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 <linux/mmu_context.h>
22 #include <asm/cputable.h>
23 #include <asm/current.h>
24 #include <asm/copro.h>
29 * Allocates space for a CXL context.
31 struct cxl_context *cxl_context_alloc(void)
33 return kzalloc(sizeof(struct cxl_context), GFP_KERNEL);
37 * Initialises a CXL context.
39 int cxl_context_init(struct cxl_context *ctx, struct cxl_afu *afu, bool master)
45 ctx->pid = NULL; /* Set in start work ioctl */
46 mutex_init(&ctx->mapping_lock);
49 if (cxl_is_power8()) {
50 spin_lock_init(&ctx->sste_lock);
53 * Allocate the segment table before we put it in the IDR so that we
54 * can always access it when dereferenced from IDR. For the same
55 * reason, the segment table is only destroyed after the context is
56 * removed from the IDR. Access to this in the IOCTL is protected by
57 * Linux filesytem symantics (can't IOCTL until open is complete).
59 i = cxl_alloc_sst(ctx);
64 INIT_WORK(&ctx->fault_work, cxl_handle_fault);
66 init_waitqueue_head(&ctx->wq);
67 spin_lock_init(&ctx->lock);
69 ctx->irq_bitmap = NULL;
70 ctx->pending_irq = false;
71 ctx->pending_fault = false;
72 ctx->pending_afu_err = false;
74 INIT_LIST_HEAD(&ctx->irq_names);
75 INIT_LIST_HEAD(&ctx->extra_irq_contexts);
78 * When we have to destroy all contexts in cxl_context_detach_all() we
79 * end up with afu_release_irqs() called from inside a
80 * idr_for_each_entry(). Hence we need to make sure that anything
81 * dereferenced from this IDR is ok before we allocate the IDR here.
82 * This clears out the IRQ ranges to ensure this.
84 for (i = 0; i < CXL_IRQ_RANGES; i++)
85 ctx->irqs.range[i] = 0;
87 mutex_init(&ctx->status_mutex);
92 * Allocating IDR! We better make sure everything's setup that
93 * dereferences from it.
95 mutex_lock(&afu->contexts_lock);
96 idr_preload(GFP_KERNEL);
97 i = idr_alloc(&ctx->afu->contexts_idr, ctx, ctx->afu->adapter->min_pe,
98 ctx->afu->num_procs, GFP_NOWAIT);
100 mutex_unlock(&afu->contexts_lock);
105 if (cpu_has_feature(CPU_FTR_HVMODE)) {
106 ctx->elem = &ctx->afu->native->spa[i];
107 ctx->external_pe = ctx->pe;
109 ctx->external_pe = -1; /* assigned when attaching */
111 ctx->pe_inserted = false;
114 * take a ref on the afu so that it stays alive at-least till
115 * this context is reclaimed inside reclaim_ctx.
121 void cxl_context_set_mapping(struct cxl_context *ctx,
122 struct address_space *mapping)
124 mutex_lock(&ctx->mapping_lock);
125 ctx->mapping = mapping;
126 mutex_unlock(&ctx->mapping_lock);
129 static int cxl_mmap_fault(struct vm_fault *vmf)
131 struct vm_area_struct *vma = vmf->vma;
132 struct cxl_context *ctx = vma->vm_file->private_data;
135 offset = vmf->pgoff << PAGE_SHIFT;
137 pr_devel("%s: pe: %i address: 0x%lx offset: 0x%llx\n",
138 __func__, ctx->pe, vmf->address, offset);
140 if (ctx->afu->current_mode == CXL_MODE_DEDICATED) {
141 area = ctx->afu->psn_phys;
142 if (offset >= ctx->afu->adapter->ps_size)
143 return VM_FAULT_SIGBUS;
145 area = ctx->psn_phys;
146 if (offset >= ctx->psn_size)
147 return VM_FAULT_SIGBUS;
150 mutex_lock(&ctx->status_mutex);
152 if (ctx->status != STARTED) {
153 mutex_unlock(&ctx->status_mutex);
154 pr_devel("%s: Context not started, failing problem state access\n", __func__);
155 if (ctx->mmio_err_ff) {
157 ctx->ff_page = alloc_page(GFP_USER);
160 memset(page_address(ctx->ff_page), 0xff, PAGE_SIZE);
162 get_page(ctx->ff_page);
163 vmf->page = ctx->ff_page;
164 vma->vm_page_prot = pgprot_cached(vma->vm_page_prot);
167 return VM_FAULT_SIGBUS;
170 vm_insert_pfn(vma, vmf->address, (area + offset) >> PAGE_SHIFT);
172 mutex_unlock(&ctx->status_mutex);
174 return VM_FAULT_NOPAGE;
177 static const struct vm_operations_struct cxl_mmap_vmops = {
178 .fault = cxl_mmap_fault,
182 * Map a per-context mmio space into the given vma.
