2 * PCI Backend Operations - respond to PCI requests from Frontend
4 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
17 module_param(verbose_request
, int, 0644);
19 static irqreturn_t
xen_pcibk_guest_interrupt(int irq
, void *dev_id
);
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.
25 static void xen_pcibk_control_isr(struct pci_dev
*dev
, int reset
)
27 struct xen_pcibk_dev_data
*dev_data
;
31 dev_data
= pci_get_drvdata(dev
);
35 /* We don't deal with bridges */
36 if (dev
->hdr_type
!= PCI_HEADER_TYPE_NORMAL
)
40 dev_data
->enable_intx
= 0;
41 dev_data
->ack_intr
= 0;
43 enable
= dev_data
->enable_intx
;
45 /* Asked to disable, but ISR isn't runnig */
46 if (!enable
&& !dev_data
->isr_on
)
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.
54 dev_data
->irq
= dev
->irq
;
57 * SR-IOV devices in all use MSI-X and have no legacy
58 * interrupts, so inhibit creating a fake IRQ handler for them.
60 if (dev_data
->irq
== 0)
63 dev_dbg(&dev
->dev
, "%s: #%d %s %s%s %s-> %s\n",
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");
73 rc
= request_irq(dev_data
->irq
,
74 xen_pcibk_guest_interrupt
, IRQF_SHARED
,
75 dev_data
->irq_name
, dev
);
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
);
83 free_irq(dev_data
->irq
, dev
);
86 dev_data
->isr_on
= enable
;
87 dev_data
->ack_intr
= enable
;
89 dev_dbg(&dev
->dev
, "%s: #%d %s %s%s %s\n",
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"));
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)
103 void xen_pcibk_reset_device(struct pci_dev
*dev
)
107 xen_pcibk_control_isr(dev
, 1 /* reset device */);
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
);
119 if (pci_is_enabled(dev
))
120 pci_disable_device(dev
);
122 pci_write_config_word(dev
, PCI_COMMAND
, 0);
124 dev
->is_busmaster
= 0;
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
);
131 dev
->is_busmaster
= 0;
136 #ifdef CONFIG_PCI_MSI
138 int xen_pcibk_enable_msi(struct xen_pcibk_device
*pdev
,
139 struct pci_dev
*dev
, struct xen_pci_op
*op
)
141 struct xen_pcibk_dev_data
*dev_data
;
144 if (unlikely(verbose_request
))
145 printk(KERN_DEBUG DRV_NAME
": %s: enable MSI\n", pci_name(dev
));
147 status
= pci_enable_msi(dev
);
150 pr_warn_ratelimited("%s: error enabling MSI for guest %u: err %d\n",
151 pci_name(dev
), pdev
->xdev
->otherend_id
,
154 return XEN_PCI_ERR_op_failed
;
157 /* The value the guest needs is actually the IDT vector, not the
158 * the local domain's IRQ number. */
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
),
165 dev_data
= pci_get_drvdata(dev
);
167 dev_data
->ack_intr
= 0;
173 int xen_pcibk_disable_msi(struct xen_pcibk_device
*pdev
,
174 struct pci_dev
*dev
, struct xen_pci_op
*op
)
176 struct xen_pcibk_dev_data
*dev_data
;
178 if (unlikely(verbose_request
))
179 printk(KERN_DEBUG DRV_NAME
": %s: disable MSI\n",
181 pci_disable_msi(dev
);
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
),
187 dev_data
= pci_get_drvdata(dev
);
189 dev_data
->ack_intr
= 1;
194 int xen_pcibk_enable_msix(struct xen_pcibk_device
*pdev
,
195 struct pci_dev
*dev
, struct xen_pci_op
*op
)
197 struct xen_pcibk_dev_data
*dev_data
;
199 struct msix_entry
*entries
;
201 if (unlikely(verbose_request
))
202 printk(KERN_DEBUG DRV_NAME
": %s: enable MSI-X\n",
204 if (op
->value
> SH_INFO_MAX_VEC
)
207 entries
= kmalloc(op
->value
* sizeof(*entries
), GFP_KERNEL
);
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
;
216 result
= pci_enable_msix_exact(dev
, entries
, op
->value
);
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: " \
227 op
->msix_entries
[i
].vector
);
230 pr_warn_ratelimited("%s: error enabling MSI-X for guest %u: err %d!\n",
231 pci_name(dev
), pdev
->xdev
->otherend_id
,
236 dev_data
= pci_get_drvdata(dev
);
238 dev_data
->ack_intr
= 0;
240 return result
> 0 ? 0 : result
;
244 int xen_pcibk_disable_msix(struct xen_pcibk_device
*pdev
,
245 struct pci_dev
*dev
, struct xen_pci_op
*op
)
247 struct xen_pcibk_dev_data
*dev_data
;
248 if (unlikely(verbose_request
))
249 printk(KERN_DEBUG DRV_NAME
": %s: disable MSI-X\n",
251 pci_disable_msix(dev
);
254 * SR-IOV devices (which don't have any legacy IRQ) have
255 * an undefined IRQ value of zero.
