brintos

brintos / linux-shallow public Read only

0
0
Text · 23.2 KiB · 3a45879 Raw
927 lines · c
1// SPDX-License-Identifier: GPL-2.02/*3 * PCI Message Signaled Interrupt (MSI)4 *5 * Copyright (C) 2003-2004 Intel6 * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)7 * Copyright (C) 2016 Christoph Hellwig.8 */9#include <linux/bitfield.h>10#include <linux/err.h>11#include <linux/export.h>12#include <linux/irq.h>13 14#include "../pci.h"15#include "msi.h"16 17int pci_msi_enable = 1;18int pci_msi_ignore_mask;19 20/**21 * pci_msi_supported - check whether MSI may be enabled on a device22 * @dev: pointer to the pci_dev data structure of MSI device function23 * @nvec: how many MSIs have been requested?24 *25 * Look at global flags, the device itself, and its parent buses26 * to determine if MSI/-X are supported for the device. If MSI/-X is27 * supported return 1, else return 0.28 **/29static int pci_msi_supported(struct pci_dev *dev, int nvec)30{31	struct pci_bus *bus;32 33	/* MSI must be globally enabled and supported by the device */34	if (!pci_msi_enable)35		return 0;36 37	if (!dev || dev->no_msi)38		return 0;39 40	/*41	 * You can't ask to have 0 or less MSIs configured.42	 *  a) it's stupid ..43	 *  b) the list manipulation code assumes nvec >= 1.44	 */45	if (nvec < 1)46		return 0;47 48	/*49	 * Any bridge which does NOT route MSI transactions from its50	 * secondary bus to its primary bus must set NO_MSI flag on51	 * the secondary pci_bus.52	 *53	 * The NO_MSI flag can either be set directly by:54	 * - arch-specific PCI host bus controller drivers (deprecated)55	 * - quirks for specific PCI bridges56	 *57	 * or indirectly by platform-specific PCI host bridge drivers by58	 * advertising the 'msi_domain' property, which results in59	 * the NO_MSI flag when no MSI domain is found for this bridge60	 * at probe time.61	 */62	for (bus = dev->bus; bus; bus = bus->parent)63		if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)64			return 0;65 66	return 1;67}68 69static void pcim_msi_release(void *pcidev)70{71	struct pci_dev *dev = pcidev;72 73	dev->is_msi_managed = false;74	pci_free_irq_vectors(dev);75}76 77/*78 * Needs to be separate from pcim_release to prevent an ordering problem79 * vs. msi_device_data_release() in the MSI core code.80 */81static int pcim_setup_msi_release(struct pci_dev *dev)82{83	int ret;84 85	if (!pci_is_managed(dev) || dev->is_msi_managed)86		return 0;87 88	ret = devm_add_action(&dev->dev, pcim_msi_release, dev);89	if (ret)90		return ret;91 92	dev->is_msi_managed = true;93	return 0;94}95 96/*97 * Ordering vs. devres: msi device data has to be installed first so that98 * pcim_msi_release() is invoked before it on device release.99 */100static int pci_setup_msi_context(struct pci_dev *dev)101{102	int ret = msi_setup_device_data(&dev->dev);103 104	if (ret)105		return ret;106 107	return pcim_setup_msi_release(dev);108}109 110/*111 * Helper functions for mask/unmask and MSI message handling112 */113 114void pci_msi_update_mask(struct msi_desc *desc, u32 clear, u32 set)115{116	raw_spinlock_t *lock = &to_pci_dev(desc->dev)->msi_lock;117	unsigned long flags;118 119	if (!desc->pci.msi_attrib.can_mask)120		return;121 122	raw_spin_lock_irqsave(lock, flags);123	desc->pci.msi_mask &= ~clear;124	desc->pci.msi_mask |= set;125	pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->pci.mask_pos,126			       desc->pci.msi_mask);127	raw_spin_unlock_irqrestore(lock, flags);128}129 130/**131 * pci_msi_mask_irq - Generic IRQ chip callback to mask PCI/MSI interrupts132 * @data:	pointer to irqdata associated to that interrupt133 */134void pci_msi_mask_irq(struct irq_data *data)135{136	struct msi_desc *desc = irq_data_get_msi_desc(data);137 138	__pci_msi_mask_desc(desc, BIT(data->irq - desc->irq));139}140EXPORT_SYMBOL_GPL(pci_msi_mask_irq);141 142/**143 * pci_msi_unmask_irq - Generic IRQ chip callback to unmask PCI/MSI interrupts144 * @data:	pointer to irqdata associated to that interrupt145 */146void pci_msi_unmask_irq(struct irq_data *data)147{148	struct msi_desc *desc = irq_data_get_msi_desc(data);149 150	__pci_msi_unmask_desc(desc, BIT(data->irq - desc->irq));151}152EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);153 154void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)155{156	struct pci_dev *dev = msi_desc_to_pci_dev(entry);157 158	BUG_ON(dev->current_state != PCI_D0);159 160	if (entry->pci.msi_attrib.is_msix) {161		void __iomem *base = pci_msix_desc_addr(entry);162 163		if (WARN_ON_ONCE(entry->pci.msi_attrib.is_virtual))164			return;165 166		msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);167		msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);168		msg->data = readl(base + PCI_MSIX_ENTRY_DATA);169	} else {170		int pos = dev->msi_cap;171		u16 data;172 173		pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,174				      &msg->address_lo);175		if (entry->pci.msi_attrib.is_64) {176			pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,177					      &msg->address_hi);178			pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);179		} else {180			msg->address_hi = 0;181			pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);182		}183		msg->data = data;184	}185}186 187static inline void pci_write_msg_msi(struct pci_dev *dev, struct msi_desc *desc,188				     struct msi_msg *msg)189{190	int pos = dev->msi_cap;191	u16 msgctl;192 193	pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);194	msgctl &= ~PCI_MSI_FLAGS_QSIZE;195	msgctl |= FIELD_PREP(PCI_MSI_FLAGS_QSIZE, desc->pci.msi_attrib.multiple);196	pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);197 198	pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, msg->address_lo);199	if (desc->pci.msi_attrib.is_64) {200		pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,  msg->address_hi);201		pci_write_config_word(dev, pos + PCI_MSI_DATA_64, msg->data);202	} else {203		pci_write_config_word(dev, pos + PCI_MSI_DATA_32, msg->data);204	}205	/* Ensure that the writes are visible in the device */206	pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);207}208 209static inline void pci_write_msg_msix(struct msi_desc *desc, struct msi_msg *msg)210{211	void __iomem *base = pci_msix_desc_addr(desc);212	u32 ctrl = desc->pci.msix_ctrl;213	bool unmasked = !(ctrl & PCI_MSIX_ENTRY_CTRL_MASKBIT);214 215	if (desc->pci.msi_attrib.is_virtual)216		return;217	/*218	 * The specification mandates that the entry is masked219	 * when the message is modified:220	 *221	 * "If software changes the Address or Data value of an222	 * entry while the entry is unmasked, the result is223	 * undefined."224	 */225	if (unmasked)226		pci_msix_write_vector_ctrl(desc, ctrl | PCI_MSIX_ENTRY_CTRL_MASKBIT);227 228	writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);229	writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);230	writel(msg->data, base + PCI_MSIX_ENTRY_DATA);231 232	if (unmasked)233		pci_msix_write_vector_ctrl(desc, ctrl);234 235	/* Ensure that the writes are visible in the device */236	readl(base + PCI_MSIX_ENTRY_DATA);237}238 239void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)240{241	struct pci_dev *dev = msi_desc_to_pci_dev(entry);242 243	if (dev->current_state != PCI_D0 || pci_dev_is_disconnected(dev)) {244		/* Don't touch the hardware now */245	} else if (entry->pci.msi_attrib.is_msix) {246		