1/*
2 * Copyright 2014 IBM Corp.
3 *
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.
8 */
9
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>
15#include <linux/fs.h>
16#include <linux/mm.h>
17#include <linux/debugfs.h>
18#include <linux/slab.h>
19#include <linux/idr.h>
20#include <asm/cputable.h>
21#include <asm/current.h>
22#include <asm/copro.h>
23
24#include "cxl.h"
25
26/*
27 * Allocates space for a CXL context.
28 */
29struct cxl_context *cxl_context_alloc(void)
30{
31	return kzalloc(sizeof(struct cxl_context), GFP_KERNEL);
32}
33
34/*
35 * Initialises a CXL context.
36 */
37int cxl_context_init(struct cxl_context *ctx, struct cxl_afu *afu, bool master,
38		     struct address_space *mapping)
39{
40	int i;
41
42	spin_lock_init(&ctx->sste_lock);
43	ctx->afu = afu;
44	ctx->master = master;
45	ctx->pid = NULL; /* Set in start work ioctl */
46	mutex_init(&ctx->mapping_lock);
47	ctx->mapping = mapping;
48
49	/*
50	 * Allocate the segment table before we put it in the IDR so that we
51	 * can always access it when dereferenced from IDR. For the same
52	 * reason, the segment table is only destroyed after the context is
53	 * removed from the IDR.  Access to this in the IOCTL is protected by
54	 * Linux filesytem symantics (can't IOCTL until open is complete).
55	 */
56	i = cxl_alloc_sst(ctx);
57	if (i)
58		return i;
59
60	INIT_WORK(&ctx->fault_work, cxl_handle_fault);
61
62	init_waitqueue_head(&ctx->wq);
63	spin_lock_init(&ctx->lock);
64
65	ctx->irq_bitmap = NULL;
66	ctx->pending_irq = false;
67	ctx->pending_fault = false;
68	ctx->pending_afu_err = false;
69
70	/*
71	 * When we have to destroy all contexts in cxl_context_detach_all() we
72	 * end up with afu_release_irqs() called from inside a
73	 * idr_for_each_entry(). Hence we need to make sure that anything
74	 * dereferenced from this IDR is ok before we allocate the IDR here.
75	 * This clears out the IRQ ranges to ensure this.
76	 */
77	for (i = 0; i < CXL_IRQ_RANGES; i++)
78		ctx->irqs.range[i] = 0;
79
80	mutex_init(&ctx->status_mutex);
81
82	ctx->status = OPENED;
83
84	/*
85	 * Allocating IDR! We better make sure everything's setup that
86	 * dereferences from it.
87	 */
88	mutex_lock(&afu->contexts_lock);
89	idr_preload(GFP_KERNEL);
90	i = idr_alloc(&ctx->afu->contexts_idr, ctx, 0,
91		      ctx->afu->num_procs, GFP_NOWAIT);
92	idr_preload_end();
93	mutex_unlock(&afu->contexts_lock);
94	if (i < 0)
95		return i;
96
97	ctx->pe = i;
98	ctx->elem = &ctx->afu->spa[i];
99	ctx->pe_inserted = false;
100	return 0;
101}
102
103static int cxl_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
104{
105	struct cxl_context *ctx = vma->vm_file->private_data;
106	unsigned long address = (unsigned long)vmf->virtual_address;
107	u64 area, offset;
108
109	offset = vmf->pgoff << PAGE_SHIFT;
110
111	pr_devel("%s: pe: %i address: 0x%lx offset: 0x%llx\n",
112			__func__, ctx->pe, address, offset);
113
114	if (ctx->afu->current_mode == CXL_MODE_DEDICATED) {
115		area = ctx->afu->psn_phys;
116		if (offset >= ctx->afu->adapter->ps_size)
117			return VM_FAULT_SIGBUS;
118	} else {
119		area = ctx->psn_phys;
120		if (offset >= ctx->psn_size)
121			return VM_FAULT_SIGBUS;
122	}
123
124	mutex_lock(&ctx->status_mutex);
125
126	if (ctx->status != STARTED) {
127		mutex_unlock(&ctx->status_mutex);
128		pr_devel("%s: Context not started, failing problem state access\n", __func__);
129		if (ctx->mmio_err_ff) {
130			if (!ctx->ff_page) {
131				ctx->ff_page = alloc_page(GFP_USER);
132				if (!ctx->ff_page)
133					return VM_FAULT_OOM;
134				memset(page_address(ctx->ff_page), 0xff, PAGE_SIZE);
135			}
136			get_page(ctx->ff_page);
137			vmf->page = ctx->ff_page;
138			vma->vm_page_prot = pgprot_cached(vma->vm_page_prot);
139			return 0;
140		}
141		return VM_FAULT_SIGBUS;
142	}
143
144	vm_insert_pfn(vma, address, (area + offset) >> PAGE_SHIFT);
145
146	mutex_unlock(&ctx->status_mutex);
147
148	return VM_FAULT_NOPAGE;
149}
150
151static const struct vm_operations_struct cxl_mmap_vmops = {
152	.fault = cxl_mmap_fault,
153};
154
155/*
156 * Map a per-context mmio space into the given vma.