257 op
->value
= dev
->irq
? xen_pirq_from_irq(dev
->irq
) : 0;
258 if (unlikely(verbose_request
))
259 printk(KERN_DEBUG DRV_NAME
": %s: MSI-X: %d\n", pci_name(dev
),
261 dev_data
= pci_get_drvdata(dev
);
263 dev_data
->ack_intr
= 1;
268 * Now the same evtchn is used for both pcifront conf_read_write request
269 * as well as pcie aer front end ack. We use a new work_queue to schedule
270 * xen_pcibk conf_read_write service for avoiding confict with aer_core
271 * do_recovery job which also use the system default work_queue
273 void xen_pcibk_test_and_schedule_op(struct xen_pcibk_device
*pdev
)
275 /* Check that frontend is requesting an operation and that we are not
276 * already processing a request */
277 if (test_bit(_XEN_PCIF_active
, (unsigned long *)&pdev
->sh_info
->flags
)
278 && !test_and_set_bit(_PDEVF_op_active
, &pdev
->flags
)) {
279 queue_work(xen_pcibk_wq
, &pdev
->op_work
);
281 /*_XEN_PCIB_active should have been cleared by pcifront. And also make
282 sure xen_pcibk is waiting for ack by checking _PCIB_op_pending*/
283 if (!test_bit(_XEN_PCIB_active
, (unsigned long *)&pdev
->sh_info
->flags
)
284 && test_bit(_PCIB_op_pending
, &pdev
->flags
)) {
285 wake_up(&xen_pcibk_aer_wait_queue
);
289 /* Performing the configuration space reads/writes must not be done in atomic
290 * context because some of the pci_* functions can sleep (mostly due to ACPI
291 * use of semaphores). This function is intended to be called from a work
292 * queue in process context taking a struct xen_pcibk_device as a parameter */
294 void xen_pcibk_do_op(struct work_struct
*data
)
296 struct xen_pcibk_device
*pdev
=
297 container_of(data
, struct xen_pcibk_device
, op_work
);
299 struct xen_pcibk_dev_data
*dev_data
= NULL
;
300 struct xen_pci_op
*op
= &pdev
->sh_info
->op
;
303 dev
= xen_pcibk_get_pci_dev(pdev
, op
->domain
, op
->bus
, op
->devfn
);
306 op
->err
= XEN_PCI_ERR_dev_not_found
;
308 dev_data
= pci_get_drvdata(dev
);
310 test_intx
= dev_data
->enable_intx
;
312 case XEN_PCI_OP_conf_read
:
313 op
->err
= xen_pcibk_config_read(dev
,
314 op
->offset
, op
->size
, &op
->value
);
316 case XEN_PCI_OP_conf_write
:
317 op
->err
= xen_pcibk_config_write(dev
,
318 op
->offset
, op
->size
, op
->value
);
320 #ifdef CONFIG_PCI_MSI
321 case XEN_PCI_OP_enable_msi
:
322 op
->err
= xen_pcibk_enable_msi(pdev
, dev
, op
);
324 case XEN_PCI_OP_disable_msi
:
325 op
->err
= xen_pcibk_disable_msi(pdev
, dev
, op
);
327 case XEN_PCI_OP_enable_msix
:
328 op
->err
= xen_pcibk_enable_msix(pdev
, dev
, op
);
330 case XEN_PCI_OP_disable_msix
:
331 op
->err
= xen_pcibk_disable_msix(pdev
, dev
, op
);
335 op
->err
= XEN_PCI_ERR_not_implemented
;
339 if (!op
->err
&& dev
&& dev_data
) {
340 /* Transition detected */
341 if ((dev_data
->enable_intx
!= test_intx
))
342 xen_pcibk_control_isr(dev
, 0 /* no reset */);
344 /* Tell the driver domain that we're done. */
346 clear_bit(_XEN_PCIF_active
, (unsigned long *)&pdev
->sh_info
->flags
);
347 notify_remote_via_irq(pdev
->evtchn_irq
);
349 /* Mark that we're done. */
350 smp_mb__before_clear_bit(); /* /after/ clearing PCIF_active */
351 clear_bit(_PDEVF_op_active
, &pdev
->flags
);
352 smp_mb__after_clear_bit(); /* /before/ final check for work */
354 /* Check to see if the driver domain tried to start another request in
355 * between clearing _XEN_PCIF_active and clearing _PDEVF_op_active.
357 xen_pcibk_test_and_schedule_op(pdev
);
360 irqreturn_t
xen_pcibk_handle_event(int irq
, void *dev_id
)
362 struct xen_pcibk_device
*pdev
= dev_id
;
364 xen_pcibk_test_and_schedule_op(pdev
);
368 static irqreturn_t
xen_pcibk_guest_interrupt(int irq
, void *dev_id
)
370 struct pci_dev
*dev
= (struct pci_dev
*)dev_id
;
371 struct xen_pcibk_dev_data
*dev_data
= pci_get_drvdata(dev
);
373 if (dev_data
->isr_on
&& dev_data
->ack_intr
) {
375 if ((dev_data
->handled
% 1000) == 0) {
376 if (xen_test_irq_shared(irq
)) {
377 pr_info("%s IRQ line is not shared "
378 "with other domains. Turning ISR off\n",
380 dev_data
->ack_intr
= 0;