xref: /openbmc/linux/drivers/s390/cio/vfio_ccw_cp.c (revision c2863febd88bb2d0028eebcf0ee94b49859a06d3)
1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0
20a19e61eSDong Jia Shi /*
30a19e61eSDong Jia Shi  * channel program interfaces
40a19e61eSDong Jia Shi  *
50a19e61eSDong Jia Shi  * Copyright IBM Corp. 2017
60a19e61eSDong Jia Shi  *
70a19e61eSDong Jia Shi  * Author(s): Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com>
80a19e61eSDong Jia Shi  *            Xiao Feng Ren <renxiaof@linux.vnet.ibm.com>
90a19e61eSDong Jia Shi  */
100a19e61eSDong Jia Shi 
11725b94d7SJared Rossi #include <linux/ratelimit.h>
120a19e61eSDong Jia Shi #include <linux/mm.h>
130a19e61eSDong Jia Shi #include <linux/slab.h>
14*c2863febSNicolin Chen #include <linux/highmem.h>
150a19e61eSDong Jia Shi #include <linux/iommu.h>
160a19e61eSDong Jia Shi #include <linux/vfio.h>
170a19e61eSDong Jia Shi #include <asm/idals.h>
180a19e61eSDong Jia Shi 
190a19e61eSDong Jia Shi #include "vfio_ccw_cp.h"
200a587956SJason Gunthorpe #include "vfio_ccw_private.h"
210a19e61eSDong Jia Shi 
2213314605SNicolin Chen struct page_array {
2313314605SNicolin Chen 	/* Array that stores pages need to pin. */
2413314605SNicolin Chen 	dma_addr_t		*pa_iova;
2580c57f7aSDong Jia Shi 	/* Array that receives PFNs of the pages pinned. */
260a19e61eSDong Jia Shi 	unsigned long		*pa_pfn;
275c1cfb1cSDong Jia Shi 	/* Number of pages pinned from @pa_iova. */
280a19e61eSDong Jia Shi 	int			pa_nr;
290a19e61eSDong Jia Shi };
300a19e61eSDong Jia Shi 
310a19e61eSDong Jia Shi struct ccwchain {
320a19e61eSDong Jia Shi 	struct list_head	next;
330a19e61eSDong Jia Shi 	struct ccw1		*ch_ccw;
340a19e61eSDong Jia Shi 	/* Guest physical address of the current chain. */
350a19e61eSDong Jia Shi 	u64			ch_iova;
360a19e61eSDong Jia Shi 	/* Count of the valid ccws in chain. */
370a19e61eSDong Jia Shi 	int			ch_len;
380a19e61eSDong Jia Shi 	/* Pinned PAGEs for the original data. */
3913314605SNicolin Chen 	struct page_array	*ch_pa;
400a19e61eSDong Jia Shi };
410a19e61eSDong Jia Shi 
420a19e61eSDong Jia Shi /*
4313314605SNicolin Chen  * page_array_alloc() - alloc memory for page array
4413314605SNicolin Chen  * @pa: page_array on which to perform the operation
455c1cfb1cSDong Jia Shi  * @iova: target guest physical address
465c1cfb1cSDong Jia Shi  * @len: number of bytes that should be pinned from @iova
470a19e61eSDong Jia Shi  *
4813314605SNicolin Chen  * Attempt to allocate memory for page array.
490a19e61eSDong Jia Shi  *
5013314605SNicolin Chen  * Usage of page_array:
5113314605SNicolin Chen  * We expect (pa_nr == 0) and (pa_iova == NULL), any field in
525c1cfb1cSDong Jia Shi  * this structure will be filled in by this function.
530a19e61eSDong Jia Shi  *
540a19e61eSDong Jia Shi  * Returns:
5513314605SNicolin Chen  *         0 if page array is allocated
5613314605SNicolin Chen  *   -EINVAL if pa->pa_nr is not initially zero, or pa->pa_iova is not NULL
57e4f3f18bSEric Farman  *   -ENOMEM if alloc failed
580a19e61eSDong Jia Shi  */
5913314605SNicolin Chen static int page_array_alloc(struct page_array *pa, u64 iova, unsigned int len)
600a19e61eSDong Jia Shi {
61e4f3f18bSEric Farman 	int i;
620a19e61eSDong Jia Shi 
6313314605SNicolin Chen 	if (pa->pa_nr || pa->pa_iova)
640a19e61eSDong Jia Shi 		return -EINVAL;
650a19e61eSDong Jia Shi 
660a19e61eSDong Jia Shi 	pa->pa_nr = ((iova & ~PAGE_MASK) + len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
670a19e61eSDong Jia Shi 	if (!pa->pa_nr)
680a19e61eSDong Jia Shi 		return -EINVAL;
690a19e61eSDong Jia Shi 
7013314605SNicolin Chen 	pa->pa_iova = kcalloc(pa->pa_nr,
7113314605SNicolin Chen 			      sizeof(*pa->pa_iova) + sizeof(*pa->pa_pfn),
720a19e61eSDong Jia Shi 			      GFP_KERNEL);
7313314605SNicolin Chen 	if (unlikely(!pa->pa_iova)) {
74c1ab6926SFarhan Ali 		pa->pa_nr = 0;
750a19e61eSDong Jia Shi 		return -ENOMEM;
76c1ab6926SFarhan Ali 	}
7713314605SNicolin Chen 	pa->pa_pfn = (unsigned long *)&pa->pa_iova[pa->pa_nr];
780a19e61eSDong Jia Shi 
7913314605SNicolin Chen 	pa->pa_iova[0] = iova;
80c34a12e6SEric Farman 	pa->pa_pfn[0] = -1ULL;
81c34a12e6SEric Farman 	for (i = 1; i < pa->pa_nr; i++) {
8213314605SNicolin Chen 		pa->pa_iova[i] = pa->pa_iova[i - 1] + PAGE_SIZE;
83c34a12e6SEric Farman 		pa->pa_pfn[i] = -1ULL;
84c34a12e6SEric Farman 	}
850a19e61eSDong Jia Shi 
86e4f3f18bSEric Farman 	return 0;
87e4f3f18bSEric Farman }
88e4f3f18bSEric Farman 
89e4f3f18bSEric Farman /*
9013314605SNicolin Chen  * page_array_unpin() - Unpin user pages in memory
9113314605SNicolin Chen  * @pa: page_array on which to perform the operation
92cfedb3d5SNicolin Chen  * @vdev: the vfio device to perform the operation
93cfedb3d5SNicolin Chen  * @pa_nr: number of user pages to unpin
94cfedb3d5SNicolin Chen  *
95cfedb3d5SNicolin Chen  * Only unpin if any pages were pinned to begin with, i.e. pa_nr > 0,
96cfedb3d5SNicolin Chen  * otherwise only clear pa->pa_nr
97cfedb3d5SNicolin Chen  */
9813314605SNicolin Chen static void page_array_unpin(struct page_array *pa,
99cfedb3d5SNicolin Chen 			     struct vfio_device *vdev, int pa_nr)
100cfedb3d5SNicolin Chen {
101cfedb3d5SNicolin Chen 	int unpinned = 0, npage = 1;
102cfedb3d5SNicolin Chen 
103cfedb3d5SNicolin Chen 	while (unpinned < pa_nr) {
10413314605SNicolin Chen 		dma_addr_t *first = &pa->pa_iova[unpinned];
10513314605SNicolin Chen 		dma_addr_t *last = &first[npage];
106cfedb3d5SNicolin Chen 
107cfedb3d5SNicolin Chen 		if (unpinned + npage < pa_nr &&
10813314605SNicolin Chen 		    *first + npage * PAGE_SIZE == *last) {
109cfedb3d5SNicolin Chen 			npage++;
110cfedb3d5SNicolin Chen 			continue;
111cfedb3d5SNicolin Chen 		}
112cfedb3d5SNicolin Chen 
11313314605SNicolin Chen 		vfio_unpin_pages(vdev, *first, npage);
114cfedb3d5SNicolin Chen 		unpinned += npage;
115cfedb3d5SNicolin Chen 		npage = 1;
116cfedb3d5SNicolin Chen 	}
117cfedb3d5SNicolin Chen 
118cfedb3d5SNicolin Chen 	pa->pa_nr = 0;
119cfedb3d5SNicolin Chen }
120cfedb3d5SNicolin Chen 
121cfedb3d5SNicolin Chen /*
12213314605SNicolin Chen  * page_array_pin() - Pin user pages in memory
12313314605SNicolin Chen  * @pa: page_array on which to perform the operation
124e4f3f18bSEric Farman  * @mdev: the mediated device to perform pin operations
125e4f3f18bSEric Farman  *
126e4f3f18bSEric Farman  * Returns number of pages pinned upon success.
