1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright(c) 2020 Intel Corporation. All rights reserved. */
3 #include <linux/io-64-nonatomic-lo-hi.h>
4 #include <linux/moduleparam.h>
5 #include <linux/module.h>
6 #include <linux/delay.h>
7 #include <linux/sizes.h>
8 #include <linux/mutex.h>
9 #include <linux/list.h>
10 #include <linux/pci.h>
11 #include <linux/aer.h>
21 * This implements the PCI exclusive functionality for a CXL device as it is
22 * defined by the Compute Express Link specification. CXL devices may surface
23 * certain functionality even if it isn't CXL enabled. While this driver is
24 * focused around the PCI specific aspects of a CXL device, it binds to the
25 * specific CXL memory device class code, and therefore the implementation of
26 * cxl_pci is focused around CXL memory devices.
28 * The driver has several responsibilities, mainly:
29 * - Create the memX device and register on the CXL bus.
30 * - Enumerate device's register interface and map them.
31 * - Registers nvdimm bridge device with cxl_core.
32 * - Registers a CXL mailbox with cxl_core.
35 #define cxl_doorbell_busy(cxlds) \
36 (readl((cxlds)->regs.mbox + CXLDEV_MBOX_CTRL_OFFSET) & \
37 CXLDEV_MBOX_CTRL_DOORBELL)
39 /* CXL 2.0 - 8.2.8.4 */
40 #define CXL_MAILBOX_TIMEOUT_MS (2 * HZ)
43 * CXL 2.0 ECN "Add Mailbox Ready Time" defines a capability field to
44 * dictate how long to wait for the mailbox to become ready. The new
45 * field allows the device to tell software the amount of time to wait
46 * before mailbox ready. This field per the spec theoretically allows
47 * for up to 255 seconds. 255 seconds is unreasonably long, its longer
48 * than the maximum SATA port link recovery wait. Default to 60 seconds
49 * until someone builds a CXL device that needs more time in practice.
51 static unsigned short mbox_ready_timeout = 60;
52 module_param(mbox_ready_timeout, ushort, 0644);
53 MODULE_PARM_DESC(mbox_ready_timeout, "seconds to wait for mailbox ready");
55 static int cxl_pci_mbox_wait_for_doorbell(struct cxl_dev_state *cxlds)
57 const unsigned long start = jiffies;
58 unsigned long end = start;
60 while (cxl_doorbell_busy(cxlds)) {
63 if (time_after(end, start + CXL_MAILBOX_TIMEOUT_MS)) {
64 /* Check again in case preempted before timeout test */
65 if (!cxl_doorbell_busy(cxlds))
72 dev_dbg(cxlds->dev, "Doorbell wait took %dms",
73 jiffies_to_msecs(end) - jiffies_to_msecs(start));
77 #define cxl_err(dev, status, msg) \
78 dev_err_ratelimited(dev, msg ", device state %s%s\n", \
79 status & CXLMDEV_DEV_FATAL ? " fatal" : "", \
80 status & CXLMDEV_FW_HALT ? " firmware-halt" : "")
82 #define cxl_cmd_err(dev, cmd, status, msg) \
83 dev_err_ratelimited(dev, msg " (opcode: %#x), device state %s%s\n", \
85 status & CXLMDEV_DEV_FATAL ? " fatal" : "", \
86 status & CXLMDEV_FW_HALT ? " firmware-halt" : "")
89 struct cxl_dev_state *cxlds;
92 static int cxl_request_irq(struct cxl_dev_state *cxlds, int irq,
93 irq_handler_t handler, irq_handler_t thread_fn)
95 struct device *dev = cxlds->dev;
96 struct cxl_dev_id *dev_id;
98 /* dev_id must be globally unique and must contain the cxlds */
99 dev_id = devm_kzalloc(dev, sizeof(*dev_id), GFP_KERNEL);
102 dev_id->cxlds = cxlds;
