1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright(c) 2020 Intel Corporation. All rights reserved. */
3 #include <linux/unaligned.h>
4 #include <linux/io-64-nonatomic-lo-hi.h>
5 #include <linux/moduleparam.h>
6 #include <linux/module.h>
7 #include <linux/delay.h>
8 #include <linux/sizes.h>
9 #include <linux/mutex.h>
10 #include <linux/list.h>
11 #include <linux/pci.h>
12 #include <linux/aer.h>
14 #include <cxl/mailbox.h>
23 * This implements the PCI exclusive functionality for a CXL device as it is
24 * defined by the Compute Express Link specification. CXL devices may surface
25 * certain functionality even if it isn't CXL enabled. While this driver is
26 * focused around the PCI specific aspects of a CXL device, it binds to the
27 * specific CXL memory device class code, and therefore the implementation of
28 * cxl_pci is focused around CXL memory devices.
30 * The driver has several responsibilities, mainly:
31 * - Create the memX device and register on the CXL bus.
32 * - Enumerate device's register interface and map them.
33 * - Registers nvdimm bridge device with cxl_core.
34 * - Registers a CXL mailbox with cxl_core.
37 #define cxl_doorbell_busy(cxlds) \
38 (readl((cxlds)->regs.mbox + CXLDEV_MBOX_CTRL_OFFSET) & \
39 CXLDEV_MBOX_CTRL_DOORBELL)
41 /* CXL 2.0 - 8.2.8.4 */
42 #define CXL_MAILBOX_TIMEOUT_MS (2 * HZ)
45 * CXL 2.0 ECN "Add Mailbox Ready Time" defines a capability field to
46 * dictate how long to wait for the mailbox to become ready. The new
47 * field allows the device to tell software the amount of time to wait
48 * before mailbox ready. This field per the spec theoretically allows
49 * for up to 255 seconds. 255 seconds is unreasonably long, its longer
50 * than the maximum SATA port link recovery wait. Default to 60 seconds
51 * until someone builds a CXL device that needs more time in practice.
53 static unsigned short mbox_ready_timeout
= 60;
54 module_param(mbox_ready_timeout
, ushort
, 0644);
55 MODULE_PARM_DESC(mbox_ready_timeout
, "seconds to wait for mailbox ready");
57 static int cxl_pci_mbox_wait_for_doorbell(struct cxl_dev_state
*cxlds
)
59 const unsigned long start
= jiffies
;
60 unsigned long end
= start
;
62 while (cxl_doorbell_busy(cxlds
)) {
65 if (time_after(end
, start
+ CXL_MAILBOX_TIMEOUT_MS
)) {
66 /* Check again in case preempted before timeout test */
67 if (!cxl_doorbell_busy(cxlds
))
74 dev_dbg(cxlds
->dev
, "Doorbell wait took %dms",
75 jiffies_to_msecs(end
) - jiffies_to_msecs(start
));
79 #define cxl_err(dev, status, msg) \
80 dev_err_ratelimited(dev, msg ", device state %s%s\n", \
81 status & CXLMDEV_DEV_FATAL ? " fatal" : "", \
82 status & CXLMDEV_FW_HALT ? " firmware-halt" : "")
84 #define cxl_cmd_err(dev, cmd, status, msg) \
85 dev_err_ratelimited(dev, msg " (opcode: %#x), device state %s%s\n", \
87 status & CXLMDEV_DEV_FATAL ? " fatal" : "", \
88 status & CXLMDEV_FW_HALT ? " firmware-halt" : "")
91 * Threaded irq dev_id's must be globally unique. cxl_dev_id provides a unique
92 * wrapper object for each irq within the same cxlds.
95 struct cxl_dev_state
*cxlds
;
98 static int cxl_request_irq(struct cxl_dev_state
*cxlds
, int irq
,
99 irq_handler_t thread_fn
)
101 struct device
*dev
= cxlds
->dev
;
102 struct cxl_dev_id
*dev_id
;
104 dev_id
= devm_kzalloc(dev
, sizeof(*dev_id
), GFP_KERNEL
);
107 dev_id
->cxlds
= cxlds
;
109 return devm_request_threaded_irq(dev
, irq
, NULL
, thread_fn
,
110 IRQF_SHARED
| IRQF_ONESHOT
, NULL
,
114 static bool cxl_mbox_background_complete(struct cxl_dev_state
*cxlds
)
118 reg
= readq(cxlds
->regs
.mbox
+ CXLDEV_MBOX_BG_CMD_STATUS_OFFSET
);
119 return FIELD_GET(CXLDEV_MBOX_BG_CMD_COMMAND_PCT_MASK
, reg
) == 100;
122 static irqreturn_t
cxl_pci_mbox_irq(int irq
, void *id
)
126 struct cxl_dev_id
*dev_id
= id
;
127 struct cxl_dev_state
*cxlds
= dev_id
->cxlds
;
128 struct cxl_mailbox
*cxl_mbox
= &cxlds
->cxl_mbox
;
129 struct cxl_memdev_state
*mds
= to_cxl_memdev_state(cxlds
);
131 if (!cxl_mbox_background_complete(cxlds
))
134 reg
= readq(cxlds
->regs
.mbox
+ CXLDEV_MBOX_BG_CMD_STATUS_OFFSET
);
135 opcode
= FIELD_GET(CXLDEV_MBOX_BG_CMD_COMMAND_OPCODE_MASK
, reg
);
136 if (opcode
== CXL_MBOX_OP_SANITIZE
) {
137 mutex_lock(&cxl_mbox
->mbox_mutex
);
138 if (mds
->security
.sanitize_node
)
139 mod_delayed_work(system_wq
, &mds
->security
.poll_dwork
, 0);
140 mutex_unlock(&cxl_mbox
->mbox_mutex
);
142 /* short-circuit the wait in __cxl_pci_mbox_send_cmd() */
143 rcuwait_wake_up(&cxl_mbox
->mbox_wait
);
150 * Sanitization operation polling mode.