127e4f3f18bSEric Farman  * If the pin request partially succeeds, or fails completely,
128e4f3f18bSEric Farman  * all pages are left unpinned and a negative error value is returned.
129e4f3f18bSEric Farman  */
13013314605SNicolin Chen static int page_array_pin(struct page_array *pa, struct vfio_device *vdev)
131e4f3f18bSEric Farman {
132cfedb3d5SNicolin Chen 	int pinned = 0, npage = 1;
133e4f3f18bSEric Farman 	int ret = 0;
134e4f3f18bSEric Farman 
135cfedb3d5SNicolin Chen 	while (pinned < pa->pa_nr) {
13613314605SNicolin Chen 		dma_addr_t *first = &pa->pa_iova[pinned];
13713314605SNicolin Chen 		dma_addr_t *last = &first[npage];
1385c1cfb1cSDong Jia Shi 
139cfedb3d5SNicolin Chen 		if (pinned + npage < pa->pa_nr &&
14013314605SNicolin Chen 		    *first + npage * PAGE_SIZE == *last) {
141cfedb3d5SNicolin Chen 			npage++;
142cfedb3d5SNicolin Chen 			continue;
143cfedb3d5SNicolin Chen 		}
144cfedb3d5SNicolin Chen 
14513314605SNicolin Chen 		ret = vfio_pin_pages(vdev, *first, npage,
146cfedb3d5SNicolin Chen 				     IOMMU_READ | IOMMU_WRITE,
147cfedb3d5SNicolin Chen 				     &pa->pa_pfn[pinned]);
1485c1cfb1cSDong Jia Shi 		if (ret < 0) {
1495c1cfb1cSDong Jia Shi 			goto err_out;
150cfedb3d5SNicolin Chen 		} else if (ret > 0 && ret != npage) {
151cfedb3d5SNicolin Chen 			pinned += ret;
1520a19e61eSDong Jia Shi 			ret = -EINVAL;
1535c1cfb1cSDong Jia Shi 			goto err_out;
1545c1cfb1cSDong Jia Shi 		}
155cfedb3d5SNicolin Chen 		pinned += npage;
156cfedb3d5SNicolin Chen 		npage = 1;
157cfedb3d5SNicolin Chen 	}
1580a19e61eSDong Jia Shi 
1590a19e61eSDong Jia Shi 	return ret;
1605c1cfb1cSDong Jia Shi 
1615c1cfb1cSDong Jia Shi err_out:
16213314605SNicolin Chen 	page_array_unpin(pa, vdev, pinned);
1635c1cfb1cSDong Jia Shi 	return ret;
1645c1cfb1cSDong Jia Shi }
1655c1cfb1cSDong Jia Shi 
1665c1cfb1cSDong Jia Shi /* Unpin the pages before releasing the memory. */
16713314605SNicolin Chen static void page_array_unpin_free(struct page_array *pa, struct vfio_device *vdev)
1685c1cfb1cSDong Jia Shi {
16913314605SNicolin Chen 	page_array_unpin(pa, vdev, pa->pa_nr);
17013314605SNicolin Chen 	kfree(pa->pa_iova);
1710a19e61eSDong Jia Shi }
1720a19e61eSDong Jia Shi 
17313314605SNicolin Chen static bool page_array_iova_pinned(struct page_array *pa, unsigned long iova)
1740a19e61eSDong Jia Shi {
1750a19e61eSDong Jia Shi 	int i;
1760a19e61eSDong Jia Shi 
177e7eaf91bSEric Farman 	for (i = 0; i < pa->pa_nr; i++)
17813314605SNicolin Chen 		if (pa->pa_iova[i] == iova)
1790a19e61eSDong Jia Shi 			return true;
1800a19e61eSDong Jia Shi 
1810a19e61eSDong Jia Shi 	return false;
1820a19e61eSDong Jia Shi }
18313314605SNicolin Chen /* Create the list of IDAL words for a page_array. */
18413314605SNicolin Chen static inline void page_array_idal_create_words(struct page_array *pa,
1850a19e61eSDong Jia Shi 						unsigned long *idaws)
1860a19e61eSDong Jia Shi {
187e7eaf91bSEric Farman 	int i;
1880a19e61eSDong Jia Shi 
1890a19e61eSDong Jia Shi 	/*
1900a19e61eSDong Jia Shi 	 * Idal words (execept the first one) rely on the memory being 4k
1910a19e61eSDong Jia Shi 	 * aligned. If a user virtual address is 4K aligned, then it's
1920a19e61eSDong Jia Shi 	 * corresponding kernel physical address will also be 4K aligned. Thus
1930a19e61eSDong Jia Shi 	 * there will be no problem here to simply use the phys to create an
1940a19e61eSDong Jia Shi 	 * idaw.
1950a19e61eSDong Jia Shi 	 */
196e7eaf91bSEric Farman 
197e7eaf91bSEric Farman 	for (i = 0; i < pa->pa_nr; i++)
198e7eaf91bSEric Farman 		idaws[i] = pa->pa_pfn[i] << PAGE_SHIFT;
1998aabf0edSEric Farman 
2008aabf0edSEric Farman 	/* Adjust the first IDAW, since it may not start on a page boundary */
20113314605SNicolin Chen 	idaws[0] += pa->pa_iova[0] & (PAGE_SIZE - 1);
2020a19e61eSDong Jia Shi }
2030a19e61eSDong Jia Shi 
204dbd66558SCornelia Huck static void convert_ccw0_to_ccw1(struct ccw1 *source, unsigned long len)
2057f8e89a8SEric Farman {
2067f8e89a8SEric Farman 	struct ccw0 ccw0;
2077f8e89a8SEric Farman 	struct ccw1 *pccw1 = source;
2087f8e89a8SEric Farman 	int i;
2097f8e89a8SEric Farman 
2107f8e89a8SEric Farman 	for (i = 0; i < len; i++) {
2117f8e89a8SEric Farman 		ccw0 = *(struct ccw0 *)pccw1;
2127f8e89a8SEric Farman 		if ((pccw1->cmd_code & 0x0f) == CCW_CMD_TIC) {
2137f8e89a8SEric Farman 			pccw1->cmd_code = CCW_CMD_TIC;
2147f8e89a8SEric Farman 			pccw1->flags = 0;
2157f8e89a8SEric Farman 			pccw1->count = 0;
2167f8e89a8SEric Farman 		} else {
2177f8e89a8SEric Farman 			pccw1->cmd_code = ccw0.cmd_code;
2187f8e89a8SEric Farman 			pccw1->flags = ccw0.flags;
2197f8e89a8SEric Farman 			pccw1->count = ccw0.count;
2207f8e89a8SEric Farman 		}
2217f8e89a8SEric Farman 		pccw1->cda = ccw0.cda;
2227f8e89a8SEric Farman 		pccw1++;
2237f8e89a8SEric Farman 	}
2247f8e89a8SEric Farman }
2250a19e61eSDong Jia Shi 
2260a19e61eSDong Jia Shi /*
2270a19e61eSDong Jia Shi  * Within the domain (@mdev), copy @n bytes from a guest physical
2280a19e61eSDong Jia Shi  * address (@iova) to a host physical address (@to).