104 return devm_request_threaded_irq(dev, irq, handler, thread_fn,
105 IRQF_SHARED | IRQF_ONESHOT,
109 static bool cxl_mbox_background_complete(struct cxl_dev_state *cxlds)
113 reg = readq(cxlds->regs.mbox + CXLDEV_MBOX_BG_CMD_STATUS_OFFSET);
114 return FIELD_GET(CXLDEV_MBOX_BG_CMD_COMMAND_PCT_MASK, reg) == 100;
117 static irqreturn_t cxl_pci_mbox_irq(int irq, void *id)
121 struct cxl_dev_id *dev_id = id;
122 struct cxl_dev_state *cxlds = dev_id->cxlds;
123 struct cxl_memdev_state *mds = to_cxl_memdev_state(cxlds);
125 if (!cxl_mbox_background_complete(cxlds))
128 reg = readq(cxlds->regs.mbox + CXLDEV_MBOX_BG_CMD_STATUS_OFFSET);
129 opcode = FIELD_GET(CXLDEV_MBOX_BG_CMD_COMMAND_OPCODE_MASK, reg);
130 if (opcode == CXL_MBOX_OP_SANITIZE) {
131 if (mds->security.sanitize_node)
132 sysfs_notify_dirent(mds->security.sanitize_node);
134 dev_dbg(cxlds->dev, "Sanitization operation ended\n");
136 /* short-circuit the wait in __cxl_pci_mbox_send_cmd() */
137 rcuwait_wake_up(&mds->mbox_wait);
144 * Sanitization operation polling mode.
146 static void cxl_mbox_sanitize_work(struct work_struct *work)
148 struct cxl_memdev_state *mds =
149 container_of(work, typeof(*mds), security.poll_dwork.work);
150 struct cxl_dev_state *cxlds = &mds->cxlds;
152 mutex_lock(&mds->mbox_mutex);
153 if (cxl_mbox_background_complete(cxlds)) {
154 mds->security.poll_tmo_secs = 0;
155 if (mds->security.sanitize_node)
156 sysfs_notify_dirent(mds->security.sanitize_node);
158 dev_dbg(cxlds->dev, "Sanitization operation ended\n");
160 int timeout = mds->security.poll_tmo_secs + 10;
162 mds->security.poll_tmo_secs = min(15 * 60, timeout);
163 queue_delayed_work(system_wq, &mds->security.poll_dwork,
166 mutex_unlock(&mds->mbox_mutex);
170 * __cxl_pci_mbox_send_cmd() - Execute a mailbox command
171 * @mds: The memory device driver data
172 * @mbox_cmd: Command to send to the memory device.
174 * Context: Any context. Expects mbox_mutex to be held.
175 * Return: -ETIMEDOUT if timeout occurred waiting for completion. 0 on success.
176 * Caller should check the return code in @mbox_cmd to make sure it
179 * This is a generic form of the CXL mailbox send command thus only using the
180 * registers defined by the mailbox capability ID - CXL 2.0 8.2.8.4. Memory
181 * devices, and perhaps other types of CXL devices may have further information
182 * available upon error conditions. Driver facilities wishing to send mailbox
183 * commands should use the wrapper command.
185 * The CXL spec allows for up to two mailboxes. The intention is for the primary
186 * mailbox to be OS controlled and the secondary mailbox to be used by system
187 * firmware. This allows the OS and firmware to communicate with the device and
188 * not need to coordinate with each other. The driver only uses the primary
191 static int __cxl_pci_mbox_send_cmd(struct cxl_memdev_state *mds,
192 struct cxl_mbox_cmd *mbox_cmd)
194 struct cxl_dev_state *cxlds = &mds->cxlds;
195 void __iomem *payload = cxlds->regs.mbox + CXLDEV_MBOX_PAYLOAD_OFFSET;
196 struct device *dev = cxlds->dev;
197 u64 cmd_reg, status_reg;
201 lockdep_assert_held(&mds->mbox_mutex);
204 * Here are the steps from 8.2.8.4 of the CXL 2.0 spec.