152 static void cxl_mbox_sanitize_work(struct work_struct
*work
)
154 struct cxl_memdev_state
*mds
=
155 container_of(work
, typeof(*mds
), security
.poll_dwork
.work
);
156 struct cxl_dev_state
*cxlds
= &mds
->cxlds
;
157 struct cxl_mailbox
*cxl_mbox
= &cxlds
->cxl_mbox
;
159 mutex_lock(&cxl_mbox
->mbox_mutex
);
160 if (cxl_mbox_background_complete(cxlds
)) {
161 mds
->security
.poll_tmo_secs
= 0;
162 if (mds
->security
.sanitize_node
)
163 sysfs_notify_dirent(mds
->security
.sanitize_node
);
164 mds
->security
.sanitize_active
= false;
166 dev_dbg(cxlds
->dev
, "Sanitization operation ended\n");
168 int timeout
= mds
->security
.poll_tmo_secs
+ 10;
170 mds
->security
.poll_tmo_secs
= min(15 * 60, timeout
);
171 schedule_delayed_work(&mds
->security
.poll_dwork
, timeout
* HZ
);
173 mutex_unlock(&cxl_mbox
->mbox_mutex
);
177 * __cxl_pci_mbox_send_cmd() - Execute a mailbox command
178 * @cxl_mbox: CXL mailbox context
179 * @mbox_cmd: Command to send to the memory device.
181 * Context: Any context. Expects mbox_mutex to be held.
182 * Return: -ETIMEDOUT if timeout occurred waiting for completion. 0 on success.
183 * Caller should check the return code in @mbox_cmd to make sure it
186 * This is a generic form of the CXL mailbox send command thus only using the
187 * registers defined by the mailbox capability ID - CXL 2.0 8.2.8.4. Memory
188 * devices, and perhaps other types of CXL devices may have further information
189 * available upon error conditions. Driver facilities wishing to send mailbox
190 * commands should use the wrapper command.
192 * The CXL spec allows for up to two mailboxes. The intention is for the primary
193 * mailbox to be OS controlled and the secondary mailbox to be used by system
194 * firmware. This allows the OS and firmware to communicate with the device and
195 * not need to coordinate with each other. The driver only uses the primary
198 static int __cxl_pci_mbox_send_cmd(struct cxl_mailbox
*cxl_mbox
,
199 struct cxl_mbox_cmd
*mbox_cmd
)
201 struct cxl_dev_state
*cxlds
= mbox_to_cxlds(cxl_mbox
);
202 struct cxl_memdev_state
*mds
= to_cxl_memdev_state(cxlds
);
203 void __iomem
*payload
= cxlds
->regs
.mbox
+ CXLDEV_MBOX_PAYLOAD_OFFSET
;
204 struct device
*dev
= cxlds
->dev
;
205 u64 cmd_reg
, status_reg
;
209 lockdep_assert_held(&cxl_mbox
->mbox_mutex
);
212 * Here are the steps from 8.2.8.4 of the CXL 2.0 spec.
213 * 1. Caller reads MB Control Register to verify doorbell is clear
214 * 2. Caller writes Command Register
215 * 3. Caller writes Command Payload Registers if input payload is non-empty
216 * 4. Caller writes MB Control Register to set doorbell
217 * 5. Caller either polls for doorbell to be clear or waits for interrupt if configured
218 * 6. Caller reads MB Status Register to fetch Return code
219 * 7. If command successful, Caller reads Command Register to get Payload Length
220 * 8. If output payload is non-empty, host reads Command Payload Registers
222 * Hardware is free to do whatever it wants before the doorbell is rung,
223 * and isn't allowed to change anything after it clears the doorbell. As
224 * such, steps 2 and 3 can happen in any order, and steps 6, 7, 8 can
225 * also happen in any order (though some orders might not make sense).
229 if (cxl_doorbell_busy(cxlds
)) {
231 readq(cxlds
->regs
.memdev
+ CXLMDEV_STATUS_OFFSET
);
233 cxl_cmd_err(cxlds
->dev
, mbox_cmd
, md_status
,
234 "mailbox queue busy");
239 * With sanitize polling, hardware might be done and the poller still
240 * not be in sync. Ensure no new command comes in until so. Keep the
241 * hardware semantics and only allow device health status.