2290a19e61eSDong Jia Shi  */
2300a587956SJason Gunthorpe static long copy_from_iova(struct vfio_device *vdev, void *to, u64 iova,
2310a19e61eSDong Jia Shi 			   unsigned long n)
2320a19e61eSDong Jia Shi {
23313314605SNicolin Chen 	struct page_array pa = {0};
2340a19e61eSDong Jia Shi 	int i, ret;
2350a19e61eSDong Jia Shi 	unsigned long l, m;
2360a19e61eSDong Jia Shi 
23713314605SNicolin Chen 	ret = page_array_alloc(&pa, iova, n);
238e4f3f18bSEric Farman 	if (ret < 0)
2390a19e61eSDong Jia Shi 		return ret;
2400a19e61eSDong Jia Shi 
24113314605SNicolin Chen 	ret = page_array_pin(&pa, vdev);
242e4f3f18bSEric Farman 	if (ret < 0) {
24313314605SNicolin Chen 		page_array_unpin_free(&pa, vdev);
244e4f3f18bSEric Farman 		return ret;
245e4f3f18bSEric Farman 	}
246e4f3f18bSEric Farman 
2470a19e61eSDong Jia Shi 	l = n;
2480a19e61eSDong Jia Shi 	for (i = 0; i < pa.pa_nr; i++) {
249*c2863febSNicolin Chen 		struct page *page = pfn_to_page(pa.pa_pfn[i]);
250*c2863febSNicolin Chen 		void *from = kmap_local_page(page);
251*c2863febSNicolin Chen 
2520a19e61eSDong Jia Shi 		m = PAGE_SIZE;
2530a19e61eSDong Jia Shi 		if (i == 0) {
2540a19e61eSDong Jia Shi 			from += iova & (PAGE_SIZE - 1);
2550a19e61eSDong Jia Shi 			m -= iova & (PAGE_SIZE - 1);
2560a19e61eSDong Jia Shi 		}
2570a19e61eSDong Jia Shi 
2580a19e61eSDong Jia Shi 		m = min(l, m);
259*c2863febSNicolin Chen 		memcpy(to + (n - l), from, m);
260*c2863febSNicolin Chen 		kunmap_local(from);
2610a19e61eSDong Jia Shi 
2620a19e61eSDong Jia Shi 		l -= m;
2630a19e61eSDong Jia Shi 		if (l == 0)
2640a19e61eSDong Jia Shi 			break;
2650a19e61eSDong Jia Shi 	}
2660a19e61eSDong Jia Shi 
26713314605SNicolin Chen 	page_array_unpin_free(&pa, vdev);
2680a19e61eSDong Jia Shi 
2690a19e61eSDong Jia Shi 	return l;
2700a19e61eSDong Jia Shi }
2710a19e61eSDong Jia Shi 
2720a19e61eSDong Jia Shi /*
2730a19e61eSDong Jia Shi  * Helpers to operate ccwchain.
2740a19e61eSDong Jia Shi  */
2755d87fbf7SEric Farman #define ccw_is_read(_ccw) (((_ccw)->cmd_code & 0x03) == 0x02)
2765d87fbf7SEric Farman #define ccw_is_read_backward(_ccw) (((_ccw)->cmd_code & 0x0F) == 0x0C)
2775d87fbf7SEric Farman #define ccw_is_sense(_ccw) (((_ccw)->cmd_code & 0x0F) == CCW_CMD_BASIC_SENSE)
2785d87fbf7SEric Farman 
2790a19e61eSDong Jia Shi #define ccw_is_noop(_ccw) ((_ccw)->cmd_code == CCW_CMD_NOOP)
2800a19e61eSDong Jia Shi 
2810a19e61eSDong Jia Shi #define ccw_is_tic(_ccw) ((_ccw)->cmd_code == CCW_CMD_TIC)
2820a19e61eSDong Jia Shi 
2830a19e61eSDong Jia Shi #define ccw_is_idal(_ccw) ((_ccw)->flags & CCW_FLAG_IDA)
2845d87fbf7SEric Farman #define ccw_is_skip(_ccw) ((_ccw)->flags & CCW_FLAG_SKIP)
2850a19e61eSDong Jia Shi 
2860a19e61eSDong Jia Shi #define ccw_is_chain(_ccw) ((_ccw)->flags & (CCW_FLAG_CC | CCW_FLAG_DC))
2870a19e61eSDong Jia Shi 
28848bd0eeeSEric Farman /*
2895d87fbf7SEric Farman  * ccw_does_data_transfer()
2905d87fbf7SEric Farman  *
2915d87fbf7SEric Farman  * Determine whether a CCW will move any data, such that the guest pages
2925d87fbf7SEric Farman  * would need to be pinned before performing the I/O.
2935d87fbf7SEric Farman  *
2945d87fbf7SEric Farman  * Returns 1 if yes, 0 if no.
2955d87fbf7SEric Farman  */
2965d87fbf7SEric Farman static inline int ccw_does_data_transfer(struct ccw1 *ccw)
2975d87fbf7SEric Farman {
298453eac31SEric Farman 	/* If the count field is zero, then no data will be transferred */
299453eac31SEric Farman 	if (ccw->count == 0)
300453eac31SEric Farman 		return 0;
301453eac31SEric Farman 
3029b6e57e5SEric Farman 	/* If the command is a NOP, then no data will be transferred */
3039b6e57e5SEric Farman 	if (ccw_is_noop(ccw))
3049b6e57e5SEric Farman 		return 0;
3059b6e57e5SEric Farman 
3065d87fbf7SEric Farman 	/* If the skip flag is off, then data will be transferred */
3075d87fbf7SEric Farman 	if (!ccw_is_skip(ccw))
3085d87fbf7SEric Farman 		return 1;
3095d87fbf7SEric Farman 
3105d87fbf7SEric Farman 	/*
3115d87fbf7SEric Farman 	 * If the skip flag is on, it is only meaningful if the command
3125d87fbf7SEric Farman 	 * code is a read, read backward, sense, or sense ID.  In those
3135d87fbf7SEric Farman 	 * cases, no data will be transferred.
3145d87fbf7SEric Farman 	 */
3155d87fbf7SEric Farman 	if (ccw_is_read(ccw) || ccw_is_read_backward(ccw))
3165d87fbf7SEric Farman 		return 0;
3175d87fbf7SEric Farman 
3185d87fbf7SEric Farman 	if (ccw_is_sense(ccw))
3195d87fbf7SEric Farman 		return 0;
3205d87fbf7SEric Farman 
3215d87fbf7SEric Farman 	/* The skip flag is on, but it is ignored for this command code. */
3225d87fbf7SEric Farman 	return 1;
3235d87fbf7SEric Farman }
3245d87fbf7SEric Farman 
3255d87fbf7SEric Farman /*
32648bd0eeeSEric Farman  * is_cpa_within_range()
32748bd0eeeSEric Farman  *
32848bd0eeeSEric Farman  * @cpa: channel program address being questioned
32948bd0eeeSEric Farman  * @head: address of the beginning of a CCW chain
33048bd0eeeSEric Farman  * @len: number of CCWs within the chain
33148bd0eeeSEric Farman  *
33248bd0eeeSEric Farman  * Determine whether the address of a CCW (whether a new chain,
33348bd0eeeSEric Farman  * or the target of a TIC) falls within a range (including the end points).
33448bd0eeeSEric Farman  *
33548bd0eeeSEric Farman  * Returns 1 if yes, 0 if no.