205 * 1. Caller reads MB Control Register to verify doorbell is clear
206 * 2. Caller writes Command Register
207 * 3. Caller writes Command Payload Registers if input payload is non-empty
208 * 4. Caller writes MB Control Register to set doorbell
209 * 5. Caller either polls for doorbell to be clear or waits for interrupt if configured
210 * 6. Caller reads MB Status Register to fetch Return code
211 * 7. If command successful, Caller reads Command Register to get Payload Length
212 * 8. If output payload is non-empty, host reads Command Payload Registers
214 * Hardware is free to do whatever it wants before the doorbell is rung,
215 * and isn't allowed to change anything after it clears the doorbell. As
216 * such, steps 2 and 3 can happen in any order, and steps 6, 7, 8 can
217 * also happen in any order (though some orders might not make sense).
221 if (cxl_doorbell_busy(cxlds)) {
223 readq(cxlds->regs.memdev + CXLMDEV_STATUS_OFFSET);
225 cxl_cmd_err(cxlds->dev, mbox_cmd, md_status,
226 "mailbox queue busy");
231 * With sanitize polling, hardware might be done and the poller still
232 * not be in sync. Ensure no new command comes in until so. Keep the
233 * hardware semantics and only allow device health status.
235 if (mds->security.poll_tmo_secs > 0) {
236 if (mbox_cmd->opcode != CXL_MBOX_OP_GET_HEALTH_INFO)
240 cmd_reg = FIELD_PREP(CXLDEV_MBOX_CMD_COMMAND_OPCODE_MASK,
242 if (mbox_cmd->size_in) {
243 if (WARN_ON(!mbox_cmd->payload_in))
246 cmd_reg |= FIELD_PREP(CXLDEV_MBOX_CMD_PAYLOAD_LENGTH_MASK,
248 memcpy_toio(payload, mbox_cmd->payload_in, mbox_cmd->size_in);
252 writeq(cmd_reg, cxlds->regs.mbox + CXLDEV_MBOX_CMD_OFFSET);
255 dev_dbg(dev, "Sending command: 0x%04x\n", mbox_cmd->opcode);
256 writel(CXLDEV_MBOX_CTRL_DOORBELL,
257 cxlds->regs.mbox + CXLDEV_MBOX_CTRL_OFFSET);
260 rc = cxl_pci_mbox_wait_for_doorbell(cxlds);
261 if (rc == -ETIMEDOUT) {
262 u64 md_status = readq(cxlds->regs.memdev + CXLMDEV_STATUS_OFFSET);
264 cxl_cmd_err(cxlds->dev, mbox_cmd, md_status, "mailbox timeout");
269 status_reg = readq(cxlds->regs.mbox + CXLDEV_MBOX_STATUS_OFFSET);
270 mbox_cmd->return_code =
271 FIELD_GET(CXLDEV_MBOX_STATUS_RET_CODE_MASK, status_reg);
274 * Handle the background command in a synchronous manner.
276 * All other mailbox commands will serialize/queue on the mbox_mutex,
277 * which we currently hold. Furthermore this also guarantees that
278 * cxl_mbox_background_complete() checks are safe amongst each other,
279 * in that no new bg operation can occur in between.
281 * Background operations are timesliced in accordance with the nature
282 * of the command. In the event of timeout, the mailbox state is
283 * indeterminate until the next successful command submission and the
284 * driver can get back in sync with the hardware state.
286 if (mbox_cmd->return_code == CXL_MBOX_CMD_RC_BACKGROUND) {
291 * Sanitization is a special case which monopolizes the device
292 * and cannot be timesliced. Handle asynchronously instead,
293 * and allow userspace to poll(2) for completion.