pci_write_msg_msix(entry, msg);247	} else {248		pci_write_msg_msi(dev, entry, msg);249	}250 251	entry->msg = *msg;252 253	if (entry->write_msi_msg)254		entry->write_msi_msg(entry, entry->write_msi_msg_data);255}256 257void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)258{259	struct msi_desc *entry = irq_get_msi_desc(irq);260 261	__pci_write_msi_msg(entry, msg);262}263EXPORT_SYMBOL_GPL(pci_write_msi_msg);264 265 266/* PCI/MSI specific functionality */267 268static void pci_intx_for_msi(struct pci_dev *dev, int enable)269{270	if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))271		pci_intx(dev, enable);272}273 274static void pci_msi_set_enable(struct pci_dev *dev, int enable)275{276	u16 control;277 278	pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);279	control &= ~PCI_MSI_FLAGS_ENABLE;280	if (enable)281		control |= PCI_MSI_FLAGS_ENABLE;282	pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);283}284 285static int msi_setup_msi_desc(struct pci_dev *dev, int nvec,286			      struct irq_affinity_desc *masks)287{288	struct msi_desc desc;289	u16 control;290 291	/* MSI Entry Initialization */292	memset(&desc, 0, sizeof(desc));293 294	pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);295	/* Lies, damned lies, and MSIs */296	if (dev->dev_flags & PCI_DEV_FLAGS_HAS_MSI_MASKING)297		control |= PCI_MSI_FLAGS_MASKBIT;298	/* Respect XEN's mask disabling */299	if (pci_msi_ignore_mask)300		control &= ~PCI_MSI_FLAGS_MASKBIT;301 302	desc.nvec_used			= nvec;303	desc.pci.msi_attrib.is_64	= !!(control & PCI_MSI_FLAGS_64BIT);304	desc.pci.msi_attrib.can_mask	= !!(control & PCI_MSI_FLAGS_MASKBIT);305	desc.pci.msi_attrib.default_irq	= dev->irq;306	desc.pci.msi_attrib.multi_cap	= FIELD_GET(PCI_MSI_FLAGS_QMASK, control);307	desc.pci.msi_attrib.multiple	= ilog2(__roundup_pow_of_two(nvec));308	desc.affinity			= masks;309 310	if (control & PCI_MSI_FLAGS_64BIT)311		desc.pci.mask_pos = dev->msi_cap + PCI_MSI_MASK_64;312	else313		desc.pci.mask_pos = dev->msi_cap + PCI_MSI_MASK_32;314 315	/* Save the initial mask status */316	if (desc.pci.msi_attrib.can_mask)317		pci_read_config_dword(dev, desc.pci.mask_pos, &desc.pci.msi_mask);318 319	return msi_insert_msi_desc(&dev->dev, &desc);320}321 322static int msi_verify_entries(struct pci_dev *dev)323{324	struct msi_desc *entry;325 326	if (!dev->no_64bit_msi)327		return 0;328 329	msi_for_each_desc(entry, &dev->dev, MSI_DESC_ALL) {330		if (entry->msg.address_hi) {331			pci_err(dev, "arch assigned 64-bit MSI address %#x%08x but device only supports 32 bits\n",332				entry->msg.address_hi, entry->msg.address_lo);333			break;334		}335	}336	return !entry ? 0 : -EIO;337}338 339/**340 * msi_capability_init - configure device's MSI capability structure341 * @dev: pointer to the pci_dev data structure of MSI device function342 * @nvec: number of interrupts to allocate343 * @affd: description of automatic IRQ affinity assignments (may be %NULL)344 *345 * Setup the MSI capability structure of the device with the requested346 * number of interrupts.  A return value of zero indicates the successful347 * setup of an entry with the new MSI IRQ.  A negative return value indicates348 * an error, and a positive return value indicates the number of interrupts349 * which could have been allocated.350 */351static int msi_capability_init(struct pci_dev *dev, int nvec,352			       struct irq_affinity *affd)353{354	struct irq_affinity_desc *masks = NULL;355	struct msi_desc *entry, desc;356	int ret;357 358	/* Reject multi-MSI early on irq domain enabled architectures */359	if (nvec > 1 && !pci_msi_domain_supports(dev, MSI_FLAG_MULTI_PCI_MSI, ALLOW_LEGACY))360		return 1;361 362	/*363	 * Disable MSI during