33648bd0eeeSEric Farman  */
33748bd0eeeSEric Farman static inline int is_cpa_within_range(u32 cpa, u32 head, int len)
33848bd0eeeSEric Farman {
33948bd0eeeSEric Farman 	u32 tail = head + (len - 1) * sizeof(struct ccw1);
34048bd0eeeSEric Farman 
34148bd0eeeSEric Farman 	return (head <= cpa && cpa <= tail);
34248bd0eeeSEric Farman }
34348bd0eeeSEric Farman 
34448bd0eeeSEric Farman static inline int is_tic_within_range(struct ccw1 *ccw, u32 head, int len)
34548bd0eeeSEric Farman {
34648bd0eeeSEric Farman 	if (!ccw_is_tic(ccw))
34748bd0eeeSEric Farman 		return 0;
34848bd0eeeSEric Farman 
34948bd0eeeSEric Farman 	return is_cpa_within_range(ccw->cda, head, len);
35048bd0eeeSEric Farman }
35148bd0eeeSEric Farman 
3520a19e61eSDong Jia Shi static struct ccwchain *ccwchain_alloc(struct channel_program *cp, int len)
3530a19e61eSDong Jia Shi {
3540a19e61eSDong Jia Shi 	struct ccwchain *chain;
3550a19e61eSDong Jia Shi 	void *data;
3560a19e61eSDong Jia Shi 	size_t size;
3570a19e61eSDong Jia Shi 
3580a19e61eSDong Jia Shi 	/* Make ccw address aligned to 8. */
3590a19e61eSDong Jia Shi 	size = ((sizeof(*chain) + 7L) & -8L) +
3600a19e61eSDong Jia Shi 		sizeof(*chain->ch_ccw) * len +
361e7eaf91bSEric Farman 		sizeof(*chain->ch_pa) * len;
3620a19e61eSDong Jia Shi 	chain = kzalloc(size, GFP_DMA | GFP_KERNEL);
3630a19e61eSDong Jia Shi 	if (!chain)
3640a19e61eSDong Jia Shi 		return NULL;
3650a19e61eSDong Jia Shi 
3660a19e61eSDong Jia Shi 	data = (u8 *)chain + ((sizeof(*chain) + 7L) & -8L);
3670a19e61eSDong Jia Shi 	chain->ch_ccw = (struct ccw1 *)data;
3680a19e61eSDong Jia Shi 
3690a19e61eSDong Jia Shi 	data = (u8 *)(chain->ch_ccw) + sizeof(*chain->ch_ccw) * len;
37013314605SNicolin Chen 	chain->ch_pa = (struct page_array *)data;
3710a19e61eSDong Jia Shi 
3720a19e61eSDong Jia Shi 	chain->ch_len = len;
3730a19e61eSDong Jia Shi 
3740a19e61eSDong Jia Shi 	list_add_tail(&chain->next, &cp->ccwchain_list);
3750a19e61eSDong Jia Shi 
3760a19e61eSDong Jia Shi 	return chain;
3770a19e61eSDong Jia Shi }
3780a19e61eSDong Jia Shi 
3790a19e61eSDong Jia Shi static void ccwchain_free(struct ccwchain *chain)
3800a19e61eSDong Jia Shi {
3810a19e61eSDong Jia Shi 	list_del(&chain->next);
3820a19e61eSDong Jia Shi 	kfree(chain);
3830a19e61eSDong Jia Shi }
3840a19e61eSDong Jia Shi 
3850a19e61eSDong Jia Shi /* Free resource for a ccw that allocated memory for its cda. */
3860a19e61eSDong Jia Shi static void ccwchain_cda_free(struct ccwchain *chain, int idx)
3870a19e61eSDong Jia Shi {
3880a19e61eSDong Jia Shi 	struct ccw1 *ccw = chain->ch_ccw + idx;
3890a19e61eSDong Jia Shi 
3909b6e57e5SEric Farman 	if (ccw_is_tic(ccw))
391408358b5SJason J. Herne 		return;
3920a19e61eSDong Jia Shi 
3930a19e61eSDong Jia Shi 	kfree((void *)(u64)ccw->cda);
3940a19e61eSDong Jia Shi }
3950a19e61eSDong Jia Shi 
3960a19e61eSDong Jia Shi /**
3970a19e61eSDong Jia Shi  * ccwchain_calc_length - calculate the length of the ccw chain.
3980a19e61eSDong Jia Shi  * @iova: guest physical address of the target ccw chain
3990a19e61eSDong Jia Shi  * @cp: channel_program on which to perform the operation
4000a19e61eSDong Jia Shi  *
4010a19e61eSDong Jia Shi  * This is the chain length not considering any TICs.
4020a19e61eSDong Jia Shi  * You need to do a new round for each TIC target.
4030a19e61eSDong Jia Shi  *
404fb9e7880SHalil Pasic  * The program is also validated for absence of not yet supported
405fb9e7880SHalil Pasic  * indirect data addressing scenarios.
406fb9e7880SHalil Pasic  *
4070a19e61eSDong Jia Shi  * Returns: the length of the ccw chain or -errno.
4080a19e61eSDong Jia Shi  */
4090a19e61eSDong Jia Shi static int ccwchain_calc_length(u64 iova, struct channel_program *cp)
4100a19e61eSDong Jia Shi {
4111d897e47SEric Farman 	struct ccw1 *ccw = cp->guest_cp;
412ded563f3SEric Farman 	int cnt = 0;
4130a19e61eSDong Jia Shi 
4140a19e61eSDong Jia Shi 	do {
4150a19e61eSDong Jia Shi 		cnt++;
4160a19e61eSDong Jia Shi 
417fb9e7880SHalil Pasic 		/*
418fb9e7880SHalil Pasic 		 * As we don't want to fail direct addressing even if the
419fb9e7880SHalil Pasic 		 * orb specified one of the unsupported formats, we defer
420fb9e7880SHalil Pasic 		 * checking for IDAWs in unsupported formats to here.
421fb9e7880SHalil Pasic 		 */
4221d897e47SEric Farman 		if ((!cp->orb.cmd.c64 || cp->orb.cmd.i2k) && ccw_is_idal(ccw))
423fb9e7880SHalil Pasic 			return -EOPNOTSUPP;
424fb9e7880SHalil Pasic 
42548bd0eeeSEric Farman 		/*
42648bd0eeeSEric Farman 		 * We want to keep counting if the current CCW has the
42748bd0eeeSEric Farman 		 * command-chaining flag enabled, or if it is a TIC CCW
42848bd0eeeSEric Farman 		 * that loops back into the current chain.  The latter
42948bd0eeeSEric Farman 		 * is used for device orientation, where the CCW PRIOR to
43048bd0eeeSEric Farman 		 * the TIC can either jump to the TIC or a CCW immediately
43148bd0eeeSEric Farman 		 * after the TIC, depending on the results of its operation.