243 if (mds
->security
.poll_tmo_secs
> 0) {
244 if (mbox_cmd
->opcode
!= CXL_MBOX_OP_GET_HEALTH_INFO
)
248 cmd_reg
= FIELD_PREP(CXLDEV_MBOX_CMD_COMMAND_OPCODE_MASK
,
250 if (mbox_cmd
->size_in
) {
251 if (WARN_ON(!mbox_cmd
->payload_in
))
254 cmd_reg
|= FIELD_PREP(CXLDEV_MBOX_CMD_PAYLOAD_LENGTH_MASK
,
256 memcpy_toio(payload
, mbox_cmd
->payload_in
, mbox_cmd
->size_in
);
260 writeq(cmd_reg
, cxlds
->regs
.mbox
+ CXLDEV_MBOX_CMD_OFFSET
);
263 dev_dbg(dev
, "Sending command: 0x%04x\n", mbox_cmd
->opcode
);
264 writel(CXLDEV_MBOX_CTRL_DOORBELL
,
265 cxlds
->regs
.mbox
+ CXLDEV_MBOX_CTRL_OFFSET
);
268 rc
= cxl_pci_mbox_wait_for_doorbell(cxlds
);
269 if (rc
== -ETIMEDOUT
) {
270 u64 md_status
= readq(cxlds
->regs
.memdev
+ CXLMDEV_STATUS_OFFSET
);
272 cxl_cmd_err(cxlds
->dev
, mbox_cmd
, md_status
, "mailbox timeout");
277 status_reg
= readq(cxlds
->regs
.mbox
+ CXLDEV_MBOX_STATUS_OFFSET
);
278 mbox_cmd
->return_code
=
279 FIELD_GET(CXLDEV_MBOX_STATUS_RET_CODE_MASK
, status_reg
);
282 * Handle the background command in a synchronous manner.
284 * All other mailbox commands will serialize/queue on the mbox_mutex,
285 * which we currently hold. Furthermore this also guarantees that
286 * cxl_mbox_background_complete() checks are safe amongst each other,
287 * in that no new bg operation can occur in between.
289 * Background operations are timesliced in accordance with the nature
290 * of the command. In the event of timeout, the mailbox state is
291 * indeterminate until the next successful command submission and the
292 * driver can get back in sync with the hardware state.
294 if (mbox_cmd
->return_code
== CXL_MBOX_CMD_RC_BACKGROUND
) {
299 * Sanitization is a special case which monopolizes the device
300 * and cannot be timesliced. Handle asynchronously instead,
301 * and allow userspace to poll(2) for completion.
303 if (mbox_cmd
->opcode
== CXL_MBOX_OP_SANITIZE
) {
304 if (mds
->security
.sanitize_active
)
307 /* give first timeout a second */
309 mds
->security
.poll_tmo_secs
= timeout
;
310 mds
->security
.sanitize_active
= true;
311 schedule_delayed_work(&mds
->security
.poll_dwork
,
313 dev_dbg(dev
, "Sanitization operation started\n");
317 dev_dbg(dev
, "Mailbox background operation (0x%04x) started\n",
320 timeout
= mbox_cmd
->poll_interval_ms
;
321 for (i
= 0; i
< mbox_cmd
->poll_count
; i
++) {
322 if (rcuwait_wait_event_timeout(&cxl_mbox
->mbox_wait
,
323 cxl_mbox_background_complete(cxlds
),
324 TASK_UNINTERRUPTIBLE
,
325 msecs_to_jiffies(timeout
)) > 0)
329 if (!cxl_mbox_background_complete(cxlds
)) {
330 dev_err(dev
, "timeout waiting for background (%d ms)\n",
331 timeout
* mbox_cmd
->poll_count
);
335 bg_status_reg
= readq(cxlds
->regs
.mbox
+
336 CXLDEV_MBOX_BG_CMD_STATUS_OFFSET
);
337 mbox_cmd
->return_code
=
338 FIELD_GET(CXLDEV_MBOX_BG_CMD_COMMAND_RC_MASK
,
341 "Mailbox background operation (0x%04x) completed\n",
345 if (mbox_cmd
->return_code
!= CXL_MBOX_CMD_RC_SUCCESS
) {
346 dev_dbg(dev
, "Mailbox operation had an error: %s\n",
347 cxl_mbox_cmd_rc2str(mbox_cmd
));
348 return 0; /* completed but caller must check return_code */
353 cmd_reg
= readq(cxlds
->regs
.mbox
+ CXLDEV_MBOX_CMD_OFFSET
);
354 out_len
= FIELD_GET(CXLDEV_MBOX_CMD_PAYLOAD_LENGTH_MASK
, cmd_reg
);
357 if (out_len
&& mbox_cmd
->payload_out
) {
359 * Sanitize the copy. If hardware misbehaves, out_len per the
360 * spec can actually be greater than the max allowed size (21
361 * bits available but spec defined 1M max). The caller also may
362 * have requested less data than the hardware supplied even
367 n
= min3(mbox_cmd
->size_out
, cxl_mbox
->payload_size
, out_len
);
368 memcpy_fromio(mbox_cmd
->payload_out
, payload
, n
);
369 mbox_cmd
->size_out
= n
;
371 mbox_cmd
->size_out
= 0;
377 static int cxl_pci_mbox_send(struct cxl_mailbox
*cxl_mbox
,
378 struct cxl_mbox_cmd
*cmd
)
382 mutex_lock_io(&cxl_mbox
->mbox_mutex
);
383 rc
= __cxl_pci_mbox_send_cmd(cxl_mbox
, cmd
);
384 mutex_unlock(&cxl_mbox
->mbox_mutex
);
389 static int cxl_pci_setup_mailbox(struct cxl_memdev_state
*mds
, bool irq_avail
)
391 struct cxl_dev_state
*cxlds
= &mds
->cxlds
;
392 struct cxl_mailbox
*cxl_mbox
= &cxlds
->cxl_mbox
;
393 const int cap
= readl(cxlds
->regs
.mbox
+ CXLDEV_MBOX_CAPS_OFFSET
);
394 struct device
*dev
= cxlds
->dev
;
395 unsigned long timeout
;
400 timeout
= jiffies
+ mbox_ready_timeout
* HZ
;
402 md_status
= readq(cxlds
->regs
.memdev
+ CXLMDEV_STATUS_OFFSET
);
403 if (md_status
& CXLMDEV_MBOX_IF_READY
)
405 if (msleep_interruptible(100))
407 } while (!time_after(jiffies
, timeout
));
409 if (!(md_status
& CXLMDEV_MBOX_IF_READY
)) {
410 cxl_err(dev
, md_status
, "timeout awaiting mailbox ready");
415 * A command may be in flight from a previous driver instance,
416 * think kexec, do one doorbell wait so that
417 * __cxl_pci_mbox_send_cmd() can assume that it is the only
418 * source for future doorbell busy events.