setup in the hardware, but mark it enabled364	 * so that setup code can evaluate it.365	 */366	pci_msi_set_enable(dev, 0);367	dev->msi_enabled = 1;368 369	if (affd)370		masks = irq_create_affinity_masks(nvec, affd);371 372	msi_lock_descs(&dev->dev);373	ret = msi_setup_msi_desc(dev, nvec, masks);374	if (ret)375		goto fail;376 377	/* All MSIs are unmasked by default; mask them all */378	entry = msi_first_desc(&dev->dev, MSI_DESC_ALL);379	pci_msi_mask(entry, msi_multi_mask(entry));380	/*381	 * Copy the MSI descriptor for the error path because382	 * pci_msi_setup_msi_irqs() will free it for the hierarchical383	 * interrupt domain case.384	 */385	memcpy(&desc, entry, sizeof(desc));386 387	/* Configure MSI capability structure */388	ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);389	if (ret)390		goto err;391 392	ret = msi_verify_entries(dev);393	if (ret)394		goto err;395 396	/* Set MSI enabled bits	*/397	pci_intx_for_msi(dev, 0);398	pci_msi_set_enable(dev, 1);399 400	pcibios_free_irq(dev);401	dev->irq = entry->irq;402	goto unlock;403 404err:405	pci_msi_unmask(&desc, msi_multi_mask(&desc));406	pci_free_msi_irqs(dev);407fail:408	dev->msi_enabled = 0;409unlock:410	msi_unlock_descs(&dev->dev);411	kfree(masks);412	return ret;413}414 415int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,416			   struct irq_affinity *affd)417{418	int nvec;419	int rc;420 421	if (!pci_msi_supported(dev, minvec) || dev->current_state != PCI_D0)422		return -EINVAL;423 424	/* Check whether driver already requested MSI-X IRQs */425	if (dev->msix_enabled) {426		pci_info(dev, "can't enable MSI (MSI-X already enabled)\n");427		return -EINVAL;428	}429 430	if (maxvec < minvec)431		return -ERANGE;432 433	if (WARN_ON_ONCE(dev->msi_enabled))434		return -EINVAL;435 436	nvec = pci_msi_vec_count(dev);437	if (nvec < 0)438		return nvec;439	if (nvec < minvec)440		return -ENOSPC;441 442	if (nvec > maxvec)443		nvec = maxvec;444 445	rc = pci_setup_msi_context(dev);446	if (rc)447		return rc;448 449	if (!pci_setup_msi_device_domain(dev))450		return -ENODEV;451 452	for (;;) {453		if (affd) {454			nvec = irq_calc_affinity_vectors(minvec, nvec, affd);455			if (nvec < minvec)456				return -ENOSPC;457		}458 459		rc = msi_capability_init(dev, nvec, affd);460		if (rc == 0)461			return nvec;462 463		if (rc < 0)464			return rc;465		if (rc < minvec)466			return -ENOSPC;467 468		nvec = rc;469	}470}471 472/**473 * pci_msi_vec_count - Return the number of MSI vectors a device can send474 * @dev: device to report about475 *476 * This function returns the number of MSI vectors a device requested via477 * Multiple Message Capable register. It returns a negative errno if the478 * device is not capable sending MSI interrupts. Otherwise, the call succeeds479 * and returns a power of two, up to a maximum of 2^5 (32), according to the480 * MSI specification.481 **/482int pci_msi_vec_count(struct pci_dev *dev)483{484	int ret;485	u16 msgctl;486 487	if (!dev->msi_cap)488		return -EINVAL;489 490	pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);491	ret = 1 << FIELD_GET(PCI_MSI_FLAGS_QMASK, msgctl);492 493	return ret;494}495EXPORT_SYMBOL(pci_msi_vec_count);496 497/*498 * Architecture override returns true when the PCI MSI message should be499 * written by the generic restore function.500 */501bool __weak arch_restore_msi_irqs(struct pci_dev *dev)502{503	return true;504}505 506void __pci_restore_msi_state(struct pci_dev *dev)507{508	struct msi_desc *entry;509	u16 control;510 511	if (!dev->msi_enabled)512		return;513 514	entry = irq_get_msi_desc(dev->irq);515 516	pci_intx_for_msi(dev, 0);517	pci_msi_set_enable(dev, 0);518	if (arch_restore_msi_irqs(dev))519		