Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  *  Intel IO-APIC support for multi-Pentium hosts.
0004  *
0005  *  Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
0006  *
0007  *  Many thanks to Stig Venaas for trying out countless experimental
0008  *  patches and reporting/debugging problems patiently!
0009  *
0010  *  (c) 1999, Multiple IO-APIC support, developed by
0011  *  Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
0012  *      Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
0013  *  further tested and cleaned up by Zach Brown <zab@redhat.com>
0014  *  and Ingo Molnar <mingo@redhat.com>
0015  *
0016  *  Fixes
0017  *  Maciej W. Rozycki   :   Bits for genuine 82489DX APICs;
0018  *                  thanks to Eric Gilmore
0019  *                  and Rolf G. Tews
0020  *                  for testing these extensively
0021  *  Paul Diefenbaugh    :   Added full ACPI support
0022  *
0023  * Historical information which is worth to be preserved:
0024  *
0025  * - SiS APIC rmw bug:
0026  *
0027  *  We used to have a workaround for a bug in SiS chips which
0028  *  required to rewrite the index register for a read-modify-write
0029  *  operation as the chip lost the index information which was
0030  *  setup for the read already. We cache the data now, so that
0031  *  workaround has been removed.
0032  */
0033 
0034 #include <linux/mm.h>
0035 #include <linux/interrupt.h>
0036 #include <linux/irq.h>
0037 #include <linux/init.h>
0038 #include <linux/delay.h>
0039 #include <linux/sched.h>
0040 #include <linux/pci.h>
0041 #include <linux/mc146818rtc.h>
0042 #include <linux/compiler.h>
0043 #include <linux/acpi.h>
0044 #include <linux/export.h>
0045 #include <linux/syscore_ops.h>
0046 #include <linux/freezer.h>
0047 #include <linux/kthread.h>
0048 #include <linux/jiffies.h>  /* time_after() */
0049 #include <linux/slab.h>
0050 #include <linux/memblock.h>
0051 #include <linux/msi.h>
0052 
0053 #include <asm/irqdomain.h>
0054 #include <asm/io.h>
0055 #include <asm/smp.h>
0056 #include <asm/cpu.h>
0057 #include <asm/desc.h>
0058 #include <asm/proto.h>
0059 #include <asm/acpi.h>
0060 #include <asm/dma.h>
0061 #include <asm/timer.h>
0062 #include <asm/time.h>
0063 #include <asm/i8259.h>
0064 #include <asm/setup.h>
0065 #include <asm/irq_remapping.h>
0066 #include <asm/hw_irq.h>
0067 #include <asm/apic.h>
0068 #include <asm/pgtable.h>
0069 
0070 #define for_each_ioapic(idx)        \
0071     for ((idx) = 0; (idx) < nr_ioapics; (idx)++)
0072 #define for_each_ioapic_reverse(idx)    \
0073     for ((idx) = nr_ioapics - 1; (idx) >= 0; (idx)--)
0074 #define for_each_pin(idx, pin)      \
0075     for ((pin) = 0; (pin) < ioapics[(idx)].nr_registers; (pin)++)
0076 #define for_each_ioapic_pin(idx, pin)   \
0077     for_each_ioapic((idx))      \
0078         for_each_pin((idx), (pin))
0079 #define for_each_irq_pin(entry, head) \
0080     list_for_each_entry(entry, &head, list)
0081 
0082 static DEFINE_RAW_SPINLOCK(ioapic_lock);
0083 static DEFINE_MUTEX(ioapic_mutex);
0084 static unsigned int ioapic_dynirq_base;
0085 static int ioapic_initialized;
0086 
0087 struct irq_pin_list {
0088     struct list_head list;
0089     int apic, pin;
0090 };
0091 
0092 struct mp_chip_data {
0093     struct list_head        irq_2_pin;
0094     struct IO_APIC_route_entry  entry;
0095     bool                is_level;
0096     bool                active_low;
0097     bool                isa_irq;
0098     u32 count;
0099 };
0100 
0101 struct mp_ioapic_gsi {
0102     u32 gsi_base;
0103     u32 gsi_end;
0104 };
0105 
0106 static struct ioapic {
0107     /*
0108      * # of IRQ routing registers
0109      */
0110     int nr_registers;
0111     /*
0112      * Saved state during suspend/resume, or while enabling intr-remap.
0113      */
0114     struct IO_APIC_route_entry *saved_registers;
0115     /* I/O APIC config */
0116     struct mpc_ioapic mp_config;
0117     /* IO APIC gsi routing info */
0118     struct mp_ioapic_gsi  gsi_config;
0119     struct ioapic_domain_cfg irqdomain_cfg;
0120     struct irq_domain *irqdomain;
0121     struct resource *iomem_res;
0122 } ioapics[MAX_IO_APICS];
0123 
0124 #define mpc_ioapic_ver(ioapic_idx)  ioapics[ioapic_idx].mp_config.apicver
0125 
0126 int mpc_ioapic_id(int ioapic_idx)
0127 {
0128     return ioapics[ioapic_idx].mp_config.apicid;
0129 }
0130 
0131 unsigned int mpc_ioapic_addr(int ioapic_idx)
0132 {
0133     return ioapics[ioapic_idx].mp_config.apicaddr;
0134 }
0135 
0136 static inline struct mp_ioapic_gsi *mp_ioapic_gsi_routing(int ioapic_idx)
0137 {
0138     return &ioapics[ioapic_idx].gsi_config;
0139 }
0140 
0141 static inline int mp_ioapic_pin_count(int ioapic)
0142 {
0143     struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic);
0144 
0145     return gsi_cfg->gsi_end - gsi_cfg->gsi_base + 1;
0146 }
0147 
0148 static inline u32 mp_pin_to_gsi(int ioapic, int pin)
0149 {
0150     return mp_ioapic_gsi_routing(ioapic)->gsi_base + pin;
0151 }
0152 
0153 static inline bool mp_is_legacy_irq(int irq)
0154 {
0155     return irq >= 0 && irq < nr_legacy_irqs();
0156 }
0157 
0158 static inline struct irq_domain *mp_ioapic_irqdomain(int ioapic)
0159 {
0160     return ioapics[ioapic].irqdomain;
0161 }
0162 
0163 int nr_ioapics;
0164 
0165 /* The one past the highest gsi number used */
0166 u32 gsi_top;
0167 
0168 /* MP IRQ source entries */
0169 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
0170 
0171 /* # of MP IRQ source entries */
0172 int mp_irq_entries;
0173 
0174 #ifdef CONFIG_EISA
0175 int mp_bus_id_to_type[MAX_MP_BUSSES];
0176 #endif
0177 
0178 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
0179 
0180 int skip_ioapic_setup;
0181 
0182 /**
0183  * disable_ioapic_support() - disables ioapic support at runtime
0184  */
0185 void disable_ioapic_support(void)
0186 {
0187 #ifdef CONFIG_PCI
0188     noioapicquirk = 1;
0189     noioapicreroute = -1;
0190 #endif
0191     skip_ioapic_setup = 1;
0192 }
0193 
0194 static int __init parse_noapic(char *str)
0195 {
0196     /* disable IO-APIC */
0197     disable_ioapic_support();
0198     return 0;
0199 }
0200 early_param("noapic", parse_noapic);
0201 
0202 /* Will be called in mpparse/ACPI codes for saving IRQ info */
0203 void mp_save_irq(struct mpc_intsrc *m)
0204 {
0205     int i;
0206 
0207     apic_printk(APIC_VERBOSE, "Int: type %d, pol %d, trig %d, bus %02x,"
0208         " IRQ %02x, APIC ID %x, APIC INT %02x\n",
0209         m->irqtype, m->irqflag & 3, (m->irqflag >> 2) & 3, m->srcbus,
0210         m->srcbusirq, m->dstapic, m->dstirq);
0211 
0212     for (i = 0; i < mp_irq_entries; i++) {
0213         if (!memcmp(&mp_irqs[i], m, sizeof(*m)))
0214             return;
0215     }
0216 
0217     memcpy(&mp_irqs[mp_irq_entries], m, sizeof(*m));
0218     if (++mp_irq_entries == MAX_IRQ_SOURCES)
0219         panic("Max # of irq sources exceeded!!\n");
0220 }
0221 
0222 static void alloc_ioapic_saved_registers(int idx)
0223 {
0224     size_t size;
0225 
0226     if (ioapics[idx].saved_registers)
0227         return;
0228 
0229     size = sizeof(struct IO_APIC_route_entry) * ioapics[idx].nr_registers;
0230     ioapics[idx].saved_registers = kzalloc(size, GFP_KERNEL);
0231     if (!ioapics[idx].saved_registers)
0232         pr_err("IOAPIC %d: suspend/resume impossible!\n", idx);
0233 }
0234 
0235 static void free_ioapic_saved_registers(int idx)
0236 {
0237     kfree(ioapics[idx].saved_registers);
0238     ioapics[idx].saved_registers = NULL;
0239 }
0240 
0241 int __init arch_early_ioapic_init(void)
0242 {
0243     int i;
0244 
0245     if (!nr_legacy_irqs())
0246         io_apic_irqs = ~0UL;
0247 
0248     for_each_ioapic(i)
0249         alloc_ioapic_saved_registers(i);
0250 
0251     return 0;
0252 }
0253 
0254 struct io_apic {
0255     unsigned int index;
0256     unsigned int unused[3];
0257     unsigned int data;
0258     unsigned int unused2[11];
0259     unsigned int eoi;
0260 };
0261 
0262 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
0263 {
0264     return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
0265         + (mpc_ioapic_addr(idx) & ~PAGE_MASK);
0266 }
0267 
0268 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
0269 {
0270     struct io_apic __iomem *io_apic = io_apic_base(apic);
0271     writel(vector, &io_apic->eoi);
0272 }
0273 
0274 unsigned int native_io_apic_read(unsigned int apic, unsigned int reg)
0275 {
0276     struct io_apic __iomem *io_apic = io_apic_base(apic);
0277     writel(reg, &io_apic->index);
0278     return readl(&io_apic->data);
0279 }
0280 
0281 static void io_apic_write(unsigned int apic, unsigned int reg,
0282               unsigned int value)
0283 {
0284     struct io_apic __iomem *io_apic = io_apic_base(apic);
0285 
0286     writel(reg, &io_apic->index);
0287     writel(value, &io_apic->data);
0288 }
0289 
0290 static struct IO_APIC_route_entry __ioapic_read_entry(int apic, int pin)
0291 {
0292     struct IO_APIC_route_entry entry;
0293 
0294     entry.w1 = io_apic_read(apic, 0x10 + 2 * pin);
0295     entry.w2 = io_apic_read(apic, 0x11 + 2 * pin);
0296 
0297     return entry;
0298 }
0299 
0300 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
0301 {
0302     struct IO_APIC_route_entry entry;
0303     unsigned long flags;
0304 
0305     raw_spin_lock_irqsave(&ioapic_lock, flags);
0306     entry = __ioapic_read_entry(apic, pin);
0307     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
0308 
0309     return entry;
0310 }
0311 
0312 /*
0313  * When we write a new IO APIC routing entry, we need to write the high
0314  * word first! If the mask bit in the low word is clear, we will enable
0315  * the interrupt, and we need to make sure the entry is fully populated
0316  * before that happens.
0317  */
0318 static void __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
0319 {
0320     io_apic_write(apic, 0x11 + 2*pin, e.w2);
0321     io_apic_write(apic, 0x10 + 2*pin, e.w1);
0322 }
0323 
0324 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
0325 {
0326     unsigned long flags;
0327 
0328     raw_spin_lock_irqsave(&ioapic_lock, flags);
0329     __ioapic_write_entry(apic, pin, e);
0330     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
0331 }
0332 
0333 /*
0334  * When we mask an IO APIC routing entry, we need to write the low
0335  * word first, in order to set the mask bit before we change the
0336  * high bits!
0337  */
0338 static void ioapic_mask_entry(int apic, int pin)
0339 {
0340     struct IO_APIC_route_entry e = { .masked = true };
0341     unsigned long flags;
0342 
0343     raw_spin_lock_irqsave(&ioapic_lock, flags);
0344     io_apic_write(apic, 0x10 + 2*pin, e.w1);
0345     io_apic_write(apic, 0x11 + 2*pin, e.w2);
0346     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
0347 }
0348 
0349 /*
0350  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
0351  * shared ISA-space IRQs, so we have to support them. We are super
0352  * fast in the common case, and fast for shared ISA-space IRQs.
0353  */
0354 static int __add_pin_to_irq_node(struct mp_chip_data *data,
0355                  int node, int apic, int pin)
0356 {
0357     struct irq_pin_list *entry;
0358 
0359     /* don't allow duplicates */
0360     for_each_irq_pin(entry, data->irq_2_pin)
0361         if (entry->apic == apic && entry->pin == pin)
0362             return 0;
0363 
0364     entry = kzalloc_node(sizeof(struct irq_pin_list), GFP_ATOMIC, node);
0365     if (!entry) {
0366         pr_err("can not alloc irq_pin_list (%d,%d,%d)\n",
0367                node, apic, pin);
0368         return -ENOMEM;
0369     }
0370     entry->apic = apic;
0371     entry->pin = pin;
0372     list_add_tail(&entry->list, &data->irq_2_pin);
0373 
0374     return 0;
0375 }
0376 
0377 static void __remove_pin_from_irq(struct mp_chip_data *data, int apic, int pin)
0378 {
0379     struct irq_pin_list *tmp, *entry;
0380 
0381     list_for_each_entry_safe(entry, tmp, &data->irq_2_pin, list)
0382         if (entry->apic == apic && entry->pin == pin) {
0383             list_del(&entry->list);
0384             kfree(entry);
0385             return;
0386         }
0387 }
0388 
0389 static void add_pin_to_irq_node(struct mp_chip_data *data,
0390                 int node, int apic, int pin)
0391 {
0392     if (__add_pin_to_irq_node(data, node, apic, pin))
0393         panic("IO-APIC: failed to add irq-pin. Can not proceed\n");
0394 }
0395 
0396 /*
0397  * Reroute an IRQ to a different pin.
