1 /*
2  * PCI Backend Operations - respond to PCI requests from Frontend
3  *
4  *   Author: Ryan Wilson <hap9@epoch.ncsc.mil>
5  */
6 
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 
9 #include <linux/module.h>
10 #include <linux/wait.h>
11 #include <linux/bitops.h>
12 #include <xen/events.h>
13 #include <linux/sched.h>
14 #include "pciback.h"
15 
16 int verbose_request;
17 module_param(verbose_request, int, 0644);
18 
19 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id);
20 
21 /* Ensure a device is has the fake IRQ handler "turned on/off" and is
22  * ready to be exported. This MUST be run after xen_pcibk_reset_device
23  * which does the actual PCI device enable/disable.
24  */
25 static void xen_pcibk_control_isr(struct pci_dev *dev, int reset)
26 {
27 	struct xen_pcibk_dev_data *dev_data;
28 	int rc;
29 	int enable = 0;
30 
31 	dev_data = pci_get_drvdata(dev);
32 	if (!dev_data)
33 		return;
34 
35 	/* We don't deal with bridges */
36 	if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL)
37 		return;
38 
39 	if (reset) {
40 		dev_data->enable_intx = 0;
41 		dev_data->ack_intr = 0;
42 	}
43 	enable =  dev_data->enable_intx;
44 
45 	/* Asked to disable, but ISR isn't runnig */
46 	if (!enable && !dev_data->isr_on)
47 		return;
48 
49 	/* Squirrel away the IRQs in the dev_data. We need this
50 	 * b/c when device transitions to MSI, the dev->irq is
51 	 * overwritten with the MSI vector.
52 	 */
53 	if (enable)
54 		dev_data->irq = dev->irq;
55 
56 	/*
57 	 * SR-IOV devices in all use MSI-X and have no legacy
58 	 * interrupts, so inhibit creating a fake IRQ handler for them.
59 	 */
60 	if (dev_data->irq == 0)
61 		goto out;
62 
63 	dev_dbg(&dev->dev, "%s: #%d %s %s%s %s-> %s\n",
64 		dev_data->irq_name,
65 		dev_data->irq,
66 		pci_is_enabled(dev) ? "on" : "off",
67 		dev->msi_enabled ? "MSI" : "",
68 		dev->msix_enabled ? "MSI/X" : "",
69 		dev_data->isr_on ? "enable" : "disable",
70 		enable ? "enable" : "disable");
71 
72 	if (enable) {
73 		rc = request_irq(dev_data->irq,
74 				xen_pcibk_guest_interrupt, IRQF_SHARED,
75 				dev_data->irq_name, dev);
76 		if (rc) {
77 			dev_err(&dev->dev, "%s: failed to install fake IRQ " \
78 				"handler for IRQ %d! (rc:%d)\n",
79 				dev_data->irq_name, dev_data->irq, rc);
80 			goto out;
81 		}
82 	} else {
83 		free_irq(dev_data->irq, dev);
84 		dev_data->irq = 0;
85 	}
86 	dev_data->isr_on = enable;
87 	dev_data->ack_intr = enable;
88 out:
89 	dev_dbg(&dev->dev, "%s: #%d %s %s%s %s\n",
90 		dev_data->irq_name,
91 		dev_data->irq,
92 		pci_is_enabled(dev) ? "on" : "off",
93 		dev->msi_enabled ? "MSI" : "",
94 		dev->msix_enabled ? "MSI/X" : "",
95 		enable ? (dev_data->isr_on ? "enabled" : "failed to enable") :
96 			(dev_data->isr_on ? "failed to disable" : "disabled"));
97 }
98 
99 /* Ensure a device is "turned off" and ready to be exported.
