Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
KVM: s390: pci: provide routines for enabling/disabling interrupt forwarding
These routines will be wired into a kvm ioctl in order to respond to
requests to enable / disable a device for Adapter Event Notifications /
Adapter Interuption Forwarding.
Reviewed-by: Christian Borntraeger <borntraeger@linux.ibm.com>
Acked-by: Niklas Schnelle <schnelle@linux.ibm.com>
Signed-off-by: Matthew Rosato <mjrosato@linux.ibm.com>
Link: https://lore.kernel.org/r/20220606203325.110625-16-mjrosato@linux.ibm.com
Signed-off-by: Christian Borntraeger <borntraeger@linux.ibm.com>
authored by
Matthew Rosato
and committed by
Christian Borntraeger
3c5a1b6f
3f4bbb43
+251
-1
+247
arch/s390/kvm/pci.c
+247
arch/s390/kvm/pci.c
···
11
11
#include <linux/pci.h>
12
12
#include <asm/pci.h>
13
13
#include <asm/pci_insn.h>
14
+
#include <asm/pci_io.h>
14
15
#include "pci.h"
15
16
16
17
struct zpci_aift *aift;
···
150
149
kfree(aift->kzdev);
151
150
unlock:
152
151
mutex_unlock(&aift->aift_lock);
152
+
return rc;
153
+
}
154
+
155
+
/* Modify PCI: Register floating adapter interruption forwarding */
156
+
static int kvm_zpci_set_airq(struct zpci_dev *zdev)
157
+
{
158
+
u64 req = ZPCI_CREATE_REQ(zdev->fh, 0, ZPCI_MOD_FC_REG_INT);
159
+
struct zpci_fib fib = {};
160
+
u8 status;
161
+
162
+
fib.fmt0.isc = zdev->kzdev->fib.fmt0.isc;
163
+
fib.fmt0.sum = 1; /* enable summary notifications */
164
+
fib.fmt0.noi = airq_iv_end(zdev->aibv);
165
+
fib.fmt0.aibv = virt_to_phys(zdev->aibv->vector);
166
+
fib.fmt0.aibvo = 0;
167
+
fib.fmt0.aisb = virt_to_phys(aift->sbv->vector + (zdev->aisb / 64) * 8);
168
+
fib.fmt0.aisbo = zdev->aisb & 63;
169
+
fib.gd = zdev->gisa;
170
+
171
+
return zpci_mod_fc(req, &fib, &status) ? -EIO : 0;
172
+
}
173
+
174
+
/* Modify PCI: Unregister floating adapter interruption forwarding */
175
+
static int kvm_zpci_clear_airq(struct zpci_dev *zdev)
176
+
{
177
+
u64 req = ZPCI_CREATE_REQ(zdev->fh, 0, ZPCI_MOD_FC_DEREG_INT);
178
+
struct zpci_fib fib = {};
179
+
u8 cc, status;
180
+
181
+
fib.gd = zdev->gisa;
182
+
183
+
cc = zpci_mod_fc(req, &fib, &status);
184
+
if (cc == 3 || (cc == 1 && status == 24))
185
+
/* Function already gone or IRQs already deregistered. */
186
+
cc = 0;
187
+
188
+
return cc ? -EIO : 0;
189
+
}
190
+
191
+
static inline void unaccount_mem(unsigned long nr_pages)
192
+
{
193
+
struct user_struct *user = get_uid(current_user());
194
+
195
+
if (user)
196
+
atomic_long_sub(nr_pages, &user->locked_vm);
197
+
if (current->mm)
198
+
atomic64_sub(nr_pages, ¤t->mm->pinned_vm);
199
+
}
200
+
201
+
static inline int account_mem(unsigned long nr_pages)
202
+
{
203
+
struct user_struct *user = get_uid(current_user());
204
+
unsigned long page_limit, cur_pages, new_pages;
205
+
206
+
page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
207
+
208
+
do {
209
+
cur_pages = atomic_long_read(&user->locked_vm);
210
+
new_pages = cur_pages + nr_pages;
211
+
if (new_pages > page_limit)
212
+
return -ENOMEM;
213
+
