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/module.h>
11 #include <linux/kernel.h>
12 #include <linux/bitmap.h>
13 #include <linux/sched.h>
14 #include <linux/pid.h>
17 #include <linux/debugfs.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/sched/mm.h>
21 #include <asm/cputable.h>
22 #include <asm/current.h>
23 #include <asm/copro.h>
28 * Allocates space for a CXL context.
30 struct cxl_context *cxl_context_alloc(void)
32 return kzalloc(sizeof(struct cxl_context), GFP_KERNEL);
36 * Initialises a CXL context.
38 int cxl_context_init(struct cxl_context *ctx, struct cxl_afu *afu, bool master)
44 ctx->pid = NULL; /* Set in start work ioctl */
45 mutex_init(&ctx->mapping_lock);
48 if (cxl_is_power8()) {
49 spin_lock_init(&ctx->sste_lock);
52 * Allocate the segment table before we put it in the IDR so that we
53 * can always access it when dereferenced from IDR. For the same
54 * reason, the segment table is only destroyed after the context is
55 * removed from the IDR. Access to this in the IOCTL is protected by
56 * Linux filesytem symantics (can't IOCTL until open is complete).
58 i = cxl_alloc_sst(ctx);
63 INIT_WORK(&ctx->fault_work, cxl_handle_fault);
65 init_waitqueue_head(&ctx->wq);
66 spin_lock_init(&ctx->lock);
68 ctx->irq_bitmap = NULL;
69 ctx->pending_irq = false;
70 ctx->pending_fault = false;
71 ctx->pending_afu_err = false;
73 INIT_LIST_HEAD(&ctx->irq_names);
74 INIT_LIST_HEAD(&ctx->extra_irq_contexts);
77 * When we have to destroy all contexts in cxl_context_detach_all() we
78 * end up with afu_release_irqs() called from inside a
79 * idr_for_each_entry(). Hence we need to make sure that anything
80 * dereferenced from this IDR is ok before we allocate the IDR here.
81 * This clears out the IRQ ranges to ensure this.
83 for (i = 0; i < CXL_IRQ_RANGES; i++)
84 ctx->irqs.range[i] = 0;
86 mutex_init(&ctx->status_mutex);
91 * Allocating IDR! We better make sure everything's setup that
92 * dereferences from it.
94 mutex_lock(&afu->contexts_lock);
95 idr_preload(GFP_KERNEL);
96 i = idr_alloc(&ctx->afu->contexts_idr, ctx, ctx->afu->adapter->min_pe,
97 ctx->afu->num_procs, GFP_NOWAIT);
99 mutex_unlock(&afu->contexts_lock);
104 if (cpu_has_feature(CPU_FTR_HVMODE)) {
105 ctx->elem = &ctx->afu->native->spa[i];
106 ctx->external_pe = ctx->pe;
108 ctx->external_pe = -1; /* assigned when attaching */
110 ctx->pe_inserted = false;
113 * take a ref on the afu so that it stays alive at-least till
114 * this context is reclaimed inside reclaim_ctx.
120 void cxl_context_set_mapping(struct cxl_context *ctx,
121 struct address_space *mapping)
123 mutex_lock(&ctx->mapping_lock);
124 ctx->mapping = mapping;
125 mutex_unlock(&ctx->mapping_lock);
128 static int cxl_mmap_fault(struct vm_fault *vmf)
130 struct vm_area_struct *vma = vmf->vma;
131 struct cxl_context *ctx = vma->vm_file->private_data;
134 offset = vmf->pgoff << PAGE_SHIFT;
136 pr_devel("%s: pe: %i address: 0x%lx offset: 0x%llx\n",
137 __func__, ctx->pe, vmf->address, offset);
139 if (ctx->afu->current_mode == CXL_MODE_DEDICATED) {
140 area = ctx->afu->psn_phys;
141 if (offset >= ctx->afu->adapter->ps_size)
142 return VM_FAULT_SIGBUS;
144 area = ctx->psn_phys;
145 if (offset >= ctx->psn_size)
146 return VM_FAULT_SIGBUS;
149 mutex_lock(&ctx->status_mutex);
151 if (ctx->status != STARTED) {
152 mutex_unlock(&ctx->status_mutex);
153 pr_devel("%s: Context not started, failing problem state access\n", __func__);
154 if (ctx->mmio_err_ff) {
156 ctx->ff_page = alloc_page(GFP_USER);
159 memset(page_address(ctx->ff_page), 0xff, PAGE_SIZE);
161 get_page(ctx->ff_page);
162 vmf->page = ctx->ff_page;
163 vma->vm_page_prot = pgprot_cached(vma->vm_page_prot);
166 return VM_FAULT_SIGBUS;
169 vm_insert_pfn(vma, vmf->address, (area + offset) >> PAGE_SHIFT);
171 mutex_unlock(&ctx->status_mutex);
173 return VM_FAULT_NOPAGE;
176 static const struct vm_operations_struct cxl_mmap_vmops = {
177 .fault = cxl_mmap_fault,
181 * Map a per-context mmio space into the given vma.