__pci_write_msi_msg(entry, &entry->msg);520 521	pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);522	pci_msi_update_mask(entry, 0, 0);523	control &= ~PCI_MSI_FLAGS_QSIZE;524	control |= PCI_MSI_FLAGS_ENABLE |525		   FIELD_PREP(PCI_MSI_FLAGS_QSIZE, entry->pci.msi_attrib.multiple);526	pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);527}528 529void pci_msi_shutdown(struct pci_dev *dev)530{531	struct msi_desc *desc;532 533	if (!pci_msi_enable || !dev || !dev->msi_enabled)534		return;535 536	pci_msi_set_enable(dev, 0);537	pci_intx_for_msi(dev, 1);538	dev->msi_enabled = 0;539 540	/* Return the device with MSI unmasked as initial states */541	desc = msi_first_desc(&dev->dev, MSI_DESC_ALL);542	if (!WARN_ON_ONCE(!desc))543		pci_msi_unmask(desc, msi_multi_mask(desc));544 545	/* Restore dev->irq to its default pin-assertion IRQ */546	dev->irq = desc->pci.msi_attrib.default_irq;547	pcibios_alloc_irq(dev);548}549 550/* PCI/MSI-X specific functionality */551 552static void pci_msix_clear_and_set_ctrl(struct pci_dev *dev, u16 clear, u16 set)553{554	u16 ctrl;555 556	pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &ctrl);557	ctrl &= ~clear;558	ctrl |= set;559	pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, ctrl);560}561 562static void __iomem *msix_map_region(struct pci_dev *dev,563				     unsigned int nr_entries)564{565	resource_size_t phys_addr;566	u32 table_offset;567	unsigned long flags;568	u8 bir;569 570	pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,571			      &table_offset);572	bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);573	flags = pci_resource_flags(dev, bir);574	if (!flags || (flags & IORESOURCE_UNSET))575		return NULL;576 577	table_offset &= PCI_MSIX_TABLE_OFFSET;578	phys_addr = pci_resource_start(dev, bir) + table_offset;579 580	return ioremap(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);581}582 583/**584 * msix_prepare_msi_desc - Prepare a half initialized MSI descriptor for operation585 * @dev:	The PCI device for which the descriptor is prepared586 * @desc:	The MSI descriptor for preparation587 *588 * This is separate from msix_setup_msi_descs() below to handle dynamic589 * allocations for MSI-X after initial enablement.590 *591 * Ideally the whole MSI-X setup would work that way, but there is no way to592 * support this for the legacy arch_setup_msi_irqs() mechanism and for the593 * fake irq domains like the x86 XEN one. Sigh...594 *595 * The descriptor is zeroed and only @desc::msi_index and @desc::affinity596 * are set. When called from msix_setup_msi_descs() then the is_virtual597 * attribute is initialized as well.598 *599 * Fill in the rest.600 */601void msix_prepare_msi_desc(struct pci_dev *dev, struct msi_desc *desc)602{603	desc->nvec_used				= 1;604	desc->pci.msi_attrib.is_msix		= 1;605	desc->pci.msi_attrib.is_64		= 1;606	desc->pci.msi_attrib.default_irq	= dev->irq;607	desc->pci.mask_base			= dev->msix_base;608	desc->pci.msi_attrib.can_mask		= !pci_msi_ignore_mask &&609						  !desc->pci.msi_attrib.is_virtual;610 611	if (desc->pci.msi_attrib.can_mask) {612		void __iomem *addr = pci_msix_desc_addr(desc);613 614		desc->pci.msix_ctrl = readl(addr + PCI_MSIX_ENTRY_VECTOR_CTRL);615	}616}617 618static int msix_setup_msi_descs(struct pci_dev *dev, struct msix_entry *entries,619				int nvec, struct irq_affinity_desc *masks)620{621	int ret = 0, i, vec_count = pci_msix_vec_count(dev);622	struct irq_affinity_desc *curmsk;623	struct msi_desc desc;624 625	memset(&desc, 0, sizeof(desc));626 627	for (i = 0, curmsk = masks; i < nvec; i++, curmsk++) {628		desc.msi_index = entries ? entries[i].entry : i;629		desc.affinity = masks ? curmsk : NULL;630		