295 if (mbox_cmd->opcode == CXL_MBOX_OP_SANITIZE) {
296 if (mds->security.poll) {
297 /* give first timeout a second */
299 mds->security.poll_tmo_secs = timeout;
300 queue_delayed_work(system_wq,
301 &mds->security.poll_dwork,
305 dev_dbg(dev, "Sanitization operation started\n");
309 dev_dbg(dev, "Mailbox background operation (0x%04x) started\n",
312 timeout = mbox_cmd->poll_interval_ms;
313 for (i = 0; i < mbox_cmd->poll_count; i++) {
314 if (rcuwait_wait_event_timeout(&mds->mbox_wait,
315 cxl_mbox_background_complete(cxlds),
316 TASK_UNINTERRUPTIBLE,
317 msecs_to_jiffies(timeout)) > 0)
321 if (!cxl_mbox_background_complete(cxlds)) {
322 dev_err(dev, "timeout waiting for background (%d ms)\n",
323 timeout * mbox_cmd->poll_count);
327 bg_status_reg = readq(cxlds->regs.mbox +
328 CXLDEV_MBOX_BG_CMD_STATUS_OFFSET);
329 mbox_cmd->return_code =
330 FIELD_GET(CXLDEV_MBOX_BG_CMD_COMMAND_RC_MASK,
333 "Mailbox background operation (0x%04x) completed\n",
337 if (mbox_cmd->return_code != CXL_MBOX_CMD_RC_SUCCESS) {
338 dev_dbg(dev, "Mailbox operation had an error: %s\n",
339 cxl_mbox_cmd_rc2str(mbox_cmd));
340 return 0; /* completed but caller must check return_code */
345 cmd_reg = readq(cxlds->regs.mbox + CXLDEV_MBOX_CMD_OFFSET);
346 out_len = FIELD_GET(CXLDEV_MBOX_CMD_PAYLOAD_LENGTH_MASK, cmd_reg);
349 if (out_len && mbox_cmd->payload_out) {
351 * Sanitize the copy. If hardware misbehaves, out_len per the
352 * spec can actually be greater than the max allowed size (21
353 * bits available but spec defined 1M max). The caller also may
354 * have requested less data than the hardware supplied even
359 n = min3(mbox_cmd->size_out, mds->payload_size, out_len);
360 memcpy_fromio(mbox_cmd->payload_out, payload, n);
361 mbox_cmd->size_out = n;
363 mbox_cmd->size_out = 0;
369 static int cxl_pci_mbox_send(struct cxl_memdev_state *mds,
370 struct cxl_mbox_cmd *cmd)
374 mutex_lock_io(&mds->mbox_mutex);
375 rc = __cxl_pci_mbox_send_cmd(mds, cmd);
376 mutex_unlock(&mds->mbox_mutex);
381 static int cxl_pci_setup_mailbox(struct cxl_memdev_state *mds)
383 struct cxl_dev_state *cxlds = &mds->cxlds;
384 const int cap = readl(cxlds->regs.mbox + CXLDEV_MBOX_CAPS_OFFSET);
385 struct device *dev = cxlds->dev;
386 unsigned long timeout;
389 timeout = jiffies + mbox_ready_timeout * HZ;
391 md_status = readq(cxlds->regs.memdev + CXLMDEV_STATUS_OFFSET);
392 if (md_status & CXLMDEV_MBOX_IF_READY)
394 if (msleep_interruptible(100))
396 } while (!time_after(jiffies, timeout));
398 if (!(md_status & CXLMDEV_MBOX_IF_READY)) {
399 cxl_err(dev, md_status, "timeout awaiting mailbox ready");
404 * A command may be in flight from a previous driver instance,
405 * think kexec, do one doorbell wait so that
406 * __cxl_pci_mbox_send_cmd() can assume that it is the only
407 * source for future doorbell busy events.
409 if (cxl_pci_mbox_wait_for_doorbell(cxlds) != 0) {
410 cxl_err(dev, md_status, "timeout awaiting mailbox idle");
414 mds->mbox_send = cxl_pci_mbox_send;
416 1 << FIELD_GET(CXLDEV_MBOX_CAP_PAYLOAD_SIZE_MASK, cap);
419 * CXL 2.0 8.2.8.4.3 Mailbox Capabilities Register
421 * If the size is too small, mandatory commands will not work and so
422 * there's no point in going forward. If the size is too large, there's
423 * no harm is soft limiting it.