0398  */
0399 static void __init replace_pin_at_irq_node(struct mp_chip_data *data, int node,
0400                        int oldapic, int oldpin,
0401                        int newapic, int newpin)
0402 {
0403     struct irq_pin_list *entry;
0404 
0405     for_each_irq_pin(entry, data->irq_2_pin) {
0406         if (entry->apic == oldapic && entry->pin == oldpin) {
0407             entry->apic = newapic;
0408             entry->pin = newpin;
0409             /* every one is different, right? */
0410             return;
0411         }
0412     }
0413 
0414     /* old apic/pin didn't exist, so just add new ones */
0415     add_pin_to_irq_node(data, node, newapic, newpin);
0416 }
0417 
0418 static void io_apic_modify_irq(struct mp_chip_data *data, bool masked,
0419                    void (*final)(struct irq_pin_list *entry))
0420 {
0421     struct irq_pin_list *entry;
0422 
0423     data->entry.masked = masked;
0424 
0425     for_each_irq_pin(entry, data->irq_2_pin) {
0426         io_apic_write(entry->apic, 0x10 + 2 * entry->pin, data->entry.w1);
0427         if (final)
0428             final(entry);
0429     }
0430 }
0431 
0432 static void io_apic_sync(struct irq_pin_list *entry)
0433 {
0434     /*
0435      * Synchronize the IO-APIC and the CPU by doing
0436      * a dummy read from the IO-APIC
0437      */
0438     struct io_apic __iomem *io_apic;
0439 
0440     io_apic = io_apic_base(entry->apic);
0441     readl(&io_apic->data);
0442 }
0443 
0444 static void mask_ioapic_irq(struct irq_data *irq_data)
0445 {
0446     struct mp_chip_data *data = irq_data->chip_data;
0447     unsigned long flags;
0448 
0449     raw_spin_lock_irqsave(&ioapic_lock, flags);
0450     io_apic_modify_irq(data, true, &io_apic_sync);
0451     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
0452 }
0453 
0454 static void __unmask_ioapic(struct mp_chip_data *data)
0455 {
0456     io_apic_modify_irq(data, false, NULL);
0457 }
0458 
0459 static void unmask_ioapic_irq(struct irq_data *irq_data)
0460 {
0461     struct mp_chip_data *data = irq_data->chip_data;
0462     unsigned long flags;
0463 
0464     raw_spin_lock_irqsave(&ioapic_lock, flags);
0465     __unmask_ioapic(data);
0466     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
0467 }
0468 
0469 /*
0470  * IO-APIC versions below 0x20 don't support EOI register.
0471  * For the record, here is the information about various versions:
0472  *     0Xh     82489DX
0473  *     1Xh     I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
0474  *     2Xh     I/O(x)APIC which is PCI 2.2 Compliant
0475  *     30h-FFh Reserved
0476  *
0477  * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
0478  * version as 0x2. This is an error with documentation and these ICH chips
0479  * use io-apic's of version 0x20.
0480  *
0481  * For IO-APIC's with EOI register, we use that to do an explicit EOI.
0482  * Otherwise, we simulate the EOI message manually by changing the trigger
0483  * mode to edge and then back to level, with RTE being masked during this.
0484  */
0485 static void __eoi_ioapic_pin(int apic, int pin, int vector)
0486 {
0487     if (mpc_ioapic_ver(apic) >= 0x20) {
0488         io_apic_eoi(apic, vector);
0489     } else {
0490         struct IO_APIC_route_entry entry, entry1;
0491 
0492         entry = entry1 = __ioapic_read_entry(apic, pin);
0493 
0494         /*
0495          * Mask the entry and change the trigger mode to edge.
0496          */
0497         entry1.masked = true;
0498         entry1.is_level = false;
0499 
0500         __ioapic_write_entry(apic, pin, entry1);
0501 
0502         /*
0503          * Restore the previous level triggered entry.
0504          */
0505         __ioapic_write_entry(apic, pin, entry);
0506     }
0507 }
0508 
0509 static void eoi_ioapic_pin(int vector, struct mp_chip_data *data)
0510 {
0511     unsigned long flags;
0512     struct irq_pin_list *entry;
0513 
0514     raw_spin_lock_irqsave(&ioapic_lock, flags);
0515     for_each_irq_pin(entry, data->irq_2_pin)
0516         __eoi_ioapic_pin(entry->apic, entry->pin, vector);
0517     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
0518 }
0519 
0520 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
0521 {
0522     struct IO_APIC_route_entry entry;
0523 
0524     /* Check delivery_mode to be sure we're not clearing an SMI pin */
0525     entry = ioapic_read_entry(apic, pin);
0526     if (entry.delivery_mode == APIC_DELIVERY_MODE_SMI)
0527         return;
0528 
0529     /*
0530      * Make sure the entry is masked and re-read the contents to check
0531      * if it is a level triggered pin and if the remote-IRR is set.
0532      */
0533     if (!entry.masked) {
0534         entry.masked = true;
0535         ioapic_write_entry(apic, pin, entry);
0536         entry = ioapic_read_entry(apic, pin);
0537     }
0538 
0539     if (entry.irr) {
0540         unsigned long flags;
0541 
0542         /*
0543          * Make sure the trigger mode is set to level. Explicit EOI
0544          * doesn't clear the remote-IRR if the trigger mode is not
0545          * set to level.
0546          */
0547         if (!entry.is_level) {
0548             entry.is_level = true;
0549             ioapic_write_entry(apic, pin, entry);
0550         }
0551         raw_spin_lock_irqsave(&ioapic_lock, flags);
0552         __eoi_ioapic_pin(apic, pin, entry.vector);
0553         raw_spin_unlock_irqrestore(&ioapic_lock, flags);
0554     }
0555 
0556     /*
0557      * Clear the rest of the bits in the IO-APIC RTE except for the mask
0558      * bit.
0559      */
0560     ioapic_mask_entry(apic, pin);
0561     entry = ioapic_read_entry(apic, pin);
0562     if (entry.irr)
0563         pr_err("Unable to reset IRR for apic: %d, pin :%d\n",
0564                mpc_ioapic_id(apic), pin);
0565 }
0566 
0567 void clear_IO_APIC (void)
0568 {
0569     int apic, pin;
0570 
0571     for_each_ioapic_pin(apic, pin)
0572         clear_IO_APIC_pin(apic, pin);
0573 }
0574 
0575 #ifdef CONFIG_X86_32
0576 /*
0577  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
0578  * specific CPU-side IRQs.
0579  */
0580 
0581 #define MAX_PIRQS 8
0582 static int pirq_entries[MAX_PIRQS] = {
0583     [0 ... MAX_PIRQS - 1] = -1
0584 };
0585 
0586 static int __init ioapic_pirq_setup(char *str)
0587 {
0588     int i, max;
0589     int ints[MAX_PIRQS+1];
0590 
0591     get_options(str, ARRAY_SIZE(ints), ints);
0592 
0593     apic_printk(APIC_VERBOSE, KERN_INFO
0594             "PIRQ redirection, working around broken MP-BIOS.\n");
0595     max = MAX_PIRQS;
0596     if (ints[0] < MAX_PIRQS)
0597         max = ints[0];
0598 
0599     for (i = 0; i < max; i++) {
0600         apic_printk(APIC_VERBOSE, KERN_DEBUG
0601                 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
0602         /*
0603          * PIRQs are mapped upside down, usually.
0604          */
0605         pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
0606     }
0607     return 1;
0608 }
0609 
0610 __setup("pirq=", ioapic_pirq_setup);
0611 #endif /* CONFIG_X86_32 */
0612 
0613 /*
0614  * Saves all the IO-APIC RTE's
0615  */
0616 int save_ioapic_entries(void)
0617 {
0618     int apic, pin;
0619     int err = 0;
0620 
0621     for_each_ioapic(apic) {
0622         if (!ioapics[apic].saved_registers) {
0623             err = -ENOMEM;
0624             continue;
0625         }
0626 
0627         for_each_pin(apic, pin)
0628             ioapics[apic].saved_registers[pin] =
0629                 ioapic_read_entry(apic, pin);
0630     }
0631 
0632     return err;
0633 }
0634 
0635 /*
0636  * Mask all IO APIC entries.
0637  */
0638 void mask_ioapic_entries(void)
0639 {
0640     int apic, pin;
0641 
0642     for_each_ioapic(apic) {
0643         if (!ioapics[apic].saved_registers)
0644             continue;
0645 
0646         for_each_pin(apic, pin) {
0647             struct IO_APIC_route_entry entry;
0648 
0649             entry = ioapics[apic].saved_registers[pin];
0650             if (!entry.masked) {
0651                 entry.masked = true;
0652                 ioapic_write_entry(apic, pin, entry);
0653             }
0654         }
0655     }
0656 }
0657 
0658 /*
0659  * Restore IO APIC entries which was saved in the ioapic structure.
0660  */
0661 int restore_ioapic_entries(void)
0662 {
0663     int apic, pin;
0664 
0665     for_each_ioapic(apic) {
0666         if (!ioapics[apic].saved_registers)
0667             continue;
0668 
0669         for_each_pin(apic, pin)
0670             ioapic_write_entry(apic, pin,
0671                        ioapics[apic].saved_registers[pin]);
0672     }
0673     return 0;
0674 }
0675 
0676 /*
0677  * Find the IRQ entry number of a certain pin.
0678  */
0679 static int find_irq_entry(int ioapic_idx, int pin, int type)
0680 {
0681     int i;
0682 
0683     for (i = 0; i < mp_irq_entries; i++)
0684         if (mp_irqs[i].irqtype == type &&
0685             (mp_irqs[i].dstapic == mpc_ioapic_id(ioapic_idx) ||
0686              mp_irqs[i].dstapic == MP_APIC_ALL) &&
0687             mp_irqs[i].dstirq == pin)
0688             return i;
0689 
0690     return -1;
0691 }
0692 
0693 /*
0694  * Find the pin to which IRQ[irq] (ISA) is connected
0695  */
0696 static int __init find_isa_irq_pin(int irq, int type)
0697 {
0698     int i;
0699 
0700     for (i = 0; i < mp_irq_entries; i++) {
0701         int lbus = mp_irqs[i].srcbus;
0702 
0703         if (test_bit(lbus, mp_bus_not_pci) &&
0704             (mp_irqs[i].irqtype == type) &&
0705             (mp_irqs[i].srcbusirq == irq))
0706 
0707             return mp_irqs[i].dstirq;
0708     }
0709     return -1;
0710 }
0711 
0712 static int __init find_isa_irq_apic(int irq, int type)
0713 {
0714     int i;
0715 
0716     for (i = 0; i < mp_irq_entries; i++) {
0717         int lbus = mp_irqs[i].srcbus;
0718 
0719         if (test_bit(lbus, mp_bus_not_pci) &&
0720             (mp_irqs[i].irqtype == type) &&
0721             (mp_irqs[i].srcbusirq == irq))
0722             break;
0723     }
0724 
0725     if (i < mp_irq_entries) {
0726         int ioapic_idx;
0727 
0728         for_each_ioapic(ioapic_idx)
0729             if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic)
0730                 return ioapic_idx;
0731     }
0732 
0733     return -1;
0734 }
0735 
0736 static bool irq_active_low(int idx)
0737 {
0738     int bus = mp_irqs[idx].srcbus;
0739 
0740     /*
0741      * Determine IRQ line polarity (high active or low active):
0742      */
0743     switch (mp_irqs[idx].irqflag & MP_IRQPOL_MASK) {
0744     case MP_IRQPOL_DEFAULT:
0745         /*
0746          * Conforms to spec, ie. bus-type dependent polarity.  PCI
0747          * defaults to low active. [E]ISA defaults to high active.
0748          */
0749         return !test_bit(bus, mp_bus_not_pci);
0750     case MP_IRQPOL_ACTIVE_HIGH:
0751         return false;
0752     case MP_IRQPOL_RESERVED:
0753         pr_warn("IOAPIC: Invalid polarity: 2, defaulting to low\n");
0754         fallthrough;
0755     case MP_IRQPOL_ACTIVE_LOW:
0756     default: /* Pointless default required due to do gcc stupidity */
0757         return true;
0758     }
0759 }
0760 
0761 #ifdef CONFIG_EISA
0762 /*
0763  * EISA Edge/Level control register, ELCR
0764  */
0765 static bool EISA_ELCR(unsigned int irq)
0766 {
0767     if (irq < nr_legacy_irqs()) {
0768         unsigned int port = PIC_ELCR1 + (irq >> 3);
0769         return (inb(port) >> (irq & 7)) & 1;
0770     }
0771     apic_printk(APIC_VERBOSE, KERN_INFO
0772             "Broken MPtable reports ISA irq %d\n", irq);
0773     return false;
0774 }
0775 
0776 /*
0777  * EISA interrupts are always active high and can be edge or level
0778  * triggered depending on the ELCR value.  If an interrupt is listed as
0779  * EISA conforming in the MP table, that means its trigger type must be
0780  * read in from the ELCR.
0781  */
0782 static bool eisa_irq_is_level(int idx, int bus, bool level)
0783 {
0784     switch (mp_bus_id_to_type[bus]) {
0785     case MP_BUS_PCI:
0786     case MP_BUS_ISA:
0787         return level;
0788     case MP_BUS_EISA:
0789         return EISA_ELCR(mp_irqs[idx].srcbusirq);
0790     }
0791     pr_warn("IOAPIC: Invalid srcbus: %d defaulting to level\n", bus);
0792     return true;
0793 }
0794 #else
0795 static inline int eisa_irq_is_level(int idx, int bus, bool level)
0796 {
0797     return level;
0798 }
0799 #endif
0800 
0801 static bool irq_is_level(int idx)
0802 {
0803     int bus = mp_irqs[idx].srcbus;
0804     bool level;
0805 
0806     /*
0807      * Determine IRQ trigger mode (edge or level sensitive):
0808      */
0809     switch (mp_irqs[idx].irqflag & MP_IRQTRIG_MASK) {
0810     case MP_IRQTRIG_DEFAULT:
0811         /*
0812          * Conforms to spec, ie. bus-type dependent trigger
0813          * mode. PCI defaults to level, ISA to edge.
