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
16int verbose_request;
17module_param(verbose_request, int, 0644);
18
19static 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 */
25static 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;
88out:
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 */
103void 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
137static
138int 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
172static
173int 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
193static
194int 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	u16 cmd;
201
202	if (unlikely(verbose_request))
203		printk(KERN_DEBUG DRV_NAME ": %s: enable MSI-X\n",
204		       pci_name(dev));
205
206	if (op->value > SH_INFO_MAX_VEC)
207		return -EINVAL;
208
209	if (dev->msix_enabled)
210		return -EALREADY;
211
212	/*
213	 * PCI_COMMAND_MEMORY must be enabled, otherwise we may not be able
214	 * to access the BARs where the MSI-X entries reside.
215	 * But VF devices are unique in which the PF needs to be checked.
216	 */
217	pci_read_config_word(pci_physfn(dev), PCI_COMMAND, &cmd);
218	if (dev->msi_enabled || !(cmd & PCI_COMMAND_MEMORY))
219		return -ENXIO;
220
221	entries = kmalloc(op->value * sizeof(*entries), GFP_KERNEL);
222	if (entries == NULL)
223		return -ENOMEM;
224
225	for (i = 0; i < op->value; i++) {
226		entries[i].entry = op->msix_entries[i].entry;
227		entries[i].vector = op->msix_entries[i].vector;
228	}
229
230	result = pci_enable_msix_exact(dev, entries, op->value);
231	if (result == 0) {
232		for (i = 0; i < op->value; i++) {
233			op->msix_entries[i].entry = entries[i].entry;
234			if (entries[i].vector) {
235				op->msix_entries[i].vector =
236					xen_pirq_from_irq(entries[i].vector);
237				if (unlikely(verbose_request))
238					printk(KERN_DEBUG DRV_NAME ": %s: " \
239						"MSI-X[%d]: %d\n",
240						pci_name(dev), i,
241						op->msix_entries[i].vector);
242			}
243		}
244	} else
245		pr_warn_ratelimited("%s: error enabling MSI-X for guest %u: err %d!\n",
246				    pci_name(dev), pdev->xdev->otherend_id,
247				    result);
248	kfree(entries);
249
250	op->value = result;
251	dev_data = pci_get_drvdata(dev);
252	if (dev_data)
253		dev_data->ack_intr = 0;
254
255	return result > 0 ? 0 : result;
256}
257
258static
259int xen_pcibk_disable_msix(struct xen_pcibk_device *pdev,
260			   struct pci_dev *dev, struct xen_pci_op *op)
261{
262	struct xen_pcibk_dev_data *dev_data;
263	if (unlikely(verbose_request))
264		printk(KERN_DEBUG DRV_NAME ": %s: disable MSI-X\n",
265			pci_name(dev));
266	pci_disable_msix(dev);
267
268	/*
269	 * SR-IOV devices (which don't have any legacy IRQ) have
270	 * an undefined IRQ value of zero.
271	 */
272	op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
273	if (unlikely(verbose_request))
274		printk(KERN_DEBUG DRV_NAME ": %s: MSI-X: %d\n", pci_name(dev),
275			op->value);
276	dev_data = pci_get_drvdata(dev);
277	if (dev_data)
278		dev_data->ack_intr = 1;
279	return 0;
280}
281#endif
282/*
283* Now the same evtchn is used for both pcifront conf_read_write request
284* as well as pcie aer front end ack. We use a new work_queue to schedule
285* xen_pcibk conf_read_write service for avoiding confict with aer_core
286* do_recovery job which also use the system default work_queue
287*/
288void xen_pcibk_test_and_schedule_op(struct xen_pcibk_device *pdev)
289{
290	/* Check that frontend is requesting an operation and that we are not
291	 * already processing a request */
292	if (test_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags)
293	    && !test_and_set_bit(_PDEVF_op_active, &pdev->flags)) {
294		queue_work(xen_pcibk_wq, &pdev->op_work);
295	}
296	/*_XEN_PCIB_active should have been cleared by pcifront. And also make
297	sure xen_pcibk is waiting for ack by checking _PCIB_op_pending*/
