Date: Thu, 1 Nov 2018 18:34:27 +0000 (UTC) From: John Baldwin <jhb@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-11@freebsd.org Subject: svn commit: r340016 - in stable/11: sys/sys sys/x86/acpica sys/x86/include sys/x86/iommu sys/x86/isa sys/x86/x86 sys/x86/xen sys/xen usr.bin/vmstat Message-ID: <201811011834.wA1IYRqc015449@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: jhb Date: Thu Nov 1 18:34:26 2018 New Revision: 340016 URL: https://svnweb.freebsd.org/changeset/base/340016 Log: MFC 338360,338415,338624,338630,338631,338725: Dynamic x86 IRQ layout. 338360: Dynamically allocate IRQ ranges on x86. Previously, x86 used static ranges of IRQ values for different types of I/O interrupts. Interrupt pins on I/O APICs and 8259A PICs used IRQ values from 0 to 254. MSI interrupts used a compile-time-defined range starting at 256, and Xen event channels used a compile-time-defined range after MSI. Some recent systems have more than 255 I/O APIC interrupt pins which resulted in those IRQ values overflowing into the MSI range triggering an assertion failure. Replace statically assigned ranges with dynamic ranges. Do a single pass computing the sizes of the IRQ ranges (PICs, MSI, Xen) to determine the total number of IRQs required. Allocate the interrupt source and interrupt count arrays dynamically once this pass has completed. To minimize runtime complexity these arrays are only sized once during bootup. The PIC range is determined by the PICs present in the system. The MSI and Xen ranges continue to use a fixed size, though this does make it possible to turn the MSI range size into a tunable in the future. As a result, various places are updated to use dynamic limits instead of constants. In addition, the vmstat(8) utility has been taught to understand that some kernels may treat 'intrcnt' and 'intrnames' as pointers rather than arrays when extracting interrupt stats from a crashdump. This is determined by the presence (vs absence) of a global 'nintrcnt' symbol. This change reverts r189404 which worked around a buggy BIOS which enumerated an I/O APIC twice (using the same memory mapped address for both entries but using an IRQ base of 256 for one entry and a valid IRQ base for the second entry). Making the "base" of MSI IRQ values dynamic avoids the panic that r189404 worked around, and there may now be valid I/O APICs with an IRQ base above 256 which this workaround would incorrectly skip. If in the future the issue reported in PR 130483 reoccurs, we will have to add a pass over the I/O APIC entries in the MADT to detect duplicates using the memory mapped address and use some strategy to choose the "correct" one. While here, reserve room in intrcnts for the Hyper-V counters. 338415: Fix build of x86 UP kernels after dynamic IRQ changes in r338360. 338624: msi: remove the check that interrupt sources have been added When running as a specific type of Xen guest the hypervisor won't provide any emulated IO-APICs or legacy PICs at all, thus hitting the following assert in the MSI code: panic: Assertion num_io_irqs > 0 failed at /usr/src/sys/x86/x86/msi.c:334 cpuid = 0 time = 1 KDB: stack backtrace: db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame 0xffffffff826ffa70 vpanic() at vpanic+0x1a3/frame 0xffffffff826ffad0 panic() at panic+0x43/frame 0xffffffff826ffb30 msi_init() at msi_init+0xed/frame 0xffffffff826ffb40 apic_setup_io() at apic_setup_io+0x72/frame 0xffffffff826ffb50 mi_startup() at mi_startup+0x118/frame 0xffffffff826ffb70 start_kernel() at start_kernel+0x10 Fix this by removing the assert in the MSI code, since it's possible to get to the MSI initialization without having registered any other interrupt sources. 338630: lapic: skip setting intrcnt if lapic is not present Instead of panicking. Legacy PVH mode doesn't provide a lapic, and since native_lapic_intrcnt is called unconditionally this would cause the assert to trigger. Change the assert into a continue in order to take into account the possibility of systems without a lapic. 338631: xen: legacy PVH fixes for the new interrupt count Register interrupts using the PIC pic_register_sources method instead of doing it in apic_setup_io. This is now required, since the internal interrupt structures are not yet setup when calling apic_setup_io. 338725: Fix a regression in r338360 when booting an x86 machine without APIC. The atpic_register_sources callback tries to avoid registering interrupt sources that would collide with an I/O APIC. However, the previous implementation was failing to register IRQs 8-15 since the slave PIC saw valid IRQs from the master and assumed an I/O APIC was present. To fix, go back to registering all 8259A interrupt sources in one loop when the master's register_sources method is invoked. PR: 229429, 130483, 231291 Modified: stable/11/sys/sys/interrupt.h stable/11/sys/x86/acpica/madt.c stable/11/sys/x86/include/apicvar.h stable/11/sys/x86/include/intr_machdep.h stable/11/sys/x86/iommu/intel_intrmap.c stable/11/sys/x86/isa/atpic.c