425 mds->payload_size = min_t(size_t, mds->payload_size, SZ_1M);
426 if (mds->payload_size < 256) {
427 dev_err(dev, "Mailbox is too small (%zub)",
432 dev_dbg(dev, "Mailbox payload sized %zu", mds->payload_size);
434 rcuwait_init(&mds->mbox_wait);
436 if (cap & CXLDEV_MBOX_CAP_BG_CMD_IRQ) {
439 struct pci_dev *pdev = to_pci_dev(cxlds->dev);
441 msgnum = FIELD_GET(CXLDEV_MBOX_CAP_IRQ_MSGNUM_MASK, cap);
442 irq = pci_irq_vector(pdev, msgnum);
446 if (cxl_request_irq(cxlds, irq, cxl_pci_mbox_irq, NULL))
449 /* enable background command mbox irq support */
450 ctrl = readl(cxlds->regs.mbox + CXLDEV_MBOX_CTRL_OFFSET);
451 ctrl |= CXLDEV_MBOX_CTRL_BG_CMD_IRQ;
452 writel(ctrl, cxlds->regs.mbox + CXLDEV_MBOX_CTRL_OFFSET);
458 mds->security.poll = true;
459 INIT_DELAYED_WORK(&mds->security.poll_dwork, cxl_mbox_sanitize_work);
461 dev_dbg(cxlds->dev, "Mailbox interrupts are unsupported");
466 * Assume that any RCIEP that emits the CXL memory expander class code
469 static bool is_cxl_restricted(struct pci_dev *pdev)
471 return pci_pcie_type(pdev) == PCI_EXP_TYPE_RC_END;
474 static int cxl_rcrb_get_comp_regs(struct pci_dev *pdev,
475 struct cxl_register_map *map)
477 struct cxl_port *port;
478 struct cxl_dport *dport;
479 resource_size_t component_reg_phys;
481 *map = (struct cxl_register_map) {
483 .resource = CXL_RESOURCE_NONE,
486 port = cxl_pci_find_port(pdev, &dport);
488 return -EPROBE_DEFER;
490 component_reg_phys = cxl_rcd_component_reg_phys(&pdev->dev, dport);
492 put_device(&port->dev);
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 rc = cxl_rcrb_get_comp_regs(pdev, map);
522 return cxl_setup_regs(map);
525 static int cxl_pci_ras_unmask(struct pci_dev *pdev)
527 struct cxl_dev_state *cxlds = pci_get_drvdata(pdev);
529 u32 orig_val, val, mask;
533 if (!cxlds->regs.ras) {
534 dev_dbg(&pdev->dev, "No RAS registers.\n");
538 /* BIOS has PCIe AER error control */
539 if (!pcie_aer_is_native(pdev))
542 rc = pcie_capability_read_word(pdev, PCI_EXP_DEVCTL, &cap);
546 if (cap & PCI_EXP_DEVCTL_URRE) {
547 addr = cxlds->regs.ras + CXL_RAS_UNCORRECTABLE_MASK_OFFSET;
548 orig_val = readl(addr);
550 mask = CXL_RAS_UNCORRECTABLE_MASK_MASK |
551 CXL_RAS_UNCORRECTABLE_MASK_F256B_MASK;
552 val = orig_val & ~mask;
555 "Uncorrectable RAS Errors Mask: %#x -> %#x\n",
559 if (cap & PCI_EXP_DEVCTL_CERE) {
560 addr = cxlds->regs.ras + CXL_RAS_CORRECTABLE_MASK_OFFSET;
561 orig_val = readl(addr);
562 val = orig_val & ~CXL_RAS_CORRECTABLE_MASK_MASK;
564 dev_dbg(&pdev->dev, "Correctable RAS Errors Mask: %#x -> %#x\n",
571 static void free_event_buf(void *buf)
577 * There is a single buffer for reading event logs from the mailbox. All logs
578 * share this buffer protected by the mds->event_log_lock.