100  * (Also see xen_pcibk_config_reset to ensure virtual configuration space is
101  * ready to be re-exported)
102  */
103 void xen_pcibk_reset_device(struct pci_dev *dev)
104 {
105 	u16 cmd;
106 
107 	xen_pcibk_control_isr(dev, 1 /* reset device */);
108 
109 	/* Disable devices (but not bridges) */
110 	if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
111 #ifdef CONFIG_PCI_MSI
112 		/* The guest could have been abruptly killed without
113 		 * disabling MSI/MSI-X interrupts.*/
114 		if (dev->msix_enabled)
115 			pci_disable_msix(dev);
116 		if (dev->msi_enabled)
117 			pci_disable_msi(dev);
118 #endif
119 		if (pci_is_enabled(dev))
120 			pci_disable_device(dev);
121 
122 		pci_write_config_word(dev, PCI_COMMAND, 0);
123 
124 		dev->is_busmaster = 0;
125 	} else {
126 		pci_read_config_word(dev, PCI_COMMAND, &cmd);
127 		if (cmd & (PCI_COMMAND_INVALIDATE)) {
128 			cmd &= ~(PCI_COMMAND_INVALIDATE);
129 			pci_write_config_word(dev, PCI_COMMAND, cmd);
130 
131 			dev->is_busmaster = 0;
132 		}
133 	}
134 }
135 
136 #ifdef CONFIG_PCI_MSI
137 static
138 int xen_pcibk_enable_msi(struct xen_pcibk_device *pdev,
139 			 struct pci_dev *dev, struct xen_pci_op *op)
140 {
141 	struct xen_pcibk_dev_data *dev_data;
142 	int status;
143 
144 	if (unlikely(verbose_request))
145 		printk(KERN_DEBUG DRV_NAME ": %s: enable MSI\n", pci_name(dev));
146 
147 	status = pci_enable_msi(dev);
148 
149 	if (status) {
150 		pr_warn_ratelimited("%s: error enabling MSI for guest %u: err %d\n",
151 				    pci_name(dev), pdev->xdev->otherend_id,
152 				    status);
153 		op->value = 0;
154 		return XEN_PCI_ERR_op_failed;
155 	}
156 
157 	/* The value the guest needs is actually the IDT vector, not the
158 	 * the local domain's IRQ number. */
159 
160 	op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
161 	if (unlikely(verbose_request))
162 		printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
163 			op->value);
164 
165 	dev_data = pci_get_drvdata(dev);
166 	if (dev_data)
167 		dev_data->ack_intr = 0;
168 
169 	return 0;
170 }
171 
172 static
173 int xen_pcibk_disable_msi(struct xen_pcibk_device *pdev,
174 			  struct pci_dev *dev, struct xen_pci_op *op)
175 {
176 	struct xen_pcibk_dev_data *dev_data;
177 
178 	if (unlikely(verbose_request))
179 		printk(KERN_DEBUG DRV_NAME ": %s: disable MSI\n",
180 		       pci_name(dev));
181 	pci_disable_msi(dev);
182 
183 	op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
184 	if (unlikely(verbose_request))
185 		printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
186 			op->value);
187 	dev_data = pci_get_drvdata(dev);
188 	if (dev_data)
189 		dev_data->ack_intr = 1;
190 	return 0;
191 }
192 
193 static
194 int xen_pcibk_enable_msix(struct xen_pcibk_device *pdev,
195 			  struct pci_dev *dev, struct xen_pci_op *op)
196 {
197 	struct xen_pcibk_dev_data *dev_data;
198 	int i, result;
199 	struct msix_entry *entries;
200 
201 	if (unlikely(verbose_request))
202 		printk(KERN_DEBUG DRV_NAME ": %s: enable MSI-X\n",
203 		       pci_name(dev));
204 	if (op->value > SH_INFO_MAX_VEC)
205 		return -EINVAL;
206 
207 	entries = kmalloc(op->value * sizeof(*entries), GFP_KERNEL);
208 	if (entries == NULL)
209 		return -ENOMEM;
210 
211 	for (i = 0; i < op->value; i++) {
212 		entries[i].entry = op->msix_entries[i].entry;
213 		entries[i].vector = op->msix_entries[i].vector;
214 	}
215 
216 	result = pci_enable_msix_exact(dev, entries, op->value);
217 	if (result == 0) {
218 		for (i = 0; i < op->value; i++) {
219 			op->msix_entries[i].entry = entries[i].entry;
220 			if (entries[i].vector) {
221 				op->msix_entries[i].vector =
222 					xen_pirq_from_irq(entries[i].vector);
223 				if (unlikely(verbose_request))
224 					printk(KERN_DEBUG DRV_NAME ": %s: " \
225 						"MSI-X[%d]: %d\n",
226 						pci_name(dev), i,
227 						op->msix_entries[i].vector);
228 			}
229 		}
230 	} else
231 		pr_warn_ratelimited("%s: error enabling MSI-X for guest %u: err %d!\n",
232 				    pci_name(dev), pdev->xdev->otherend_id,
233 				    result);
234 	kfree(entries);
235 
236 	op->value = result;
237 	dev_data = pci_get_drvdata(dev);
238 	if (dev_data)
239 		dev_data->ack_intr = 0;
240 
241 	return result > 0 ? 0 : result;
242 }
243 
244 static
245 int xen_pcibk_disable_msix(struct xen_pcibk_device *pdev,
246 			   struct pci_dev *dev, struct xen_pci_op *op)
247 {
248 	struct xen_pcibk_dev_data *dev_data;
249 	if (unlikely(verbose_request))
250 		printk(KERN_DEBUG DRV_NAME ": %s: disable MSI-X\n",
251 			pci_name(dev));
252 	pci_disable_msix(dev);
253 
254 	/*
255 	 * SR-IOV devices (which don't have any legacy IRQ) have
256 	 * an undefined IRQ value of zero.
