X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fdevices%2Fapic.c;h=91c6824cd68d836abb7741398c3c871d47a16cf4;hb=eb5eb0f70b84003972a96060c76bd7efad340be4;hp=0ce5c7720d6e1a01522694ed53a3fc57b4e5cb37;hpb=f737ee43ce5cef6dbebf40940482811bfd3169b8;p=palacios.git diff --git a/palacios/src/devices/apic.c b/palacios/src/devices/apic.c index 0ce5c77..91c6824 100644 --- a/palacios/src/devices/apic.c +++ b/palacios/src/devices/apic.c @@ -22,13 +22,35 @@ #include #include #include +#include +#include +#include -#ifndef DEBUG_APIC +#ifndef CONFIG_DEBUG_APIC #undef PrintDebug #define PrintDebug(fmt, args...) #endif +#ifdef CONFIG_DEBUG_APIC +static char * shorthand_str[] = { + "(no shorthand)", + "(self)", + "(all)", + "(all-but-me)", +}; + +static char * deliverymode_str[] = { + "(fixed)", + "(lowest priority)", + "(SMI)", + "(reserved)", + "(NMI)", + "(INIT)", + "(Start Up)", + "(ExtInt)", +}; +#endif typedef enum { APIC_TMR_INT, APIC_THERM_INT, APIC_PERF_INT, APIC_LINT0_INT, APIC_LINT1_INT, APIC_ERR_INT } apic_irq_type_t; @@ -40,7 +62,7 @@ typedef enum { APIC_TMR_INT, APIC_THERM_INT, APIC_PERF_INT, #define APIC_EXTINT_DELIVERY 0x7 -#define BASE_ADDR_MSR 0x0000001B +#define BASE_ADDR_MSR 0x0000001B #define DEFAULT_BASE_ADDR 0xfee00000 #define APIC_ID_OFFSET 0x020 @@ -113,30 +135,33 @@ typedef enum { APIC_TMR_INT, APIC_THERM_INT, APIC_PERF_INT, #define EXT_INT_LOC_VEC_TBL_OFFSET2 0x520 // 0x500 - 0x530 #define EXT_INT_LOC_VEC_TBL_OFFSET3 0x530 // 0x500 - 0x530 - - struct apic_msr { union { - uint64_t val; + uint64_t value; struct { - uchar_t rsvd; - uint_t bootstrap_cpu : 1; - uint_t rsvd2 : 2; - uint_t apic_enable : 1; - ullong_t base_addr : 40; - uint_t rsvd3 : 12; + uint8_t rsvd; + uint8_t bootstrap_cpu : 1; + uint8_t rsvd2 : 2; + uint8_t apic_enable : 1; + uint64_t base_addr : 40; + uint32_t rsvd3 : 12; } __attribute__((packed)); } __attribute__((packed)); } __attribute__((packed)); +typedef enum {INIT_ST, + SIPI, + STARTED} ipi_state_t; + +struct apic_dev_state; struct apic_state { addr_t base_addr; /* MSRs */ - v3_msr_t base_addr_msr; + struct apic_msr base_addr_msr; /* memory map registers */ @@ -168,28 +193,59 @@ struct apic_state { uint32_t tmr_init_cnt; + struct local_vec_tbl_reg ext_intr_vec_tbl[4]; uint32_t rem_rd_data; - uchar_t int_req_reg[32]; - uchar_t int_svc_reg[32]; - uchar_t int_en_reg[32]; - uchar_t trig_mode_reg[32]; - - uint32_t eoi; + ipi_state_t ipi_state; + + uint8_t int_req_reg[32]; + uint8_t int_svc_reg[32]; + uint8_t int_en_reg[32]; + uint8_t trig_mode_reg[32]; + struct guest_info * core; + struct v3_timer * timer; + + uint32_t eoi; + + v3_lock_t lock; }; -static void init_apic_state(struct apic_state * apic) { + + +struct apic_dev_state { + int num_apics; + + struct apic_state apics[0]; +} __attribute__((packed)); + + + +static int apic_read(struct guest_info * core, addr_t guest_addr, void * dst, uint_t length, void * priv_data); +static int apic_write(struct guest_info * core, addr_t guest_addr, void * src, uint_t length, void * priv_data); + +static