desc.pci.msi_attrib.is_virtual = desc.msi_index >= vec_count;631 632		msix_prepare_msi_desc(dev, &desc);633 634		ret = msi_insert_msi_desc(&dev->dev, &desc);635		if (ret)636			break;637	}638	return ret;639}640 641static void msix_update_entries(struct pci_dev *dev, struct msix_entry *entries)642{643	struct msi_desc *desc;644 645	if (entries) {646		msi_for_each_desc(desc, &dev->dev, MSI_DESC_ALL) {647			entries->vector = desc->irq;648			entries++;649		}650	}651}652 653static void msix_mask_all(void __iomem *base, int tsize)654{655	u32 ctrl = PCI_MSIX_ENTRY_CTRL_MASKBIT;656	int i;657 658	if (pci_msi_ignore_mask)659		return;660 661	for (i = 0; i < tsize; i++, base += PCI_MSIX_ENTRY_SIZE)662		writel(ctrl, base + PCI_MSIX_ENTRY_VECTOR_CTRL);663}664 665static int msix_setup_interrupts(struct pci_dev *dev, struct msix_entry *entries,666				 int nvec, struct irq_affinity *affd)667{668	struct irq_affinity_desc *masks = NULL;669	int ret;670 671	if (affd)672		masks = irq_create_affinity_masks(nvec, affd);673 674	msi_lock_descs(&dev->dev);675	ret = msix_setup_msi_descs(dev, entries, nvec, masks);676	if (ret)677		goto out_free;678 679	ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);680	if (ret)681		goto out_free;682 683	/* Check if all MSI entries honor device restrictions */684	ret = msi_verify_entries(dev);685	if (ret)686		goto out_free;687 688	msix_update_entries(dev, entries);689	goto out_unlock;690 691out_free:692	pci_free_msi_irqs(dev);693out_unlock:694	msi_unlock_descs(&dev->dev);695	kfree(masks);696	return ret;697}698 699/**700 * msix_capability_init - configure device's MSI-X capability701 * @dev: pointer to the pci_dev data structure of MSI-X device function702 * @entries: pointer to an array of struct msix_entry entries703 * @nvec: number of @entries704 * @affd: Optional pointer to enable automatic affinity assignment705 *706 * Setup the MSI-X capability structure of device function with a707 * single MSI-X IRQ. A return of zero indicates the successful setup of708 * requested MSI-X entries with allocated IRQs or non-zero for otherwise.709 **/710static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,711				int nvec, struct irq_affinity *affd)712{713	int ret, tsize;714	u16 control;715 716	/*717	 * Some devices require MSI-X to be enabled before the MSI-X718	 * registers can be accessed.  Mask all the vectors to prevent719	 * interrupts coming in before they're fully set up.720	 */721	pci_msix_clear_and_set_ctrl(dev, 0, PCI_MSIX_FLAGS_MASKALL |722				    PCI_MSIX_FLAGS_ENABLE);723 724	/* Mark it enabled so setup functions can query it */725	dev->msix_enabled = 1;726 727	pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);728	/* Request & Map MSI-X table region */729	tsize = msix_table_size(control);730	dev->msix_base = msix_map_region(dev, tsize);731	if (!dev->msix_base) {732		ret = -ENOMEM;733		goto out_disable;734	}735 736	ret = msix_setup_interrupts(dev, entries, nvec, affd);737	if (ret)738		goto out_disable;739 740	/* Disable INTX */741	pci_intx_for_msi(dev, 0);742 743	/*744	 * Ensure that all table entries are masked to prevent745	 * stale entries from firing in a crash kernel.746	 *747	 * Done late to deal with a broken Marvell NVME device748	 * which takes the MSI-X mask bits into account even749	 * when MSI-X is disabled, which prevents MSI delivery.750	 */751	msix_mask_all(dev->msix_base, tsize);752	pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);753 754	pcibios_free_irq(dev);755	return 0;756 757out_disable:758	dev->msix_enabled = 0;759	pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE, 0);760 761	return ret;762}763 764static bool pci_msix_validate_entries(struct pci_dev *dev, struct