420 if (cxl_pci_mbox_wait_for_doorbell(cxlds
) != 0) {
421 cxl_err(dev
, md_status
, "timeout awaiting mailbox idle");
425 cxl_mbox
->mbox_send
= cxl_pci_mbox_send
;
426 cxl_mbox
->payload_size
=
427 1 << FIELD_GET(CXLDEV_MBOX_CAP_PAYLOAD_SIZE_MASK
, cap
);
430 * CXL 2.0 8.2.8.4.3 Mailbox Capabilities Register
432 * If the size is too small, mandatory commands will not work and so
433 * there's no point in going forward. If the size is too large, there's
434 * no harm is soft limiting it.
436 cxl_mbox
->payload_size
= min_t(size_t, cxl_mbox
->payload_size
, SZ_1M
);
437 if (cxl_mbox
->payload_size
< 256) {
438 dev_err(dev
, "Mailbox is too small (%zub)",
439 cxl_mbox
->payload_size
);
443 dev_dbg(dev
, "Mailbox payload sized %zu", cxl_mbox
->payload_size
);
445 INIT_DELAYED_WORK(&mds
->security
.poll_dwork
, cxl_mbox_sanitize_work
);
447 /* background command interrupts are optional */
448 if (!(cap
& CXLDEV_MBOX_CAP_BG_CMD_IRQ
) || !irq_avail
)
451 msgnum
= FIELD_GET(CXLDEV_MBOX_CAP_IRQ_MSGNUM_MASK
, cap
);
452 irq
= pci_irq_vector(to_pci_dev(cxlds
->dev
), msgnum
);
456 if (cxl_request_irq(cxlds
, irq
, cxl_pci_mbox_irq
))
459 dev_dbg(cxlds
->dev
, "Mailbox interrupts enabled\n");
460 /* enable background command mbox irq support */
461 ctrl
= readl(cxlds
->regs
.mbox
+ CXLDEV_MBOX_CTRL_OFFSET
);
462 ctrl
|= CXLDEV_MBOX_CTRL_BG_CMD_IRQ
;
463 writel(ctrl
, cxlds
->regs
.mbox
+ CXLDEV_MBOX_CTRL_OFFSET
);
469 * Assume that any RCIEP that emits the CXL memory expander class code
472 static bool is_cxl_restricted(struct pci_dev
*pdev
)
474 return pci_pcie_type(pdev
) == PCI_EXP_TYPE_RC_END
;
477 static int cxl_rcrb_get_comp_regs(struct pci_dev
*pdev
,
478 struct cxl_register_map
*map
,
479 struct cxl_dport
*dport
)
481 resource_size_t component_reg_phys
;
483 *map
= (struct cxl_register_map
) {
485 .resource
= CXL_RESOURCE_NONE
,
488 struct cxl_port
*port
__free(put_cxl_port
) =
489 cxl_pci_find_port(pdev
, &dport
);
491 return -EPROBE_DEFER
;
493 component_reg_phys
= cxl_rcd_component_reg_phys(&pdev
->dev
, dport
);
494 if (component_reg_phys
== CXL_RESOURCE_NONE
)
497 map
->resource
= component_reg_phys
;
498 map
->reg_type
= CXL_REGLOC_RBI_COMPONENT
;
499 map
->max_size
= CXL_COMPONENT_REG_BLOCK_SIZE
;
504 static int cxl_pci_setup_regs(struct pci_dev
*pdev
, enum cxl_regloc_type type
,
505 struct cxl_register_map
*map
)
509 rc
= cxl_find_regblock(pdev
, type
, map
);
512 * If the Register Locator DVSEC does not exist, check if it
513 * is an RCH and try to extract the Component Registers from
516 if (rc
&& type
== CXL_REGLOC_RBI_COMPONENT
&& is_cxl_restricted(pdev
)) {
517 struct cxl_dport
*dport
;
518 struct cxl_port
*port
__free(put_cxl_port
) =
519 cxl_pci_find_port(pdev
, &dport
);
521 return -EPROBE_DEFER
;
523 rc
= cxl_rcrb_get_comp_regs(pdev
, map
, dport
);
527 rc
= cxl_dport_map_rcd_linkcap(pdev
, dport
);
535 return cxl_setup_regs(map
);
538 static int cxl_pci_ras_unmask(struct pci_dev
*pdev
)
540 struct cxl_dev_state
*cxlds
= pci_get_drvdata(pdev
);
542 u32 orig_val
, val
, mask
;
546 if (!cxlds
->regs
.ras
) {
547 dev_dbg(&pdev
->dev
, "No RAS registers.\n");
551 /* BIOS has PCIe AER error control */
552 if (!pcie_aer_is_native(pdev
))
555 rc
= pcie_capability_read_word(pdev
, PCI_EXP_DEVCTL
, &cap
);
559 if (cap
& PCI_EXP_DEVCTL_URRE
) {
560 addr
= cxlds
->regs
.ras
+ CXL_RAS_UNCORRECTABLE_MASK_OFFSET
;
561 orig_val
= readl(addr
);
563 mask
= CXL_RAS_UNCORRECTABLE_MASK_MASK
|
564 CXL_RAS_UNCORRECTABLE_MASK_F256B_MASK
;
565 val
= orig_val
& ~mask
;
568 "Uncorrectable RAS Errors Mask: %#x -> %#x\n",
572 if (cap
& PCI_EXP_DEVCTL_CERE
) {
573 addr
= cxlds
->regs
.ras
+ CXL_RAS_CORRECTABLE_MASK_OFFSET
;
574 orig_val
= readl(addr
);
575 val
= orig_val
& ~CXL_RAS_CORRECTABLE_MASK_MASK
;
577 dev_dbg(&pdev
->dev
, "Correctable RAS Errors Mask: %#x -> %#x\n",
584 static void free_event_buf(void *buf
)
590 * There is a single buffer for reading event logs from the mailbox. All logs
591 * share this buffer protected by the mds->event_log_lock.