void init_apic_state(struct apic_state * apic, uint32_t id) { apic->base_addr = DEFAULT_BASE_ADDR; - apic->base_addr_msr.value = 0x0000000000000900LL; + + if (id == 0) { + // boot processor, enabled + apic->base_addr_msr.value = 0x0000000000000900LL; + } else { + // ap processor, enabled + apic->base_addr_msr.value = 0x0000000000000800LL; + } + + // same base address regardless of ap or main apic->base_addr_msr.value |= ((uint64_t)DEFAULT_BASE_ADDR); - PrintDebug("Sizeof Interrupt Request Register %d, should be 32\n", - (uint_t)sizeof(apic->int_req_reg)); + PrintDebug("apic %u: (init_apic_state): msr=0x%llx\n",id, apic->base_addr_msr.value); + + PrintDebug("apic %u: (init_apic_state): Sizeof Interrupt Request Register %d, should be 32\n", + id, (uint_t)sizeof(apic->int_req_reg)); memset(apic->int_req_reg, 0, sizeof(apic->int_req_reg)); memset(apic->int_svc_reg, 0, sizeof(apic->int_svc_reg)); @@ -201,9 +257,9 @@ static void init_apic_state(struct apic_state * apic) { apic->tmr_init_cnt = 0x00000000; apic->tmr_cur_cnt = 0x00000000; - // TODO: - // We need to figure out what the APIC ID is.... - apic->lapic_id.val = 0x00000000; + apic->lapic_id.val = id; + + apic->ipi_state = INIT_ST; // The P6 has 6 LVT entries, so we set the value to (6-1)... apic->apic_ver.val = 0x80050010; @@ -223,30 +279,61 @@ static void init_apic_state(struct apic_state * apic) { apic->lint1_vec_tbl.val = 0x00010000; apic->err_vec_tbl.val = 0x00010000; apic->tmr_div_cfg.val = 0x00000000; + //apic->ext_apic_feature.val = 0x00000007; apic->ext_apic_feature.val = 0x00040007; apic->ext_apic_ctrl.val = 0x00000000; apic->spec_eoi.val = 0x00000000; + + v3_lock_init(&(apic->lock)); } -static int read_apic_msr(uint_t msr, v3_msr_t * dst, void * priv_data) { - struct vm_device * dev = (struct vm_device *)priv_data; - struct apic_state * apic = (struct apic_state *)dev->private_data; - PrintError("READING APIC BASE ADDR: HI=%x LO=%x\n", apic->base_addr_msr.hi, apic->base_addr_msr.lo); +static int read_apic_msr(struct guest_info * core, uint_t msr, v3_msr_t * dst, void * priv_data) { + struct apic_dev_state * apic_dev = (struct apic_dev_state *)priv_data; + struct apic_state * apic = &(apic_dev->apics[core->cpu_id]); - return -1; + PrintDebug("apic %u: core %u: MSR read\n", apic->lapic_id.val, core->cpu_id); + v3_lock(apic->lock); + dst->value = apic->base_addr; + v3_unlock(apic->lock); + return 0; } -static int write_apic_msr(uint_t msr, v3_msr_t src, void * priv_data) { - // struct vm_device * dev = (struct vm_device *)priv_data; - // struct apic_state * apic = (struct apic_state *)dev->private_data; +static int write_apic_msr(struct guest_info * core, uint_t msr, v3_msr_t src, void * priv_data) { + struct apic_dev_state * apic_dev = (struct apic_dev_state *)priv_data; + struct apic_state * apic = &(apic_dev->apics[core->cpu_id]); + struct v3_mem_region * old_reg = v3_get_mem_region(core->vm_info, core->cpu_id, apic->base_addr); - PrintError("WRITING APIC BASE ADDR: HI=%x LO=%x\n", src.hi, src.lo); - return -1; + PrintDebug("apic %u: core %u: MSR write\n", apic->lapic_id.val, core->cpu_id); + + if (old_reg == NULL) { + // uh oh... + PrintError("apic %u: core %u: APIC Base address region does not exit...