43248bd0eeeSEric Farman 		 */
43348bd0eeeSEric Farman 		if (!ccw_is_chain(ccw) && !is_tic_within_range(ccw, iova, cnt))
4340a19e61eSDong Jia Shi 			break;
4350a19e61eSDong Jia Shi 
4360a19e61eSDong Jia Shi 		ccw++;
4370a19e61eSDong Jia Shi 	} while (cnt < CCWCHAIN_LEN_MAX + 1);
4380a19e61eSDong Jia Shi 
4390a19e61eSDong Jia Shi 	if (cnt == CCWCHAIN_LEN_MAX + 1)
4400a19e61eSDong Jia Shi 		cnt = -EINVAL;
4410a19e61eSDong Jia Shi 
4420a19e61eSDong Jia Shi 	return cnt;
4430a19e61eSDong Jia Shi }
4440a19e61eSDong Jia Shi 
4450a19e61eSDong Jia Shi static int tic_target_chain_exists(struct ccw1 *tic, struct channel_program *cp)
4460a19e61eSDong Jia Shi {
4470a19e61eSDong Jia Shi 	struct ccwchain *chain;
4482904337fSEric Farman 	u32 ccw_head;
4490a19e61eSDong Jia Shi 
4500a19e61eSDong Jia Shi 	list_for_each_entry(chain, &cp->ccwchain_list, next) {
4510a19e61eSDong Jia Shi 		ccw_head = chain->ch_iova;
4522904337fSEric Farman 		if (is_cpa_within_range(tic->cda, ccw_head, chain->ch_len))
4530a19e61eSDong Jia Shi 			return 1;
4540a19e61eSDong Jia Shi 	}
4550a19e61eSDong Jia Shi 
4560a19e61eSDong Jia Shi 	return 0;
4570a19e61eSDong Jia Shi }
4580a19e61eSDong Jia Shi 
4590a19e61eSDong Jia Shi static int ccwchain_loop_tic(struct ccwchain *chain,
4600a19e61eSDong Jia Shi 			     struct channel_program *cp);
4610a19e61eSDong Jia Shi 
462363fe5f7SEric Farman static int ccwchain_handle_ccw(u32 cda, struct channel_program *cp)
4630a19e61eSDong Jia Shi {
4640a587956SJason Gunthorpe 	struct vfio_device *vdev =
4650a587956SJason Gunthorpe 		&container_of(cp, struct vfio_ccw_private, cp)->vdev;
4660a19e61eSDong Jia Shi 	struct ccwchain *chain;
4678b515be5SFarhan Ali 	int len, ret;
4680a19e61eSDong Jia Shi 
469ded563f3SEric Farman 	/* Copy 2K (the most we support today) of possible CCWs */
4700a587956SJason Gunthorpe 	len = copy_from_iova(vdev, cp->guest_cp, cda,
4715223bee8SEric Farman 			     CCWCHAIN_LEN_MAX * sizeof(struct ccw1));
472ded563f3SEric Farman 	if (len)
473ded563f3SEric Farman 		return len;
474ded563f3SEric Farman 
4757f8e89a8SEric Farman 	/* Convert any Format-0 CCWs to Format-1 */
4767f8e89a8SEric Farman 	if (!cp->orb.cmd.fmt)
477c382cbc6SEric Farman 		convert_ccw0_to_ccw1(cp->guest_cp, CCWCHAIN_LEN_MAX);
4787f8e89a8SEric Farman 
479ded563f3SEric Farman 	/* Count the CCWs in the current chain */
480363fe5f7SEric Farman 	len = ccwchain_calc_length(cda, cp);
4810a19e61eSDong Jia Shi 	if (len < 0)
4820a19e61eSDong Jia Shi 		return len;
4830a19e61eSDong Jia Shi 
4840a19e61eSDong Jia Shi 	/* Need alloc a new chain for this one. */
4850a19e61eSDong Jia Shi 	chain = ccwchain_alloc(cp, len);
4860a19e61eSDong Jia Shi 	if (!chain)
4870a19e61eSDong Jia Shi 		return -ENOMEM;
488363fe5f7SEric Farman 	chain->ch_iova = cda;
4890a19e61eSDong Jia Shi 
49062465902SEric Farman 	/* Copy the actual CCWs into the new chain */
49162465902SEric Farman 	memcpy(chain->ch_ccw, cp->guest_cp, len * sizeof(struct ccw1));
4920a19e61eSDong Jia Shi 
4930a19e61eSDong Jia Shi 	/* Loop for tics on this new chain. */
4948b515be5SFarhan Ali 	ret = ccwchain_loop_tic(chain, cp);
4958b515be5SFarhan Ali 
4968b515be5SFarhan Ali 	if (ret)
4978b515be5SFarhan Ali 		ccwchain_free(chain);
4988b515be5SFarhan Ali 
4998b515be5SFarhan Ali 	return ret;
5000a19e61eSDong Jia Shi }
5010a19e61eSDong Jia Shi 
5020a19e61eSDong Jia Shi /* Loop for TICs. */
5030a19e61eSDong Jia Shi static int ccwchain_loop_tic(struct ccwchain *chain, struct channel_program *cp)
5040a19e61eSDong Jia Shi {
5050a19e61eSDong Jia Shi 	struct ccw1 *tic;
5060a19e61eSDong Jia Shi 	int i, ret;
5070a19e61eSDong Jia Shi 
5080a19e61eSDong Jia Shi 	for (i = 0; i < chain->ch_len; i++) {
5090a19e61eSDong Jia Shi 		tic = chain->ch_ccw + i;
5100a19e61eSDong Jia Shi 
5110a19e61eSDong Jia Shi 		if (!ccw_is_tic(tic))
5120a19e61eSDong Jia Shi 			continue;
5130a19e61eSDong Jia Shi 
514e64bd689SEric Farman 		/* May transfer to an existing chain. */
515e64bd689SEric Farman 		if (tic_target_chain_exists(tic, cp))
516e64bd689SEric Farman 			continue;
517e64bd689SEric Farman 
518363fe5f7SEric Farman 		/* Build a ccwchain for the next segment */
519363fe5f7SEric Farman 		ret = ccwchain_handle_ccw(tic->cda, cp);
5200a19e61eSDong Jia Shi 		if (ret)
5210a19e61eSDong Jia Shi 			return ret;
5220a19e61eSDong Jia Shi 	}
5230a19e61eSDong Jia Shi 
5240a19e61eSDong Jia Shi 	return 0;
5250a19e61eSDong Jia Shi }
5260a19e61eSDong Jia Shi 
5270a19e61eSDong Jia Shi static int ccwchain_fetch_tic(struct ccwchain *chain,
5280a19e61eSDong Jia Shi 			      int idx,
5290a19e61eSDong Jia Shi 			      struct channel_program *cp)
5300a19e61eSDong Jia Shi {
5310a19e61eSDong Jia Shi 	struct ccw1 *ccw = chain->ch_ccw + idx;
5320a19e61eSDong Jia Shi 	struct ccwchain *iter;
5332904337fSEric Farman 	u32 ccw_head;
5340a19e61eSDong Jia Shi 
5350a19e61eSDong Jia Shi 	list_for_each_entry(iter, &cp->ccwchain_list, next) {
5360a19e61eSDong Jia Shi 		ccw_head = iter->ch_iova;
5372904337fSEric Farman 		if (is_cpa_within_range(ccw->cda, ccw_head, iter->ch_len)) {
538c389377cSJason J. Herne 			ccw->cda = (__u32) (addr_t) (((char *)iter->ch_ccw) +
5390a19e61eSDong Jia Shi 						     (ccw->cda - ccw_head));
5400a19e61eSDong Jia Shi 			return 0;
5410a19e61eSDong Jia Shi 		}
5420a19e61eSDong Jia Shi 	}
5430a19e61eSDong Jia Shi 
5440a19e61eSDong Jia Shi 	return -EFAULT;
5450a19e61eSDong Jia Shi }
5460a19e61eSDong Jia Shi 
5470a19e61eSDong Jia Shi static int ccwchain_fetch_direct(struct ccwchain *chain,
5480a19e61eSDong Jia Shi 				 int idx,
5490a19e61eSDong Jia Shi 				 struct channel_program *cp)
5500a19e61eSDong Jia Shi {
5510a587956SJason Gunthorpe 	struct vfio_device *vdev =
5520a587956SJason Gunthorpe 		&container_of(cp, struct vfio_ccw_private, cp)->vdev;
5530a19e61eSDong Jia Shi 	struct ccw1 *ccw;
55413314605SNicolin Chen 	struct page_array *pa;
55501aa26c6SEric Farman 	u64 iova;
5560a19e61eSDong Jia Shi 	unsigned long *idaws;
5576238f921SDong Jia Shi 	int ret;
558453eac31SEric Farman 	int bytes = 1;
55901aa26c6SEric Farman 	int idaw_nr, idal_len;
56001aa26c6SEric Farman 	int i;
5610a19e61eSDong Jia Shi 
5620a19e61eSDong Jia Shi 	ccw = chain->ch_ccw + idx;
5630a19e61eSDong Jia Shi 
564e8573b39SEric Farman 	if (ccw->count)
565453eac31SEric Farman 		bytes = ccw->count;
566e8573b39SEric Farman 
567e8573b39SEric Farman 	/* Calculate size of IDAL */
56801aa26c6SEric Farman 	if (ccw_is_idal(ccw)) {
56901aa26c6SEric Farman 		/* Read first IDAW to see if it's 4K-aligned or not. */
57001aa26c6SEric Farman 		/* All subsequent IDAws will be 4K-aligned. */
5710a587956SJason Gunthorpe 		ret = copy_from_iova(vdev, &iova, ccw->cda, sizeof(iova));
57201aa26c6SEric Farman 		if (ret)
57301aa26c6SEric Farman 			return ret;
57401aa26c6SEric Farman 	} else {
57501aa26c6SEric Farman 		iova = ccw->cda;
57601aa26c6SEric Farman 	}
57701aa26c6SEric Farman 	idaw_nr = idal_nr_words((void *)iova, bytes);
57801aa26c6SEric Farman 	idal_len = idaw_nr * sizeof(*idaws);
579e8573b39SEric Farman 
580e8573b39SEric Farman 	/* Allocate an IDAL from host storage */
581e8573b39SEric Farman 	idaws = kcalloc(idaw_nr, sizeof(*idaws), GFP_DMA | GFP_KERNEL);
582e8573b39SEric Farman 	if (!idaws) {
583e8573b39SEric Farman 		ret = -ENOMEM;
584e8573b39SEric Farman 		goto out_init;
5854cebc5d6SDong Jia Shi 	}
5864cebc5d6SDong Jia Shi 
5870a19e61eSDong Jia Shi 	/*
58813314605SNicolin Chen 	 * Allocate an array of pages to pin/translate.