0814          */
0815         level = !test_bit(bus, mp_bus_not_pci);
0816         /* Take EISA into account */
0817         return eisa_irq_is_level(idx, bus, level);
0818     case MP_IRQTRIG_EDGE:
0819         return false;
0820     case MP_IRQTRIG_RESERVED:
0821         pr_warn("IOAPIC: Invalid trigger mode 2 defaulting to level\n");
0822         fallthrough;
0823     case MP_IRQTRIG_LEVEL:
0824     default: /* Pointless default required due to do gcc stupidity */
0825         return true;
0826     }
0827 }
0828 
0829 static int __acpi_get_override_irq(u32 gsi, bool *trigger, bool *polarity)
0830 {
0831     int ioapic, pin, idx;
0832 
0833     if (skip_ioapic_setup)
0834         return -1;
0835 
0836     ioapic = mp_find_ioapic(gsi);
0837     if (ioapic < 0)
0838         return -1;
0839 
0840     pin = mp_find_ioapic_pin(ioapic, gsi);
0841     if (pin < 0)
0842         return -1;
0843 
0844     idx = find_irq_entry(ioapic, pin, mp_INT);
0845     if (idx < 0)
0846         return -1;
0847 
0848     *trigger = irq_is_level(idx);
0849     *polarity = irq_active_low(idx);
0850     return 0;
0851 }
0852 
0853 #ifdef CONFIG_ACPI
0854 int acpi_get_override_irq(u32 gsi, int *is_level, int *active_low)
0855 {
0856     *is_level = *active_low = 0;
0857     return __acpi_get_override_irq(gsi, (bool *)is_level,
0858                        (bool *)active_low);
0859 }
0860 #endif
0861 
0862 void ioapic_set_alloc_attr(struct irq_alloc_info *info, int node,
0863                int trigger, int polarity)
0864 {
0865     init_irq_alloc_info(info, NULL);
0866     info->type = X86_IRQ_ALLOC_TYPE_IOAPIC;
0867     info->ioapic.node = node;
0868     info->ioapic.is_level = trigger;
0869     info->ioapic.active_low = polarity;
0870     info->ioapic.valid = 1;
0871 }
0872 
0873 static void ioapic_copy_alloc_attr(struct irq_alloc_info *dst,
0874                    struct irq_alloc_info *src,
0875                    u32 gsi, int ioapic_idx, int pin)
0876 {
0877     bool level, pol_low;
0878 
0879     copy_irq_alloc_info(dst, src);
0880     dst->type = X86_IRQ_ALLOC_TYPE_IOAPIC;
0881     dst->devid = mpc_ioapic_id(ioapic_idx);
0882     dst->ioapic.pin = pin;
0883     dst->ioapic.valid = 1;
0884     if (src && src->ioapic.valid) {
0885         dst->ioapic.node = src->ioapic.node;
0886         dst->ioapic.is_level = src->ioapic.is_level;
0887         dst->ioapic.active_low = src->ioapic.active_low;
0888     } else {
0889         dst->ioapic.node = NUMA_NO_NODE;
0890         if (__acpi_get_override_irq(gsi, &level, &pol_low) >= 0) {
0891             dst->ioapic.is_level = level;
0892             dst->ioapic.active_low = pol_low;
0893         } else {
0894             /*
0895              * PCI interrupts are always active low level
0896              * triggered.
0897              */
0898             dst->ioapic.is_level = true;
0899             dst->ioapic.active_low = true;
0900         }
0901     }
0902 }
0903 
0904 static int ioapic_alloc_attr_node(struct irq_alloc_info *info)
0905 {
0906     return (info && info->ioapic.valid) ? info->ioapic.node : NUMA_NO_NODE;
0907 }
0908 
0909 static void mp_register_handler(unsigned int irq, bool level)
0910 {
0911     irq_flow_handler_t hdl;
0912     bool fasteoi;
0913 
0914     if (level) {
0915         irq_set_status_flags(irq, IRQ_LEVEL);
0916         fasteoi = true;
0917     } else {
0918         irq_clear_status_flags(irq, IRQ_LEVEL);
0919         fasteoi = false;
0920     }
0921 
0922     hdl = fasteoi ? handle_fasteoi_irq : handle_edge_irq;
0923     __irq_set_handler(irq, hdl, 0, fasteoi ? "fasteoi" : "edge");
0924 }
0925 
0926 static bool mp_check_pin_attr(int irq, struct irq_alloc_info *info)
0927 {
0928     struct mp_chip_data *data = irq_get_chip_data(irq);
0929 
0930     /*
0931      * setup_IO_APIC_irqs() programs all legacy IRQs with default trigger
0932      * and polarity attributes. So allow the first user to reprogram the
0933      * pin with real trigger and polarity attributes.
0934      */
0935     if (irq < nr_legacy_irqs() && data->count == 1) {
0936         if (info->ioapic.is_level != data->is_level)
0937             mp_register_handler(irq, info->ioapic.is_level);
0938         data->entry.is_level = data->is_level = info->ioapic.is_level;
0939         data->entry.active_low = data->active_low = info->ioapic.active_low;
0940     }
0941 
0942     return data->is_level == info->ioapic.is_level &&
0943            data->active_low == info->ioapic.active_low;
0944 }
0945 
0946 static int alloc_irq_from_domain(struct irq_domain *domain, int ioapic, u32 gsi,
0947                  struct irq_alloc_info *info)
0948 {
0949     bool legacy = false;
0950     int irq = -1;
0951     int type = ioapics[ioapic].irqdomain_cfg.type;
0952 
0953     switch (type) {
0954     case IOAPIC_DOMAIN_LEGACY:
0955         /*
0956          * Dynamically allocate IRQ number for non-ISA IRQs in the first
0957          * 16 GSIs on some weird platforms.
0958          */
0959         if (!ioapic_initialized || gsi >= nr_legacy_irqs())
0960             irq = gsi;
0961         legacy = mp_is_legacy_irq(irq);
0962         break;
0963     case IOAPIC_DOMAIN_STRICT:
0964         irq = gsi;
0965         break;
0966     case IOAPIC_DOMAIN_DYNAMIC:
0967         break;
0968     default:
0969         WARN(1, "ioapic: unknown irqdomain type %d\n", type);
0970         return -1;
0971     }
0972 
0973     return __irq_domain_alloc_irqs(domain, irq, 1,
0974                        ioapic_alloc_attr_node(info),
0975                        info, legacy, NULL);
0976 }
0977 
0978 /*
0979  * Need special handling for ISA IRQs because there may be multiple IOAPIC pins
0980  * sharing the same ISA IRQ number and irqdomain only supports 1:1 mapping
0981  * between IOAPIC pin and IRQ number. A typical IOAPIC has 24 pins, pin 0-15 are
0982  * used for legacy IRQs and pin 16-23 are used for PCI IRQs (PIRQ A-H).
0983  * When ACPI is disabled, only legacy IRQ numbers (IRQ0-15) are available, and
0984  * some BIOSes may use MP Interrupt Source records to override IRQ numbers for
0985  * PIRQs instead of reprogramming the interrupt routing logic. Thus there may be
0986  * multiple pins sharing the same legacy IRQ number when ACPI is disabled.
0987  */
0988 static int alloc_isa_irq_from_domain(struct irq_domain *domain,
0989                      int irq, int ioapic, int pin,
0990                      struct irq_alloc_info *info)
0991 {
0992     struct mp_chip_data *data;
0993     struct irq_data *irq_data = irq_get_irq_data(irq);
0994     int node = ioapic_alloc_attr_node(info);
0995 
0996     /*
0997      * Legacy ISA IRQ has already been allocated, just add pin to
0998      * the pin list associated with this IRQ and program the IOAPIC
0999      * entry. The IOAPIC entry
1000      */
1001     if (irq_data && irq_data->parent_data) {
1002         if (!mp_check_pin_attr(irq, info))
1003             return -EBUSY;
1004         if (__add_pin_to_irq_node(irq_data->chip_data, node, ioapic,
1005                       info->ioapic.pin))
1006             return -ENOMEM;
1007     } else {
1008         info->flags |= X86_IRQ_ALLOC_LEGACY;
1009         irq = __irq_domain_alloc_irqs(domain, irq, 1, node, info, true,
1010                           NULL);
1011         if (irq >= 0) {
1012             irq_data = irq_domain_get_irq_data(domain, irq);
1013             data = irq_data->chip_data;
1014             data->isa_irq = true;
1015         }
1016     }
1017 
1018     return irq;
1019 }
1020 
1021 static int mp_map_pin_to_irq(u32 gsi, int idx, int ioapic, int pin,
1022                  unsigned int flags, struct irq_alloc_info *info)
1023 {
1024     int irq;
1025     bool legacy = false;
1026     struct irq_alloc_info tmp;
1027     struct mp_chip_data *data;
1028     struct irq_domain *domain = mp_ioapic_irqdomain(ioapic);
1029 
1030     if (!domain)
1031         return -ENOSYS;
1032 
1033     if (idx >= 0 && test_bit(mp_irqs[idx].srcbus, mp_bus_not_pci)) {
1034         irq = mp_irqs[idx].srcbusirq;
1035         legacy = mp_is_legacy_irq(irq);
1036         /*
1037          * IRQ2 is unusable for historical reasons on systems which
1038          * have a legacy PIC. See the comment vs. IRQ2 further down.
1039          *
1040          * If this gets removed at some point then the related code
1041          * in lapic_assign_system_vectors() needs to be adjusted as
1042          * well.
1043          */
1044         if (legacy && irq == PIC_CASCADE_IR)
1045             return -EINVAL;
1046     }
1047 
1048     mutex_lock(&ioapic_mutex);
1049     if (!(flags & IOAPIC_MAP_ALLOC)) {
1050         if (!legacy) {
1051             irq = irq_find_mapping(domain, pin);
1052             if (irq == 0)
1053                 irq = -ENOENT;
1054         }
1055     } else {
1056         ioapic_copy_alloc_attr(&tmp, info, gsi, ioapic, pin);
1057         if (legacy)
1058             irq = alloc_isa_irq_from_domain(domain, irq,
1059                             ioapic, pin, &tmp);
1060         else if ((irq = irq_find_mapping(domain, pin)) == 0)
1061             irq = alloc_irq_from_domain(domain, ioapic, gsi, &tmp);
1062         else if (!mp_check_pin_attr(irq, &tmp))
1063             irq = -EBUSY;
1064         if (irq >= 0) {
1065             data = irq_get_chip_data(irq);
1066             data->count++;
1067         }
1068     }
1069     mutex_unlock(&ioapic_mutex);
1070 
1071     return irq;
1072 }
1073 
1074 static int pin_2_irq(int idx, int ioapic, int pin, unsigned int flags)
1075 {
1076     u32 gsi = mp_pin_to_gsi(ioapic, pin);
1077 
1078     /*
1079      * Debugging check, we are in big trouble if this message pops up!
1080      */
1081     if (mp_irqs[idx].dstirq != pin)
1082         pr_err("broken BIOS or MPTABLE parser, ayiee!!\n");
1083 
1084 #ifdef CONFIG_X86_32
1085     /*
1086      * PCI IRQ command line redirection. Yes, limits are hardcoded.
1087      */
1088     if ((pin >= 16) && (pin <= 23)) {
1089         if (pirq_entries[pin-16] != -1) {
1090             if (!pirq_entries[pin-16]) {
1091                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1092                         "disabling PIRQ%d\n", pin-16);
1093             } else {
1094                 int irq = pirq_entries[pin-16];
1095                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1096                         "using PIRQ%d -> IRQ %d\n",
1097                         pin-16, irq);
1098                 return irq;
1099             }
1100         }
1101     }
1102 #endif
1103 
1104     return  mp_map_pin_to_irq(gsi, idx, ioapic, pin, flags, NULL);
1105 }
1106 
1107 int mp_map_gsi_to_irq(u32 gsi, unsigned int flags, struct irq_alloc_info *info)
1108 {
1109     int ioapic, pin, idx;
1110 
1111     ioapic = mp_find_ioapic(gsi);
1112     if (ioapic < 0)
1113         return -ENODEV;
1114 
1115     pin = mp_find_ioapic_pin(ioapic, gsi);
1116     idx = find_irq_entry(ioapic, pin, mp_INT);
1117     if ((flags & IOAPIC_MAP_CHECK) && idx < 0)
1118         return -ENODEV;
1119 
1120     return mp_map_pin_to_irq(gsi, idx, ioapic, pin, flags, info);
1121 }
1122 
1123 void mp_unmap_irq(int irq)
1124 {
1125     struct irq_data *irq_data = irq_get_irq_data(irq);
1126     struct mp_chip_data *data;
1127 
1128     if (!irq_data || !irq_data->domain)
1129         return;
1130 
1131     data = irq_data->chip_data;
1132     if (!data || data->isa_irq)
1133         return;
1134 
1135     mutex_lock(&ioapic_mutex);
1136     if (--data->count == 0)
1137         irq_domain_free_irqs(irq, 1);
1138     mutex_unlock(&ioapic_mutex);
1139 }
1140 
1141 /*
1142  * Find a specific PCI IRQ entry.
1143  * Not an __init, possibly needed by modules
1144  */
1145 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
1146 {
1147     int irq, i, best_ioapic = -1, best_idx = -1;
1148 
1149     apic_printk(APIC_DEBUG,
1150             "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1151             bus, slot, pin);
1152     if (test_bit(bus, mp_bus_not_pci)) {
1153         apic_printk(APIC_VERBOSE,
1154                 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1155         return -1;
1156     }
1157 
1158     for (i = 0; i < mp_irq_entries; i++) {
1159         int lbus = mp_irqs[i].srcbus;
1160         int ioapic_idx, found = 0;
1161 
1162         if (bus != lbus || mp_irqs[i].irqtype != mp_INT ||
1163             slot != ((mp_irqs[i].srcbusirq >> 2) & 0x1f))
1164             continue;
1165 
1166         for_each_ioapic(ioapic_idx)
1167             if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic ||
1168                 mp_irqs[i].dstapic == MP_APIC_ALL) {
1169                 found = 1;
1170                 break;
1171             }
1172         if (!found)
1173             continue;
1174 
1175         /* Skip ISA IRQs */
1176         irq = pin_2_irq(i, ioapic_idx, mp_irqs[i].dstirq, 0);
1177         if (irq > 0 && !IO_APIC_IRQ(irq))
1178             continue;
1179 
1180         if (pin == (mp_irqs[i].srcbusirq & 3)) {
1181             best_idx = i;
1182             best_ioapic = ioapic_idx;
1183             goto out;
1184         }
1185 
1186         /*
1187          * Use the first all-but-pin matching entry as a
1188          * best-guess fuzzy result for broken mptables.