\n", + apic->lapic_id.val, core->cpu_id); + return -1; + } + + v3_lock(apic->lock); + + v3_delete_mem_region(core->vm_info, old_reg); + + apic->base_addr = src.value; + + if (v3_hook_full_mem(core->vm_info, core->cpu_id, apic->base_addr, + apic->base_addr + PAGE_SIZE_4KB, + apic_read, apic_write, apic_dev) == -1) { + PrintError("apic %u: core %u: Could not hook new APIC Base address\n", + apic->lapic_id.val, core->cpu_id); + v3_unlock(apic->lock); + return -1; + } + + v3_unlock(apic->lock); + return 0; } @@ -254,21 +341,30 @@ static int write_apic_msr(uint_t msr, v3_msr_t src, void * priv_data) { static int activate_apic_irq(struct apic_state * apic, uint32_t irq_num) { int major_offset = (irq_num & ~0x00000007) >> 3; int minor_offset = irq_num & 0x00000007; - uchar_t * req_location = apic->int_req_reg + major_offset; - uchar_t * en_location = apic->int_en_reg + major_offset; - uchar_t flag = 0x1 << minor_offset; + uint8_t * req_location = apic->int_req_reg + major_offset; + uint8_t * en_location = apic->int_en_reg + major_offset; + uint8_t flag = 0x1 << minor_offset; + + if (irq_num <= 15) { - PrintError("Attempting to raise an invalid interrupt: %d\n", irq_num); + PrintError("apic %u: core %d: Attempting to raise an invalid interrupt: %d\n", + apic->lapic_id.val, apic->core->cpu_id, irq_num); return -1; } - PrintDebug("Raising APIC IRQ %d\n", irq_num); + + PrintDebug("apic %u: core %d: Raising APIC IRQ %d\n", apic->lapic_id.val, apic->core->cpu_id, irq_num); + + if (*req_location & flag) { + PrintDebug("Interrupt %d coallescing\n", irq_num); + } if (*en_location & flag) { *req_location |= flag; } else { - PrintDebug("Interrupt not enabled... %.2x\n", *en_location); + PrintDebug("apic %u: core %d: Interrupt not enabled... %.2x\n", + apic->lapic_id.val, apic->core->cpu_id,*en_location); return 0; } @@ -277,30 +373,79 @@ static int activate_apic_irq(struct apic_state * apic, uint32_t irq_num) { -static int apic_do_eoi(struct apic_state * apic) { +static int get_highest_isr(struct apic_state * apic) { int i = 0, j = 0; // We iterate backwards to find the highest priority for (i = 31; i >= 0; i--) { - uchar_t * svc_major = apic->int_svc_reg + i; + uint8_t * svc_major = apic->int_svc_reg + i; if ((*svc_major) & 0xff) { for (j = 7; j >= 0; j--) { - uchar_t flag = 0x1 << j; + uint8_t flag = 0x1 << j; if ((*svc_major) & flag) { - *svc_major &= ~flag; + return ((i * 8) + j); + } + } + } + } -#ifdef CRAY_XT - if (((i * 8) + j) == 238) { - V3_ACK_IRQ(238); - } -#endif - return 0; + return -1; +} + + + +static int get_highest_irr(struct apic_state * apic) { + int i = 0, j = 0; + + // We iterate backwards to find the highest priority + for (i = 31; i >= 0; i--) { + uint8_t * req_major = apic->int_req_reg + i; + + if ((*req_major) & 0xff) { + for (j = 7; j >= 