58901aa26c6SEric Farman 	 * The number of pages is actually the count of the idaws
59001aa26c6SEric Farman 	 * required for the data transfer, since we only only support
59101aa26c6SEric Farman 	 * 4K IDAWs today.
5920a19e61eSDong Jia Shi 	 */
593e7eaf91bSEric Farman 	pa = chain->ch_pa + idx;
59413314605SNicolin Chen 	ret = page_array_alloc(pa, iova, bytes);
595e4f3f18bSEric Farman 	if (ret < 0)
596e8573b39SEric Farman 		goto out_free_idaws;
597e4f3f18bSEric Farman 
59801aa26c6SEric Farman 	if (ccw_is_idal(ccw)) {
59901aa26c6SEric Farman 		/* Copy guest IDAL into host IDAL */
6000a587956SJason Gunthorpe 		ret = copy_from_iova(vdev, idaws, ccw->cda, idal_len);
60101aa26c6SEric Farman 		if (ret)
60201aa26c6SEric Farman 			goto out_unpin;
60301aa26c6SEric Farman 
60401aa26c6SEric Farman 		/*
60513314605SNicolin Chen 		 * Copy guest IDAWs into page_array, in case the memory they
60601aa26c6SEric Farman 		 * occupy is not contiguous.
60701aa26c6SEric Farman 		 */
60801aa26c6SEric Farman 		for (i = 0; i < idaw_nr; i++)
60913314605SNicolin Chen 			pa->pa_iova[i] = idaws[i];
61001aa26c6SEric Farman 	} else {
61101aa26c6SEric Farman 		/*
61213314605SNicolin Chen 		 * No action is required here; the iova addresses in page_array
61313314605SNicolin Chen 		 * were initialized sequentially in page_array_alloc() beginning
61401aa26c6SEric Farman 		 * with the contents of ccw->cda.
61501aa26c6SEric Farman 		 */
61601aa26c6SEric Farman 	}
61701aa26c6SEric Farman 
6185d87fbf7SEric Farman 	if (ccw_does_data_transfer(ccw)) {
61913314605SNicolin Chen 		ret = page_array_pin(pa, vdev);
6206238f921SDong Jia Shi 		if (ret < 0)
621806212f9SEric Farman 			goto out_unpin;
6225d87fbf7SEric Farman 	} else {
623e7eaf91bSEric Farman 		pa->pa_nr = 0;
6245d87fbf7SEric Farman 	}
6250a19e61eSDong Jia Shi 
6260a19e61eSDong Jia Shi 	ccw->cda = (__u32) virt_to_phys(idaws);
6270a19e61eSDong Jia Shi 	ccw->flags |= CCW_FLAG_IDA;
6280a19e61eSDong Jia Shi 
62913314605SNicolin Chen 	/* Populate the IDAL with pinned/translated addresses from page */
63013314605SNicolin Chen 	page_array_idal_create_words(pa, idaws);
6310a19e61eSDong Jia Shi 
6320a19e61eSDong Jia Shi 	return 0;
6336238f921SDong Jia Shi 
6346238f921SDong Jia Shi out_unpin:
63513314605SNicolin Chen 	page_array_unpin_free(pa, vdev);
636e8573b39SEric Farman out_free_idaws:
637e8573b39SEric Farman 	kfree(idaws);
6386238f921SDong Jia Shi out_init:
6396238f921SDong Jia Shi 	ccw->cda = 0;
6406238f921SDong Jia Shi 	return ret;
6410a19e61eSDong Jia Shi }
6420a19e61eSDong Jia Shi 
6430a19e61eSDong Jia Shi /*
6440a19e61eSDong Jia Shi  * Fetch one ccw.
6450a19e61eSDong Jia Shi  * To reduce memory copy, we'll pin the cda page in memory,
6460a19e61eSDong Jia Shi  * and to get rid of the cda 2G limitiaion of ccw1, we'll translate
6470a19e61eSDong Jia Shi  * direct ccws to idal ccws.
6480a19e61eSDong Jia Shi  */
6490a19e61eSDong Jia Shi static int ccwchain_fetch_one(struct ccwchain *chain,
6500a19e61eSDong Jia Shi 			      int idx,
6510a19e61eSDong Jia Shi 			      struct channel_program *cp)
6520a19e61eSDong Jia Shi {
6530a19e61eSDong Jia Shi 	struct ccw1 *ccw = chain->ch_ccw + idx;
6540a19e61eSDong Jia Shi 
6550a19e61eSDong Jia Shi 	if (ccw_is_tic(ccw))
6560a19e61eSDong Jia Shi 		return ccwchain_fetch_tic(chain, idx, cp);
6570a19e61eSDong Jia Shi 
6580a19e61eSDong Jia Shi 	return ccwchain_fetch_direct(chain, idx, cp);
6590a19e61eSDong Jia Shi }
6600a19e61eSDong Jia Shi 
6610a19e61eSDong Jia Shi /**
6620a19e61eSDong Jia Shi  * cp_init() - allocate ccwchains for a channel program.
6630a19e61eSDong Jia Shi  * @cp: channel_program on which to perform the operation
6640a19e61eSDong Jia Shi  * @mdev: the mediated device to perform pin/unpin operations
6650a19e61eSDong Jia Shi  * @orb: control block for the channel program from the guest
6660a19e61eSDong Jia Shi  *
6670a19e61eSDong Jia Shi  * This creates one or more ccwchain(s), and copies the raw data of
6680a19e61eSDong Jia Shi  * the target channel program from @orb->cmd.iova to the new ccwchain(s).
6690a19e61eSDong Jia Shi  *
6700a19e61eSDong Jia Shi  * Limitations:
671725b94d7SJared Rossi  * 1. Supports idal(c64) ccw chaining.
672725b94d7SJared Rossi  * 2. Supports 4k idaw.
6730a19e61eSDong Jia Shi  *
6740a19e61eSDong Jia Shi  * Returns:
6750a19e61eSDong Jia Shi  *   %0 on success and a negative error value on failure.
6760a19e61eSDong Jia Shi  */
6770a587956SJason Gunthorpe int cp_init(struct channel_program *cp, union orb *orb)
6780a19e61eSDong Jia Shi {
6790a587956SJason Gunthorpe 	struct vfio_device *vdev =
6800a587956SJason Gunthorpe 		&container_of(cp, struct vfio_ccw_private, cp)->vdev;
681725b94d7SJared Rossi 	/* custom ratelimit used to avoid flood during guest IPL */
682725b94d7SJared Rossi 	static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 1);
68399afcb05SEric Farman 	int ret;
6840a19e61eSDong Jia Shi 
685c6c82e0cSEric Farman 	/* this is an error in the caller */
686c6c82e0cSEric Farman 	if (cp->initialized)
687c6c82e0cSEric Farman 		return -EBUSY;
688c6c82e0cSEric Farman 
6890a19e61eSDong Jia Shi 	/*
690725b94d7SJared Rossi 	 * We only support prefetching the channel program. We assume all channel
691725b94d7SJared Rossi 	 * programs executed by supported guests likewise support prefetching.
692725b94d7SJared Rossi 	 * Executing a channel program that does not specify prefetching will
693725b94d7SJared Rossi 	 * typically not cause an error, but a warning is issued to help identify
694725b94d7SJared Rossi 	 * the problem if something does break.