593 static int cxl_mem_alloc_event_buf(struct cxl_memdev_state
*mds
)
595 struct cxl_mailbox
*cxl_mbox
= &mds
->cxlds
.cxl_mbox
;
596 struct cxl_get_event_payload
*buf
;
598 buf
= kvmalloc(cxl_mbox
->payload_size
, GFP_KERNEL
);
601 mds
->event
.buf
= buf
;
603 return devm_add_action_or_reset(mds
->cxlds
.dev
, free_event_buf
, buf
);
606 static bool cxl_alloc_irq_vectors(struct pci_dev
*pdev
)
611 * Per CXL 3.0 3.1.1 CXL.io Endpoint a function on a CXL device must
612 * not generate INTx messages if that function participates in
613 * CXL.cache or CXL.mem.
615 * Additionally pci_alloc_irq_vectors() handles calling
616 * pci_free_irq_vectors() automatically despite not being called
617 * pcim_*. See pci_setup_msi_context().
619 nvecs
= pci_alloc_irq_vectors(pdev
, 1, CXL_PCI_DEFAULT_MAX_VECTORS
,
620 PCI_IRQ_MSIX
| PCI_IRQ_MSI
);
622 dev_dbg(&pdev
->dev
, "Failed to alloc irq vectors: %d\n", nvecs
);
628 static irqreturn_t
cxl_event_thread(int irq
, void *id
)
630 struct cxl_dev_id
*dev_id
= id
;
631 struct cxl_dev_state
*cxlds
= dev_id
->cxlds
;
632 struct cxl_memdev_state
*mds
= to_cxl_memdev_state(cxlds
);
637 * CXL 3.0 8.2.8.3.1: The lower 32 bits are the status;
638 * ignore the reserved upper 32 bits
640 status
= readl(cxlds
->regs
.status
+ CXLDEV_DEV_EVENT_STATUS_OFFSET
);
641 /* Ignore logs unknown to the driver */
642 status
&= CXLDEV_EVENT_STATUS_ALL
;
645 cxl_mem_get_event_records(mds
, status
);
652 static int cxl_event_req_irq(struct cxl_dev_state
*cxlds
, u8 setting
)
654 struct pci_dev
*pdev
= to_pci_dev(cxlds
->dev
);
657 if (FIELD_GET(CXLDEV_EVENT_INT_MODE_MASK
, setting
) != CXL_INT_MSI_MSIX
)
660 irq
= pci_irq_vector(pdev
,
661 FIELD_GET(CXLDEV_EVENT_INT_MSGNUM_MASK
, setting
));
665 return cxl_request_irq(cxlds
, irq
, cxl_event_thread
);
668 static int cxl_event_get_int_policy(struct cxl_memdev_state
*mds
,
669 struct cxl_event_interrupt_policy
*policy
)
671 struct cxl_mailbox
*cxl_mbox
= &mds
->cxlds
.cxl_mbox
;
672 struct cxl_mbox_cmd mbox_cmd
= {
673 .opcode
= CXL_MBOX_OP_GET_EVT_INT_POLICY
,
674 .payload_out
= policy
,
675 .size_out
= sizeof(*policy
),
679 rc
= cxl_internal_send_cmd(cxl_mbox
, &mbox_cmd
);
681 dev_err(mds
->cxlds
.dev
,
682 "Failed to get event interrupt policy : %d", rc
);
687 static int cxl_event_config_msgnums(struct cxl_memdev_state
*mds
,
688 struct cxl_event_interrupt_policy
*policy
)
690 struct cxl_mailbox
*cxl_mbox
= &mds
->cxlds
.cxl_mbox
;
691 struct cxl_mbox_cmd mbox_cmd
;
694 *policy
= (struct cxl_event_interrupt_policy
) {
695 .info_settings
= CXL_INT_MSI_MSIX
,
696 .warn_settings
= CXL_INT_MSI_MSIX
,
697 .failure_settings
= CXL_INT_MSI_MSIX
,
698 .fatal_settings
= CXL_INT_MSI_MSIX
,
701 mbox_cmd
= (struct cxl_mbox_cmd
) {
702 .opcode
= CXL_MBOX_OP_SET_EVT_INT_POLICY
,
703 .payload_in
= policy
,
704 .size_in
= sizeof(*policy
),
707 rc
= cxl_internal_send_cmd(cxl_mbox
, &mbox_cmd
);
709 dev_err(mds
->cxlds
.dev
, "Failed to set event interrupt policy : %d",
714 /* Retrieve final interrupt settings */
715 return cxl_event_get_int_policy(mds
, policy
);
718 static int cxl_event_irqsetup(struct cxl_memdev_state
*mds
)
720 struct cxl_dev_state
*cxlds
= &mds
->cxlds
;
721 struct cxl_event_interrupt_policy policy
;
724 rc
= cxl_event_config_msgnums(mds
, &policy
);
728 rc
= cxl_event_req_irq(cxlds
, policy
.info_settings
);
730 dev_err(cxlds
->dev
, "Failed to get interrupt for event Info log\n");
734 rc
= cxl_event_req_irq(cxlds
, policy
.warn_settings
);
736 dev_err(cxlds
->dev
, "Failed to get interrupt for event Warn log\n");
740 rc
= cxl_event_req_irq(cxlds
, policy
.failure_settings
);
742 dev_err(cxlds
->dev
, "Failed to get interrupt for event Failure log\n");
746 rc
= cxl_event_req_irq(cxlds
, policy
.fatal_settings
);
748 dev_err(cxlds
->dev