1189          */
1190         if (best_idx < 0) {
1191             best_idx = i;
1192             best_ioapic = ioapic_idx;
1193         }
1194     }
1195     if (best_idx < 0)
1196         return -1;
1197 
1198 out:
1199     return pin_2_irq(best_idx, best_ioapic, mp_irqs[best_idx].dstirq,
1200              IOAPIC_MAP_ALLOC);
1201 }
1202 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1203 
1204 static struct irq_chip ioapic_chip, ioapic_ir_chip;
1205 
1206 static void __init setup_IO_APIC_irqs(void)
1207 {
1208     unsigned int ioapic, pin;
1209     int idx;
1210 
1211     apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1212 
1213     for_each_ioapic_pin(ioapic, pin) {
1214         idx = find_irq_entry(ioapic, pin, mp_INT);
1215         if (idx < 0)
1216             apic_printk(APIC_VERBOSE,
1217                     KERN_DEBUG " apic %d pin %d not connected\n",
1218                     mpc_ioapic_id(ioapic), pin);
1219         else
1220             pin_2_irq(idx, ioapic, pin,
1221                   ioapic ? 0 : IOAPIC_MAP_ALLOC);
1222     }
1223 }
1224 
1225 void ioapic_zap_locks(void)
1226 {
1227     raw_spin_lock_init(&ioapic_lock);
1228 }
1229 
1230 static void io_apic_print_entries(unsigned int apic, unsigned int nr_entries)
1231 {
1232     struct IO_APIC_route_entry entry;
1233     char buf[256];
1234     int i;
1235 
1236     printk(KERN_DEBUG "IOAPIC %d:\n", apic);
1237     for (i = 0; i <= nr_entries; i++) {
1238         entry = ioapic_read_entry(apic, i);
1239         snprintf(buf, sizeof(buf),
1240              " pin%02x, %s, %s, %s, V(%02X), IRR(%1d), S(%1d)",
1241              i,
1242              entry.masked ? "disabled" : "enabled ",
1243              entry.is_level ? "level" : "edge ",
1244              entry.active_low ? "low " : "high",
1245              entry.vector, entry.irr, entry.delivery_status);
1246         if (entry.ir_format) {
1247             printk(KERN_DEBUG "%s, remapped, I(%04X),  Z(%X)\n",
1248                    buf,
1249                    (entry.ir_index_15 << 15) | entry.ir_index_0_14,
1250                 entry.ir_zero);
1251         } else {
1252             printk(KERN_DEBUG "%s, %s, D(%02X%02X), M(%1d)\n", buf,
1253                    entry.dest_mode_logical ? "logical " : "physical",
1254                    entry.virt_destid_8_14, entry.destid_0_7,
1255                    entry.delivery_mode);
1256         }
1257     }
1258 }
1259 
1260 static void __init print_IO_APIC(int ioapic_idx)
1261 {
1262     union IO_APIC_reg_00 reg_00;
1263     union IO_APIC_reg_01 reg_01;
1264     union IO_APIC_reg_02 reg_02;
1265     union IO_APIC_reg_03 reg_03;
1266     unsigned long flags;
1267 
1268     raw_spin_lock_irqsave(&ioapic_lock, flags);
1269     reg_00.raw = io_apic_read(ioapic_idx, 0);
1270     reg_01.raw = io_apic_read(ioapic_idx, 1);
1271     if (reg_01.bits.version >= 0x10)
1272         reg_02.raw = io_apic_read(ioapic_idx, 2);
1273     if (reg_01.bits.version >= 0x20)
1274         reg_03.raw = io_apic_read(ioapic_idx, 3);
1275     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1276 
1277     printk(KERN_DEBUG "IO APIC #%d......\n", mpc_ioapic_id(ioapic_idx));
1278     printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1279     printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
1280     printk(KERN_DEBUG ".......    : Delivery Type: %X\n", reg_00.bits.delivery_type);
1281     printk(KERN_DEBUG ".......    : LTS          : %X\n", reg_00.bits.LTS);
1282 
1283     printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1284     printk(KERN_DEBUG ".......     : max redirection entries: %02X\n",
1285         reg_01.bits.entries);
1286 
1287     printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
1288     printk(KERN_DEBUG ".......     : IO APIC version: %02X\n",
1289         reg_01.bits.version);
1290 
1291     /*
1292      * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1293      * but the value of reg_02 is read as the previous read register
1294      * value, so ignore it if reg_02 == reg_01.
1295      */
1296     if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1297         printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1298         printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
1299     }
1300 
1301     /*
1302      * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1303      * or reg_03, but the value of reg_0[23] is read as the previous read
1304      * register value, so ignore it if reg_03 == reg_0[12].
1305      */
1306     if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1307         reg_03.raw != reg_01.raw) {
1308         printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1309         printk(KERN_DEBUG ".......     : Boot DT    : %X\n", reg_03.bits.boot_DT);
1310     }
1311 
1312     printk(KERN_DEBUG ".... IRQ redirection table:\n");
1313     io_apic_print_entries(ioapic_idx, reg_01.bits.entries);
1314 }
1315 
1316 void __init print_IO_APICs(void)
1317 {
1318     int ioapic_idx;
1319     unsigned int irq;
1320 
1321     printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1322     for_each_ioapic(ioapic_idx)
1323         printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1324                mpc_ioapic_id(ioapic_idx),
1325                ioapics[ioapic_idx].nr_registers);
1326 
1327     /*
1328      * We are a bit conservative about what we expect.  We have to
1329      * know about every hardware change ASAP.
1330      */
1331     printk(KERN_INFO "testing the IO APIC.......................\n");
1332 
1333     for_each_ioapic(ioapic_idx)
1334         print_IO_APIC(ioapic_idx);
1335 
1336     printk(KERN_DEBUG "IRQ to pin mappings:\n");
1337     for_each_active_irq(irq) {
1338         struct irq_pin_list *entry;
1339         struct irq_chip *chip;
1340         struct mp_chip_data *data;
1341 
1342         chip = irq_get_chip(irq);
1343         if (chip != &ioapic_chip && chip != &ioapic_ir_chip)
1344             continue;
1345         data = irq_get_chip_data(irq);
1346         if (!data)
1347             continue;
1348         if (list_empty(&data->irq_2_pin))
1349             continue;
1350 
1351         printk(KERN_DEBUG "IRQ%d ", irq);
1352         for_each_irq_pin(entry, data->irq_2_pin)
1353             pr_cont("-> %d:%d", entry->apic, entry->pin);
1354         pr_cont("\n");
1355     }
1356 
1357     printk(KERN_INFO ".................................... done.\n");
1358 }
1359 
1360 /* Where if anywhere is the i8259 connect in external int mode */
1361 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1362 
1363 void __init enable_IO_APIC(void)
1364 {
1365     int i8259_apic, i8259_pin;
1366     int apic, pin;
1367 
1368     if (skip_ioapic_setup)
1369         nr_ioapics = 0;
1370 
1371     if (!nr_legacy_irqs() || !nr_ioapics)
1372         return;
1373 
1374     for_each_ioapic_pin(apic, pin) {
1375         /* See if any of the pins is in ExtINT mode */
1376         struct IO_APIC_route_entry entry = ioapic_read_entry(apic, pin);
1377 
1378         /* If the interrupt line is enabled and in ExtInt mode
1379          * I have found the pin where the i8259 is connected.
1380          */
1381         if (!entry.masked &&
1382             entry.delivery_mode == APIC_DELIVERY_MODE_EXTINT) {
1383             ioapic_i8259.apic = apic;
1384             ioapic_i8259.pin  = pin;
1385             goto found_i8259;
1386         }
1387     }
1388  found_i8259:
1389     /* Look to see what if the MP table has reported the ExtINT */
1390     /* If we could not find the appropriate pin by looking at the ioapic
1391      * the i8259 probably is not connected the ioapic but give the
1392      * mptable a chance anyway.
1393      */
1394     i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1395     i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1396     /* Trust the MP table if nothing is setup in the hardware */
1397     if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1398         printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1399         ioapic_i8259.pin  = i8259_pin;
1400         ioapic_i8259.apic = i8259_apic;
1401     }
1402     /* Complain if the MP table and the hardware disagree */
1403     if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1404         (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1405     {
1406         printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1407     }
1408 
1409     /*
1410      * Do not trust the IO-APIC being empty at bootup
1411      */
1412     clear_IO_APIC();
1413 }
1414 
1415 void native_restore_boot_irq_mode(void)
1416 {
1417     /*
1418      * If the i8259 is routed through an IOAPIC
1419      * Put that IOAPIC in virtual wire mode
1420      * so legacy interrupts can be delivered.
1421      */
1422     if (ioapic_i8259.pin != -1) {
1423         struct IO_APIC_route_entry entry;
1424         u32 apic_id = read_apic_id();
1425 
1426         memset(&entry, 0, sizeof(entry));
1427         entry.masked        = false;
1428         entry.is_level      = false;
1429         entry.active_low    = false;
1430         entry.dest_mode_logical = false;
1431         entry.delivery_mode = APIC_DELIVERY_MODE_EXTINT;
1432         entry.destid_0_7    = apic_id & 0xFF;
1433         entry.virt_destid_8_14  = apic_id >> 8;
1434 
1435         /*
1436          * Add it to the IO-APIC irq-routing table:
1437          */
1438         ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1439     }
1440 
1441     if (boot_cpu_has(X86_FEATURE_APIC) || apic_from_smp_config())
1442         disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1443 }
1444 
1445 void restore_boot_irq_mode(void)
1446 {
1447     if (!nr_legacy_irqs())
1448         return;
1449 
1450     x86_apic_ops.restore();
1451 }
1452 
1453 #ifdef CONFIG_X86_32
1454 /*
1455  * function to set the IO-APIC physical IDs based on the
1456  * values stored in the MPC table.
1457  *
1458  * by Matt Domsch <Matt_Domsch@dell.com>  Tue Dec 21 12:25:05 CST 1999
1459  */
1460 void __init setup_ioapic_ids_from_mpc_nocheck(void)
1461 {
1462     union IO_APIC_reg_00 reg_00;
1463     physid_mask_t phys_id_present_map;
1464     int ioapic_idx;
1465     int i;
1466     unsigned char old_id;
1467     unsigned long flags;
1468 
1469     /*
1470      * This is broken; anything with a real cpu count has to
1471      * circumvent this idiocy regardless.
1472      */
1473     apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map);
1474 
1475     /*
1476      * Set the IOAPIC ID to the value stored in the MPC table.
1477      */
1478     for_each_ioapic(ioapic_idx) {
1479         /* Read the register 0 value */
1480         raw_spin_lock_irqsave(&ioapic_lock, flags);
1481         reg_00.raw = io_apic_read(ioapic_idx, 0);
1482         raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1483 
1484         old_id = mpc_ioapic_id(ioapic_idx);
1485 
1486         if (mpc_ioapic_id(ioapic_idx) >= get_physical_broadcast()) {
1487             printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1488                 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1489             printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1490                 reg_00.bits.ID);
1491             ioapics[ioapic_idx].mp_config.apicid = reg_00.bits.ID;
1492         }
1493 
1494         /*
1495          * Sanity check, is the ID really free? Every APIC in a
1496          * system must have a unique ID or we get lots of nice
1497          * 'stuck on smp_invalidate_needed IPI wait' messages.
1498          */
1499         if (apic->check_apicid_used(&phys_id_present_map,
1500                         mpc_ioapic_id(ioapic_idx))) {
1501             printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1502                 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1503             for (i = 0; i < get_physical_broadcast(); i++)
1504                 if (!physid_isset(i, phys_id_present_map))
1505                     break;
1506             if (i >= get_physical_broadcast())
1507                 panic("Max APIC ID exceeded!\n");
1508             printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1509                 i);
1510             physid_set(i, phys_id_present_map);
1511             ioapics[ioapic_idx].mp_config.apicid = i;
1512         } else {
1513             physid_mask_t tmp;
1514             apic->apicid_to_cpu_present(mpc_ioapic_id(ioapic_idx),
1515                             &tmp);
1516             apic_printk(APIC_VERBOSE, "Setting %d in the "
1517                     "phys_id_present_map\n",
1518                     mpc_ioapic_id(ioapic_idx));
1519             physids_or(phys_id_present_map, phys_id_present_map, tmp);
1520         }
1521 
1522         /*
1523          * We need to adjust the IRQ routing table
1524          * if the ID changed.
1525          */
1526         if (old_id != mpc_ioapic_id(ioapic_idx))
1527             for (i = 0; i < mp_irq_entries; i++)
1528                 if (mp_irqs[i].dstapic == old_id)
1529                     mp_irqs[i].dstapic
1530                         = mpc_ioapic_id(ioapic_idx);
1531 
1532         /*
1533          * Update the ID register according to the right value
1534          * from the MPC table if they are different.
1535          */
1536         if (mpc_ioapic_id(ioapic_idx) == reg_00.bits.ID)
1537             continue;
1538 
1539         apic_printk(APIC_VERBOSE, KERN_INFO
1540             "...changing IO-APIC physical APIC ID to %d ...",
1541             mpc_ioapic_id(ioapic_idx));
1542 
1543         reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
1544         raw_spin_lock_irqsave(&ioapic_lock, flags);
1545         io_apic_write(ioapic_idx, 0, reg_00.raw);
1546         raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1547 
1548         /*
1549          * Sanity check
1550          */
1551         raw_spin_lock_irqsave(&ioapic_lock, flags);
1552         reg_00.raw = io_apic_read(ioapic_idx, 0);
1553         raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1554         if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx))
1555             pr_cont("could not set ID!\n");
1556         else
1557             apic_printk(APIC_VERBOSE, " ok.\n");
1558     }
1559 }
1560 
1561 void __init setup_ioapic_ids_from_mpc(void)
1562 {
1563 
1564     if (acpi_ioapic)
1565         return;
1566     /*
1567      * Don't check I/O APIC IDs for xAPIC systems.  They have
1568      * no meaning without the serial APIC bus.