298	if (!test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
299	    && test_bit(_PCIB_op_pending, &pdev->flags)) {
300		wake_up(&xen_pcibk_aer_wait_queue);
301	}
302}
303
304/* Performing the configuration space reads/writes must not be done in atomic
305 * context because some of the pci_* functions can sleep (mostly due to ACPI
306 * use of semaphores). This function is intended to be called from a work
307 * queue in process context taking a struct xen_pcibk_device as a parameter */
308
309void xen_pcibk_do_op(struct work_struct *data)
310{
311	struct xen_pcibk_device *pdev =
312		container_of(data, struct xen_pcibk_device, op_work);
313	struct pci_dev *dev;
314	struct xen_pcibk_dev_data *dev_data = NULL;
315	struct xen_pci_op *op = &pdev->op;
316	int test_intx = 0;
317#ifdef CONFIG_PCI_MSI
318	unsigned int nr = 0;
319#endif
320
321	*op = pdev->sh_info->op;
322	barrier();
323	dev = xen_pcibk_get_pci_dev(pdev, op->domain, op->bus, op->devfn);
324
325	if (dev == NULL)
326		op->err = XEN_PCI_ERR_dev_not_found;
327	else {
328		dev_data = pci_get_drvdata(dev);
329		if (dev_data)
330			test_intx = dev_data->enable_intx;
331		switch (op->cmd) {
332		case XEN_PCI_OP_conf_read:
333			op->err = xen_pcibk_config_read(dev,
334				  op->offset, op->size, &op->value);
335			break;
336		case XEN_PCI_OP_conf_write:
337			op->err = xen_pcibk_config_write(dev,
338				  op->offset, op->size,	op->value);
339			break;
340#ifdef CONFIG_PCI_MSI
341		case XEN_PCI_OP_enable_msi:
342			op->err = xen_pcibk_enable_msi(pdev, dev, op);
343			break;
344		case XEN_PCI_OP_disable_msi:
345			op->err = xen_pcibk_disable_msi(pdev, dev, op);
346			break;
347		case XEN_PCI_OP_enable_msix:
348			nr = op->value;
349			op->err = xen_pcibk_enable_msix(pdev, dev, op);
350			break;
351		case XEN_PCI_OP_disable_msix:
352			op->err = xen_pcibk_disable_msix(pdev, dev, op);
353			break;
354#endif
355		default:
356			op->err = XEN_PCI_ERR_not_implemented;
357			break;
358		}
359	}
360	if (!op->err && dev && dev_data) {
361		/* Transition detected */
362		if ((dev_data->enable_intx != test_intx))
363			xen_pcibk_control_isr(dev, 0 /* no reset */);
364	}
365	pdev->sh_info->op.err = op->err;
366	pdev->sh_info->op.value = op->value;
367#ifdef CONFIG_PCI_MSI
368	if (op->cmd == XEN_PCI_OP_enable_msix && op->err == 0) {
369		unsigned int i;
370
371		for (i = 0; i < nr; i++)
372			pdev->sh_info->op.msix_entries[i].vector =
373				op->msix_entries[i].vector;
374	}
375#endif
376	/* Tell the driver domain that we're done. */
377	wmb();
378	clear_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
379	notify_remote_via_irq(pdev->evtchn_irq);
380
381	/* Mark that we're done. */
382	smp_mb__before_atomic(); /* /after/ clearing PCIF_active */
383	clear_bit(_PDEVF_op_active, &pdev->flags);
384	smp_mb__after_atomic(); /* /before/ final check for work */
385
386	/* Check to see if the driver domain tried to start another request in
387	 * between clearing _XEN_PCIF_active and clearing _PDEVF_op_active.
388	*/
389	xen_pcibk_test_and_schedule_op(pdev);
390}
391
392irqreturn_t xen_pcibk_handle_event(int irq, void *dev_id)
393{
394	struct xen_pcibk_device *pdev = dev_id;
395
396	xen_pcibk_test_and_schedule_op(pdev);
397
398	return IRQ_HANDLED;
399}
400static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id)
401{
402	struct pci_dev *dev = (struct pci_dev *)dev_id;
403	struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
404
405	if (dev_data->isr_on && dev_data->ack_intr) {
406		dev_data->handled++;
407		if ((dev_data->handled % 1000) == 0) {
408			if (xen_test_irq_shared(irq)) {
409				pr_info("%s IRQ line is not shared "
410					"with other domains. Turning ISR off\n",
411					 dev_data->irq_name);
412				dev_data->ack_intr = 0;
413			}
414		}
415		return IRQ_HANDLED;
416	}
417	return IRQ_NONE;
418}
419