, "Failed to get interrupt for event Fatal log\n");
755 static bool cxl_event_int_is_fw(u8 setting
)
757 u8 mode
= FIELD_GET(CXLDEV_EVENT_INT_MODE_MASK
, setting
);
759 return mode
== CXL_INT_FW
;
762 static int cxl_event_config(struct pci_host_bridge
*host_bridge
,
763 struct cxl_memdev_state
*mds
, bool irq_avail
)
765 struct cxl_event_interrupt_policy policy
;
769 * When BIOS maintains CXL error reporting control, it will process
770 * event records. Only one agent can do so.
772 if (!host_bridge
->native_cxl_error
)
776 dev_info(mds
->cxlds
.dev
, "No interrupt support, disable event processing.\n");
780 rc
= cxl_event_get_int_policy(mds
, &policy
);
784 if (cxl_event_int_is_fw(policy
.info_settings
) ||
785 cxl_event_int_is_fw(policy
.warn_settings
) ||
786 cxl_event_int_is_fw(policy
.failure_settings
) ||
787 cxl_event_int_is_fw(policy
.fatal_settings
)) {
788 dev_err(mds
->cxlds
.dev
,
789 "FW still in control of Event Logs despite _OSC settings\n");
793 rc
= cxl_mem_alloc_event_buf(mds
);
797 rc
= cxl_event_irqsetup(mds
);
801 cxl_mem_get_event_records(mds
, CXLDEV_EVENT_STATUS_ALL
);
806 static int cxl_pci_type3_init_mailbox(struct cxl_dev_state
*cxlds
)
811 * Fail the init if there's no mailbox. For a type3 this is out of spec.
813 if (!cxlds
->reg_map
.device_map
.mbox
.valid
)
816 rc
= cxl_mailbox_init(&cxlds
->cxl_mbox
, cxlds
->dev
);
823 static ssize_t
rcd_pcie_cap_emit(struct device
*dev
, u16 offset
, char *buf
, size_t width
)
825 struct cxl_dev_state
*cxlds
= dev_get_drvdata(dev
);
826 struct cxl_memdev
*cxlmd
= cxlds
->cxlmd
;
827 struct device
*root_dev
;
828 struct cxl_dport
*dport
;
829 struct cxl_port
*root
__free(put_cxl_port
) =
830 cxl_mem_find_port(cxlmd
, &dport
);
835 root_dev
= root
->uport_dev
;
839 guard(device
)(root_dev
);
840 if (!root_dev
->driver
)
845 return sysfs_emit(buf
, "%#x\n",
846 readw(dport
->regs
.rcd_pcie_cap
+ offset
));
848 return sysfs_emit(buf
, "%#x\n",
849 readl(dport
->regs
.rcd_pcie_cap
+ offset
));
855 static ssize_t
rcd_link_cap_show(struct device
*dev
,
856 struct device_attribute
*attr
, char *buf
)
858 return rcd_pcie_cap_emit(dev
, PCI_EXP_LNKCAP
, buf
, sizeof(u32
));
860 static DEVICE_ATTR_RO(rcd_link_cap
);
862 static ssize_t
rcd_link_ctrl_show(struct device
*dev
,
863 struct device_attribute
*attr
, char *buf
)
865 return rcd_pcie_cap_emit(dev
, PCI_EXP_LNKCTL
, buf
, sizeof(u16
));
867 static DEVICE_ATTR_RO(rcd_link_ctrl
);
869 static ssize_t
rcd_link_status_show(struct device
*dev
,
870 struct device_attribute
*attr
, char *buf
)
872 return rcd_pcie_cap_emit(dev
, PCI_EXP_LNKSTA
, buf
, sizeof(u16
));
874 static DEVICE_ATTR_RO(rcd_link_status
);
876 static struct attribute
*cxl_rcd_attrs
[] = {
877 &dev_attr_rcd_link_cap
.attr
,
878 &dev_attr_rcd_link_ctrl
.attr
,
879 &dev_attr_rcd_link_status
.attr
,
883 static umode_t
cxl_rcd_visible(struct kobject
*kobj
, struct attribute
*a
, int n
)
885 struct device
*dev
= kobj_to_dev(kobj
);
886 struct pci_dev
*pdev
= to_pci_dev(dev
);
888 if (is_cxl_restricted(pdev
))
894 static struct attribute_group cxl_rcd_group
= {
895 .attrs
= cxl_rcd_attrs
,
896 .is_visible
= cxl_rcd_visible
,
898 __ATTRIBUTE_GROUPS(cxl_rcd
);
900 static int cxl_pci_probe(struct pci_dev
*pdev
, const struct pci_device_id
*id
)
902 struct pci_host_bridge
*host_bridge
= pci_find_host_bridge(pdev
->bus
);
903 struct cxl_memdev_state
*mds
;
904 struct cxl_dev_state
*cxlds
;
905 struct cxl_register_map map
;
906 struct cxl_memdev
*cxlmd
;
907 int i
, rc
, pmu_count
;
911 * Double check the anonymous union trickery in struct cxl_regs
912 * FIXME switch to struct_group()
914 BUILD_BUG_ON(offsetof(struct cxl_regs
, memdev
) !=
915 offsetof(struct cxl_regs
, device_regs
.memdev
));
917 rc
= pcim_enable_device(pdev
);
920 pci_set_master(pdev
);
922 mds