183 int cxl_context_iomap(struct cxl_context *ctx, struct vm_area_struct *vma)
185 u64 start = vma->vm_pgoff << PAGE_SHIFT;
186 u64 len = vma->vm_end - vma->vm_start;
188 if (ctx->afu->current_mode == CXL_MODE_DEDICATED) {
189 if (start + len > ctx->afu->adapter->ps_size)
192 if (cxl_is_power9()) {
194 * Make sure there is a valid problem state
195 * area space for this AFU.
197 if (ctx->master && !ctx->afu->psa) {
198 pr_devel("AFU doesn't support mmio space\n");
202 /* Can't mmap until the AFU is enabled */
203 if (!ctx->afu->enabled)
207 if (start + len > ctx->psn_size)
210 /* Make sure there is a valid per process space for this AFU */
211 if ((ctx->master && !ctx->afu->psa) || (!ctx->afu->pp_psa)) {
212 pr_devel("AFU doesn't support mmio space\n");
216 /* Can't mmap until the AFU is enabled */
217 if (!ctx->afu->enabled)
221 pr_devel("%s: mmio physical: %llx pe: %i master:%i\n", __func__,
222 ctx->psn_phys, ctx->pe , ctx->master);
224 vma->vm_flags |= VM_IO | VM_PFNMAP;
225 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
226 vma->vm_ops = &cxl_mmap_vmops;
231 * Detach a context from the hardware. This disables interrupts and doesn't
232 * return until all outstanding interrupts for this context have completed. The
233 * hardware should no longer access *ctx after this has returned.
235 int __detach_context(struct cxl_context *ctx)
237 enum cxl_context_status status;
239 mutex_lock(&ctx->status_mutex);
240 status = ctx->status;
241 ctx->status = CLOSED;
242 mutex_unlock(&ctx->status_mutex);
243 if (status != STARTED)
246 /* Only warn if we detached while the link was OK.
247 * If detach fails when hw is down, we don't care.
249 WARN_ON(cxl_ops->detach_process(ctx) &&
250 cxl_ops->link_ok(ctx->afu->adapter, ctx->afu));
251 flush_work(&ctx->fault_work); /* Only needed for dedicated process */
254 * Wait until no further interrupts are presented by the PSL
257 if (cxl_ops->irq_wait)
258 cxl_ops->irq_wait(ctx);
260 /* release the reference to the group leader and mm handling pid */
265 /* Decrease the attached context count on the adapter */
266 cxl_adapter_context_put(ctx->afu->adapter);
268 /* Decrease the mm count on the context */
269 cxl_context_mm_count_put(ctx);
276 * Detach the given context from the AFU. This doesn't actually
277 * free the context but it should stop the context running in hardware
278 * (ie. prevent this context from generating any further interrupts
279 * so that it can be freed).
281 void cxl_context_detach(struct cxl_context *ctx)
285 rc = __detach_context(ctx);
289 afu_release_irqs(ctx, ctx);
290 wake_up_all(&ctx->wq);
294 * Detach all contexts on the given AFU.
296 void cxl_context_detach_all(struct cxl_afu *afu)
298 struct cxl_context *ctx;
301 mutex_lock(&afu->contexts_lock);
302 idr_for_each_entry(&afu->contexts_idr, ctx, tmp) {
304 * Anything done in here needs to be setup before the IDR is
305 * created and torn down after the IDR removed
307 cxl_context_detach(ctx);
310 * We are force detaching - remove any active PSA mappings so
311 * userspace cannot interfere with the card if it comes back.
312 * Easiest way to exercise this is to unbind and rebind the
313 * driver via sysfs while it is in use.
315 mutex_lock(&ctx->mapping_lock);
317 unmap_mapping_range(ctx->mapping, 0, 0, 1);
318 mutex_unlock(&ctx->mapping_lock);
320 mutex_unlock(&afu->contexts_lock);
323 static void reclaim_ctx(struct rcu_head *rcu)
325 struct cxl_context *ctx = container_of(rcu, struct cxl_context, rcu);
328 free_page((u64)ctx->sstp);
330 __free_page(ctx->ff_page);
333 kfree(ctx->irq_bitmap);
335 /* Drop ref to the afu device taken during cxl_context_init */
336 cxl_afu_put(ctx->afu);
341 void cxl_context_free(struct cxl_context *ctx)
343 if (ctx->kernelapi && ctx->mapping)
344 cxl_release_mapping(ctx);
345 mutex_lock(&ctx->afu->contexts_lock);
346 idr_remove(&ctx->afu->contexts_idr, ctx->pe);
347 mutex_unlock(&ctx->afu->contexts_lock);
348 call_rcu(&ctx->rcu, reclaim_ctx);
351 void cxl_context_mm_count_get(struct cxl_context *ctx)
354 atomic_inc(&ctx->mm->mm_count);
357 void cxl_context_mm_count_put(struct cxl_context *ctx)