stable/11/sys/x86/x86/intr_machdep.c stable/11/sys/x86/x86/io_apic.c stable/11/sys/x86/x86/local_apic.c stable/11/sys/x86/x86/msi.c stable/11/sys/x86/x86/nexus.c stable/11/sys/x86/xen/pvcpu_enum.c stable/11/sys/x86/xen/xen_intr.c stable/11/sys/x86/xen/xen_msi.c stable/11/sys/x86/xen/xen_nexus.c stable/11/sys/xen/xen_intr.h stable/11/usr.bin/vmstat/vmstat.c Directory Properties: stable/11/ (props changed) Modified: stable/11/sys/sys/interrupt.h ============================================================================== --- stable/11/sys/sys/interrupt.h Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/sys/interrupt.h Thu Nov 1 18:34:26 2018 (r340016) @@ -149,8 +149,13 @@ extern struct intr_event *clk_intr_event; extern void *vm_ih; /* Counts and names for statistics (defined in MD code). */ +#if defined(__amd64__) || defined(__i386__) +extern u_long *intrcnt; /* counts for for each device and stray */ +extern char *intrnames; /* string table containing device names */ +#else extern u_long intrcnt[]; /* counts for for each device and stray */ extern char intrnames[]; /* string table containing device names */ +#endif extern size_t sintrcnt; /* size of intrcnt table */ extern size_t sintrnames; /* size of intrnames table */ Modified: stable/11/sys/x86/acpica/madt.c ============================================================================== --- stable/11/sys/x86/acpica/madt.c Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/x86/acpica/madt.c Thu Nov 1 18:34:26 2018 (r340016) @@ -370,10 +370,6 @@ madt_parse_apics(ACPI_SUBTABLE_HEADER *entry, void *ar apic->Id); if (ioapics[apic->Id].io_apic != NULL) panic("%s: Double APIC ID %u", __func__, apic->Id); - if (apic->GlobalIrqBase >= FIRST_MSI_INT) { - printf("MADT: Ignoring bogus I/O APIC ID %u", apic->Id); - break; - } ioapics[apic->Id].io_apic = ioapic_create(apic->Address, apic->Id, apic->GlobalIrqBase); ioapics[apic->Id].io_vector = apic->GlobalIrqBase; Modified: stable/11/sys/x86/include/apicvar.h ============================================================================== --- stable/11/sys/x86/include/apicvar.h Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/x86/include/apicvar.h Thu Nov 1 18:34:26 2018 (r340016) @@ -152,10 +152,10 @@ #define APIC_BUS_PCI 2 #define APIC_BUS_MAX APIC_BUS_PCI -#define IRQ_EXTINT (NUM_IO_INTS + 1) -#define IRQ_NMI (NUM_IO_INTS + 2) -#define IRQ_SMI (NUM_IO_INTS + 3) -#define IRQ_DISABLED (NUM_IO_INTS + 4) +#define IRQ_EXTINT -1 +#define IRQ_NMI -2 +#define IRQ_SMI -3 +#define IRQ_DISABLED -4 /* * An APIC enumerator is a psuedo bus driver that enumerates APIC's including Modified: stable/11/sys/x86/include/intr_machdep.h ============================================================================== --- stable/11/sys/x86/include/intr_machdep.h Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/x86/include/intr_machdep.h Thu Nov 1 18:34:26 2018 (r340016) @@ -32,55 +32,41 @@ #ifdef _KERNEL /* - * The maximum number of I/O interrupts we allow. This number is rather - * arbitrary as it is just the maximum IRQ resource value. The interrupt - * source for a given IRQ maps that I/O interrupt to device interrupt - * source whether it be a pin on an interrupt controller or an MSI interrupt. - * The 16 ISA IRQs are assigned fixed IDT vectors, but all other device - * interrupts allocate IDT vectors on demand. Currently we have 191 IDT - * vectors available for device interrupts. On many systems with I/O APICs, - * a lot of the IRQs are not used, so this number can be much larger than - * 191 and still be safe since only interrupt sources in actual use will - * allocate IDT vectors. + * Values used in determining the allocation of IRQ values among + * different types of I/O interrupts. These values are used as + * indices into a interrupt source array to map I/O interrupts to a + * device interrupt source whether it be a pin on an interrupt + * controller or an MSI interrupt. The 16 ISA IRQs are assigned fixed + * IDT vectors, but all other device interrupts allocate IDT vectors + * on demand. Currently we have 191 IDT vectors available for device + * interrupts on each CPU. On many systems with I/O APICs, a lot of + * the IRQs are not used, so the total number of IRQ values reserved + * can exceed the number of available IDT slots. * - * The first 255 IRQs (0 - 254) are reserved for ISA IRQs and PCI intline IRQs. - * IRQ values from 256 to 767 are used by MSI. When running under the Xen - * Hypervisor, IRQ values from 768 to 4863 are available for binding to - * event channel events. We leave 255 unused to avoid confusion since 255 is - * used in PCI to indicate an invalid IRQ. + * The first 16 IRQs (0 - 15) are reserved for ISA IRQs. Interrupt + * pins on I/O APICs for non-ISA interrupts use IRQ values starting at + * IRQ 17. This layout matches the GSI numbering used by ACPI so that + * IRQ values returned by ACPI methods such as _CRS can be used + * directly by the ACPI bus driver. + * + * MSI interrupts allocate a block