6950a19e61eSDong Jia Shi 	 */
696725b94d7SJared Rossi 	if (!orb->cmd.pfch && __ratelimit(&ratelimit_state))
6970a587956SJason Gunthorpe 		dev_warn(
6980a587956SJason Gunthorpe 			vdev->dev,
6990a587956SJason Gunthorpe 			"Prefetching channel program even though prefetch not specified in ORB");
7000a19e61eSDong Jia Shi 
7010a19e61eSDong Jia Shi 	INIT_LIST_HEAD(&cp->ccwchain_list);
7020a19e61eSDong Jia Shi 	memcpy(&cp->orb, orb, sizeof(*orb));
7030a19e61eSDong Jia Shi 
70499afcb05SEric Farman 	/* Build a ccwchain for the first CCW segment */
70599afcb05SEric Farman 	ret = ccwchain_handle_ccw(orb->cmd.cpa, cp);
70699afcb05SEric Farman 
707c9f597a4SFarhan Ali 	if (!ret) {
708c9f597a4SFarhan Ali 		cp->initialized = true;
709c9f597a4SFarhan Ali 
710c9f597a4SFarhan Ali 		/* It is safe to force: if it was not set but idals used
711c9f597a4SFarhan Ali 		 * ccwchain_calc_length would have returned an error.
712fb9e7880SHalil Pasic 		 */
713fb9e7880SHalil Pasic 		cp->orb.cmd.c64 = 1;
714c9f597a4SFarhan Ali 	}
71571189f26SCornelia Huck 
7160a19e61eSDong Jia Shi 	return ret;
7170a19e61eSDong Jia Shi }
7180a19e61eSDong Jia Shi 
7190a19e61eSDong Jia Shi 
7200a19e61eSDong Jia Shi /**
7210a19e61eSDong Jia Shi  * cp_free() - free resources for channel program.
7220a19e61eSDong Jia Shi  * @cp: channel_program on which to perform the operation
7230a19e61eSDong Jia Shi  *
7240a19e61eSDong Jia Shi  * This unpins the memory pages and frees the memory space occupied by
7250a19e61eSDong Jia Shi  * @cp, which must have been returned by a previous call to cp_init().
7260a19e61eSDong Jia Shi  * Otherwise, undefined behavior occurs.
7270a19e61eSDong Jia Shi  */
7280a19e61eSDong Jia Shi void cp_free(struct channel_program *cp)
7290a19e61eSDong Jia Shi {
7300a587956SJason Gunthorpe 	struct vfio_device *vdev =
7310a587956SJason Gunthorpe 		&container_of(cp, struct vfio_ccw_private, cp)->vdev;
732812271b9SEric Farman 	struct ccwchain *chain, *temp;
733812271b9SEric Farman 	int i;
734812271b9SEric Farman 
735812271b9SEric Farman 	if (!cp->initialized)
736812271b9SEric Farman 		return;
737812271b9SEric Farman 
738812271b9SEric Farman 	cp->initialized = false;
739812271b9SEric Farman 	list_for_each_entry_safe(chain, temp, &cp->ccwchain_list, next) {
740812271b9SEric Farman 		for (i = 0; i < chain->ch_len; i++) {
74113314605SNicolin Chen 			page_array_unpin_free(chain->ch_pa + i, vdev);
742812271b9SEric Farman 			ccwchain_cda_free(chain, i);
743812271b9SEric Farman 		}
744812271b9SEric Farman 		ccwchain_free(chain);
745812271b9SEric Farman 	}
7460a19e61eSDong Jia Shi }
7470a19e61eSDong Jia Shi 
7480a19e61eSDong Jia Shi /**
7490a19e61eSDong Jia Shi  * cp_prefetch() - translate a guest physical address channel program to
7500a19e61eSDong Jia Shi  *                 a real-device runnable channel program.
7510a19e61eSDong Jia Shi  * @cp: channel_program on which to perform the operation
7520a19e61eSDong Jia Shi  *
7530a19e61eSDong Jia Shi  * This function translates the guest-physical-address channel program
7540a19e61eSDong Jia Shi  * and stores the result to ccwchain list. @cp must have been
7550a19e61eSDong Jia Shi  * initialized by a previous call with cp_init(). Otherwise, undefined
7560a19e61eSDong Jia Shi  * behavior occurs.
757d66a7355SHalil Pasic  * For each chain composing the channel program:
758d66a7355SHalil Pasic  * - On entry ch_len holds the count of CCWs to be translated.
759d66a7355SHalil Pasic  * - On exit ch_len is adjusted to the count of successfully translated CCWs.
760d66a7355SHalil Pasic  * This allows cp_free to find in ch_len the count of CCWs to free in a chain.
7610a19e61eSDong Jia Shi  *
7620a19e61eSDong Jia Shi  * The S/390 CCW Translation APIS (prefixed by 'cp_') are introduced
7630a19e61eSDong Jia Shi  * as helpers to do ccw chain translation inside the kernel. Basically
7640a19e61eSDong Jia Shi  * they accept a channel program issued by a virtual machine, and
7650a19e61eSDong Jia Shi  * translate the channel program to a real-device runnable channel
7660a19e61eSDong Jia Shi  * program.
7670a19e61eSDong Jia Shi  *
7680a19e61eSDong Jia Shi  * These APIs will copy the ccws into kernel-space buffers, and update
7690a19e61eSDong Jia Shi  * the guest phsical addresses with their corresponding host physical
7700a19e61eSDong Jia Shi  * addresses.  Then channel I/O device drivers could issue the
7710a19e61eSDong Jia Shi  * translated channel program to real devices to perform an I/O
7720a19e61eSDong Jia Shi  * operation.
7730a19e61eSDong Jia Shi  *
7740a19e61eSDong Jia Shi  * These interfaces are designed to support translation only for
7750a19e61eSDong Jia Shi  * channel programs, which are generated and formatted by a
7760a19e61eSDong Jia Shi  * guest. Thus this will make it possible for things like VFIO to
7770a19e61eSDong Jia Shi  * leverage the interfaces to passthrough a channel I/O mediated
7780a19e61eSDong Jia Shi  * device in QEMU.
7790a19e61eSDong Jia Shi  *
7800a19e61eSDong Jia Shi  * We support direct ccw chaining by translating them to idal ccws.
7810a19e61eSDong Jia Shi  *
7820a19e61eSDong Jia Shi  * Returns:
7830a19e61eSDong Jia Shi  *   %0 on success and a negative error value on failure.
7840a19e61eSDong Jia Shi  */
7850a19e61eSDong Jia Shi int cp_prefetch(struct channel_program *cp)
7860a19e61eSDong Jia Shi {
7870a19e61eSDong Jia Shi 	struct ccwchain *chain;
7880a19e61eSDong Jia Shi 	int len, idx, ret;
7890a19e61eSDong Jia Shi 
79071189f26SCornelia Huck 	/* this is an error in the caller */
79171189f26SCornelia Huck 	if (!cp->initialized)
79271189f26SCornelia Huck 		return -EINVAL;
79371189f26SCornelia Huck 
7940a19e61eSDong Jia Shi 	list_for_each_entry(chain, &cp->ccwchain_list, next) {
7950a19e61eSDong Jia Shi 		len = chain->ch_len;
7960a19e61eSDong Jia Shi 		for (idx = 0; idx < len; idx++) {
7970a19e61eSDong Jia Shi 			ret = ccwchain_fetch_one(chain, idx, cp);
7980a19e61eSDong Jia Shi 			if (ret)
799d66a7355SHalil Pasic 				goto out_err;
8000a19e61eSDong Jia Shi 		}
8010a19e61eSDong Jia Shi 	}
8020a19e61eSDong Jia Shi 
8030a19e61eSDong Jia Shi 	return 0;
804d66a7355SHalil Pasic out_err:
805d66a7355SHalil Pasic 	/* Only cleanup the chain elements that were actually translated. */
806d66a7355SHalil Pasic 	chain->ch_len = idx;
807d66a7355SHalil Pasic 	list_for_each_entry_continue(chain, &cp->ccwchain_list, next) {
808d66a7355SHalil Pasic 		chain->ch_len = 0;
809d66a7355SHalil Pasic 	}
810d66a7355SHalil Pasic 	return ret;
8110a19e61eSDong Jia Shi }
8120a19e61eSDong Jia Shi 
8130a19e61eSDong Jia Shi /**
8140a19e61eSDong Jia Shi  * cp_get_orb() - get the orb of the channel program
8150a19e61eSDong Jia Shi  * @cp: channel_program on which to perform the operation
8160a19e61eSDong Jia Shi  * @intparm: new intparm for the returned orb
8170a19e61eSDong Jia Shi  * @lpm: candidate value of the logical-path mask for the returned orb
8180a19e61eSDong Jia Shi  *
8190a19e61eSDong Jia Shi  * This function returns the address of the updated orb of the channel
8200a19e61eSDong Jia Shi  * program. Channel I/O device drivers could use this orb to issue a
8210a19e61eSDong Jia Shi  * ssch.