1569      */
1570     if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
1571         || APIC_XAPIC(boot_cpu_apic_version))
1572         return;
1573     setup_ioapic_ids_from_mpc_nocheck();
1574 }
1575 #endif
1576 
1577 int no_timer_check __initdata;
1578 
1579 static int __init notimercheck(char *s)
1580 {
1581     no_timer_check = 1;
1582     return 1;
1583 }
1584 __setup("no_timer_check", notimercheck);
1585 
1586 static void __init delay_with_tsc(void)
1587 {
1588     unsigned long long start, now;
1589     unsigned long end = jiffies + 4;
1590 
1591     start = rdtsc();
1592 
1593     /*
1594      * We don't know the TSC frequency yet, but waiting for
1595      * 40000000000/HZ TSC cycles is safe:
1596      * 4 GHz == 10 jiffies
1597      * 1 GHz == 40 jiffies
1598      */
1599     do {
1600         rep_nop();
1601         now = rdtsc();
1602     } while ((now - start) < 40000000000ULL / HZ &&
1603         time_before_eq(jiffies, end));
1604 }
1605 
1606 static void __init delay_without_tsc(void)
1607 {
1608     unsigned long end = jiffies + 4;
1609     int band = 1;
1610 
1611     /*
1612      * We don't know any frequency yet, but waiting for
1613      * 40940000000/HZ cycles is safe:
1614      * 4 GHz == 10 jiffies
1615      * 1 GHz == 40 jiffies
1616      * 1 << 1 + 1 << 2 +...+ 1 << 11 = 4094
1617      */
1618     do {
1619         __delay(((1U << band++) * 10000000UL) / HZ);
1620     } while (band < 12 && time_before_eq(jiffies, end));
1621 }
1622 
1623 /*
1624  * There is a nasty bug in some older SMP boards, their mptable lies
1625  * about the timer IRQ. We do the following to work around the situation:
1626  *
1627  *  - timer IRQ defaults to IO-APIC IRQ
1628  *  - if this function detects that timer IRQs are defunct, then we fall
1629  *    back to ISA timer IRQs
1630  */
1631 static int __init timer_irq_works(void)
1632 {
1633     unsigned long t1 = jiffies;
1634 
1635     if (no_timer_check)
1636         return 1;
1637 
1638     local_irq_enable();
1639     if (boot_cpu_has(X86_FEATURE_TSC))
1640         delay_with_tsc();
1641     else
1642         delay_without_tsc();
1643 
1644     /*
1645      * Expect a few ticks at least, to be sure some possible
1646      * glue logic does not lock up after one or two first
1647      * ticks in a non-ExtINT mode.  Also the local APIC
1648      * might have cached one ExtINT interrupt.  Finally, at
1649      * least one tick may be lost due to delays.
1650      */
1651 
1652     local_irq_disable();
1653 
1654     /* Did jiffies advance? */
1655     return time_after(jiffies, t1 + 4);
1656 }
1657 
1658 /*
1659  * In the SMP+IOAPIC case it might happen that there are an unspecified
1660  * number of pending IRQ events unhandled. These cases are very rare,
1661  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1662  * better to do it this way as thus we do not have to be aware of
1663  * 'pending' interrupts in the IRQ path, except at this point.
1664  */
1665 /*
1666  * Edge triggered needs to resend any interrupt
1667  * that was delayed but this is now handled in the device
1668  * independent code.
1669  */
1670 
1671 /*
1672  * Starting up a edge-triggered IO-APIC interrupt is
1673  * nasty - we need to make sure that we get the edge.
1674  * If it is already asserted for some reason, we need
1675  * return 1 to indicate that is was pending.
1676  *
1677  * This is not complete - we should be able to fake
1678  * an edge even if it isn't on the 8259A...
1679  */
1680 static unsigned int startup_ioapic_irq(struct irq_data *data)
1681 {
1682     int was_pending = 0, irq = data->irq;
1683     unsigned long flags;
1684 
1685     raw_spin_lock_irqsave(&ioapic_lock, flags);
1686     if (irq < nr_legacy_irqs()) {
1687         legacy_pic->mask(irq);
1688         if (legacy_pic->irq_pending(irq))
1689             was_pending = 1;
1690     }
1691     __unmask_ioapic(data->chip_data);
1692     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1693 
1694     return was_pending;
1695 }
1696 
1697 atomic_t irq_mis_count;
1698 
1699 #ifdef CONFIG_GENERIC_PENDING_IRQ
1700 static bool io_apic_level_ack_pending(struct mp_chip_data *data)
1701 {
1702     struct irq_pin_list *entry;
1703     unsigned long flags;
1704 
1705     raw_spin_lock_irqsave(&ioapic_lock, flags);
1706     for_each_irq_pin(entry, data->irq_2_pin) {
1707         struct IO_APIC_route_entry e;
1708         int pin;
1709 
1710         pin = entry->pin;
1711         e.w1 = io_apic_read(entry->apic, 0x10 + pin*2);
1712         /* Is the remote IRR bit set? */
1713         if (e.irr) {
1714             raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1715             return true;
1716         }
1717     }
1718     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1719 
1720     return false;
1721 }
1722 
1723 static inline bool ioapic_prepare_move(struct irq_data *data)
1724 {
1725     /* If we are moving the IRQ we need to mask it */
1726     if (unlikely(irqd_is_setaffinity_pending(data))) {
1727         if (!irqd_irq_masked(data))
1728             mask_ioapic_irq(data);
1729         return true;
1730     }
1731     return false;
1732 }
1733 
1734 static inline void ioapic_finish_move(struct irq_data *data, bool moveit)
1735 {
1736     if (unlikely(moveit)) {
1737         /* Only migrate the irq if the ack has been received.
1738          *
1739          * On rare occasions the broadcast level triggered ack gets
1740          * delayed going to ioapics, and if we reprogram the
1741          * vector while Remote IRR is still set the irq will never
1742          * fire again.
1743          *
1744          * To prevent this scenario we read the Remote IRR bit
1745          * of the ioapic.  This has two effects.
1746          * - On any sane system the read of the ioapic will
1747          *   flush writes (and acks) going to the ioapic from
1748          *   this cpu.
1749          * - We get to see if the ACK has actually been delivered.
1750          *
1751          * Based on failed experiments of reprogramming the
1752          * ioapic entry from outside of irq context starting
1753          * with masking the ioapic entry and then polling until
1754          * Remote IRR was clear before reprogramming the
1755          * ioapic I don't trust the Remote IRR bit to be
1756          * completely accurate.
1757          *
1758          * However there appears to be no other way to plug
1759          * this race, so if the Remote IRR bit is not
1760          * accurate and is causing problems then it is a hardware bug
1761          * and you can go talk to the chipset vendor about it.
1762          */
1763         if (!io_apic_level_ack_pending(data->chip_data))
1764             irq_move_masked_irq(data);
1765         /* If the IRQ is masked in the core, leave it: */
1766         if (!irqd_irq_masked(data))
1767             unmask_ioapic_irq(data);
1768     }
1769 }
1770 #else
1771 static inline bool ioapic_prepare_move(struct irq_data *data)
1772 {
1773     return false;
1774 }
1775 static inline void ioapic_finish_move(struct irq_data *data, bool moveit)
1776 {
1777 }
1778 #endif
1779 
1780 static void ioapic_ack_level(struct irq_data *irq_data)
1781 {
1782     struct irq_cfg *cfg = irqd_cfg(irq_data);
1783     unsigned long v;
1784     bool moveit;
1785     int i;
1786 
1787     irq_complete_move(cfg);
1788     moveit = ioapic_prepare_move(irq_data);
1789 
1790     /*
1791      * It appears there is an erratum which affects at least version 0x11
1792      * of I/O APIC (that's the 82093AA and cores integrated into various
1793      * chipsets).  Under certain conditions a level-triggered interrupt is
1794      * erroneously delivered as edge-triggered one but the respective IRR
1795      * bit gets set nevertheless.  As a result the I/O unit expects an EOI
1796      * message but it will never arrive and further interrupts are blocked
1797      * from the source.  The exact reason is so far unknown, but the
1798      * phenomenon was observed when two consecutive interrupt requests
1799      * from a given source get delivered to the same CPU and the source is
1800      * temporarily disabled in between.
1801      *
1802      * A workaround is to simulate an EOI message manually.  We achieve it
1803      * by setting the trigger mode to edge and then to level when the edge
1804      * trigger mode gets detected in the TMR of a local APIC for a
1805      * level-triggered interrupt.  We mask the source for the time of the
1806      * operation to prevent an edge-triggered interrupt escaping meanwhile.
1807      * The idea is from Manfred Spraul.  --macro
1808      *
1809      * Also in the case when cpu goes offline, fixup_irqs() will forward
1810      * any unhandled interrupt on the offlined cpu to the new cpu
1811      * destination that is handling the corresponding interrupt. This
1812      * interrupt forwarding is done via IPI's. Hence, in this case also
1813      * level-triggered io-apic interrupt will be seen as an edge
1814      * interrupt in the IRR. And we can't rely on the cpu's EOI
1815      * to be broadcasted to the IO-APIC's which will clear the remoteIRR
1816      * corresponding to the level-triggered interrupt. Hence on IO-APIC's
1817      * supporting EOI register, we do an explicit EOI to clear the
1818      * remote IRR and on IO-APIC's which don't have an EOI register,
1819      * we use the above logic (mask+edge followed by unmask+level) from
1820      * Manfred Spraul to clear the remote IRR.
1821      */
1822     i = cfg->vector;
1823     v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1824 
1825     /*
1826      * We must acknowledge the irq before we move it or the acknowledge will
1827      * not propagate properly.
1828      */
1829     ack_APIC_irq();
1830 
1831     /*
1832      * Tail end of clearing remote IRR bit (either by delivering the EOI
1833      * message via io-apic EOI register write or simulating it using
1834      * mask+edge followed by unmask+level logic) manually when the
1835      * level triggered interrupt is seen as the edge triggered interrupt
1836      * at the cpu.
1837      */
1838     if (!(v & (1 << (i & 0x1f)))) {
1839         atomic_inc(&irq_mis_count);
1840         eoi_ioapic_pin(cfg->vector, irq_data->chip_data);
1841     }
1842 
1843     ioapic_finish_move(irq_data, moveit);
1844 }
1845 
1846 static void ioapic_ir_ack_level(struct irq_data *irq_data)
1847 {
1848     struct mp_chip_data *data = irq_data->chip_data;
1849 
1850     /*
1851      * Intr-remapping uses pin number as the virtual vector
1852      * in the RTE. Actual vector is programmed in
1853      * intr-remapping table entry. Hence for the io-apic
1854      * EOI we use the pin number.
1855      */
1856     apic_ack_irq(irq_data);
1857     eoi_ioapic_pin(data->entry.vector, data);
1858 }
1859 
1860 /*
1861  * The I/OAPIC is just a device for generating MSI messages from legacy
1862  * interrupt pins. Various fields of the RTE translate into bits of the
1863  * resulting MSI which had a historical meaning.
1864  *
1865  * With interrupt remapping, many of those bits have different meanings
1866  * in the underlying MSI, but the way that the I/OAPIC transforms them
1867  * from its RTE to the MSI message is the same. This function allows
1868  * the parent IRQ domain to compose the MSI message, then takes the
1869  * relevant bits to put them in the appropriate places in the RTE in
1870  * order to generate that message when the IRQ happens.
1871  *
1872  * The setup here relies on a preconfigured route entry (is_level,
1873  * active_low, masked) because the parent domain is merely composing the
1874  * generic message routing information which is used for the MSI.
1875  */
1876 static void ioapic_setup_msg_from_msi(struct irq_data *irq_data,
1877                       struct IO_APIC_route_entry *entry)
1878 {
1879     struct msi_msg msg;
1880 
1881     /* Let the parent domain compose the MSI message */
1882     irq_chip_compose_msi_msg(irq_data, &msg);
1883 
1884     /*
1885      * - Real vector
1886      * - DMAR/IR: 8bit subhandle (ioapic.pin)
1887      * - AMD/IR:  8bit IRTE index
1888      */
1889     entry->vector           = msg.arch_data.vector;
1890     /* Delivery mode (for DMAR/IR all 0) */
1891     entry->delivery_mode        = msg.arch_data.delivery_mode;
1892     /* Destination mode or DMAR/IR index bit 15 */
1893     entry->dest_mode_logical    = msg.arch_addr_lo.dest_mode_logical;
1894     /* DMAR/IR: 1, 0 for all other modes */
1895     entry->ir_format        = msg.arch_addr_lo.dmar_format;
1896     /*
1897      * - DMAR/IR: index bit 0-14.
1898      *
1899      * - Virt: If the host supports x2apic without a virtualized IR
1900      *     unit then bit 0-6 of dmar_index_0_14 are providing bit
1901      *     8-14 of the destination id.
1902      *
1903      * All other modes have bit 0-6 of dmar_index_0_14 cleared and the
1904      * topmost 8 bits are destination id bit 0-7 (entry::destid_0_7).
1905      */
1906     entry->ir_index_0_14        = msg.arch_addr_lo.dmar_index_0_14;
1907 }
1908 
1909 static void ioapic_configure_entry(struct irq_data *irqd)
1910 {
1911     struct mp_chip_data *mpd = irqd->chip_data;
1912     struct irq_pin_list *entry;
1913 
1914     ioapic_setup_msg_from_msi(irqd, &mpd->entry);
1915 
1916     for_each_irq_pin(entry, mpd->irq_2_pin)
1917         __ioapic_write_entry(entry->apic, entry->pin, mpd->entry);
1918 }
1919 
1920 static int ioapic_set_affinity(struct irq_data *irq_data,
1921                    const struct cpumask *mask, bool force)
1922 {
1923     struct irq_data *parent = irq_data->parent_data;
1924     unsigned long flags;
1925     int ret;
1926 
1927     ret = parent->chip->irq_set_affinity(parent, mask, force);
1928     raw_spin_lock_irqsave(&ioapic_lock, flags);
1929     if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE)
1930         ioapic_configure_entry(irq_data);
1931     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1932 
1933     return ret;
1934 }
1935 
1936 /*
1937  * Interrupt shutdown masks the ioapic pin, but the interrupt might already
1938  * be in flight, but not yet serviced by the target CPU. That means
1939  * __synchronize_hardirq() would return and claim that everything is calmed
1940  * down. So free_irq() would proceed and deactivate the interrupt and free
1941  * resources.