= cxl_memdev_state_create(&pdev
->dev
);
926 pci_set_drvdata(pdev
, cxlds
);
928 cxlds
->rcd
= is_cxl_restricted(pdev
);
929 cxlds
->serial
= pci_get_dsn(pdev
);
930 cxlds
->cxl_dvsec
= pci_find_dvsec_capability(
931 pdev
, PCI_VENDOR_ID_CXL
, CXL_DVSEC_PCIE_DEVICE
);
932 if (!cxlds
->cxl_dvsec
)
934 "Device DVSEC not present, skip CXL.mem init\n");
936 rc
= cxl_pci_setup_regs(pdev
, CXL_REGLOC_RBI_MEMDEV
, &map
);
940 rc
= cxl_map_device_regs(&map
, &cxlds
->regs
.device_regs
);
945 * If the component registers can't be found, the cxl_pci driver may
946 * still be useful for management functions so don't return an error.
948 rc
= cxl_pci_setup_regs(pdev
, CXL_REGLOC_RBI_COMPONENT
,
951 dev_warn(&pdev
->dev
, "No component registers (%d)\n", rc
);
952 else if (!cxlds
->reg_map
.component_map
.ras
.valid
)
953 dev_dbg(&pdev
->dev
, "RAS registers not found\n");
955 rc
= cxl_map_component_regs(&cxlds
->reg_map
, &cxlds
->regs
.component
,
956 BIT(CXL_CM_CAP_CAP_ID_RAS
));
958 dev_dbg(&pdev
->dev
, "Failed to map RAS capability.\n");
960 rc
= cxl_pci_type3_init_mailbox(cxlds
);
964 rc
= cxl_await_media_ready(cxlds
);
966 cxlds
->media_ready
= true;
968 dev_warn(&pdev
->dev
, "Media not active (%d)\n", rc
);
970 irq_avail
= cxl_alloc_irq_vectors(pdev
);
972 rc
= cxl_pci_setup_mailbox(mds
, irq_avail
);
976 rc
= cxl_enumerate_cmds(mds
);
980 rc
= cxl_set_timestamp(mds
);
984 rc
= cxl_poison_state_init(mds
);
988 rc
= cxl_dev_state_identify(mds
);
992 rc
= cxl_mem_create_range_info(mds
);
996 cxlmd
= devm_cxl_add_memdev(&pdev
->dev
, cxlds
);
998 return PTR_ERR(cxlmd
);
1000 rc
= devm_cxl_setup_fw_upload(&pdev
->dev
, mds
);
1004 rc
= devm_cxl_sanitize_setup_notifier(&pdev
->dev
, cxlmd
);
1008 pmu_count
= cxl_count_regblock(pdev
, CXL_REGLOC_RBI_PMU
);
1009 for (i
= 0; i
< pmu_count
; i
++) {
1010 struct cxl_pmu_regs pmu_regs
;
1012 rc
= cxl_find_regblock_instance(pdev
, CXL_REGLOC_RBI_PMU
, &map
, i
);
1014 dev_dbg(&pdev
->dev
, "Could not find PMU regblock\n");
1018 rc
= cxl_map_pmu_regs(&map
, &pmu_regs
);
1020 dev_dbg(&pdev
->dev
, "Could not map PMU regs\n");
1024 rc
= devm_cxl_pmu_add(cxlds
->dev
, &pmu_regs
, cxlmd
->id
, i
, CXL_PMU_MEMDEV
);
1026 dev_dbg(&pdev
->dev
, "Could not add PMU instance\n");
1031 rc
= cxl_event_config(host_bridge
, mds
, irq_avail
);
1035 rc
= cxl_pci_ras_unmask(pdev
);
1037 dev_dbg(&pdev
->dev
, "No RAS reporting unmasked\n");
1039 pci_save_state(pdev
);
1044 static const struct pci_device_id cxl_mem_pci_tbl
[] = {
1045 /* PCI class code for CXL.mem Type-3 Devices */
1046 { PCI_DEVICE_CLASS((PCI_CLASS_MEMORY_CXL
<< 8 | CXL_MEMORY_PROGIF
), ~0)},
1047 { /* terminate list */ },
1049 MODULE_DEVICE_TABLE(pci
, cxl_mem_pci_tbl
);
1051 static pci_ers_result_t
cxl_slot_reset(struct pci_dev
*pdev
)
1053 struct cxl_dev_state
*cxlds
= pci_get_drvdata(pdev
);
1054 struct cxl_memdev
*cxlmd
= cxlds
->cxlmd
;
1055 struct device
*dev
= &cxlmd
->dev
;
1057 dev_info(&pdev
->dev
, "%s: restart CXL.mem after slot reset\n",
1059 pci_restore_state(pdev
);
1060 if (device_attach(dev
) <= 0)
1061 return PCI_ERS_RESULT_DISCONNECT
;
1062 return PCI_ERS_RESULT_RECOVERED
;
1065 static void cxl_error_resume(struct pci_dev
*pdev
)
1067 struct cxl_dev_state
*cxlds
= pci_get_drvdata(pdev
);
1068 struct cxl_memdev
*cxlmd
= cxlds
->cxlmd
;
1069 struct device
*dev
= &cxlmd
->dev
;
1071 dev_info(&pdev
->dev
, "%s: error resume %s\n", dev_name(dev
),
1072 dev
->driver
? "successful" : "failed");
1075 static void cxl_reset_done(struct pci_dev
*pdev
)
1077 struct cxl_dev_state
*cxlds
= pci_get_drvdata(pdev
);
1078 struct cxl_memdev
*cxlmd
= cxlds
->cxlmd
;
1079 struct device
*dev
= &pdev
->dev
;
1082 * FLR does not expect to touch the HDM decoders and related
1083 * registers. SBR, however, will wipe all device configurations.