8220a19e61eSDong Jia Shi  */
8230a19e61eSDong Jia Shi union orb *cp_get_orb(struct channel_program *cp, u32 intparm, u8 lpm)
8240a19e61eSDong Jia Shi {
8250a19e61eSDong Jia Shi 	union orb *orb;
8260a19e61eSDong Jia Shi 	struct ccwchain *chain;
8270a19e61eSDong Jia Shi 	struct ccw1 *cpa;
8280a19e61eSDong Jia Shi 
82971189f26SCornelia Huck 	/* this is an error in the caller */
83071189f26SCornelia Huck 	if (!cp->initialized)
83171189f26SCornelia Huck 		return NULL;
83271189f26SCornelia Huck 
8330a19e61eSDong Jia Shi 	orb = &cp->orb;
8340a19e61eSDong Jia Shi 
8350a19e61eSDong Jia Shi 	orb->cmd.intparm = intparm;
8360a19e61eSDong Jia Shi 	orb->cmd.fmt = 1;
8370a19e61eSDong Jia Shi 	orb->cmd.key = PAGE_DEFAULT_KEY >> 4;
8380a19e61eSDong Jia Shi 
8390a19e61eSDong Jia Shi 	if (orb->cmd.lpm == 0)
8400a19e61eSDong Jia Shi 		orb->cmd.lpm = lpm;
8410a19e61eSDong Jia Shi 
8420a19e61eSDong Jia Shi 	chain = list_first_entry(&cp->ccwchain_list, struct ccwchain, next);
8430a19e61eSDong Jia Shi 	cpa = chain->ch_ccw;
8440a19e61eSDong Jia Shi 	orb->cmd.cpa = (__u32) __pa(cpa);
8450a19e61eSDong Jia Shi 
8460a19e61eSDong Jia Shi 	return orb;
8470a19e61eSDong Jia Shi }
8480a19e61eSDong Jia Shi 
8490a19e61eSDong Jia Shi /**
8500a19e61eSDong Jia Shi  * cp_update_scsw() - update scsw for a channel program.
8510a19e61eSDong Jia Shi  * @cp: channel_program on which to perform the operation
8520a19e61eSDong Jia Shi  * @scsw: I/O results of the channel program and also the target to be
8530a19e61eSDong Jia Shi  *        updated
8540a19e61eSDong Jia Shi  *
8550a19e61eSDong Jia Shi  * @scsw contains the I/O results of the channel program that pointed
8560a19e61eSDong Jia Shi  * to by @cp. However what @scsw->cpa stores is a host physical
8570a19e61eSDong Jia Shi  * address, which is meaningless for the guest, which is waiting for
8580a19e61eSDong Jia Shi  * the I/O results.
8590a19e61eSDong Jia Shi  *
8600a19e61eSDong Jia Shi  * This function updates @scsw->cpa to its coressponding guest physical
8610a19e61eSDong Jia Shi  * address.
8620a19e61eSDong Jia Shi  */
8630a19e61eSDong Jia Shi void cp_update_scsw(struct channel_program *cp, union scsw *scsw)
8640a19e61eSDong Jia Shi {
8650a19e61eSDong Jia Shi 	struct ccwchain *chain;
8660a19e61eSDong Jia Shi 	u32 cpa = scsw->cmd.cpa;
8672904337fSEric Farman 	u32 ccw_head;
8680a19e61eSDong Jia Shi 
86971189f26SCornelia Huck 	if (!cp->initialized)
87071189f26SCornelia Huck 		return;
87171189f26SCornelia Huck 
8720a19e61eSDong Jia Shi 	/*
8730a19e61eSDong Jia Shi 	 * LATER:
8740a19e61eSDong Jia Shi 	 * For now, only update the cmd.cpa part. We may need to deal with
8750a19e61eSDong Jia Shi 	 * other portions of the schib as well, even if we don't return them
8760a19e61eSDong Jia Shi 	 * in the ioctl directly. Path status changes etc.
8770a19e61eSDong Jia Shi 	 */
8780a19e61eSDong Jia Shi 	list_for_each_entry(chain, &cp->ccwchain_list, next) {
8790a19e61eSDong Jia Shi 		ccw_head = (u32)(u64)chain->ch_ccw;
88015f0eb3dSEric Farman 		/*
88115f0eb3dSEric Farman 		 * On successful execution, cpa points just beyond the end
88215f0eb3dSEric Farman 		 * of the chain.
88315f0eb3dSEric Farman 		 */
88415f0eb3dSEric Farman 		if (is_cpa_within_range(cpa, ccw_head, chain->ch_len + 1)) {
8850a19e61eSDong Jia Shi 			/*
8860a19e61eSDong Jia Shi 			 * (cpa - ccw_head) is the offset value of the host
8870a19e61eSDong Jia Shi 			 * physical ccw to its chain head.
8880a19e61eSDong Jia Shi 			 * Adding this value to the guest physical ccw chain
8890a19e61eSDong Jia Shi 			 * head gets us the guest cpa.
8900a19e61eSDong Jia Shi 			 */
8910a19e61eSDong Jia Shi 			cpa = chain->ch_iova + (cpa - ccw_head);
8920a19e61eSDong Jia Shi 			break;
8930a19e61eSDong Jia Shi 		}
8940a19e61eSDong Jia Shi 	}
8950a19e61eSDong Jia Shi 
8960a19e61eSDong Jia Shi 	scsw->cmd.cpa = cpa;
8970a19e61eSDong Jia Shi }
8980a19e61eSDong Jia Shi 
8990a19e61eSDong Jia Shi /**
9000a19e61eSDong Jia Shi  * cp_iova_pinned() - check if an iova is pinned for a ccw chain.
901364e3f90SSebastian Ott  * @cp: channel_program on which to perform the operation
9020a19e61eSDong Jia Shi  * @iova: the iova to check
9030a19e61eSDong Jia Shi  *
9040a19e61eSDong Jia Shi  * If the @iova is currently pinned for the ccw chain, return true;
9050a19e61eSDong Jia Shi  * else return false.
9060a19e61eSDong Jia Shi  */
9070a19e61eSDong Jia Shi bool cp_iova_pinned(struct channel_program *cp, u64 iova)
9080a19e61eSDong Jia Shi {
9090a19e61eSDong Jia Shi 	struct ccwchain *chain;
9100a19e61eSDong Jia Shi 	int i;
9110a19e61eSDong Jia Shi 
91271189f26SCornelia Huck 	if (!cp->initialized)
91371189f26SCornelia Huck 		return false;
91471189f26SCornelia Huck 
9150a19e61eSDong Jia Shi 	list_for_each_entry(chain, &cp->ccwchain_list, next) {
9160a19e61eSDong Jia Shi 		for (i = 0; i < chain->ch_len; i++)
91713314605SNicolin Chen 			if (page_array_iova_pinned(chain->ch_pa + i, iova))
9180a19e61eSDong Jia Shi 				return true;
9190a19e61eSDong Jia Shi 	}
9200a19e61eSDong Jia Shi 
9210a19e61eSDong Jia Shi 	return false;
9220a19e61eSDong Jia Shi }
923