1942  *
1943  * Once the target CPU comes around to service it it will find a cleared
1944  * vector and complain. While the spurious interrupt is harmless, the full
1945  * release of resources might prevent the interrupt from being acknowledged
1946  * which keeps the hardware in a weird state.
1947  *
1948  * Verify that the corresponding Remote-IRR bits are clear.
1949  */
1950 static int ioapic_irq_get_chip_state(struct irq_data *irqd,
1951                    enum irqchip_irq_state which,
1952                    bool *state)
1953 {
1954     struct mp_chip_data *mcd = irqd->chip_data;
1955     struct IO_APIC_route_entry rentry;
1956     struct irq_pin_list *p;
1957 
1958     if (which != IRQCHIP_STATE_ACTIVE)
1959         return -EINVAL;
1960 
1961     *state = false;
1962     raw_spin_lock(&ioapic_lock);
1963     for_each_irq_pin(p, mcd->irq_2_pin) {
1964         rentry = __ioapic_read_entry(p->apic, p->pin);
1965         /*
1966          * The remote IRR is only valid in level trigger mode. It's
1967          * meaning is undefined for edge triggered interrupts and
1968          * irrelevant because the IO-APIC treats them as fire and
1969          * forget.
1970          */
1971         if (rentry.irr && rentry.is_level) {
1972             *state = true;
1973             break;
1974         }
1975     }
1976     raw_spin_unlock(&ioapic_lock);
1977     return 0;
1978 }
1979 
1980 static struct irq_chip ioapic_chip __read_mostly = {
1981     .name           = "IO-APIC",
1982     .irq_startup        = startup_ioapic_irq,
1983     .irq_mask       = mask_ioapic_irq,
1984     .irq_unmask     = unmask_ioapic_irq,
1985     .irq_ack        = irq_chip_ack_parent,
1986     .irq_eoi        = ioapic_ack_level,
1987     .irq_set_affinity   = ioapic_set_affinity,
1988     .irq_retrigger      = irq_chip_retrigger_hierarchy,
1989     .irq_get_irqchip_state  = ioapic_irq_get_chip_state,
1990     .flags          = IRQCHIP_SKIP_SET_WAKE |
1991                   IRQCHIP_AFFINITY_PRE_STARTUP,
1992 };
1993 
1994 static struct irq_chip ioapic_ir_chip __read_mostly = {
1995     .name           = "IR-IO-APIC",
1996     .irq_startup        = startup_ioapic_irq,
1997     .irq_mask       = mask_ioapic_irq,
1998     .irq_unmask     = unmask_ioapic_irq,
1999     .irq_ack        = irq_chip_ack_parent,
2000     .irq_eoi        = ioapic_ir_ack_level,
2001     .irq_set_affinity   = ioapic_set_affinity,
2002     .irq_retrigger      = irq_chip_retrigger_hierarchy,
2003     .irq_get_irqchip_state  = ioapic_irq_get_chip_state,
2004     .flags          = IRQCHIP_SKIP_SET_WAKE |
2005                   IRQCHIP_AFFINITY_PRE_STARTUP,
2006 };
2007 
2008 static inline void init_IO_APIC_traps(void)
2009 {
2010     struct irq_cfg *cfg;
2011     unsigned int irq;
2012 
2013     for_each_active_irq(irq) {
2014         cfg = irq_cfg(irq);
2015         if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2016             /*
2017              * Hmm.. We don't have an entry for this,
2018              * so default to an old-fashioned 8259
2019              * interrupt if we can..
2020              */
2021             if (irq < nr_legacy_irqs())
2022                 legacy_pic->make_irq(irq);
2023             else
2024                 /* Strange. Oh, well.. */
2025                 irq_set_chip(irq, &no_irq_chip);
2026         }
2027     }
2028 }
2029 
2030 /*
2031  * The local APIC irq-chip implementation:
2032  */
2033 
2034 static void mask_lapic_irq(struct irq_data *data)
2035 {
2036     unsigned long v;
2037 
2038     v = apic_read(APIC_LVT0);
2039     apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2040 }
2041 
2042 static void unmask_lapic_irq(struct irq_data *data)
2043 {
2044     unsigned long v;
2045 
2046     v = apic_read(APIC_LVT0);
2047     apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2048 }
2049 
2050 static void ack_lapic_irq(struct irq_data *data)
2051 {
2052     ack_APIC_irq();
2053 }
2054 
2055 static struct irq_chip lapic_chip __read_mostly = {
2056     .name       = "local-APIC",
2057     .irq_mask   = mask_lapic_irq,
2058     .irq_unmask = unmask_lapic_irq,
2059     .irq_ack    = ack_lapic_irq,
2060 };
2061 
2062 static void lapic_register_intr(int irq)
2063 {
2064     irq_clear_status_flags(irq, IRQ_LEVEL);
2065     irq_set_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2066                       "edge");
2067 }
2068 
2069 /*
2070  * This looks a bit hackish but it's about the only one way of sending
2071  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
2072  * not support the ExtINT mode, unfortunately.  We need to send these
2073  * cycles as some i82489DX-based boards have glue logic that keeps the
2074  * 8259A interrupt line asserted until INTA.  --macro
2075  */
2076 static inline void __init unlock_ExtINT_logic(void)
2077 {
2078     int apic, pin, i;
2079     struct IO_APIC_route_entry entry0, entry1;
2080     unsigned char save_control, save_freq_select;
2081     u32 apic_id;
2082 
2083     pin  = find_isa_irq_pin(8, mp_INT);
2084     if (pin == -1) {
2085         WARN_ON_ONCE(1);
2086         return;
2087     }
2088     apic = find_isa_irq_apic(8, mp_INT);
2089     if (apic == -1) {
2090         WARN_ON_ONCE(1);
2091         return;
2092     }
2093 
2094     entry0 = ioapic_read_entry(apic, pin);
2095     clear_IO_APIC_pin(apic, pin);
2096 
2097     apic_id = hard_smp_processor_id();
2098     memset(&entry1, 0, sizeof(entry1));
2099 
2100     entry1.dest_mode_logical    = true;
2101     entry1.masked           = false;
2102     entry1.destid_0_7       = apic_id & 0xFF;
2103     entry1.virt_destid_8_14     = apic_id >> 8;
2104     entry1.delivery_mode        = APIC_DELIVERY_MODE_EXTINT;
2105     entry1.active_low       = entry0.active_low;
2106     entry1.is_level         = false;
2107     entry1.vector = 0;
2108 
2109     ioapic_write_entry(apic, pin, entry1);
2110 
2111     save_control = CMOS_READ(RTC_CONTROL);
2112     save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2113     CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2114            RTC_FREQ_SELECT);
2115     CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2116 
2117     i = 100;
2118     while (i-- > 0) {
2119         mdelay(10);
2120         if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2121             i -= 10;
2122     }
2123 
2124     CMOS_WRITE(save_control, RTC_CONTROL);
2125     CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2126     clear_IO_APIC_pin(apic, pin);
2127 
2128     ioapic_write_entry(apic, pin, entry0);
2129 }
2130 
2131 static int disable_timer_pin_1 __initdata;
2132 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2133 static int __init disable_timer_pin_setup(char *arg)
2134 {
2135     disable_timer_pin_1 = 1;
2136     return 0;
2137 }
2138 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2139 
2140 static int mp_alloc_timer_irq(int ioapic, int pin)
2141 {
2142     int irq = -1;
2143     struct irq_domain *domain = mp_ioapic_irqdomain(ioapic);
2144 
2145     if (domain) {
2146         struct irq_alloc_info info;
2147 
2148         ioapic_set_alloc_attr(&info, NUMA_NO_NODE, 0, 0);
2149         info.devid = mpc_ioapic_id(ioapic);
2150         info.ioapic.pin = pin;
2151         mutex_lock(&ioapic_mutex);
2152         irq = alloc_isa_irq_from_domain(domain, 0, ioapic, pin, &info);
2153         mutex_unlock(&ioapic_mutex);
2154     }
2155 
2156     return irq;
2157 }
2158 
2159 /*
2160  * This code may look a bit paranoid, but it's supposed to cooperate with
2161  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
2162  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
2163  * fanatically on his truly buggy board.
2164  *
2165  * FIXME: really need to revamp this for all platforms.
2166  */
2167 static inline void __init check_timer(void)
2168 {
2169     struct irq_data *irq_data = irq_get_irq_data(0);
2170     struct mp_chip_data *data = irq_data->chip_data;
2171     struct irq_cfg *cfg = irqd_cfg(irq_data);
2172     int node = cpu_to_node(0);
2173     int apic1, pin1, apic2, pin2;
2174     int no_pin1 = 0;
2175 
2176     if (!global_clock_event)
2177         return;
2178 
2179     local_irq_disable();
2180 
2181     /*
2182      * get/set the timer IRQ vector:
2183      */
2184     legacy_pic->mask(0);
2185 
2186     /*
2187      * As IRQ0 is to be enabled in the 8259A, the virtual
2188      * wire has to be disabled in the local APIC.  Also
2189      * timer interrupts need to be acknowledged manually in
2190      * the 8259A for the i82489DX when using the NMI
2191      * watchdog as that APIC treats NMIs as level-triggered.
2192      * The AEOI mode will finish them in the 8259A
2193      * automatically.
2194      */
2195     apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2196     legacy_pic->init(1);
2197 
2198     pin1  = find_isa_irq_pin(0, mp_INT);
2199     apic1 = find_isa_irq_apic(0, mp_INT);
2200     pin2  = ioapic_i8259.pin;
2201     apic2 = ioapic_i8259.apic;
2202 
2203     apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2204             "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2205             cfg->vector, apic1, pin1, apic2, pin2);
2206 
2207     /*
2208      * Some BIOS writers are clueless and report the ExtINTA
2209      * I/O APIC input from the cascaded 8259A as the timer
2210      * interrupt input.  So just in case, if only one pin
2211      * was found above, try it both directly and through the
2212      * 8259A.
2213      */
2214     if (pin1 == -1) {
2215         panic_if_irq_remap("BIOS bug: timer not connected to IO-APIC");
2216         pin1 = pin2;
2217         apic1 = apic2;
2218         no_pin1 = 1;
2219     } else if (pin2 == -1) {
2220         pin2 = pin1;
2221         apic2 = apic1;
2222     }
2223 
2224     if (pin1 != -1) {
2225         /* Ok, does IRQ0 through the IOAPIC work? */
2226         if (no_pin1) {
2227             mp_alloc_timer_irq(apic1, pin1);
2228         } else {
2229             /*
2230              * for edge trigger, it's already unmasked,
2231              * so only need to unmask if it is level-trigger
2232              * do we really have level trigger timer?
2233              */
2234             int idx = find_irq_entry(apic1, pin1, mp_INT);
2235 
2236             if (idx != -1 && irq_is_level(idx))
2237                 unmask_ioapic_irq(irq_get_irq_data(0));
2238         }
2239         irq_domain_deactivate_irq(irq_data);
2240         irq_domain_activate_irq(irq_data, false);
2241         if (timer_irq_works()) {
2242             if (disable_timer_pin_1 > 0)
2243                 clear_IO_APIC_pin(0, pin1);
2244             goto out;
2245         }
2246         panic_if_irq_remap("timer doesn't work through Interrupt-remapped IO-APIC");
2247         clear_IO_APIC_pin(apic1, pin1);
2248         if (!no_pin1)
2249             apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2250                     "8254 timer not connected to IO-APIC\n");
2251 
2252         apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2253                 "(IRQ0) through the 8259A ...\n");
2254         apic_printk(APIC_QUIET, KERN_INFO
2255                 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2256         /*
2257          * legacy devices should be connected to IO APIC #0
2258          */
2259         replace_pin_at_irq_node(data, node, apic1, pin1, apic2, pin2);
2260         irq_domain_deactivate_irq(irq_data);
2261         irq_domain_activate_irq(irq_data, false);
2262         legacy_pic->unmask(0);
2263         if (timer_irq_works()) {
2264             apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2265             goto out;
2266         }
2267         /*
2268          * Cleanup, just in case ...
