2 * Copyright 2014 IBM Corp.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
10 #include <linux/pci.h>
11 #include <linux/slab.h>
12 #include <linux/anon_inodes.h>
13 #include <linux/file.h>
16 #include <asm/pnv-pci.h>
17 #include <linux/msi.h>
21 struct cxl_context
*cxl_dev_context_init(struct pci_dev
*dev
)
23 struct address_space
*mapping
;
25 struct cxl_context
*ctx
;
28 afu
= cxl_pci_to_afu(dev
);
32 ctx
= cxl_context_alloc();
38 ctx
->kernelapi
= true;
41 * Make our own address space since we won't have one from the
42 * filesystem like the user api has, and even if we do associate a file
43 * with this context we don't want to use the global anonymous inode's
44 * address space as that can invalidate unrelated users:
46 mapping
= kmalloc(sizeof(struct address_space
), GFP_KERNEL
);
51 address_space_init_once(mapping
);
53 /* Make it a slave context. We can promote it later? */
54 rc
= cxl_context_init(ctx
, afu
, false, mapping
);
67 EXPORT_SYMBOL_GPL(cxl_dev_context_init
);
69 struct cxl_context
*cxl_get_context(struct pci_dev
*dev
)
71 return dev
->dev
.archdata
.cxl_ctx
;
73 EXPORT_SYMBOL_GPL(cxl_get_context
);
75 int cxl_release_context(struct cxl_context
*ctx
)
77 if (ctx
->status
>= STARTED
)
80 cxl_context_free(ctx
);
84 EXPORT_SYMBOL_GPL(cxl_release_context
);
86 static irq_hw_number_t
cxl_find_afu_irq(struct cxl_context
*ctx
, int num
)
91 for (r
= 0; r
< CXL_IRQ_RANGES
; r
++) {
92 range
= ctx
->irqs
.range
[r
];
94 return ctx
->irqs
.offset
[r
] + num
;
101 int _cxl_next_msi_hwirq(struct pci_dev
*pdev
, struct cxl_context
**ctx
, int *afu_irq
)
103 if (*ctx
== NULL
|| *afu_irq
== 0) {
105 *ctx
= cxl_get_context(pdev
);
108 if (*afu_irq
> cxl_get_max_irqs_per_process(pdev
)) {
109 *ctx
= list_next_entry(*ctx
, extra_irq_contexts
);
113 return cxl_find_afu_irq(*ctx
, *afu_irq
);
115 /* Exported via cxl_base */
117 int cxl_set_priv(struct cxl_context
*ctx
, void *priv
)
126 EXPORT_SYMBOL_GPL(cxl_set_priv
);
128 void *cxl_get_priv(struct cxl_context
*ctx
)
131 return ERR_PTR(-EINVAL
);
135 EXPORT_SYMBOL_GPL(cxl_get_priv
);
137 int cxl_allocate_afu_irqs(struct cxl_context
*ctx
, int num
)
140 irq_hw_number_t hwirq
;
143 num
= ctx
->afu
->pp_irqs
;
144 res
= afu_allocate_irqs(ctx
, num
);
148 if (!cpu_has_feature(CPU_FTR_HVMODE
)) {
149 /* In a guest, the PSL interrupt is not multiplexed. It was
150 * allocated above, and we need to set its handler
152 hwirq
= cxl_find_afu_irq(ctx
, 0);
154 cxl_map_irq(ctx
->afu
->adapter
, hwirq
, cxl_ops
->psl_interrupt
, ctx
, "psl");
157 if (ctx
->status
== STARTED
) {
158 if (cxl_ops
->update_ivtes
)
159 cxl_ops
->update_ivtes(ctx
);
160 else WARN(1, "BUG: cxl_allocate_afu_irqs must be called prior to starting the context on this platform\n");
165 EXPORT_SYMBOL_GPL(cxl_allocate_afu_irqs
);
167 void cxl_free_afu_irqs(struct cxl_context
*ctx
)
169 irq_hw_number_t hwirq
;
172 if (!cpu_has_feature(CPU_FTR_HVMODE
)) {
173 hwirq
= cxl_find_afu_irq(ctx
, 0);
175 virq
= irq_find_mapping(NULL
, hwirq
);
177 cxl_unmap_irq(virq
, ctx
);
180 afu_irq_name_free(ctx
);
181 cxl_ops
->release_irq_ranges(&ctx
->irqs
, ctx
->afu
->adapter
);
183 EXPORT_SYMBOL_GPL(cxl_free_afu_irqs
);
185 int cxl_map_afu_irq(struct cxl_context
*ctx
, int num
,
186 irq_handler_t handler
, void *cookie
, char *name
)
188 irq_hw_number_t hwirq
;
191 * Find interrupt we are to register.
