the operation
621 * @mdev: the mediated device to perform pin/unpin operations
622 * @orb: control block for the channel program from the guest
624 * This creates one or more ccwchain(s), and copies the raw data of
625 * the target channel program from @orb->cmd.iova to the new ccwchain(s).
628 * 1. Supports only prefetch enabled mode.
629 * 2. Supports idal(c64) ccw chaining.
630 * 3. Supports 4k idaw.
633 * %0 on success and a negative error value on failure.
635 int cp_init(struct channel_program
*cp
, struct device
*mdev
, union orb
*orb
)
641 * Only support prefetch enable mode now.
646 INIT_LIST_HEAD(&cp
->ccwchain_list
);
647 memcpy(&cp
->orb
, orb
, sizeof(*orb
));
650 /* Build a ccwchain for the first CCW segment */
651 ret
= ccwchain_handle_ccw(orb
->cmd
.cpa
, cp
);
654 cp
->initialized
= true;
656 /* It is safe to force: if it was not set but idals used
657 * ccwchain_calc_length would have returned an error.
667 * cp_free() - free resources for channel program.
668 * @cp: channel_program on which to perform the operation
670 * This unpins the memory pages and frees the memory space occupied by
671 * @cp, which must have been returned by a previous call to cp_init().
672 * Otherwise, undefined behavior occurs.
674 void cp_free(struct channel_program
*cp
)
676 struct ccwchain
*chain
, *temp
;
679 if (!cp
->initialized
)
682 cp
->initialized
= false;
683 list_for_each_entry_safe(chain
, temp
, &cp
->ccwchain_list
, next
) {
684 for (i
= 0; i
< chain
->ch_len
; i
++) {
685 pfn_array_unpin_free(chain
->ch_pa
+ i
, cp
->mdev
);
686 ccwchain_cda_free(chain
, i
);
688 ccwchain_free(chain
);
693 * cp_prefetch() - translate a guest physical address channel program to
694 * a real-device runnable channel program.
695 * @cp: channel_program on which to perform the operation
697 * This function translates the guest-physical-address channel program
698 * and stores the result to ccwchain list. @cp must have been
699 * initialized by a previous call with cp_init(). Otherwise, undefined
701 * For each chain composing the channel program:
702 * - On entry ch_len holds the count of CCWs to be translated.
703 * - On exit ch_len is adjusted to the count of successfully translated CCWs.
704 * This allows cp_free to find in ch_len the count of CCWs to free in a chain.
706 * The S/390 CCW Translation APIS (prefixed by 'cp_') are introduced
707 * as helpers to do ccw chain translation inside the kernel. Basically
708 * they accept a channel program issued by a virtual machine, and
709 * translate the channel program to a real-device runnable channel
712 * These APIs will copy the ccws into kernel-space buffers, and update
713 * the guest phsical addresses with their corresponding host physical
714 * addresses. Then channel I/O device drivers could issue the
715 * translated channel program to real devices to perform an I/O
718 * These interfaces are designed to support translation only for
719 * channel programs, which are generated and formatted by a
720 * guest. Thus this will make it possible for things like VFIO to
721 * leverage the interfaces to passthrough a channel I/O mediated
724 * We support direct ccw chaining by translating them to idal ccws.
727 * %0 on success and a negative error value on failure.
729 int cp_prefetch(struct channel_program
*cp
)
731 struct ccwchain
*chain
;
734 /* this is an error in the caller */
735 if (!cp
->initialized
)
738 list_for_each_entry(chain
, &cp
->ccwchain_list
, next
) {
740 for (idx
= 0; idx
< len
; idx
++) {
741 ret
= ccwchain_fetch_one(chain
, idx
, cp
);
749 /* Only cleanup the chain elements that were actually translated. */
751 list_for_each_entry_continue(chain
, &cp
->ccwchain_list
, next
) {
758 * cp_get_orb() - get the orb of the channel program
759 * @cp: channel_program on which to perform the operation
760 * @intparm: new intparm for the returned orb
761 * @lpm: candidate value of the logical-path mask for the returned orb
763 * This function returns the address of the updated orb of the channel
764 * program. Channel I/O device drivers could use this orb to issue a
767 union orb
*cp_get_orb(struct channel_program
*cp
, u32 intparm
, u8 lpm
)
770 struct ccwchain
*chain
;
773 /* this is an error in the caller */
774 if (!cp
->initialized
)
779 orb
->cmd
.intparm
= intparm
;
781 orb
->cmd
.key
= PAGE_DEFAULT_KEY
>> 4;
783 if (orb
->cmd
.lpm
== 0)
786 chain
= list_first_entry(&cp
->ccwchain_list
, struct ccwchain
, next
);
788 orb
->cmd
.cpa
= (__u32
) __pa(cpa
);
794 * cp_update_scsw() - update scsw for a channel program.
795 * @cp: channel_program on which to perform the operation
796 * @scsw: I/O results of the channel program and also the target to be
799 * @scsw contains the I/O results of the channel program that pointed
800 * to by @cp. However what @scsw->cpa stores is a host physical
801 * address, which is meaningless for the guest, which is waiting for
804 * This function updates @scsw->cpa to its coressponding guest physical
807 void cp_update_scsw(struct channel_program
*cp
, union scsw
*scsw
)
809 struct ccwchain
*chain
;
810 u32 cpa
= scsw
->cmd
.cpa
;
813 if (!cp
->initialized
)
818 * For now, only update the cmd.cpa part. We may need to deal with
819 * other portions of the schib as well, even if we don't return them
820 * in the ioctl directly. Path status changes etc.
822 list_for_each_entry(chain
, &cp
->ccwchain_list
, next
) {
823 ccw_head
= (u32
)(u64
)chain
->ch_ccw
;
825 * On successful execution, cpa points just beyond the end
828 if (is_cpa_within_range(cpa
, ccw_head
, chain
->ch_len
+ 1)) {
830 * (cpa - ccw_head) is the offset value of the host
831 * physical ccw to its chain head.
832 * Adding this value to the guest physical ccw chain
833 * head gets us the guest cpa.
835 cpa
= chain
->ch_iova
+ (cpa
- ccw_head
);
844 * cp_iova_pinned() - check if an iova is pinned for a ccw chain.
845 * @cp: channel_program on which to perform the operation
846 * @iova: the iova to check
848 * If the @iova is currently pinned for the ccw chain, return true;
851 bool cp_iova_pinned(struct channel_program
*cp
, u64 iova
)
853 struct ccwchain
*chain
;
856 if (!cp
->initialized
)
859 list_for_each_entry(chain
, &cp
->ccwchain_list
, next
) {
860 for (i
= 0; i
< chain
->ch_len
; i
++)
861 if (pfn_array_iova_pinned(chain
->ch_pa
+ i
, iova
))