2269          */
2270         legacy_pic->mask(0);
2271         clear_IO_APIC_pin(apic2, pin2);
2272         apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2273     }
2274 
2275     apic_printk(APIC_QUIET, KERN_INFO
2276             "...trying to set up timer as Virtual Wire IRQ...\n");
2277 
2278     lapic_register_intr(0);
2279     apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2280     legacy_pic->unmask(0);
2281 
2282     if (timer_irq_works()) {
2283         apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2284         goto out;
2285     }
2286     legacy_pic->mask(0);
2287     apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
2288     apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2289 
2290     apic_printk(APIC_QUIET, KERN_INFO
2291             "...trying to set up timer as ExtINT IRQ...\n");
2292 
2293     legacy_pic->init(0);
2294     legacy_pic->make_irq(0);
2295     apic_write(APIC_LVT0, APIC_DM_EXTINT);
2296     legacy_pic->unmask(0);
2297 
2298     unlock_ExtINT_logic();
2299 
2300     if (timer_irq_works()) {
2301         apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2302         goto out;
2303     }
2304     apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2305     if (apic_is_x2apic_enabled())
2306         apic_printk(APIC_QUIET, KERN_INFO
2307                 "Perhaps problem with the pre-enabled x2apic mode\n"
2308                 "Try booting with x2apic and interrupt-remapping disabled in the bios.\n");
2309     panic("IO-APIC + timer doesn't work!  Boot with apic=debug and send a "
2310         "report.  Then try booting with the 'noapic' option.\n");
2311 out:
2312     local_irq_enable();
2313 }
2314 
2315 /*
2316  * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2317  * to devices.  However there may be an I/O APIC pin available for
2318  * this interrupt regardless.  The pin may be left unconnected, but
2319  * typically it will be reused as an ExtINT cascade interrupt for
2320  * the master 8259A.  In the MPS case such a pin will normally be
2321  * reported as an ExtINT interrupt in the MP table.  With ACPI
2322  * there is no provision for ExtINT interrupts, and in the absence
2323  * of an override it would be treated as an ordinary ISA I/O APIC
2324  * interrupt, that is edge-triggered and unmasked by default.  We
2325  * used to do this, but it caused problems on some systems because
2326  * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2327  * the same ExtINT cascade interrupt to drive the local APIC of the
2328  * bootstrap processor.  Therefore we refrain from routing IRQ2 to
2329  * the I/O APIC in all cases now.  No actual device should request
2330  * it anyway.  --macro
2331  */
2332 #define PIC_IRQS    (1UL << PIC_CASCADE_IR)
2333 
2334 static int mp_irqdomain_create(int ioapic)
2335 {
2336     struct irq_domain *parent;
2337     int hwirqs = mp_ioapic_pin_count(ioapic);
2338     struct ioapic *ip = &ioapics[ioapic];
2339     struct ioapic_domain_cfg *cfg = &ip->irqdomain_cfg;
2340     struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2341     struct fwnode_handle *fn;
2342     struct irq_fwspec fwspec;
2343 
2344     if (cfg->type == IOAPIC_DOMAIN_INVALID)
2345         return 0;
2346 
2347     /* Handle device tree enumerated APICs proper */
2348     if (cfg->dev) {
2349         fn = of_node_to_fwnode(cfg->dev);
2350     } else {
2351         fn = irq_domain_alloc_named_id_fwnode("IO-APIC", mpc_ioapic_id(ioapic));
2352         if (!fn)
2353             return -ENOMEM;
2354     }
2355 
2356     fwspec.fwnode = fn;
2357     fwspec.param_count = 1;
2358     fwspec.param[0] = mpc_ioapic_id(ioapic);
2359 
2360     parent = irq_find_matching_fwspec(&fwspec, DOMAIN_BUS_ANY);
2361     if (!parent) {
2362         if (!cfg->dev)
2363             irq_domain_free_fwnode(fn);
2364         return -ENODEV;
2365     }
2366 
2367     ip->irqdomain = irq_domain_create_linear(fn, hwirqs, cfg->ops,
2368                          (void *)(long)ioapic);
2369 
2370     if (!ip->irqdomain) {
2371         /* Release fw handle if it was allocated above */
2372         if (!cfg->dev)
2373             irq_domain_free_fwnode(fn);
2374         return -ENOMEM;
2375     }
2376 
2377     ip->irqdomain->parent = parent;
2378 
2379     if (cfg->type == IOAPIC_DOMAIN_LEGACY ||
2380         cfg->type == IOAPIC_DOMAIN_STRICT)
2381         ioapic_dynirq_base = max(ioapic_dynirq_base,
2382                      gsi_cfg->gsi_end + 1);
2383 
2384     return 0;
2385 }
2386 
2387 static void ioapic_destroy_irqdomain(int idx)
2388 {
2389     struct ioapic_domain_cfg *cfg = &ioapics[idx].irqdomain_cfg;
2390     struct fwnode_handle *fn = ioapics[idx].irqdomain->fwnode;
2391 
2392     if (ioapics[idx].irqdomain) {
2393         irq_domain_remove(ioapics[idx].irqdomain);
2394         if (!cfg->dev)
2395             irq_domain_free_fwnode(fn);
2396         ioapics[idx].irqdomain = NULL;
2397     }
2398 }
2399 
2400 void __init setup_IO_APIC(void)
2401 {
2402     int ioapic;
2403 
2404     if (skip_ioapic_setup || !nr_ioapics)
2405         return;
2406 
2407     io_apic_irqs = nr_legacy_irqs() ? ~PIC_IRQS : ~0UL;
2408 
2409     apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
2410     for_each_ioapic(ioapic)
2411         BUG_ON(mp_irqdomain_create(ioapic));
2412 
2413     /*
2414          * Set up IO-APIC IRQ routing.
2415          */
2416     x86_init.mpparse.setup_ioapic_ids();
2417 
2418     sync_Arb_IDs();
2419     setup_IO_APIC_irqs();
2420     init_IO_APIC_traps();
2421     if (nr_legacy_irqs())
2422         check_timer();
2423 
2424     ioapic_initialized = 1;
2425 }
2426 
2427 static void resume_ioapic_id(int ioapic_idx)
2428 {
2429     unsigned long flags;
2430     union IO_APIC_reg_00 reg_00;
2431 
2432     raw_spin_lock_irqsave(&ioapic_lock, flags);
2433     reg_00.raw = io_apic_read(ioapic_idx, 0);
2434     if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx)) {
2435         reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
2436         io_apic_write(ioapic_idx, 0, reg_00.raw);
2437     }
2438     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2439 }
2440 
2441 static void ioapic_resume(void)
2442 {
2443     int ioapic_idx;
2444 
2445     for_each_ioapic_reverse(ioapic_idx)
2446         resume_ioapic_id(ioapic_idx);
2447 
2448     restore_ioapic_entries();
2449 }
2450 
2451 static struct syscore_ops ioapic_syscore_ops = {
2452     .suspend = save_ioapic_entries,
2453     .resume = ioapic_resume,
2454 };
2455 
2456 static int __init ioapic_init_ops(void)
2457 {
2458     register_syscore_ops(&ioapic_syscore_ops);
2459 
2460     return 0;
2461 }
2462 
2463 device_initcall(ioapic_init_ops);
2464 
2465 static int io_apic_get_redir_entries(int ioapic)
2466 {
2467     union IO_APIC_reg_01    reg_01;
2468     unsigned long flags;
2469 
2470     raw_spin_lock_irqsave(&ioapic_lock, flags);
2471     reg_01.raw = io_apic_read(ioapic, 1);
2472     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2473 
2474     /* The register returns the maximum index redir index
2475      * supported, which is one less than the total number of redir
2476      * entries.
2477      */
2478     return reg_01.bits.entries + 1;
2479 }
2480 
2481 unsigned int arch_dynirq_lower_bound(unsigned int from)
2482 {
2483     /*
2484      * dmar_alloc_hwirq() may be called before setup_IO_APIC(), so use
2485      * gsi_top if ioapic_dynirq_base hasn't been initialized yet.
2486      */
2487     if (!ioapic_initialized)
2488         return gsi_top;
2489     /*
2490      * For DT enabled machines ioapic_dynirq_base is irrelevant and not
2491      * updated. So simply return @from if ioapic_dynirq_base == 0.
2492      */
2493     return ioapic_dynirq_base ? : from;
2494 }
2495 
2496 #ifdef CONFIG_X86_32
2497 static int io_apic_get_unique_id(int ioapic, int apic_id)
2498 {
2499     union IO_APIC_reg_00 reg_00;
2500     static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2501     physid_mask_t tmp;
2502     unsigned long flags;
2503     int i = 0;
2504 
2505     /*
2506      * The P4 platform supports up to 256 APIC IDs on two separate APIC
2507      * buses (one for LAPICs, one for IOAPICs), where predecessors only
2508      * supports up to 16 on one shared APIC bus.
2509      *
2510      * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2511      *      advantage of new APIC bus architecture.
2512      */
2513 
2514     if (physids_empty(apic_id_map))
2515         apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map);
2516 
2517     raw_spin_lock_irqsave(&ioapic_lock, flags);
2518     reg_00.raw = io_apic_read(ioapic, 0);
2519     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2520 
2521     if (apic_id >= get_physical_broadcast()) {
2522         printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2523             "%d\n", ioapic, apic_id, reg_00.bits.ID);
2524         apic_id = reg_00.bits.ID;
2525     }
2526 
2527     /*
2528      * Every APIC in a system must have a unique ID or we get lots of nice
2529      * 'stuck on smp_invalidate_needed IPI wait' messages.
2530      */
2531     if (apic->check_apicid_used(&apic_id_map, apic_id)) {
2532 
2533         for (i = 0; i < get_physical_broadcast(); i++) {
2534             if (!apic->check_apicid_used(&apic_id_map, i))
2535                 break;
2536         }
2537 
2538         if (i == get_physical_broadcast())
2539             panic("Max apic_id exceeded!\n");
2540 
2541         printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2542             "trying %d\n", ioapic, apic_id, i);
2543 
2544         apic_id = i;
2545     }
2546 
2547     apic->apicid_to_cpu_present(apic_id, &tmp);
2548     physids_or(apic_id_map, apic_id_map, tmp);
2549 
2550     if (reg_00.bits.ID != apic_id) {
2551         reg_00.bits.ID = apic_id;
2552 
2553         raw_spin_lock_irqsave(&ioapic_lock, flags);
2554         io_apic_write(ioapic, 0, reg_00.raw);
2555         reg_00.raw = io_apic_read(ioapic, 0);
2556         raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2557 
2558         /* Sanity check */
2559         if (reg_00.bits.ID != apic_id) {
2560             pr_err("IOAPIC[%d]: Unable to change apic_id!\n",
2561                    ioapic);
2562             return -1;
2563         }
2564     }
2565 
2566     apic_printk(APIC_VERBOSE, KERN_INFO
2567             "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2568 
2569     return apic_id;
2570 }
2571 
2572 static u8 io_apic_unique_id(int idx, u8 id)
2573 {
2574     if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
2575         !APIC_XAPIC(boot_cpu_apic_version))
2576         return io_apic_get_unique_id(idx, id);
2577     else
2578         return id;
2579 }
2580 #else
2581 static u8 io_apic_unique_id(int idx, u8 id)
2582 {
2583     union IO_APIC_reg_00 reg_00;
2584     DECLARE_BITMAP(used, 256);
2585     unsigned long flags;
2586     u8 new_id;
2587     int i;
2588 
2589     bitmap_zero(used, 256);
2590     for_each_ioapic(i)
2591         __set_bit(mpc_ioapic_id(i), used);
2592 
2593     /* Hand out the requested id if available */
2594     if (!test_bit(id, used))
2595         return id;
2596 
2597     /*
2598      * Read the current id from the ioapic and keep it if
2599      * available.
2600      */
2601     raw_spin_lock_irqsave(&ioapic_lock, flags);
2602     reg_00.raw = io_apic_read(idx, 0);
2603     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2604     new_id = reg_00.bits.ID;
2605     if (!test_bit(new_id, used)) {
2606         apic_printk(APIC_VERBOSE, KERN_INFO
2607             "IOAPIC[%d]: Using reg apic_id %d instead of %d\n",
2608              idx, new_id, id);
2609         return new_id;
2610     }
2611 
2612     /*
2613      * Get the next free id and write it to the ioapic.
2614      */
2615     new_id = find_first_zero_bit(used, 256);
2616     reg_00.bits.ID = new_id;
2617     raw_spin_lock_irqsave(&ioapic_lock, flags);
2618     io_apic_write(idx, 0, reg_00.raw);
2619     reg_00.raw = io_apic_read(idx, 0);
2620     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2621     /* Sanity check */
2622     BUG_ON(reg_00.bits.ID != new_id);
2623 
2624     return new_id;
2625 }
2626 #endif
2627 
2628 static int io_apic_get_version(int ioapic)
2629 {
2630     union IO_APIC_reg_01    reg_01;
2631     unsigned long flags;
2632 
2633     raw_spin_lock_irqsave(&ioapic_lock, flags);
2634     reg_01.raw = io_apic_read(ioapic, 1);
2635     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2636 
2637     return reg_01.bits.version;
2638 }
2639 
2640 /*
2641  * This function updates target affinity of IOAPIC interrupts to include
2642  * the CPUs which came online during SMP bringup.