580 static int cxl_mem_alloc_event_buf(struct cxl_memdev_state *mds)
582 struct cxl_get_event_payload *buf;
584 buf = kvmalloc(mds->payload_size, GFP_KERNEL);
587 mds->event.buf = buf;
589 return devm_add_action_or_reset(mds->cxlds.dev, free_event_buf, buf);
592 static int cxl_alloc_irq_vectors(struct pci_dev *pdev)
597 * Per CXL 3.0 3.1.1 CXL.io Endpoint a function on a CXL device must
598 * not generate INTx messages if that function participates in
599 * CXL.cache or CXL.mem.
601 * Additionally pci_alloc_irq_vectors() handles calling
602 * pci_free_irq_vectors() automatically despite not being called
603 * pcim_*. See pci_setup_msi_context().
605 nvecs = pci_alloc_irq_vectors(pdev, 1, CXL_PCI_DEFAULT_MAX_VECTORS,
606 PCI_IRQ_MSIX | PCI_IRQ_MSI);
608 dev_dbg(&pdev->dev, "Failed to alloc irq vectors: %d\n", nvecs);
614 static irqreturn_t cxl_event_thread(int irq, void *id)
616 struct cxl_dev_id *dev_id = id;
617 struct cxl_dev_state *cxlds = dev_id->cxlds;
618 struct cxl_memdev_state *mds = to_cxl_memdev_state(cxlds);
623 * CXL 3.0 8.2.8.3.1: The lower 32 bits are the status;
624 * ignore the reserved upper 32 bits
626 status = readl(cxlds->regs.status + CXLDEV_DEV_EVENT_STATUS_OFFSET);
627 /* Ignore logs unknown to the driver */
628 status &= CXLDEV_EVENT_STATUS_ALL;
631 cxl_mem_get_event_records(mds, status);
638 static int cxl_event_req_irq(struct cxl_dev_state *cxlds, u8 setting)
640 struct pci_dev *pdev = to_pci_dev(cxlds->dev);
643 if (FIELD_GET(CXLDEV_EVENT_INT_MODE_MASK, setting) != CXL_INT_MSI_MSIX)
646 irq = pci_irq_vector(pdev,
647 FIELD_GET(CXLDEV_EVENT_INT_MSGNUM_MASK, setting));
651 return cxl_request_irq(cxlds, irq, NULL, cxl_event_thread);
654 static int cxl_event_get_int_policy(struct cxl_memdev_state *mds,
655 struct cxl_event_interrupt_policy *policy)
657 struct cxl_mbox_cmd mbox_cmd = {
658 .opcode = CXL_MBOX_OP_GET_EVT_INT_POLICY,
659 .payload_out = policy,
660 .size_out = sizeof(*policy),
664 rc = cxl_internal_send_cmd(mds, &mbox_cmd);
666 dev_err(mds->cxlds.dev,
667 "Failed to get event interrupt policy : %d", rc);
672 static int cxl_event_config_msgnums(struct cxl_memdev_state *mds,
673 struct cxl_event_interrupt_policy *policy)
675 struct cxl_mbox_cmd mbox_cmd;
678 *policy = (struct cxl_event_interrupt_policy) {
679 .info_settings = CXL_INT_MSI_MSIX,
680 .warn_settings = CXL_INT_MSI_MSIX,
681 .failure_settings = CXL_INT_MSI_MSIX,
682 .fatal_settings = CXL_INT_MSI_MSIX,
685 mbox_cmd = (struct cxl_mbox_cmd) {
686 .opcode = CXL_MBOX_OP_SET_EVT_INT_POLICY,
687 .payload_in = policy,
688 .size_in = sizeof(*policy),
691 rc = cxl_internal_send_cmd(mds, &mbox_cmd);
693 dev_err(mds->cxlds.dev, "Failed to set event interrupt policy : %d",
698 /* Retrieve final interrupt settings */
699 return cxl_event_get_int_policy(mds, policy);
702 static int cxl_event_irqsetup(struct cxl_memdev_state *mds)
704 struct cxl_dev_state *cxlds = &mds->cxlds;
705 struct cxl_event_interrupt_policy policy;
708 rc = cxl_event_config_msgnums(mds, &policy);
712 rc = cxl_event_req_irq(cxlds, policy.info_settings);
714 dev_err(cxlds->dev, "Failed to get interrupt for event Info log\n");
718 rc = cxl_event_req_irq(cxlds, policy.warn_settings);
720 dev_err(cxlds->dev, "Failed to get interrupt for event Warn log\n");
724 rc = cxl_event_req_irq(cxlds, policy.failure_settings);
726 dev_err(cxlds->dev, "Failed to get interrupt for event Failure log\n");
730 rc = cxl_event_req_irq(cxlds, policy.fatal_settings);
732 dev_err(cxlds->dev, "Failed to get interrupt for event Fatal log\n");
739 static bool cxl_event_int_is_fw(u8 setting)
741 u8 mode = FIELD_GET(CXLDEV_EVENT_INT_MODE_MASK, setting);
743 return mode == CXL_INT_FW;
746 static int cxl_event_config(struct pci_host_bridge *host_bridge,
747 struct cxl_memdev_state *mds)
749 struct cxl_event_interrupt_policy policy;
753 * When BIOS maintains CXL error reporting control, it will process
754 * event records. Only one agent can do so.