193 hwirq
= cxl_find_afu_irq(ctx
, num
);
197 return cxl_map_irq(ctx
->afu
->adapter
, hwirq
, handler
, cookie
, name
);
199 EXPORT_SYMBOL_GPL(cxl_map_afu_irq
);
201 void cxl_unmap_afu_irq(struct cxl_context
*ctx
, int num
, void *cookie
)
203 irq_hw_number_t hwirq
;
206 hwirq
= cxl_find_afu_irq(ctx
, num
);
210 virq
= irq_find_mapping(NULL
, hwirq
);
212 cxl_unmap_irq(virq
, cookie
);
214 EXPORT_SYMBOL_GPL(cxl_unmap_afu_irq
);
218 * Code here similar to afu_ioctl_start_work().
220 int cxl_start_context(struct cxl_context
*ctx
, u64 wed
,
221 struct task_struct
*task
)
226 pr_devel("%s: pe: %i\n", __func__
, ctx
->pe
);
228 mutex_lock(&ctx
->status_mutex
);
229 if (ctx
->status
== STARTED
)
230 goto out
; /* already started */
233 ctx
->pid
= get_task_pid(task
, PIDTYPE_PID
);
234 ctx
->glpid
= get_task_pid(task
->group_leader
, PIDTYPE_PID
);
236 ctx
->real_mode
= false;
241 if ((rc
= cxl_ops
->attach_process(ctx
, kernel
, wed
, 0))) {
247 ctx
->status
= STARTED
;
249 mutex_unlock(&ctx
->status_mutex
);
252 EXPORT_SYMBOL_GPL(cxl_start_context
);
254 int cxl_process_element(struct cxl_context
*ctx
)
256 return ctx
->external_pe
;
258 EXPORT_SYMBOL_GPL(cxl_process_element
);
260 /* Stop a context. Returns 0 on success, otherwise -Errno */
261 int cxl_stop_context(struct cxl_context
*ctx
)
263 return __detach_context(ctx
);
265 EXPORT_SYMBOL_GPL(cxl_stop_context
);
267 void cxl_set_master(struct cxl_context
*ctx
)
271 EXPORT_SYMBOL_GPL(cxl_set_master
);
273 int cxl_set_translation_mode(struct cxl_context
*ctx
, bool real_mode
)
275 if (ctx
->status
== STARTED
) {
277 * We could potentially update the PE and issue an update LLCMD
278 * to support this, but it doesn't seem to have a good use case
279 * since it's trivial to just create a second kernel context
280 * with different translation modes, so until someone convinces
286 ctx
->real_mode
= real_mode
;
289 EXPORT_SYMBOL_GPL(cxl_set_translation_mode
);
291 /* wrappers around afu_* file ops which are EXPORTED */
292 int cxl_fd_open(struct inode
*inode
, struct file
*file
)
294 return afu_open(inode
, file
);
296 EXPORT_SYMBOL_GPL(cxl_fd_open
);
297 int cxl_fd_release(struct inode
*inode
, struct file
*file
)
299 return afu_release(inode
, file
);
301 EXPORT_SYMBOL_GPL(cxl_fd_release
);
302 long cxl_fd_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
304 return afu_ioctl(file
, cmd
, arg
);
306 EXPORT_SYMBOL_GPL(cxl_fd_ioctl
);
307 int cxl_fd_mmap(struct file
*file
, struct vm_area_struct
*vm
)
309 return afu_mmap(file
, vm
);
311 EXPORT_SYMBOL_GPL(cxl_fd_mmap
);
312 unsigned int cxl_fd_poll(struct file
*file
, struct poll_table_struct
*poll
)
314 return afu_poll(file
, poll
);
316 EXPORT_SYMBOL_GPL(cxl_fd_poll
);
317 ssize_t
cxl_fd_read(struct file
*file
, char __user
*buf
, size_t count
,
320 return afu_read(file
, buf
, count
, off
);
322 EXPORT_SYMBOL_GPL(cxl_fd_read
);
324 #define PATCH_FOPS(NAME) if (!fops->NAME) fops->NAME = afu_fops.NAME
326 /* Get a struct file and fd for a context and attach the ops */
327 struct file
*cxl_get_fd(struct cxl_context
*ctx
, struct file_operations
*fops
,
331 int rc
, flags
, fdtmp
;
333 flags
= O_RDWR
| O_CLOEXEC
;
335 /* This code is similar to anon_inode_getfd() */
336 rc
= get_unused_fd_flags(flags
);
342 * Patch the file ops. Needs to be careful that this is rentrant safe.