2643  */
2644 #define IOAPIC_RESOURCE_NAME_SIZE 11
2645 
2646 static struct resource *ioapic_resources;
2647 
2648 static struct resource * __init ioapic_setup_resources(void)
2649 {
2650     unsigned long n;
2651     struct resource *res;
2652     char *mem;
2653     int i;
2654 
2655     if (nr_ioapics == 0)
2656         return NULL;
2657 
2658     n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
2659     n *= nr_ioapics;
2660 
2661     mem = memblock_alloc(n, SMP_CACHE_BYTES);
2662     if (!mem)
2663         panic("%s: Failed to allocate %lu bytes\n", __func__, n);
2664     res = (void *)mem;
2665 
2666     mem += sizeof(struct resource) * nr_ioapics;
2667 
2668     for_each_ioapic(i) {
2669         res[i].name = mem;
2670         res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
2671         snprintf(mem, IOAPIC_RESOURCE_NAME_SIZE, "IOAPIC %u", i);
2672         mem += IOAPIC_RESOURCE_NAME_SIZE;
2673         ioapics[i].iomem_res = &res[i];
2674     }
2675 
2676     ioapic_resources = res;
2677 
2678     return res;
2679 }
2680 
2681 static void io_apic_set_fixmap(enum fixed_addresses idx, phys_addr_t phys)
2682 {
2683     pgprot_t flags = FIXMAP_PAGE_NOCACHE;
2684 
2685     /*
2686      * Ensure fixmaps for IOAPIC MMIO respect memory encryption pgprot
2687      * bits, just like normal ioremap():
2688      */
2689     flags = pgprot_decrypted(flags);
2690 
2691     __set_fixmap(idx, phys, flags);
2692 }
2693 
2694 void __init io_apic_init_mappings(void)
2695 {
2696     unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
2697     struct resource *ioapic_res;
2698     int i;
2699 
2700     ioapic_res = ioapic_setup_resources();
2701     for_each_ioapic(i) {
2702         if (smp_found_config) {
2703             ioapic_phys = mpc_ioapic_addr(i);
2704 #ifdef CONFIG_X86_32
2705             if (!ioapic_phys) {
2706                 printk(KERN_ERR
2707                        "WARNING: bogus zero IO-APIC "
2708                        "address found in MPTABLE, "
2709                        "disabling IO/APIC support!\n");
2710                 smp_found_config = 0;
2711                 skip_ioapic_setup = 1;
2712                 goto fake_ioapic_page;
2713             }
2714 #endif
2715         } else {
2716 #ifdef CONFIG_X86_32
2717 fake_ioapic_page:
2718 #endif
2719             ioapic_phys = (unsigned long)memblock_alloc(PAGE_SIZE,
2720                                     PAGE_SIZE);
2721             if (!ioapic_phys)
2722                 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
2723                       __func__, PAGE_SIZE, PAGE_SIZE);
2724             ioapic_phys = __pa(ioapic_phys);
2725         }
2726         io_apic_set_fixmap(idx, ioapic_phys);
2727         apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n",
2728             __fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK),
2729             ioapic_phys);
2730         idx++;
2731 
2732         ioapic_res->start = ioapic_phys;
2733         ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1;
2734         ioapic_res++;
2735     }
2736 }
2737 
2738 void __init ioapic_insert_resources(void)
2739 {
2740     int i;
2741     struct resource *r = ioapic_resources;
2742 
2743     if (!r) {
2744         if (nr_ioapics > 0)
2745             printk(KERN_ERR
2746                 "IO APIC resources couldn't be allocated.\n");
2747         return;
2748     }
2749 
2750     for_each_ioapic(i) {
2751         insert_resource(&iomem_resource, r);
2752         r++;
2753     }
2754 }
2755 
2756 int mp_find_ioapic(u32 gsi)
2757 {
2758     int i;
2759 
2760     if (nr_ioapics == 0)
2761         return -1;
2762 
2763     /* Find the IOAPIC that manages this GSI. */
2764     for_each_ioapic(i) {
2765         struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(i);
2766         if (gsi >= gsi_cfg->gsi_base && gsi <= gsi_cfg->gsi_end)
2767             return i;
2768     }
2769 
2770     printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
2771     return -1;
2772 }
2773 
2774 int mp_find_ioapic_pin(int ioapic, u32 gsi)
2775 {
2776     struct mp_ioapic_gsi *gsi_cfg;
2777 
2778     if (WARN_ON(ioapic < 0))
2779         return -1;
2780 
2781     gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2782     if (WARN_ON(gsi > gsi_cfg->gsi_end))
2783         return -1;
2784 
2785     return gsi - gsi_cfg->gsi_base;
2786 }
2787 
2788 static int bad_ioapic_register(int idx)
2789 {
2790     union IO_APIC_reg_00 reg_00;
2791     union IO_APIC_reg_01 reg_01;
2792     union IO_APIC_reg_02 reg_02;
2793 
2794     reg_00.raw = io_apic_read(idx, 0);
2795     reg_01.raw = io_apic_read(idx, 1);
2796     reg_02.raw = io_apic_read(idx, 2);
2797 
2798     if (reg_00.raw == -1 && reg_01.raw == -1 && reg_02.raw == -1) {
2799         pr_warn("I/O APIC 0x%x registers return all ones, skipping!\n",
2800             mpc_ioapic_addr(idx));
2801         return 1;
2802     }
2803 
2804     return 0;
2805 }
2806 
2807 static int find_free_ioapic_entry(void)
2808 {
2809     int idx;
2810 
2811     for (idx = 0; idx < MAX_IO_APICS; idx++)
2812         if (ioapics[idx].nr_registers == 0)
2813             return idx;
2814 
2815     return MAX_IO_APICS;
2816 }
2817 
2818 /**
2819  * mp_register_ioapic - Register an IOAPIC device
2820  * @id:     hardware IOAPIC ID
2821  * @address:    physical address of IOAPIC register area
2822  * @gsi_base:   base of GSI associated with the IOAPIC
2823  * @cfg:    configuration information for the IOAPIC
2824  */
2825 int mp_register_ioapic(int id, u32 address, u32 gsi_base,
2826                struct ioapic_domain_cfg *cfg)
2827 {
2828     bool hotplug = !!ioapic_initialized;
2829     struct mp_ioapic_gsi *gsi_cfg;
2830     int idx, ioapic, entries;
2831     u32 gsi_end;
2832 
2833     if (!address) {
2834         pr_warn("Bogus (zero) I/O APIC address found, skipping!\n");
2835         return -EINVAL;
2836     }
2837     for_each_ioapic(ioapic)
2838         if (ioapics[ioapic].mp_config.apicaddr == address) {
2839             pr_warn("address 0x%x conflicts with IOAPIC%d\n",
2840                 address, ioapic);
2841             return -EEXIST;
2842         }
2843 
2844     idx = find_free_ioapic_entry();
2845     if (idx >= MAX_IO_APICS) {
2846         pr_warn("Max # of I/O APICs (%d) exceeded (found %d), skipping\n",
2847             MAX_IO_APICS, idx);
2848         return -ENOSPC;
2849     }
2850 
2851     ioapics[idx].mp_config.type = MP_IOAPIC;
2852     ioapics[idx].mp_config.flags = MPC_APIC_USABLE;
2853     ioapics[idx].mp_config.apicaddr = address;
2854 
2855     io_apic_set_fixmap(FIX_IO_APIC_BASE_0 + idx, address);
2856     if (bad_ioapic_register(idx)) {
2857         clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2858         return -ENODEV;
2859     }
2860 
2861     ioapics[idx].mp_config.apicid = io_apic_unique_id(idx, id);
2862     ioapics[idx].mp_config.apicver = io_apic_get_version(idx);
2863 
2864     /*
2865      * Build basic GSI lookup table to facilitate gsi->io_apic lookups
2866      * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
2867      */
2868     entries = io_apic_get_redir_entries(idx);
2869     gsi_end = gsi_base + entries - 1;
2870     for_each_ioapic(ioapic) {
2871         gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2872         if ((gsi_base >= gsi_cfg->gsi_base &&
2873              gsi_base <= gsi_cfg->gsi_end) ||
2874             (gsi_end >= gsi_cfg->gsi_base &&
2875              gsi_end <= gsi_cfg->gsi_end)) {
2876             pr_warn("GSI range [%u-%u] for new IOAPIC conflicts with GSI[%u-%u]\n",
2877                 gsi_base, gsi_end,
2878                 gsi_cfg->gsi_base, gsi_cfg->gsi_end);
2879             clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2880             return -ENOSPC;
2881         }
2882     }
2883     gsi_cfg = mp_ioapic_gsi_routing(idx);
2884     gsi_cfg->gsi_base = gsi_base;
2885     gsi_cfg->gsi_end = gsi_end;
2886 
2887     ioapics[idx].irqdomain = NULL;
2888     ioapics[idx].irqdomain_cfg = *cfg;
2889 
2890     /*
2891      * If mp_register_ioapic() is called during early boot stage when
2892      * walking ACPI/DT tables, it's too early to create irqdomain,
2893      * we are still using bootmem allocator. So delay it to setup_IO_APIC().
2894      */
2895     if (hotplug) {
2896         if (mp_irqdomain_create(idx)) {
2897             clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2898             return -ENOMEM;
2899         }
2900         alloc_ioapic_saved_registers(idx);
2901     }
2902 
2903     if (gsi_cfg->gsi_end >= gsi_top)
2904         gsi_top = gsi_cfg->gsi_end + 1;
2905     if (nr_ioapics <= idx)
2906         nr_ioapics = idx + 1;
2907 
2908     /* Set nr_registers to mark entry present */
2909     ioapics[idx].nr_registers = entries;
2910 
2911     pr_info("IOAPIC[%d]: apic_id %d, version %d, address 0x%x, GSI %d-%d\n",
2912         idx, mpc_ioapic_id(idx),
2913         mpc_ioapic_ver(idx), mpc_ioapic_addr(idx),
2914         gsi_cfg->gsi_base, gsi_cfg->gsi_end);
2915 
2916     return 0;
2917 }
2918 
2919 int mp_unregister_ioapic(u32 gsi_base)
2920 {
2921     int ioapic, pin;
2922     int found = 0;
2923 
2924     for_each_ioapic(ioapic)
2925         if (ioapics[ioapic].gsi_config.gsi_base == gsi_base) {
2926             found = 1;
2927             break;
2928         }
2929     if (!found) {
2930         pr_warn("can't find IOAPIC for GSI %d\n", gsi_base);
2931         return -ENODEV;
2932     }
2933 
2934     for_each_pin(ioapic, pin) {
2935         u32 gsi = mp_pin_to_gsi(ioapic, pin);
2936         int irq = mp_map_gsi_to_irq(gsi, 0, NULL);
2937         struct mp_chip_data *data;
2938 
2939         if (irq >= 0) {
2940             data = irq_get_chip_data(irq);
2941             if (data && data->count) {
2942                 pr_warn("pin%d on IOAPIC%d is still in use.\n",
2943                     pin, ioapic);
2944                 return -EBUSY;
2945             }
2946         }
2947     }
2948 
2949     /* Mark entry not present */
2950     ioapics[ioapic].nr_registers  = 0;
2951     ioapic_destroy_irqdomain(ioapic);
2952     free_ioapic_saved_registers(ioapic);
2953     if (ioapics[ioapic].iomem_res)
2954         release_resource(ioapics[ioapic].iomem_res);
2955     clear_fixmap(FIX_IO_APIC_BASE_0 + ioapic);
2956     memset(&ioapics[ioapic], 0, sizeof(ioapics[ioapic]));
2957 
2958     return 0;
2959 }
2960 
2961 int mp_ioapic_registered(u32 gsi_base)
2962 {
2963     int ioapic;
2964 
2965     for_each_ioapic(ioapic)
2966         if (ioapics[ioapic].gsi_config.gsi_base == gsi_base)
2967             return 1;
2968 
2969     return 0;
2970 }
2971 
2972 static void mp_irqdomain_get_attr(u32 gsi, struct mp_chip_data *data,
2973                   struct irq_alloc_info *info)
2974 {
2975     if (info && info->ioapic.valid) {
2976         data->is_level = info->ioapic.is_level;
2977         data->active_low = info->ioapic.active_low;
2978     } else if (__acpi_get_override_irq(gsi, &data->is_level,
2979                        &data->active_low) < 0) {
2980         /* PCI interrupts are always active low level triggered. */
2981         data->is_level = true;
2982         data->active_low = true;
2983     }
2984 }
2985 
2986 /*
2987  * Configure the I/O-APIC specific fields in the routing entry.
2988  *
2989  * This is important to setup the I/O-APIC specific bits (is_level,
2990  * active_low, masked) because the underlying parent domain will only
2991  * provide the routing information and is oblivious of the I/O-APIC
2992  * specific bits.
2993  *
2994  * The entry is just preconfigured at this point and not written into the
2995  * RTE. This happens later during activation which will fill in the actual
2996  * routing information.
2997  */
2998 static void mp_preconfigure_entry(struct mp_chip_data *data)
2999 {
3000     struct IO_APIC_route_entry *entry = &data->entry;
3001 
3002     memset(entry, 0, sizeof(*entry));
3003     entry->is_level      = data->is_level;
3004     entry->active_low    = data->active_low;
3005     /*
3006      * Mask level triggered irqs. Edge triggered irqs are masked
3007      * by the irq core code in case they fire.
3008      */
3009     entry->masked       = data->is_level;
3010 }
3011 
3012 int mp_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
3013                unsigned int nr_irqs, void *arg)
3014 {
3015     struct irq_alloc_info *info = arg;
3016     struct mp_chip_data *data;
3017     struct irq_data *irq_data;
3018     int ret, ioapic, pin;
3019     unsigned long flags;
3020 
3021     if (!info || nr_irqs > 1)
3022         return -EINVAL;
3023     irq_data = irq_domain_get_irq_data(domain, virq);
3024     if (!irq_data)
3025         return -EINVAL;
3026 
3027     ioapic = mp_irqdomain_ioapic_idx(domain);
3028     pin = info->ioapic.pin;
3029     if (irq_find_mapping(domain, (irq_hw_number_t)pin) > 0)
3030         return -EEXIST;
3031 
3032     data = kzalloc(sizeof(*data), GFP_KERNEL);
3033     if (!data)
3034         return -ENOMEM;
3035 
3036     ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, info);
3037     if (ret < 0) {
3038         kfree(data);
3039         return ret;
3040     }
3041 
3042     INIT_LIST_HEAD(&data->irq_2_pin);
3043     irq_data->hwirq = info->ioapic.pin;
3044     irq_data->chip = (domain->parent == x86_vector_domain) ?
3045               &ioapic_chip : &ioapic_ir_chip;
3046     irq_data->chip_data = data;
3047     mp_irqdomain_get_attr(mp_pin_to_gsi(ioapic, pin), data, info);
3048 
3049     add_pin_to_irq_node(data, ioapic_alloc_attr_node(info), ioapic, pin);
3050 
3051     mp_preconfigure_entry(data);
3052     mp_register_handler(virq, data->is_level);
3053 
3054     local_irq_save(flags);
3055     if (virq < nr_legacy_irqs())
3056         legacy_pic->mask(virq);
3057     local_irq_restore(flags);
3058 
3059     apic_printk(APIC_VERBOSE, KERN_DEBUG
3060             "IOAPIC[%d]: Preconfigured routing entry (%d-%d -> IRQ %d Level:%i ActiveLow:%i)\n",
3061             ioapic, mpc_ioapic_id(ioapic), pin, virq,
3062             data->is_level, data->active_low);
3063     return 0;
3064 }
3065 
3066 void mp_irqdomain_free(struct irq_domain *domain, unsigned int virq,
3067                unsigned int nr_irqs)
3068 {
3069     struct irq_data *irq_data;
3070     struct mp_chip_data *data;
3071 
3072     BUG_ON(nr_irqs != 1);
3073     irq_data = irq_domain_get_irq_data(domain, virq);
3074     if (irq_data && irq_data->chip_data) {
3075         data = irq_data->chip_data;
3076         __remove_pin_from_irq(data, mp_irqdomain_ioapic_idx(domain),
3077                       (int)irq_data->hwirq);
3078         WARN_ON(!list_empty(&data->irq_2_pin));
3079         kfree(irq_data->chip_data);
3080     }
3081     irq_domain_free_irqs_top(domain, virq, nr_irqs);
3082 }
3083 
3084 int mp_irqdomain_activate(struct irq_domain *domain,
3085               struct irq_data *irq_data, bool reserve)
3086 {
3087     unsigned long flags;
3088 
3089     raw_spin_lock_irqsave(&ioapic_lock, flags);
3090     ioapic_configure_entry(irq_data);
3091     raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3092     return 0;
3093 }
3094 
3095 void mp_irqdomain_deactivate(struct irq_domain *domain,
3096                  struct irq_data *irq_data)
3097 {
3098     /* It won't be called for IRQ with multiple IOAPIC pins associated */
3099     ioapic_mask_entry(mp_irqdomain_ioapic_idx(domain),
3100               (int)irq_data->hwirq);
3101 }
3102 
3103 int mp_irqdomain_ioapic_idx(struct irq_domain *domain)
3104 {
3105     return (int)(long)domain->host_data;
3106 }
3107 
3108 const struct irq_domain_ops mp_ioapic_irqdomain_ops = {
3109     .alloc      = mp_irqdomain_alloc,
3110     .free       = mp_irqdomain_free,
3111     .activate   = mp_irqdomain_activate,
3112     .deactivate = mp_irqdomain_deactivate,
3113 };