184 int cxl_context_iomap(struct cxl_context *ctx, struct vm_area_struct *vma)
186 u64 start = vma->vm_pgoff << PAGE_SHIFT;
187 u64 len = vma->vm_end - vma->vm_start;
189 if (ctx->afu->current_mode == CXL_MODE_DEDICATED) {
190 if (start + len > ctx->afu->adapter->ps_size)
193 if (cxl_is_power9()) {
195 * Make sure there is a valid problem state
196 * area space for this AFU.
198 if (ctx->master && !ctx->afu->psa) {
199 pr_devel("AFU doesn't support mmio space\n");
203 /* Can't mmap until the AFU is enabled */
204 if (!ctx->afu->enabled)
208 if (start + len > ctx->psn_size)
211 /* Make sure there is a valid per process space for this AFU */
212 if ((ctx->master && !ctx->afu->psa) || (!ctx->afu->pp_psa)) {
213 pr_devel("AFU doesn't support mmio space\n");
217 /* Can't mmap until the AFU is enabled */
218 if (!ctx->afu->enabled)
222 pr_devel("%s: mmio physical: %llx pe: %i master:%i\n", __func__,
223 ctx->psn_phys, ctx->pe , ctx->master);
225 vma->vm_flags |= VM_IO | VM_PFNMAP;
226 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
227 vma->vm_ops = &cxl_mmap_vmops;
232 * Detach a context from the hardware. This disables interrupts and doesn't
233 * return until all outstanding interrupts for this context have completed. The
234 * hardware should no longer access *ctx after this has returned.
236 int __detach_context(struct cxl_context *ctx)
238 enum cxl_context_status status;
240 mutex_lock(&ctx->status_mutex);
241 status = ctx->status;
242 ctx->status = CLOSED;
243 mutex_unlock(&ctx->status_mutex);
244 if (status != STARTED)
247 /* Only warn if we detached while the link was OK.
248 * If detach fails when hw is down, we don't care.
250 WARN_ON(cxl_ops->detach_process(ctx) &&
251 cxl_ops->link_ok(ctx->afu->adapter, ctx->afu));
252 flush_work(&ctx->fault_work); /* Only needed for dedicated process */
255 * Wait until no further interrupts are presented by the PSL
258 if (cxl_ops->irq_wait)
259 cxl_ops->irq_wait(ctx);
261 /* release the reference to the group leader and mm handling pid */
266 /* Decrease the attached context count on the adapter */
267 cxl_adapter_context_put(ctx->afu->adapter);
269 /* Decrease the mm count on the context */
270 cxl_context_mm_count_put(ctx);
272 mm_context_remove_copro(ctx->mm);
279 * Detach the given context from the AFU. This doesn't actually
280 * free the context but it should stop the context running in hardware
281 * (ie. prevent this context from generating any further interrupts
282 * so that it can be freed).
284 void cxl_context_detach(struct cxl_context *ctx)
288 rc = __detach_context(ctx);
292 afu_release_irqs(ctx, ctx);
293 wake_up_all(&ctx->wq);
297 * Detach all contexts on the given AFU.
299 void cxl_context_detach_all(struct cxl_afu *afu)
301 struct cxl_context *ctx;
304 mutex_lock(&afu->contexts_lock);
305 idr_for_each_entry(&afu->contexts_idr, ctx, tmp) {
307 * Anything done in here needs to be setup before the IDR is
308 * created and torn down after the IDR removed
310 cxl_context_detach(ctx);
313 * We are force detaching - remove any active PSA mappings so
314 * userspace cannot interfere with the card if it comes back.
315 * Easiest way to exercise this is to unbind and rebind the
316 * driver via sysfs while it is in use.
318 mutex_lock(&ctx->mapping_lock);
320 unmap_mapping_range(ctx->mapping, 0, 0, 1);
321 mutex_unlock(&ctx->mapping_lock);
323 mutex_unlock(&afu->contexts_lock);
326 static void reclaim_ctx(struct rcu_head *rcu)
328 struct cxl_context *ctx = container_of(rcu, struct cxl_context, rcu);
331 free_page((u64)ctx->sstp);
333 __free_page(ctx->ff_page);
336 kfree(ctx->irq_bitmap);
338 /* Drop ref to the afu device taken during cxl_context_init */
339 cxl_afu_put(ctx->afu);
344 void cxl_context_free(struct cxl_context *ctx)
346 if (ctx->kernelapi && ctx->mapping)
347 cxl_release_mapping(ctx);
348 mutex_lock(&ctx->afu->contexts_lock);
349 idr_remove(&ctx->afu->contexts_idr, ctx->pe);
350 mutex_unlock(&ctx->afu->contexts_lock);
351 call_rcu(&ctx->rcu, reclaim_ctx);
354 void cxl_context_mm_count_get(struct cxl_context *ctx)
357 atomic_inc(&ctx->mm->mm_count);
360 void cxl_context_mm_count_put(struct cxl_context *ctx)