756 if (!host_bridge->native_cxl_error)
759 rc = cxl_mem_alloc_event_buf(mds);
763 rc = cxl_event_get_int_policy(mds, &policy);
767 if (cxl_event_int_is_fw(policy.info_settings) ||
768 cxl_event_int_is_fw(policy.warn_settings) ||
769 cxl_event_int_is_fw(policy.failure_settings) ||
770 cxl_event_int_is_fw(policy.fatal_settings)) {
771 dev_err(mds->cxlds.dev,
772 "FW still in control of Event Logs despite _OSC settings\n");
776 rc = cxl_event_irqsetup(mds);
780 cxl_mem_get_event_records(mds, CXLDEV_EVENT_STATUS_ALL);
785 static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
787 struct pci_host_bridge *host_bridge = pci_find_host_bridge(pdev->bus);
788 struct cxl_memdev_state *mds;
789 struct cxl_dev_state *cxlds;
790 struct cxl_register_map map;
791 struct cxl_memdev *cxlmd;
792 int i, rc, pmu_count;
795 * Double check the anonymous union trickery in struct cxl_regs
796 * FIXME switch to struct_group()
798 BUILD_BUG_ON(offsetof(struct cxl_regs, memdev) !=
799 offsetof(struct cxl_regs, device_regs.memdev));
801 rc = pcim_enable_device(pdev);
804 pci_set_master(pdev);
806 mds = cxl_memdev_state_create(&pdev->dev);
810 pci_set_drvdata(pdev, cxlds);
812 cxlds->rcd = is_cxl_restricted(pdev);
813 cxlds->serial = pci_get_dsn(pdev);
814 cxlds->cxl_dvsec = pci_find_dvsec_capability(
815 pdev, PCI_DVSEC_VENDOR_ID_CXL, CXL_DVSEC_PCIE_DEVICE);
816 if (!cxlds->cxl_dvsec)
818 "Device DVSEC not present, skip CXL.mem init\n");
820 rc = cxl_pci_setup_regs(pdev, CXL_REGLOC_RBI_MEMDEV, &map);
824 rc = cxl_map_device_regs(&map, &cxlds->regs.device_regs);
829 * If the component registers can't be found, the cxl_pci driver may
830 * still be useful for management functions so don't return an error.