257 	 */
258 	op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
259 	if (unlikely(verbose_request))
260 		printk(KERN_DEBUG DRV_NAME ": %s: MSI-X: %d\n", pci_name(dev),
261 			op->value);
262 	dev_data = pci_get_drvdata(dev);
263 	if (dev_data)
264 		dev_data->ack_intr = 1;
265 	return 0;
266 }
267 #endif
268 /*
269 * Now the same evtchn is used for both pcifront conf_read_write request
270 * as well as pcie aer front end ack. We use a new work_queue to schedule
271 * xen_pcibk conf_read_write service for avoiding confict with aer_core
272 * do_recovery job which also use the system default work_queue
273 */
274 void xen_pcibk_test_and_schedule_op(struct xen_pcibk_device *pdev)
275 {
276 	/* Check that frontend is requesting an operation and that we are not
277 	 * already processing a request */
278 	if (test_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags)
279 	    && !test_and_set_bit(_PDEVF_op_active, &pdev->flags)) {
280 		queue_work(xen_pcibk_wq, &pdev->op_work);
281 	}
282 	/*_XEN_PCIB_active should have been cleared by pcifront. And also make
283 	sure xen_pcibk is waiting for ack by checking _PCIB_op_pending*/
284 	if (!test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
285 	    && test_bit(_PCIB_op_pending, &pdev->flags)) {
286 		wake_up(&xen_pcibk_aer_wait_queue);
287 	}
288 }
289 
290 /* Performing the configuration space reads/writes must not be done in atomic
291  * context because some of the pci_* functions can sleep (mostly due to ACPI
292  * use of semaphores). This function is intended to be called from a work
293  * queue in process context taking a struct xen_pcibk_device as a parameter */
294 
295 void xen_pcibk_do_op(struct work_struct *data)
296 {
297 	struct xen_pcibk_device *pdev =
298 		container_of(data, struct xen_pcibk_device, op_work);
299 	struct pci_dev *dev;
300 	struct xen_pcibk_dev_data *dev_data = NULL;
301 	struct xen_pci_op *op = &pdev->sh_info->op;
302 	int test_intx = 0;
303 
304 	dev = xen_pcibk_get_pci_dev(pdev, op->domain, op->bus, op->devfn);
305 
306 	if (dev == NULL)
307 		op->err = XEN_PCI_ERR_dev_not_found;
308 	else {
309 		dev_data = pci_get_drvdata(dev);
310 		if (dev_data)
311 			test_intx = dev_data->enable_intx;
312 		switch (op->cmd) {
313 		case XEN_PCI_OP_conf_read:
314 			op->err = xen_pcibk_config_read(dev,
315 				  op->offset, op->size, &op->value);
316 			break;
317 		case XEN_PCI_OP_conf_write:
318 			op->err = xen_pcibk_config_write(dev,
319 				  op->offset, op->size,	op->value);
320 			break;
321 #ifdef CONFIG_PCI_MSI
322 		case XEN_PCI_OP_enable_msi:
323 			op->err = xen_pcibk_enable_msi(pdev, dev, op);
324 			break;
325 		case XEN_PCI_OP_disable_msi:
326 			op->err = xen_pcibk_disable_msi(pdev, dev, op);
327 			break;
328 		case XEN_PCI_OP_enable_msix:
329 			op->err = xen_pcibk_enable_msix(pdev, dev, op);
330 			break;
331 		case XEN_PCI_OP_disable_msix:
332 			op->err = xen_pcibk_disable_msix(pdev, dev, op);
333 			break;
334 #endif
335 		default:
336 			op->err = XEN_PCI_ERR_not_implemented;
337 			break;
338 		}
339 	}
340 	if (!op->err && dev && dev_data) {
341 		/* Transition detected */
342 		if ((dev_data->enable_intx != test_intx))
343 			xen_pcibk_control_isr(dev, 0 /* no reset */);
344 	}
345 	/* Tell the driver domain that we're done. */
346 	wmb();
347 	clear_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
348 	notify_remote_via_irq(pdev->evtchn_irq);
349 
350 	/* Mark that we're done. */
351 	smp_mb__before_atomic(); /* /after/ clearing PCIF_active */
352 	clear_bit(_PDEVF_op_active, &pdev->flags);
353 	smp_mb__after_atomic(); /* /before/ final check for work */
354 
355 	/* Check to see if the driver domain tried to start another request in
356 	 * between clearing _XEN_PCIF_active and clearing _PDEVF_op_active.
357 	*/
358 	xen_pcibk_test_and_schedule_op(pdev);
359 }
360 
361 irqreturn_t xen_pcibk_handle_event(int irq, void *dev_id)
362 {
363 	struct xen_pcibk_device *pdev = dev_id;
364 
365 	xen_pcibk_test_and_schedule_op(pdev);
366 
367 	return IRQ_HANDLED;
368 }
369 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id)
370 {
371 	struct pci_dev *dev = (struct pci_dev *)dev_id;
372 	struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
373 
374 	if (dev_data->isr_on && dev_data->ack_intr) {
375 		dev_data->handled++;
376 		if ((dev_data->handled % 1000) == 0) {
377 			if (xen_test_irq_shared(irq)) {
378 				pr_info("%s IRQ line is not shared "
379 					"with other domains. Turning ISR off\n",
380 					 dev_data->irq_name);
381 				dev_data->ack_intr = 0;
382 			}
383 		}
384 		return IRQ_HANDLED;
385 	}
386 	return IRQ_NONE;
387 }
388