} while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
214
+
new_pages) != cur_pages);
215
+
216
+
atomic64_add(nr_pages, ¤t->mm->pinned_vm);
217
+
218
+
return 0;
219
+
}
220
+
221
+
static int kvm_s390_pci_aif_enable(struct zpci_dev *zdev, struct zpci_fib *fib,
222
+
bool assist)
223
+
{
224
+
struct page *pages[1], *aibv_page, *aisb_page = NULL;
225
+
unsigned int msi_vecs, idx;
226
+
struct zpci_gaite *gaite;
227
+
unsigned long hva, bit;
228
+
struct kvm *kvm;
229
+
phys_addr_t gaddr;
230
+
int rc = 0, gisc, npages, pcount = 0;
231
+
232
+
/*
233
+
* Interrupt forwarding is only applicable if the device is already
234
+
* enabled for interpretation
235
+
*/
236
+
if (zdev->gisa == 0)
237
+
return -EINVAL;
238
+
239
+
kvm = zdev->kzdev->kvm;
240
+
msi_vecs = min_t(unsigned int, fib->fmt0.noi, zdev->max_msi);
241
+
242
+
/* Get the associated forwarding ISC - if invalid, return the error */
243
+
gisc = kvm_s390_gisc_register(kvm, fib->fmt0.isc);
244
+
if (gisc < 0)
245
+
return gisc;
246
+
247
+
/* Replace AIBV address */
248
+
idx = srcu_read_lock(&kvm->srcu);
249
+
hva = gfn_to_hva(kvm, gpa_to_gfn((gpa_t)fib->fmt0.aibv));
250
+
npages = pin_user_pages_fast(hva, 1, FOLL_WRITE | FOLL_LONGTERM, pages);
251
+
srcu_read_unlock(&kvm->srcu, idx);
252
+
if (npages < 1) {
253
+
rc = -EIO;
254
+
goto out;
255
+
}
256
+
aibv_page = pages[0];
257
+
pcount++;
258
+
gaddr = page_to_phys(aibv_page) + (fib->fmt0.aibv & ~PAGE_MASK);
259
+
fib->fmt0.aibv = gaddr;
260
+
261
+
/* Pin the guest AISB if one was specified */
262
+
if (fib->fmt0.sum == 1) {
263
+
idx = srcu_read_lock(&kvm->srcu);
264
+
hva = gfn_to_hva(kvm, gpa_to_gfn((gpa_t)fib->fmt0.aisb));
265
+
npages = pin_user_pages_fast(hva, 1, FOLL_WRITE | FOLL_LONGTERM,
266
+
pages);
267
+
srcu_read_unlock(&kvm->srcu, idx);
268
+
if (npages < 1) {
269
+
rc = -EIO;
270
+
goto unpin1;
271
+
}
272
+
aisb_page = pages[0];
273
+
pcount++;
274
+
}
275
+
276
+
/* Account for pinned pages, roll back on failure */
277
+
if (account_mem(pcount))
278
+
goto unpin2;
279
+
280
+
/* AISB must be allocated before we can fill in GAITE */
281
+
mutex_lock(&aift->aift_lock);
282
+
bit = airq_iv_alloc_bit(aift->sbv);
283
+
if (bit == -1UL)
284
+
goto unlock;
285
+
zdev->aisb = bit; /* store the summary bit number */
286
+
zdev->aibv = airq_iv_create(msi_vecs, AIRQ_IV_DATA |
287
+
AIRQ_IV_BITLOCK |
288
+
AIRQ_IV_GUESTVEC,
289
+
phys_to_virt(fib->fmt0.aibv));
290
+
291
+
spin_lock_irq(&aift->gait_lock);
292
+
gaite = (struct zpci_gaite *)aift->gait + (zdev->aisb *
293
+
sizeof(struct zpci_gaite));
294
+
295
+
/* If assist not requested, host will get all alerts */
296
+
if (assist)
297
+
gaite->gisa = (u32)virt_to_phys(&kvm->arch.sie_page2->gisa);
298
+
else
299
+
gaite->gisa = 0;
300
+
301
+
gaite->gisc = fib->fmt0.isc;
302
+
gaite->count++;
303
+
gaite->aisbo = fib->fmt0.aisbo;
304
+
gaite->aisb = virt_to_phys(page_address(aisb_page) + (fib->fmt0.aisb &
305
+
~PAGE_MASK));
306
+
aift->kzdev[zdev->aisb] = zdev->kzdev;
307
+
spin_unlock_irq(&aift->gait_lock);
308
+
309
+