832 cxlds->component_reg_phys = CXL_RESOURCE_NONE;
833 rc = cxl_pci_setup_regs(pdev, CXL_REGLOC_RBI_COMPONENT, &map);
835 dev_warn(&pdev->dev, "No component registers (%d)\n", rc);
836 else if (!map.component_map.ras.valid)
837 dev_dbg(&pdev->dev, "RAS registers not found\n");
839 cxlds->component_reg_phys = map.resource;
841 rc = cxl_map_component_regs(&map, &cxlds->regs.component,
842 BIT(CXL_CM_CAP_CAP_ID_RAS));
844 dev_dbg(&pdev->dev, "Failed to map RAS capability.\n");
846 rc = cxl_await_media_ready(cxlds);
848 cxlds->media_ready = true;
850 dev_warn(&pdev->dev, "Media not active (%d)\n", rc);
852 rc = cxl_alloc_irq_vectors(pdev);
856 rc = cxl_pci_setup_mailbox(mds);
860 rc = cxl_enumerate_cmds(mds);
864 rc = cxl_set_timestamp(mds);
868 rc = cxl_poison_state_init(mds);
872 rc = cxl_dev_state_identify(mds);
876 rc = cxl_mem_create_range_info(mds);
880 cxlmd = devm_cxl_add_memdev(cxlds);
882 return PTR_ERR(cxlmd);
884 rc = cxl_memdev_setup_fw_upload(mds);
888 pmu_count = cxl_count_regblock(pdev, CXL_REGLOC_RBI_PMU);
889 for (i = 0; i < pmu_count; i++) {
890 struct cxl_pmu_regs pmu_regs;
892 rc = cxl_find_regblock_instance(pdev, CXL_REGLOC_RBI_PMU, &map, i);
894 dev_dbg(&pdev->dev, "Could not find PMU regblock\n");
898 rc = cxl_map_pmu_regs(pdev, &pmu_regs, &map);
900 dev_dbg(&pdev->dev, "Could not map PMU regs\n");
904 rc = devm_cxl_pmu_add(cxlds->dev, &pmu_regs, cxlmd->id, i, CXL_PMU_MEMDEV);
906 dev_dbg(&pdev->dev, "Could not add PMU instance\n");
911 rc = cxl_event_config(host_bridge, mds);
915 rc = cxl_pci_ras_unmask(pdev);
917 dev_dbg(&pdev->dev, "No RAS reporting unmasked\n");
919 pci_save_state(pdev);
924 static const struct pci_device_id cxl_mem_pci_tbl[] = {
925 /* PCI class code for CXL.mem Type-3 Devices */
926 { PCI_DEVICE_CLASS((PCI_CLASS_MEMORY_CXL << 8 | CXL_MEMORY_PROGIF), ~0)},
927 { /* terminate list */ },
929 MODULE_DEVICE_TABLE(pci, cxl_mem_pci_tbl);
931 static pci_ers_result_t cxl_slot_reset(struct pci_dev *pdev)
933 struct cxl_dev_state *cxlds = pci_get_drvdata(pdev);
934 struct cxl_memdev *cxlmd = cxlds->cxlmd;
935 struct device *dev = &cxlmd->dev;
937 dev_info(&pdev->dev, "%s: restart CXL.mem after slot reset\n",
939 pci_restore_state(pdev);
940 if (device_attach(dev) <= 0)
941 return PCI_ERS_RESULT_DISCONNECT;
942 return PCI_ERS_RESULT_RECOVERED;
945 static void cxl_error_resume(struct pci_dev *pdev)
947 struct cxl_dev_state *cxlds = pci_get_drvdata(pdev);
948 struct cxl_memdev *cxlmd = cxlds->cxlmd;
949 struct device *dev = &cxlmd->dev;
951 dev_info(&pdev->dev, "%s: error resume %s\n", dev_name(dev),
952 dev->driver ? "successful" : "failed");
955 static const struct pci_error_handlers cxl_error_handlers = {
956 .error_detected = cxl_error_detected,
957 .slot_reset = cxl_slot_reset,
958 .resume = cxl_error_resume,
959 .cor_error_detected = cxl_cor_error_detected,
962 static struct pci_driver cxl_pci_driver = {
963 .name = KBUILD_MODNAME,
964 .id_table = cxl_mem_pci_tbl,
965 .probe = cxl_pci_probe,
966 .err_handler = &cxl_error_handlers,
968 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
972 MODULE_LICENSE("GPL v2");
973 module_pci_driver(cxl_pci_driver);
974 MODULE_IMPORT_NS(CXL);