msix_entry *entries, int nvec)765{766	bool nogap;767	int i, j;768 769	if (!entries)770		return true;771 772	nogap = pci_msi_domain_supports(dev, MSI_FLAG_MSIX_CONTIGUOUS, DENY_LEGACY);773 774	for (i = 0; i < nvec; i++) {775		/* Check for duplicate entries */776		for (j = i + 1; j < nvec; j++) {777			if (entries[i].entry == entries[j].entry)778				return false;779		}780		/* Check for unsupported gaps */781		if (nogap && entries[i].entry != i)782			return false;783	}784	return true;785}786 787int __pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries, int minvec,788			    int maxvec, struct irq_affinity *affd, int flags)789{790	int hwsize, rc, nvec = maxvec;791 792	if (maxvec < minvec)793		return -ERANGE;794 795	if (dev->msi_enabled) {796		pci_info(dev, "can't enable MSI-X (MSI already enabled)\n");797		return -EINVAL;798	}799 800	if (WARN_ON_ONCE(dev->msix_enabled))801		return -EINVAL;802 803	/* Check MSI-X early on irq domain enabled architectures */804	if (!pci_msi_domain_supports(dev, MSI_FLAG_PCI_MSIX, ALLOW_LEGACY))805		return -ENOTSUPP;806 807	if (!pci_msi_supported(dev, nvec) || dev->current_state != PCI_D0)808		return -EINVAL;809 810	hwsize = pci_msix_vec_count(dev);811	if (hwsize < 0)812		return hwsize;813 814	if (!pci_msix_validate_entries(dev, entries, nvec))815		return -EINVAL;816 817	if (hwsize < nvec) {818		/* Keep the IRQ virtual hackery working */819		if (flags & PCI_IRQ_VIRTUAL)820			hwsize = nvec;821		else822			nvec = hwsize;823	}824 825	if (nvec < minvec)826		return -ENOSPC;827 828	rc = pci_setup_msi_context(dev);829	if (rc)830		return rc;831 832	if (!pci_setup_msix_device_domain(dev, hwsize))833		return -ENODEV;834 835	for (;;) {836		if (affd) {837			nvec = irq_calc_affinity_vectors(minvec, nvec, affd);838			if (nvec < minvec)839				return -ENOSPC;840		}841 842		rc = msix_capability_init(dev, entries, nvec, affd);843		if (rc == 0)844			return nvec;845 846		if (rc < 0)847			return rc;848		if (rc < minvec)849			return -ENOSPC;850 851		nvec = rc;852	}853}854 855void __pci_restore_msix_state(struct pci_dev *dev)856{857	struct msi_desc *entry;858	bool write_msg;859 860	if (!dev->msix_enabled)861		return;862 863	/* route the table */864	pci_intx_for_msi(dev, 0);865	pci_msix_clear_and_set_ctrl(dev, 0,866				PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);867 868	write_msg = arch_restore_msi_irqs(dev);869 870	msi_lock_descs(&dev->dev);871	msi_for_each_desc(entry, &dev->dev, MSI_DESC_ALL) {872		if (write_msg)873			__pci_write_msi_msg(entry, &entry->msg);874		pci_msix_write_vector_ctrl(entry, entry->pci.msix_ctrl);875	}876	msi_unlock_descs(&dev->dev);877 878	pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);879}880 881void pci_msix_shutdown(struct pci_dev *dev)882{883	struct msi_desc *desc;884 885	if (!pci_msi_enable || !dev || !dev->msix_enabled)886		return;887 888	if (pci_dev_is_disconnected(dev)) {889		dev->msix_enabled = 0;890		return;891	}892 893	/* Return the device with MSI-X masked as initial states */894	msi_for_each_desc(desc, &dev->dev, MSI_DESC_ALL)895		pci_msix_mask(desc);896 897	pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);898	pci_intx_for_msi(dev, 1);899	dev->msix_enabled = 0;900	pcibios_alloc_irq(dev);901}902 903/* Common interfaces */904 905void pci_free_msi_irqs(struct pci_dev *dev)906{907	pci_msi_teardown_msi_irqs(dev);908 909	if (dev->msix_base) {910		iounmap(dev->msix_base);911		dev->msix_base = NULL;912	}913}914 915/* Misc. infrastructure */916 917struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)918{919	return to_pci_dev(desc->dev);920}921EXPORT_SYMBOL(msi_desc_to_pci_dev);922 923void pci_no_msi(void)924{925	pci_msi_enable = 0;926}927