349 PATCH_FOPS(unlocked_ioctl
);
350 PATCH_FOPS(compat_ioctl
);
352 } else /* use default ops */
353 fops
= (struct file_operations
*)&afu_fops
;
355 file
= anon_inode_getfile("cxl", fops
, ctx
, flags
);
359 file
->f_mapping
= ctx
->mapping
;
365 put_unused_fd(fdtmp
);
368 EXPORT_SYMBOL_GPL(cxl_get_fd
);
370 struct cxl_context
*cxl_fops_get_context(struct file
*file
)
372 return file
->private_data
;
374 EXPORT_SYMBOL_GPL(cxl_fops_get_context
);
376 void cxl_set_driver_ops(struct cxl_context
*ctx
,
377 struct cxl_afu_driver_ops
*ops
)
379 WARN_ON(!ops
->fetch_event
|| !ops
->event_delivered
);
380 atomic_set(&ctx
->afu_driver_events
, 0);
381 ctx
->afu_driver_ops
= ops
;
383 EXPORT_SYMBOL_GPL(cxl_set_driver_ops
);
385 void cxl_context_events_pending(struct cxl_context
*ctx
,
386 unsigned int new_events
)
388 atomic_add(new_events
, &ctx
->afu_driver_events
);
389 wake_up_all(&ctx
->wq
);
391 EXPORT_SYMBOL_GPL(cxl_context_events_pending
);
393 int cxl_start_work(struct cxl_context
*ctx
,
394 struct cxl_ioctl_start_work
*work
)
398 /* code taken from afu_ioctl_start_work */
399 if (!(work
->flags
& CXL_START_WORK_NUM_IRQS
))
400 work
->num_interrupts
= ctx
->afu
->pp_irqs
;
401 else if ((work
->num_interrupts
< ctx
->afu
->pp_irqs
) ||
402 (work
->num_interrupts
> ctx
->afu
->irqs_max
)) {
406 rc
= afu_register_irqs(ctx
, work
->num_interrupts
);
410 rc
= cxl_start_context(ctx
, work
->work_element_descriptor
, current
);
412 afu_release_irqs(ctx
, ctx
);
418 EXPORT_SYMBOL_GPL(cxl_start_work
);
420 void __iomem
*cxl_psa_map(struct cxl_context
*ctx
)
422 if (ctx
->status
!= STARTED
)
425 pr_devel("%s: psn_phys%llx size:%llx\n",
426 __func__
, ctx
->psn_phys
, ctx
->psn_size
);
427 return ioremap(ctx
->psn_phys
, ctx
->psn_size
);
429 EXPORT_SYMBOL_GPL(cxl_psa_map
);
431 void cxl_psa_unmap(void __iomem
*addr
)
435 EXPORT_SYMBOL_GPL(cxl_psa_unmap
);
437 int cxl_afu_reset(struct cxl_context
*ctx
)
439 struct cxl_afu
*afu
= ctx
->afu
;
442 rc
= cxl_ops
->afu_reset(afu
);
446 return cxl_ops
->afu_check_and_enable(afu
);
448 EXPORT_SYMBOL_GPL(cxl_afu_reset
);
450 void cxl_perst_reloads_same_image(struct cxl_afu
*afu
,
451 bool perst_reloads_same_image
)
453 afu
->adapter
->perst_same_image
= perst_reloads_same_image
;
455 EXPORT_SYMBOL_GPL(cxl_perst_reloads_same_image
);
457 ssize_t
cxl_read_adapter_vpd(struct pci_dev
*dev
, void *buf
, size_t count
)
459 struct cxl_afu
*afu
= cxl_pci_to_afu(dev
);
463 return cxl_ops
->read_adapter_vpd(afu
->adapter
, buf
, count
);