of interrupts starting at either + * the end of the I/O APIC range or 256, whichever is higher. When + * running under the Xen Hypervisor, an additional range of IRQ values + * are available for binding to event channel events. We use 256 as + * the minimum IRQ value for MSI interrupts to attempt to leave 255 + * unused since 255 is used in PCI to indicate an invalid INTx IRQ. */ #define NUM_MSI_INTS 512 -#define FIRST_MSI_INT 256 -#ifdef XENHVM -#include <xen/xen-os.h> -#include <xen/interface/event_channel.h> -#define NUM_EVTCHN_INTS NR_EVENT_CHANNELS -#define FIRST_EVTCHN_INT \ - (FIRST_MSI_INT + NUM_MSI_INTS) -#define LAST_EVTCHN_INT \ - (FIRST_EVTCHN_INT + NUM_EVTCHN_INTS - 1) -#else -#define NUM_EVTCHN_INTS 0 -#endif -#define NUM_IO_INTS (FIRST_MSI_INT + NUM_MSI_INTS + NUM_EVTCHN_INTS) +#define MINIMUM_MSI_INT 256 +extern u_int first_msi_irq; +extern u_int num_io_irqs; + /* * Default base address for MSI messages on x86 platforms. */ #define MSI_INTEL_ADDR_BASE 0xfee00000 -/* - * - 1 ??? dummy counter. - * - 2 counters for each I/O interrupt. - * - 1 counter for each CPU for lapic timer. - * - 8 counters for each CPU for IPI counters for SMP. - */ -#ifdef SMP -#define INTRCNT_COUNT (1 + NUM_IO_INTS * 2 + (1 + 8) * MAXCPU) -#else -#define INTRCNT_COUNT (1 + NUM_IO_INTS * 2 + 1) -#endif - #ifndef LOCORE typedef void inthand_t(void); @@ -95,6 +81,7 @@ struct intsrc; * return the vector associated with this source. */ struct pic { + void (*pic_register_sources)(struct pic *); void (*pic_enable_source)(struct intsrc *); void (*pic_disable_source)(struct intsrc *, int); void (*pic_eoi_source)(struct intsrc *); @@ -180,6 +167,9 @@ int msi_map(int irq, uint64_t *addr, uint32_t *data); int msi_release(int *irqs, int count); int msix_alloc(device_t dev, int *irq); int msix_release(int irq); +#ifdef XENHVM +void xen_intr_alloc_irqs(void); +#endif #endif /* !LOCORE */ #endif /* _KERNEL */ Modified: stable/11/sys/x86/iommu/intel_intrmap.c ============================================================================== --- stable/11/sys/x86/iommu/intel_intrmap.c Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/x86/iommu/intel_intrmap.c Thu Nov 1 18:34:26 2018 (r340016) @@ -337,7 +337,7 @@ dmar_init_irt(struct dmar_unit *unit) "QI disabled, disabling interrupt remapping\n"); return (0); } - unit->irte_cnt = clp2(NUM_IO_INTS); + unit->irte_cnt = clp2(num_io_irqs); unit->irt = (dmar_irte_t *)(uintptr_t)kmem_alloc_contig(kernel_arena, unit->irte_cnt * sizeof(dmar_irte_t), M_ZERO | M_WAITOK, 0, dmar_high, PAGE_SIZE, 0, DMAR_IS_COHERENT(unit) ? Modified: stable/11/sys/x86/isa/atpic.c ============================================================================== --- stable/11/sys/x86/isa/atpic.c Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/x86/isa/atpic.c Thu Nov 1 18:34:26 2018 (r340016) @@ -101,6 +101,7 @@ inthand_t #define ATPIC(io, base, eoi) { \ .at_pic = { \ + .pic_register_sources = atpic_register_sources, \ .pic_enable_source = atpic_enable_source, \ .pic_disable_source = atpic_disable_source, \ .pic_eoi_source = (eoi), \ @@ -139,6 +140,7 @@ struct atpic_intsrc { u_long at_straycount; }; +static void atpic_register_sources(struct pic *pic); static void atpic_enable_source(struct intsrc *isrc); static void atpic_disable_source(struct intsrc *isrc, int eoi); static void atpic_eoi_master(struct intsrc *isrc); @@ -209,6 +211,42 @@ _atpic_eoi_slave(struct intsrc *isrc) } static void +atpic_register_sources(struct pic *pic) +{ + struct atpic *ap = (struct atpic *)pic; + struct atpic_intsrc *ai; + int i; + + /* + * If any of the ISA IRQs have an interrupt source already, then + * assume that the I/O APICs are being used and don't register any + * of our interrupt sources. This makes sure we don't accidentally + * use mixed mode. The "accidental" use could otherwise occur on + * machines that route the ACPI SCI interrupt to a different ISA + * IRQ (at least one machine routes it to IRQ 13) thus disabling + * that APIC ISA routing and allowing the ATPIC source for that IRQ + * to leak through. We used to depend on this feature for routing + * IRQ0 via mixed mode, but now we don't use mixed mode at all. + * + * To avoid the slave not register sources after the master + * registers its sources, register all IRQs when this function is + * called on the master. + */ + if (ap != &atpics[MASTER]) + return; + for (i = 0; i < NUM_ISA_IRQS; i++) + if (intr_lookup_source(i) != NULL) + return; + + /* Loop through all interrupt sources and add them. */ + for (i = 0, ai = atintrs; i < NUM_ISA_IRQS; i++, ai++) { + if (i == ICU_SLAVEID) + continue; + intr_register_source(&ai->at_intsrc); + } +} + +static void atpic_enable_source(struct intsrc *isrc) { struct atpic_intsrc *ai = (struct atpic_intsrc *)isrc; @@ -517,8 +555,6 @@ atpic_startup(void) static void atpic_init(void *dummy __unused) { - struct atpic_intsrc *ai; - int i; /* * Register our PICs, even if we aren't going to use any of their @@ -528,27 +564,8 @@ atpic_init(void *dummy __unused) intr_register_pic(&atpics[1].at_pic) != 0) panic("Unable to register ATPICs"); - /* - * If any of the ISA IRQs have an interrupt source already, then - * assume that the APICs are being used and don't register any - * of our interrupt sources. This makes sure we don't accidentally - * use mixed mode. The "accidental" use could otherwise occur on - * machines that route the ACPI SCI interrupt to a different ISA - * IRQ (at least one machines routes it to IRQ 13) thus disabling - * that APIC ISA routing and allowing the ATPIC source for that IRQ - * to leak through. We used to depend on this feature for routing - * IRQ0 via mixed mode, but now we don't use mixed mode at all. - */ - for (i = 0; i < NUM_ISA_IRQS; i++) - if (intr_lookup_source(i) != NULL) - return; - - /* Loop through all interrupt sources and add them. */ - for (i = 0, ai = atintrs; i < NUM_ISA_IRQS; i++, ai++) { - if (i == ICU_SLAVEID) - continue; - intr_register_source(&ai->at_intsrc); - } + if (num_io_irqs == 0) + num_io_irqs = NUM_ISA_IRQS; } SYSINIT(atpic_init, SI_SUB_INTR, SI_ORDER_FOURTH, atpic_init, NULL); Modified: stable/11/sys/x86/x86/intr_machdep.c ============================================================================== --- stable/11/sys/x86/x86/intr_machdep.c Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/x86/x86/intr_machdep.c Thu Nov 1 18:34:26 2018 (r340016) @@ -36,6 +36,7 @@ #include "opt_atpic.h" #include "opt_ddb.h" +#include "opt_smp.h" #include <sys/param.h> #include <sys/bus.h> @@ -43,6 +44,7 @@ #include <sys/ktr.h> #include <sys/kernel.h> #include <sys/lock.h> +#include <sys/malloc.h> #include <sys/mutex.h> #include <sys/proc.h> #include <sys/smp.h> @@ -74,21 +76,25 @@ typedef void (*mask_fn)(void *); static int intrcnt_index; -static struct intsrc *interrupt_sources[NUM_IO_INTS]; +static struct intsrc **interrupt_sources; static struct sx intrsrc_lock; static struct mtx intrpic_lock; static struct mtx intrcnt_lock; static TAILQ_HEAD(pics_head, pic) pics; +u_int num_io_irqs; #if defined(SMP) && !defined(EARLY_AP_STARTUP) static int assign_cpu; #endif -u_long intrcnt[INTRCNT_COUNT]; -char intrnames[INTRCNT_COUNT * (MAXCOMLEN + 1)]; +u_long *intrcnt; +char *intrnames; size_t sintrcnt = sizeof(intrcnt); size_t sintrnames = sizeof(intrnames); +int nintrcnt; +static MALLOC_DEFINE(M_INTR, "intr", "Interrupt Sources"); + static int intr_assign_cpu(void *arg, int cpu); static void intr_disable_src(void *arg); static void intr_init(void *__dummy); @@ -97,6 +103,18 @@ static void intrcnt_setname(const char *name, int inde static void intrcnt_updatename(struct intsrc *is); static void intrcnt_register(struct intsrc *is); +/* + * SYSINIT levels for SI_SUB_INTR: + * + * SI_ORDER_FIRST: Initialize locks and pics TAILQ, xen_hvm_cpu_init + * SI_ORDER_SECOND: Xen PICs + * SI_ORDER_THIRD: Add I/O APIC PICs, alloc MSI and Xen IRQ ranges + * SI_ORDER_FOURTH: Add 8259A PICs + * SI_ORDER_FOURTH + 1: Finalize interrupt count and add interrupt sources + * SI_ORDER_MIDDLE: SMP interrupt counters + * SI_ORDER_ANY: Enable interrupts on BSP + */ + static int intr_pic_registered(struct pic *pic) { @@ -132,6 +150,56 @@ intr_register_pic(struct pic *pic) } /* + * Allocate interrupt source arrays and register interrupt sources + * once the number of interrupts is known. + */ +static void +intr_init_sources(void *arg) +{ + struct pic *pic; + + MPASS(num_io_irqs > 0); + + interrupt_sources = mallocarray(num_io_irqs, sizeof(*interrupt_sources), + M_INTR, M_WAITOK | M_ZERO); + + /* + * - 1 ??? dummy counter. + * - 2 counters for each I/O interrupt. + * - 1 counter for each CPU for lapic timer. + * - 1 counter for each CPU for the Hyper-V vmbus driver. + * - 8 counters for each CPU for IPI counters for SMP. + */ + nintrcnt = 1 + num_io_irqs * 2 + mp_ncpus * 2; +#ifdef COUNT_IPIS + if (mp_ncpus > 1) + nintrcnt += 8 * mp_ncpus; +#endif + intrcnt = mallocarray(nintrcnt, sizeof(u_long), M_INTR, M_WAITOK | + M_ZERO); + intrnames = mallocarray(nintrcnt, MAXCOMLEN + 1, M_INTR, M_WAITOK | + M_ZERO); + sintrcnt = nintrcnt * sizeof(u_long); + sintrnames = nintrcnt * (MAXCOMLEN + 1); + + intrcnt_setname("???", 0); + intrcnt_index = 1; + + /* + * NB: intrpic_lock is not held here to avoid LORs due to + * malloc() in intr_register_source(). However, we are still + * single-threaded at this point in startup so the list of + * PICs shouldn't change. + */ + TAILQ_FOREACH(pic, &pics, pics) { + if (pic->pic_register_sources != NULL) + pic->pic_register_sources(pic); + } +} +SYSINIT(intr_init_sources, SI_SUB_INTR, SI_ORDER_FOURTH + 1, intr_init_sources, + NULL); + +/* * Register a new interrupt source with the global interrupt system. * The global interrupts need to be disabled when this function is * called. @@ -143,6 +211,8 @@ intr_register_source(struct intsrc *isrc) KASSERT(intr_pic_registered(isrc->is_pic), ("unregistered PIC")); vector = isrc->is_pic->pic_vector(isrc); + KASSERT(vector < num_io_irqs, ("IRQ %d too large (%u irqs)", vector, + num_io_irqs)); if (interrupt_sources[vector] != NULL) return (EEXIST); error = intr_event_create(&isrc->is_event, isrc, 0, vector, @@ -168,7 +238,7 @@ struct intsrc * intr_lookup_source(int vector) { - if (vector < 0 || vector >= nitems(interrupt_sources)) + if (vector < 0 || vector >= num_io_irqs) return (NULL); return (interrupt_sources[vector]); } @@ -362,6 +432,7 @@ intrcnt_register(struct intsrc *is) KASSERT(is->is_event != NULL, ("%s: isrc with no event", __func__)); mtx_lock_spin(&intrcnt_lock); + MPASS(intrcnt_index + 2 <= nintrcnt); is->is_index = intrcnt_index; intrcnt_index += 2; snprintf(straystr, MAXCOMLEN + 1, "stray irq%d", @@ -378,6 +449,7 @@ intrcnt_add(const char *name, u_long **countp) { mtx_lock_spin(&intrcnt_lock); + MPASS(intrcnt_index < nintrcnt); *countp = &intrcnt[intrcnt_index]; intrcnt_setname(name, intrcnt_index); intrcnt_index++; @@ -388,8 +460,6 @@ static void intr_init(void *dummy __unused) { - intrcnt_setname("???", 0); - intrcnt_index = 1; TAILQ_INIT(&pics); mtx_init(&intrpic_lock, "intrpic", NULL, MTX_DEF); sx_init(&intrsrc_lock, "intrsrc"); @@ -455,10 +525,10 @@ void intr_reprogram(void) { struct intsrc *is; - int v; + u_int v; sx_xlock(&intrsrc_lock); - for (v = 0; v < NUM_IO_INTS; v++) { + for (v = 0; v < num_io_irqs; v++) { is = interrupt_sources[v]; if (is == NULL) continue; @@ -475,14 +545,15 @@ intr_reprogram(void) DB_SHOW_COMMAND(irqs, db_show_irqs) { struct intsrc **isrc; - int i, verbose; + u_int i; + int verbose; if (strcmp(modif, "v") == 0) verbose = 1; else verbose = 0; isrc = interrupt_sources; - for (i = 0; i < NUM_IO_INTS && !db_pager_quit; i++, isrc++) + for (i = 0; i < num_io_irqs && !db_pager_quit; i++, isrc++) if (*isrc != NULL) db_dump_intr_event((*isrc)->is_event, verbose); } @@ -565,7 +636,7 @@ static void intr_shuffle_irqs(void *arg __unused) { struct intsrc *isrc; - int i; + u_int i; /* Don't bother on UP. */ if (mp_ncpus == 1) @@ -574,7 +645,7 @@ intr_shuffle_irqs(void *arg __unused) /* Round-robin assign a CPU to each enabled source. */ sx_xlock(&intrsrc_lock); assign_cpu = 1; - for (i = 0; i < NUM_IO_INTS; i++) { + for (i = 0; i < num_io_irqs; i++) { isrc = interrupt_sources[i]; if (isrc != NULL && isrc->is_handlers > 0) { /* Modified: stable/11/sys/x86/x86/io_apic.c ============================================================================== --- stable/11/sys/x86/x86/io_apic.c Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/x86/x86/io_apic.c Thu Nov 1 18:34:26 2018 (r340016) @@ -78,7 +78,7 @@ static MALLOC_DEFINE(M_IOAPIC, "io_apic", "I/O APIC st struct ioapic_intsrc { struct intsrc io_intsrc; - u_int io_irq; + int io_irq; u_int io_intpin:8; u_int io_vector:8; u_int io_cpu; @@ -110,6 +110,7 @@ static u_int ioapic_read(volatile ioapic_t *apic, int static void ioapic_write(volatile ioapic_t *apic, int reg, u_int val); static const char *ioapic_bus_string(int bus_type); static void ioapic_print_irq(struct ioapic_intsrc *intpin); +static void ioapic_register_sources(struct pic *pic); static void ioapic_enable_source(struct intsrc *isrc); static void ioapic_disable_source(struct intsrc *isrc, int eoi); static void ioapic_eoi_source(struct intsrc *isrc); @@ -126,6 +127,7 @@ static void ioapic_reprogram_intpin(struct intsrc *isr static STAILQ_HEAD(,ioapic) ioapic_list = STAILQ_HEAD_INITIALIZER(ioapic_list); struct pic ioapic_template = { + .pic_register_sources = ioapic_register_sources, .pic_enable_source = ioapic_enable_source, .pic_disable_source = ioapic_disable_source, .pic_eoi_source = ioapic_eoi_source, @@ -140,7 +142,7 @@ struct pic ioapic_template = { .pic_reprogram_pin = ioapic_reprogram_intpin, }; -static int next_ioapic_base; +static u_int next_ioapic_base; static u_int next_id; static int enable_extint; @@ -248,7 +250,7 @@ ioapic_print_irq(struct ioapic_intsrc *intpin) printf("SMI"); break; default: - printf("%s IRQ %u", ioapic_bus_string(intpin->io_bus), + printf("%s IRQ %d", ioapic_bus_string(intpin->io_bus), intpin->io_irq); } } @@ -316,7 +318,7 @@ ioapic_program_intpin(struct ioapic_intsrc *intpin) * been enabled yet, just ensure that the pin is masked. */ mtx_assert(&icu_lock, MA_OWNED); - if (intpin->io_irq == IRQ_DISABLED || (intpin->io_irq < NUM_IO_INTS && + if (intpin->io_irq == IRQ_DISABLED || (intpin->io_irq >= 0 && intpin->io_vector == 0)) { low = ioapic_read(io->io_addr, IOAPIC_REDTBL_LO(intpin->io_intpin)); @@ -649,6 +651,8 @@ ioapic_create(vm_paddr_t addr, int32_t apic_id, int in io->io_id, intbase, next_ioapic_base); io->io_intbase = intbase; next_ioapic_base = intbase + numintr; + if (next_ioapic_base > num_io_irqs) + num_io_irqs = next_ioapic_base; io->io_numintr = numintr; io->io_addr = apic; io->io_paddr = addr; @@ -757,7 +761,7 @@ ioapic_remap_vector(void *cookie, u_int pin, int vecto io = (struct ioapic *)cookie; if (pin >= io->io_numintr || vector < 0) return (EINVAL); - if (io->io_pins[pin].io_irq >= NUM_IO_INTS) + if (io->io_pins[pin].io_irq < 0) return (EINVAL); io->io_pins[pin].io_irq = vector; if (bootverbose) @@ -776,7 +780,7 @@ ioapic_set_bus(void *cookie, u_int pin, int bus_type) io = (struct ioapic *)cookie; if (pin >= io->io_numintr) return (EINVAL); - if (io->io_pins[pin].io_irq >= NUM_IO_INTS) + if (io->io_pins[pin].io_irq < 0) return (EINVAL); if (io->io_pins[pin].io_bus == bus_type) return (0); @@ -797,7 +801,7 @@ ioapic_set_nmi(void *cookie, u_int pin) return (EINVAL); if (io->io_pins[pin].io_irq == IRQ_NMI) return (0); - if (io->io_pins[pin].io_irq >= NUM_IO_INTS) + if (io->io_pins[pin].io_irq < 0) return (EINVAL); io->io_pins[pin].io_bus = APIC_BUS_UNKNOWN; io->io_pins[pin].io_irq = IRQ_NMI; @@ -820,7 +824,7 @@ ioapic_set_smi(void *cookie, u_int pin) return (EINVAL); if (io->io_pins[pin].io_irq == IRQ_SMI) return (0); - if (io->io_pins[pin].io_irq >= NUM_IO_INTS) + if (io->io_pins[pin].io_irq < 0) return (EINVAL); io->io_pins[pin].io_bus = APIC_BUS_UNKNOWN; io->io_pins[pin].io_irq = IRQ_SMI; @@ -843,7 +847,7 @@ ioapic_set_extint(void *cookie, u_int pin) return (EINVAL); if (io->io_pins[pin].io_irq == IRQ_EXTINT) return (0); - if (io->io_pins[pin].io_irq >= NUM_IO_INTS) + if (io->io_pins[pin].io_irq < 0) return (EINVAL); io->io_pins[pin].io_bus = APIC_BUS_UNKNOWN; io->io_pins[pin].io_irq = IRQ_EXTINT; @@ -868,7 +872,7 @@ ioapic_set_polarity(void *cookie, u_int pin, enum intr io = (struct ioapic *)cookie; if (pin >= io->io_numintr || pol == INTR_POLARITY_CONFORM) return (EINVAL); - if (io->io_pins[pin].io_irq >= NUM_IO_INTS) + if (io->io_pins[pin].io_irq < 0) return (EINVAL); activehi = (pol == INTR_POLARITY_HIGH); if (io->io_pins[pin].io_activehi == activehi) @@ -889,7 +893,7 @@ ioapic_set_triggermode(void *cookie, u_int pin, enum i io = (struct ioapic *)cookie; if (pin >= io->io_numintr || trigger == INTR_TRIGGER_CONFORM) return (EINVAL); - if (io->io_pins[pin].io_irq >= NUM_IO_INTS) + if (io->io_pins[pin].io_irq < 0) return (EINVAL); edgetrigger = (trigger == INTR_TRIGGER_EDGE); if (io->io_pins[pin].io_edgetrigger == edgetrigger) @@ -925,12 +929,26 @@ ioapic_register(void *cookie) /* * Reprogram pins to handle special case pins (such as NMI and - * SMI) and register valid pins as interrupt sources. + * SMI) and disable normal pins until a handler is registered. */ intr_register_pic(&io->io_pic); - for (i = 0, pin = io->io_pins; i < io->io_numintr; i++, pin++) { + for (i = 0, pin = io->io_pins; i < io->io_numintr; i++, pin++) ioapic_reprogram_intpin(&pin->io_intsrc); - if (pin->io_irq < NUM_IO_INTS) +} + +/* + * Add interrupt sources for I/O APIC interrupt pins. + */ +static void +ioapic_register_sources(struct pic *pic) +{ + struct ioapic_intsrc *pin; + struct ioapic *io; + int i; + + io = (struct ioapic *)pic; + for (i = 0, pin = io->io_pins; i < io->io_numintr; i++, pin++) { + if (pin->io_irq >= 0) intr_register_source(&pin->io_intsrc); } } Modified: stable/11/sys/x86/x86/local_apic.c ============================================================================== --- stable/11/sys/x86/x86/local_apic.c Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/x86/x86/local_apic.c Thu Nov 1 18:34:26 2018 (r340016) @@ -89,11 +89,16 @@ CTASSERT(APIC_TIMER_INT < APIC_LOCAL_INTS); CTASSERT(APIC_LOCAL_INTS == 240); CTASSERT(IPI_STOP < APIC_SPURIOUS_INT); -/* Magic IRQ values for the timer and syscalls. */ -#define IRQ_TIMER (NUM_IO_INTS + 1) -#define IRQ_SYSCALL (NUM_IO_INTS + 2) -#define IRQ_DTRACE_RET (NUM_IO_INTS + 3) -#define IRQ_EVTCHN (NUM_IO_INTS + 4) +/* + * I/O interrupts use non-negative IRQ values. These values are used + * to mark unused IDT entries or IDT entries reserved for a non-I/O + * interrupt. + */ +#define IRQ_FREE -1 +#define IRQ_TIMER -2 +#define IRQ_SYSCALL -3 +#define IRQ_DTRACE_RET -4 +#define IRQ_EVTCHN -5 enum lat_timer_mode { LAT_MODE_UNDEF = 0, @@ -644,7 +649,7 @@ native_lapic_create(u_int apic_id, int boot_cpu) lapics[apic_id].la_elvts[i].lvt_active = 0; } for (i = 0; i <= APIC_NUM_IOINTS; i++) - lapics[apic_id].la_ioint_irqs[i] = -1; + lapics[apic_id].la_ioint_irqs[i] = IRQ_FREE; lapics[apic_id].la_ioint_irqs[IDT_SYSCALL - APIC_IO_INTS] = IRQ_SYSCALL; lapics[apic_id].la_ioint_irqs[APIC_TIMER_INT - APIC_IO_INTS] = IRQ_TIMER; @@ -747,7 +752,6 @@ native_lapic_setup(int boot) uint32_t version; uint32_t maxlvt; register_t saveintr; - char buf[MAXCOMLEN + 1]; int elvt_count; int i; @@ -776,15 +780,11 @@ native_lapic_setup(int boot) LAPIC_LVT_PCINT)); } - /* Program timer LVT and setup handler. */ + /* Program timer LVT. */ la->lvt_timer_base = lvt_mode(la, APIC_LVT_TIMER, lapic_read32(LAPIC_LVT_TIMER)); la->lvt_timer_last = la->lvt_timer_base; lapic_write32(LAPIC_LVT_TIMER, la->lvt_timer_base); - if (boot) { - snprintf(buf, sizeof(buf), "cpu%d:timer", PCPU_GET(cpuid)); - intrcnt_add(buf, &la->la_timer_count); - } /* Calibrate the timer parameters using BSP. */ if (boot && IS_BSP()) { @@ -839,6 +839,25 @@ native_lapic_setup(int boot) } static void +native_lapic_intrcnt(void *dummy __unused) +{ + struct pcpu *pc; + struct lapic *la; + char buf[MAXCOMLEN + 1]; + + STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) { + la = &lapics[pc->pc_apic_id]; + if (!la->la_present) + continue; + + snprintf(buf, sizeof(buf), "cpu%d:timer", pc->pc_cpuid); + intrcnt_add(buf, &la->la_timer_count); + } +} +SYSINIT(native_lapic_intrcnt, SI_SUB_INTR, SI_ORDER_MIDDLE, native_lapic_intrcnt, + NULL); + +static void native_lapic_reenable_pmc(void) { #ifdef HWPMC_HOOKS @@ -1488,7 +1507,7 @@ native_apic_alloc_vector(u_int apic_id, u_int irq) { u_int vector; - KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq)); + KASSERT(irq < num_io_irqs, ("Invalid IRQ %u", irq)); /* * Search for a free vector. Currently we just use a very simple @@ -1496,7 +1515,7 @@ native_apic_alloc_vector(u_int apic_id, u_int irq) */ mtx_lock_spin(&icu_lock); for (vector = 0; vector < APIC_NUM_IOINTS; vector++) { - if (lapics[apic_id].la_ioint_irqs[vector] != -1) + if (lapics[apic_id].la_ioint_irqs[vector] != IRQ_FREE) continue; lapics[apic_id].la_ioint_irqs[vector] = irq; mtx_unlock_spin(&icu_lock); @@ -1522,7 +1541,7 @@ native_apic_alloc_vectors(u_int apic_id, u_int *irqs, KASSERT(align >= count, ("align < count")); #ifdef INVARIANTS for (run = 0; run < count; run++) - KASSERT(irqs[run] < NUM_IO_INTS, ("Invalid IRQ %u at index %u", + KASSERT(irqs[run] < num_io_irqs, ("Invalid IRQ %u at index %u", irqs[run], run)); #endif @@ -1536,7 +1555,7 @@ native_apic_alloc_vectors(u_int apic_id, u_int *irqs, for (vector = 0; vector < APIC_NUM_IOINTS; vector++) { /* Vector is in use, end run. */ - if (lapics[apic_id].la_ioint_irqs[vector] != -1) { + if (lapics[apic_id].la_ioint_irqs[vector] != IRQ_FREE) { run = 0; first = 0; continue; @@ -1617,7 +1636,7 @@ native_apic_free_vector(u_int apic_id, u_int vector, u KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL && vector <= APIC_IO_INTS + APIC_NUM_IOINTS, ("Vector %u does not map to an IRQ line", vector)); - KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq)); + KASSERT(irq < num_io_irqs, ("Invalid IRQ %u", irq)); KASSERT(lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS] == irq, ("IRQ mismatch")); #ifdef KDTRACE_HOOKS @@ -1638,7 +1657,7 @@ native_apic_free_vector(u_int apic_id, u_int vector, u thread_unlock(td); } mtx_lock_spin(&icu_lock); - lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS] = -1; + lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS] = IRQ_FREE; mtx_unlock_spin(&icu_lock); if (!rebooting) { thread_lock(td); @@ -1689,7 +1708,7 @@ DB_SHOW_COMMAND(apic, db_show_apic) db_printf("Interrupts bound to lapic %u\n", apic_id); for (i = 0; i < APIC_NUM_IOINTS + 1 && !db_pager_quit; i++) { irq = lapics[apic_id].la_ioint_irqs[i]; - if (irq == -1 || irq == IRQ_SYSCALL) + if (irq == IRQ_FREE || irq == IRQ_SYSCALL) continue; #ifdef KDTRACE_HOOKS if (irq == IRQ_DTRACE_RET) @@ -1702,7 +1721,7 @@ DB_SHOW_COMMAND(apic, db_show_apic) db_printf("vec 0x%2x -> ", i + APIC_IO_INTS); if (irq == IRQ_TIMER) db_printf("lapic timer\n"); - else if (irq < NUM_IO_INTS) { + else if (irq < num_io_irqs) { isrc = intr_lookup_source(irq); if (isrc == NULL || verbose == 0) db_printf("IRQ %u\n", irq); @@ -1926,6 +1945,10 @@ apic_setup_io(void *dummy __unused) /* Enable the MSI "pic". */ init_ops.msi_init(); + +#ifdef XENHVM + xen_intr_alloc_irqs(); +#endif } SYSINIT(apic_setup_io, SI_SUB_INTR, SI_ORDER_THIRD, apic_setup_io, NULL); Modified: stable/11/sys/x86/x86/msi.c ============================================================================== --- stable/11/sys/x86/x86/msi.c Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/x86/x86/msi.c Thu Nov 1 18:34:26 2018 (r340016) @@ -118,7 +118,7 @@ struct msi_intsrc { u_int msi_cpu; /* Local APIC ID. (g) */ u_int msi_count:8; /* Messages in this group. (g) */ u_int msi_maxcount:8; /* Alignment for this group. (g) */ - int *msi_irqs; /* Group's IRQ list. (g) */ + u_int *msi_irqs; /* Group's IRQ list. (g) */ u_int msi_remap_cookie; }; @@ -149,6 +149,8 @@ struct pic msi_pic = { .pic_reprogram_pin = NULL, }; +u_int first_msi_irq; + #ifdef SMP /** * Xen hypervisors prior to 4.6.0 do not properly handle updates to @@ -166,7 +168,7 @@ SYSCTL_INT(_machdep, OID_AUTO, disable_msix_migration, #endif static int msi_enabled; -static int msi_last_irq; +static u_int msi_last_irq; static struct mtx msi_lock; static void @@ -327,6 +329,9 @@ msi_init(void) } #endif + first_msi_irq = max(MINIMUM_MSI_INT, num_io_irqs); + num_io_irqs = first_msi_irq + NUM_MSI_INTS; + msi_enabled = 1; intr_register_pic(&msi_pic); mtx_init(&msi_lock, "msi", NULL, MTX_DEF); @@ -343,7 +348,7 @@ msi_create_source(void) mtx_unlock(&msi_lock); return; } - irq = msi_last_irq + FIRST_MSI_INT; + irq = msi_last_irq + first_msi_irq; msi_last_irq++; mtx_unlock(&msi_lock); @@ -361,8 +366,8 @@ int msi_alloc(device_t dev, int count, int maxcount, int *irqs) { struct msi_intsrc *msi, *fsrc; - u_int cpu; - int cnt, i, *mirqs, vector; + u_int cpu, *mirqs; + int cnt, i, vector; #ifdef ACPI_DMAR u_int cookies[count]; int error; @@ -380,7 +385,7 @@ again: /* Try to find 'count' free IRQs. */ cnt = 0; - for (i = FIRST_MSI_INT; i < FIRST_MSI_INT + NUM_MSI_INTS; i++) { + for (i = first_msi_irq; i < first_msi_irq + NUM_MSI_INTS; i++) { msi = (struct msi_intsrc *)intr_lookup_source(i); /* End of allocated sources, so break. */ @@ -399,7 +404,7 @@ again: /* Do we need to create some new sources? */ if (cnt < count) { /* If we would exceed the max, give up. */ - if (i + (count - cnt) >= FIRST_MSI_INT + NUM_MSI_INTS) { + if (i + (count - cnt) >= first_msi_irq + NUM_MSI_INTS) { mtx_unlock(&msi_lock); free(mirqs, M_MSI); return (ENXIO); @@ -574,8 +579,8 @@ msi_map(int irq, uint64_t *addr, uint32_t *data) #ifdef ACPI_DMAR if (!msi->msi_msix) { - for (k = msi->msi_count - 1, i = FIRST_MSI_INT; k > 0 && - i < FIRST_MSI_INT + NUM_MSI_INTS; i++) { + for (k = msi->msi_count - 1, i = first_msi_irq; k > 0 && + i < first_msi_irq + NUM_MSI_INTS; i++) { if (i == msi->msi_irq) continue; msi1 = (struct msi_intsrc *)intr_lookup_source(i); @@ -622,7 +627,7 @@ again: mtx_lock(&msi_lock); /* Find a free IRQ. */ - for (i = FIRST_MSI_INT; i < FIRST_MSI_INT + NUM_MSI_INTS; i++) { + for (i = first_msi_irq; i < first_msi_irq + NUM_MSI_INTS; i++) { msi = (struct msi_intsrc *)intr_lookup_source(i); /* End of allocated sources, so break. */ @@ -637,7 +642,7 @@ again: /* Do we need to create a new source? */ if (msi == NULL) { /* If we would exceed the max, give up. */ - if (i + 1 >= FIRST_MSI_INT + NUM_MSI_INTS) { + if (i + 1 >= first_msi_irq + NUM_MSI_INTS) { mtx_unlock(&msi_lock); return (ENXIO); } Modified: stable/11/sys/x86/x86/nexus.c ============================================================================== --- stable/11/sys/x86/x86/nexus.c Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/x86/x86/nexus.c Thu Nov 1 18:34:26 2018 (r340016) @@ -226,7 +226,7 @@ nexus_init_resources(void) irq_rman.rm_start = 0; irq_rman.rm_type = RMAN_ARRAY; irq_rman.rm_descr = "Interrupt request lines"; - irq_rman.rm_end = NUM_IO_INTS - 1; + irq_rman.rm_end = num_io_irqs - 1; if (rman_init(&irq_rman)) panic("nexus_init_resources irq_rman"); @@ -234,7 +234,7 @@ nexus_init_resources(void) * We search for regions of existing IRQs and add those to the IRQ * resource manager. */ - for (irq = 0; irq < NUM_IO_INTS; irq++) + for (irq = 0; irq < num_io_irqs; irq++) if (intr_lookup_source(irq) != NULL) if (rman_manage_region(&irq_rman, irq, irq) != 0) panic("nexus_init_resources irq_rman add"); Modified: stable/11/sys/x86/xen/pvcpu_enum.c ============================================================================== --- stable/11/sys/x86/xen/pvcpu_enum.c Thu Nov 1 18:19:10 2018 (r340015) +++ stable/11/sys/x86/xen/pvcpu_enum.c Thu Nov 1 18:34:26 2018 (r340016) @@ -179,52 +179,65 @@ xenpv_setup_io(void) { if (xen_initial_domain()) { - int i, ret; + /* + * NB: we could iterate over the MADT IOAPIC entries in order + * to figure out the exact number of IOAPIC interrupts, but + * this is legacy code so just keep using the previous + * behaviour and assume a maximum of 256 interrupts. + */ + num_io_irqs = max(MINIMUM_MSI_INT - 1, num_io_irqs); - /* Map MADT */ - madt_physaddr = acpi_find_table(ACPI_SIG_MADT); - madt = acpi_map_table(madt_physaddr, ACPI_SIG_MADT); - madt_length = madt->Header.Length; + acpi_SetDefaultIntrModel(ACPI_INTR_APIC); + } + return (0); +} - /* Try to initialize ACPI so that we can access the FADT. */ - i = acpi_Startup(); - if (ACPI_FAILURE(i)) { - printf("MADT: ACPI Startup failed with %s\n", - AcpiFormatException(i)); - printf("Try disabling either ACPI or apic support.\n"); - panic("Using MADT but ACPI doesn't work"); - } +void +xenpv_register_pirqs(struct pic *pic __unused) +{ + unsigned int i; + int ret; - /* Run through the table to see if there are any overrides. */ - madt_walk_table(madt_parse_ints, NULL); + /* Map MADT */ + madt_physaddr = acpi_find_table(ACPI_SIG_MADT); + madt = acpi_map_table(madt_physaddr, ACPI_SIG_MADT); + madt_length = madt->Header.Length; - /* - * If there was not an explicit override entry for the SCI, - * force it to use level trigger and active-low polarity. - */ - if (!madt_found_sci_override) { - printf( - "MADT: Forcing active-low polarity and level trigger for SCI\n"); - ret = xen_register_pirq(AcpiGbl_FADT.SciInterrupt, - INTR_TRIGGER_LEVEL, INTR_POLARITY_LOW); - if (ret != 0) - panic("Unable to register SCI IRQ"); - } + /* Try to initialize ACPI so that we can access the FADT. */ + ret = acpi_Startup(); + if (ACPI_FAILURE(ret)) { + printf("MADT: ACPI Startup failed with %s\n", + AcpiFormatException(ret)); + printf("Try disabling either ACPI or apic support.\n"); + panic("Using MADT but ACPI doesn't work"); + } - /* Register legacy ISA IRQs */ - for (i = 1; i < 16; i++) { - if (intr_lookup_source(i) != NULL) - continue; - ret = xen_register_pirq(i, INTR_TRIGGER_EDGE, - INTR_POLARITY_LOW); - if (ret != 0 && bootverbose) - printf("Unable to register legacy IRQ#%d: %d\n", - i, ret); - } + /* Run through the table to see if there are any overrides. */ *** DIFF OUTPUT TRUNCATED AT 1000 LINES ***
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201811011834.wA1IYRqc015449>