157 */
158int cxl_context_iomap(struct cxl_context *ctx, struct vm_area_struct *vma)
159{
160	u64 start = vma->vm_pgoff << PAGE_SHIFT;
161	u64 len = vma->vm_end - vma->vm_start;
162
163	if (ctx->afu->current_mode == CXL_MODE_DEDICATED) {
164		if (start + len > ctx->afu->adapter->ps_size)
165			return -EINVAL;
166	} else {
167		if (start + len > ctx->psn_size)
168			return -EINVAL;
169	}
170
171	if (ctx->afu->current_mode != CXL_MODE_DEDICATED) {
172		/* make sure there is a valid per process space for this AFU */
173		if ((ctx->master && !ctx->afu->psa) || (!ctx->afu->pp_psa)) {
174			pr_devel("AFU doesn't support mmio space\n");
175			return -EINVAL;
176		}
177
178		/* Can't mmap until the AFU is enabled */
179		if (!ctx->afu->enabled)
180			return -EBUSY;
181	}
182
183	pr_devel("%s: mmio physical: %llx pe: %i master:%i\n", __func__,
184		 ctx->psn_phys, ctx->pe , ctx->master);
185
186	vma->vm_flags |= VM_IO | VM_PFNMAP;
187	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
188	vma->vm_ops = &cxl_mmap_vmops;
189	return 0;
190}
191
192/*
193 * Detach a context from the hardware. This disables interrupts and doesn't
194 * return until all outstanding interrupts for this context have completed. The
195 * hardware should no longer access *ctx after this has returned.
196 */
197int __detach_context(struct cxl_context *ctx)
198{
199	enum cxl_context_status status;
200
201	mutex_lock(&ctx->status_mutex);
202	status = ctx->status;
203	ctx->status = CLOSED;
204	mutex_unlock(&ctx->status_mutex);
205	if (status != STARTED)
206		return -EBUSY;
207
208	/* Only warn if we detached while the link was OK.
209	 * If detach fails when hw is down, we don't care.
210	 */
211	WARN_ON(cxl_detach_process(ctx) &&
212		cxl_adapter_link_ok(ctx->afu->adapter));
213	flush_work(&ctx->fault_work); /* Only needed for dedicated process */
214	put_pid(ctx->pid);
215	cxl_ctx_put();
216	return 0;
217}
218
219/*
220 * Detach the given context from the AFU. This doesn't actually
221 * free the context but it should stop the context running in hardware
222 * (ie. prevent this context from generating any further interrupts
223 * so that it can be freed).
224 */
225void cxl_context_detach(struct cxl_context *ctx)
226{
227	int rc;
228
229	rc = __detach_context(ctx);
230	if (rc)
231		return;
232
233	afu_release_irqs(ctx, ctx);
234	wake_up_all(&ctx->wq);
235}
236
237/*
238 * Detach all contexts on the given AFU.
239 */
240void cxl_context_detach_all(struct cxl_afu *afu)
241{
242	struct cxl_context *ctx;
243	int tmp;
244
245	mutex_lock(&afu->contexts_lock);
246	idr_for_each_entry(&afu->contexts_idr, ctx, tmp) {
247		/*
248		 * Anything done in here needs to be setup before the IDR is
249		 * created and torn down after the IDR removed
250		 */
251		cxl_context_detach(ctx);
252
253		/*
254		 * We are force detaching - remove any active PSA mappings so
255		 * userspace cannot interfere with the card if it comes back.
256		 * Easiest way to exercise this is to unbind and rebind the
257		 * driver via sysfs while it is in use.
258		 */
259		mutex_lock(&ctx->mapping_lock);
260		if (ctx->mapping)
261			unmap_mapping_range(ctx->mapping, 0, 0, 1);
262		mutex_unlock(&ctx->mapping_lock);
263	}
264	mutex_unlock(&afu->contexts_lock);
265}
266
267static void reclaim_ctx(struct rcu_head *rcu)
268{
269	struct cxl_context *ctx = container_of(rcu, struct cxl_context, rcu);
270
271	free_page((u64)ctx->sstp);
272	if (ctx->ff_page)
273		__free_page(ctx->ff_page);
274	ctx->sstp = NULL;
275	if (ctx->kernelapi)
276		kfree(ctx->mapping);
277
278	if (ctx->irq_bitmap)
279		kfree(ctx->irq_bitmap);
280
281	kfree(ctx);
282}
283
284void cxl_context_free(struct cxl_context *ctx)
285{
286	mutex_lock(&ctx->afu->contexts_lock);
287	idr_remove(&ctx->afu->contexts_idr, ctx->pe);
288	mutex_unlock(&ctx->afu->contexts_lock);
289	call_rcu(&ctx->rcu, reclaim_ctx);
290}
291