465 EXPORT_SYMBOL_GPL(cxl_read_adapter_vpd
);
467 int cxl_set_max_irqs_per_process(struct pci_dev
*dev
, int irqs
)
469 struct cxl_afu
*afu
= cxl_pci_to_afu(dev
);
473 if (irqs
> afu
->adapter
->user_irqs
)
476 /* Limit user_irqs to prevent the user increasing this via sysfs */
477 afu
->adapter
->user_irqs
= irqs
;
478 afu
->irqs_max
= irqs
;
482 EXPORT_SYMBOL_GPL(cxl_set_max_irqs_per_process
);
484 int cxl_get_max_irqs_per_process(struct pci_dev
*dev
)
486 struct cxl_afu
*afu
= cxl_pci_to_afu(dev
);
490 return afu
->irqs_max
;
492 EXPORT_SYMBOL_GPL(cxl_get_max_irqs_per_process
);
495 * This is a special interrupt allocation routine called from the PHB's MSI
496 * setup function. When capi interrupts are allocated in this manner they must
497 * still be associated with a running context, but since the MSI APIs have no
498 * way to specify this we use the default context associated with the device.
500 * The Mellanox CX4 has a hardware limitation that restricts the maximum AFU
501 * interrupt number, so in order to overcome this their driver informs us of
502 * the restriction by setting the maximum interrupts per context, and we
503 * allocate additional contexts as necessary so that we can keep the AFU
504 * interrupt number within the supported range.
506 int _cxl_cx4_setup_msi_irqs(struct pci_dev
*pdev
, int nvec
, int type
)
508 struct cxl_context
*ctx
, *new_ctx
, *default_ctx
;
512 ctx
= default_ctx
= cxl_get_context(pdev
);
513 if (WARN_ON(!default_ctx
))
517 while (remaining
> 0) {
518 rc
= cxl_allocate_afu_irqs(ctx
, min(remaining
, ctx
->afu
->irqs_max
));
520 pr_warn("%s: Failed to find enough free MSIs\n", pci_name(pdev
));
523 remaining
-= ctx
->afu
->irqs_max
;
525 if (ctx
!= default_ctx
&& default_ctx
->status
== STARTED
) {
526 WARN_ON(cxl_start_context(ctx
,
527 be64_to_cpu(default_ctx
->elem
->common
.wed
),
532 new_ctx
= cxl_dev_context_init(pdev
);
534 pr_warn("%s: Failed to allocate enough contexts for MSIs\n", pci_name(pdev
));
537 list_add(&new_ctx
->extra_irq_contexts
, &ctx
->extra_irq_contexts
);
544 /* Exported via cxl_base */
546 void _cxl_cx4_teardown_msi_irqs(struct pci_dev
*pdev
)
548 struct cxl_context
*ctx
, *pos
, *tmp
;
550 ctx
= cxl_get_context(pdev
);
554 cxl_free_afu_irqs(ctx
);
555 list_for_each_entry_safe(pos
, tmp
, &ctx
->extra_irq_contexts
, extra_irq_contexts
) {
556 cxl_stop_context(pos
);
557 cxl_free_afu_irqs(pos
);
558 list_del(&pos
->extra_irq_contexts
);
559 cxl_release_context(pos
);
562 /* Exported via cxl_base */