1084 * Issue a warning if there was an active decoder before the reset
1085 * that no longer exists.
1087 guard(device
)(&cxlmd
->dev
);
1088 if (cxlmd
->endpoint
&&
1089 cxl_endpoint_decoder_reset_detected(cxlmd
->endpoint
)) {
1090 dev_crit(dev
, "SBR happened without memory regions removal.\n");
1091 dev_crit(dev
, "System may be unstable if regions hosted system memory.\n");
1092 add_taint(TAINT_USER
, LOCKDEP_STILL_OK
);
1096 static const struct pci_error_handlers cxl_error_handlers
= {
1097 .error_detected
= cxl_error_detected
,
1098 .slot_reset
= cxl_slot_reset
,
1099 .resume
= cxl_error_resume
,
1100 .cor_error_detected
= cxl_cor_error_detected
,
1101 .reset_done
= cxl_reset_done
,
1104 static struct pci_driver cxl_pci_driver
= {
1105 .name
= KBUILD_MODNAME
,
1106 .id_table
= cxl_mem_pci_tbl
,
1107 .probe
= cxl_pci_probe
,
1108 .err_handler
= &cxl_error_handlers
,
1109 .dev_groups
= cxl_rcd_groups
,
1111 .probe_type
= PROBE_PREFER_ASYNCHRONOUS
,
1115 #define CXL_EVENT_HDR_FLAGS_REC_SEVERITY GENMASK(1, 0)
1116 static void cxl_handle_cper_event(enum cxl_event_type ev_type
,
1117 struct cxl_cper_event_rec
*rec
)
1119 struct cper_cxl_event_devid
*device_id
= &rec
->hdr
.device_id
;
1120 struct pci_dev
*pdev
__free(pci_dev_put
) = NULL
;
1121 enum cxl_event_log_type log_type
;
1122 struct cxl_dev_state
*cxlds
;
1126 pr_debug("CPER event %d for device %u:%u:%u.%u\n", ev_type
,
1127 device_id
->segment_num
, device_id
->bus_num
,
1128 device_id
->device_num
, device_id
->func_num
);
1130 devfn
= PCI_DEVFN(device_id
->device_num
, device_id
->func_num
);
1131 pdev
= pci_get_domain_bus_and_slot(device_id
->segment_num
,
1132 device_id
->bus_num
, devfn
);
1136 guard(device
)(&pdev
->dev
);
1137 if (pdev
->driver
!= &cxl_pci_driver
)
1140 cxlds
= pci_get_drvdata(pdev
);
1144 /* Fabricate a log type */
1145 hdr_flags
= get_unaligned_le24(rec
->event
.generic
.hdr
.flags
);
1146 log_type
= FIELD_GET(CXL_EVENT_HDR_FLAGS_REC_SEVERITY
, hdr_flags
);
1148 cxl_event_trace_record(cxlds
->cxlmd
, log_type
, ev_type
,
1149 &uuid_null
, &rec
->event
);
1152 static void cxl_cper_work_fn(struct work_struct
*work
)
1154 struct cxl_cper_work_data wd
;
1156 while (cxl_cper_kfifo_get(&wd
))
1157 cxl_handle_cper_event(wd
.event_type
, &wd
.rec
);
1159 static DECLARE_WORK(cxl_cper_work
, cxl_cper_work_fn
);
1161 static int __init
cxl_pci_driver_init(void)
1165 rc
= pci_register_driver(&cxl_pci_driver
);
1169 rc
= cxl_cper_register_work(&cxl_cper_work
);
1171 pci_unregister_driver(&cxl_pci_driver
);
1176 static void __exit
cxl_pci_driver_exit(void)
1178 cxl_cper_unregister_work(&cxl_cper_work
);
1179 cancel_work_sync(&cxl_cper_work
);
1180 pci_unregister_driver(&cxl_pci_driver
);
1183 module_init(cxl_pci_driver_init
);
1184 module_exit(cxl_pci_driver_exit
);
1185 MODULE_DESCRIPTION("CXL: PCI manageability");
1186 MODULE_LICENSE("GPL v2");
1187 MODULE_IMPORT_NS("CXL");