/* Update guest FIB for re-issue */
310
+
fib->fmt0.aisbo = zdev->aisb & 63;
311
+
fib->fmt0.aisb = virt_to_phys(aift->sbv->vector + (zdev->aisb / 64) * 8);
312
+
fib->fmt0.isc = gisc;
313
+
314
+
/* Save some guest fib values in the host for later use */
315
+
zdev->kzdev->fib.fmt0.isc = fib->fmt0.isc;
316
+
zdev->kzdev->fib.fmt0.aibv = fib->fmt0.aibv;
317
+
mutex_unlock(&aift->aift_lock);
318
+
319
+
/* Issue the clp to setup the irq now */
320
+
rc = kvm_zpci_set_airq(zdev);
321
+
return rc;
322
+
323
+
unlock:
324
+
mutex_unlock(&aift->aift_lock);
325
+
unpin2:
326
+
if (fib->fmt0.sum == 1)
327
+
unpin_user_page(aisb_page);
328
+
unpin1:
329
+
unpin_user_page(aibv_page);
330
+
out:
331
+
return rc;
332
+
}
333
+
334
+
static int kvm_s390_pci_aif_disable(struct zpci_dev *zdev, bool force)
335
+
{
336
+
struct kvm_zdev *kzdev = zdev->kzdev;
337
+
struct zpci_gaite *gaite;
338
+
struct page *vpage = NULL, *spage = NULL;
339
+
int rc, pcount = 0;
340
+
u8 isc;
341
+
342
+
if (zdev->gisa == 0)
343
+
return -EINVAL;
344
+
345
+
mutex_lock(&aift->aift_lock);
346
+
347
+
/*
348
+
* If the clear fails due to an error, leave now unless we know this
349
+
* device is about to go away (force) -- In that case clear the GAITE
350
+
* regardless.
351
+
*/
352
+
rc = kvm_zpci_clear_airq(zdev);
353
+
if (rc && !force)
354
+
goto out;
355
+
356
+
if (zdev->kzdev->fib.fmt0.aibv == 0)
357
+
goto out;
358
+
spin_lock_irq(&aift->gait_lock);
359
+
gaite = (struct zpci_gaite *)aift->gait + (zdev->aisb *
360
+
sizeof(struct zpci_gaite));
361
+
isc = gaite->gisc;
362
+
gaite->count--;
363
+
if (gaite->count == 0) {
364
+
/* Release guest AIBV and AISB */
365
+
vpage = phys_to_page(kzdev->fib.fmt0.aibv);
366
+
if (gaite->aisb != 0)
367
+
spage = phys_to_page(gaite->aisb);
368
+
/* Clear the GAIT entry */
369
+
gaite->aisb = 0;
370
+
gaite->gisc = 0;
371
+
gaite->aisbo = 0;
372
+
gaite->gisa = 0;
373
+
aift->kzdev[zdev->aisb] = 0;
374
+
/* Clear zdev info */
375
+
airq_iv_free_bit(aift->sbv, zdev->aisb);
376
+
airq_iv_release(zdev->aibv);
377
+
zdev->aisb = 0;
378
+
zdev->aibv = NULL;
379
+
}
380
+
spin_unlock_irq(&aift->gait_lock);
381
+
kvm_s390_gisc_unregister(kzdev->kvm, isc);
382
+
kzdev->fib.fmt0.isc = 0;
383
+
kzdev->fib.fmt0.aibv = 0;
384
+
385
+
if (vpage) {
386
+
unpin_user_page(vpage);
387
+
pcount++;
388
+
}
389
+
if (spage) {
390
+
unpin_user_page(spage);
391
+
pcount++;
392
+
}
393
+
if (pcount > 0)
394
+
unaccount_mem(pcount);
395
+
out:
396
+
mutex_unlock(&aift->aift_lock);
397
+
153
398
return rc;
154
399
}
155
400
+1
arch/s390/kvm/pci.h
+1
arch/s390/kvm/pci.h
+1
arch/s390/pci/pci_insn.c
+1
arch/s390/pci/pci_insn.c
+2
-1
include/linux/sched/user.h
+2
-1
include/linux/sched/user.h
···
24
24
kuid_t uid;
25
25
26
26
#if defined(CONFIG_PERF_EVENTS) || defined(CONFIG_BPF_SYSCALL) || \
27
-
defined(CONFIG_NET) || defined(CONFIG_IO_URING)
27
+
defined(CONFIG_NET) || defined(CONFIG_IO_URING) || \
28
+
defined(CONFIG_VFIO_PCI_ZDEV_KVM)
28
29
atomic_long_t locked_vm;
29
30
#endif
30
31
#ifdef CONFIG_WATCH_QUEUE