0; j--) { + uint8_t flag = 0x1 << j; + if ((*req_major) & flag) { + return ((i * 8) + j); } } } } + return -1; +} + + + + +static int apic_do_eoi(struct apic_state * apic) { + int isr_irq = get_highest_isr(apic); + + if (isr_irq != -1) { + int major_offset = (isr_irq & ~0x00000007) >> 3; + int minor_offset = isr_irq & 0x00000007; + uint8_t flag = 0x1 << minor_offset; + uint8_t * svc_location = apic->int_svc_reg + major_offset; + + PrintDebug("apic %u: core ?: Received APIC EOI for IRQ %d\n", apic->lapic_id.val,isr_irq); + + *svc_location &= ~flag; + +#ifdef CONFIG_CRAY_XT + + if ((isr_irq == 238) || + (isr_irq == 239)) { + PrintDebug("apic %u: core ?: Acking IRQ %d\n", apic->lapic_id.val,isr_irq); + } + + if (isr_irq == 238) { + V3_ACK_IRQ(238); + } +#endif + } else { + //PrintError("apic %u: core ?: Spurious EOI...\n",apic->lapic_id.val); + } + return 0; } @@ -343,13 +488,13 @@ static int activate_internal_irq(struct apic_state * apic, apic_irq_type_t int_t masked = apic->err_vec_tbl.mask; break; default: - PrintError("Invalid APIC interrupt type\n"); + PrintError("apic %u: core ?: Invalid APIC interrupt type\n", apic->lapic_id.val); return -1; } // interrupt is masked, don't send if (masked == 1) { - PrintDebug("Inerrupt is masked\n"); + PrintDebug("apic %u: core ?: Inerrupt is masked\n", apic->lapic_id.val); return 0; } @@ -357,25 +502,311 @@ static int activate_internal_irq(struct apic_state * apic, apic_irq_type_t int_t //PrintDebug("Activating internal APIC IRQ %d\n", vec_num); return activate_apic_irq(apic, vec_num); } else { - PrintError("Unhandled Delivery Mode\n"); + PrintError("apic %u: core ?: Unhandled Delivery Mode\n", apic->lapic_id.val); + return -1; + } +} + + + +static inline int should_deliver_cluster_ipi(struct guest_info * dst_core, + struct apic_state * dst_apic, uint8_t mda) { + + if ( ((mda & 0xf0) == (dst_apic->log_dst.dst_log_id & 0xf0)) && // (I am in the cluster and + ((mda & 0x0f) & (dst_apic->log_dst.dst_log_id & 0x0f)) ) { // I am in the set) + + PrintDebug("apic %u core %u: accepting clustered IRQ (mda 0x%x == log_dst 0x%x)\n", + dst_apic->lapic_id.val, dst_core->cpu_id, mda, + dst_apic->log_dst.dst_log_id); + + return 1; + } else { + PrintDebug("apic %u core %u: rejecting clustered IRQ (mda 0x%x != log_dst 0x%x)\n", + dst_apic->lapic_id.val, dst_core->cpu_id, mda, + dst_apic->log_dst.dst_log_id); + return 0; + } +} + +static inline int should_deliver_flat_ipi(struct guest_info * dst_core, + struct apic_state * dst_apic, uint8_t mda) { + + if (dst_apic->log_dst.dst_log_id & mda) { // I am in the set + + PrintDebug("apic %u core %u: accepting flat IRQ (mda 0x%x == log_dst 0x%x)\n", + dst_apic->lapic_id.val, dst_core->cpu_id, mda, + dst_apic->log_dst.dst_log_id); + return 1; + } else { + PrintDebug("apic %u core %u: rejecting flat IRQ (mda 0x%x != log_dst 0x%x)\n", + dst_apic->lapic_id.val, dst_core->cpu_id, mda, + dst_apic->log_dst.dst_log_id); + return 0; + } +} + + + +static int should_deliver_ipi(struct guest_info * dst_core, + struct apic_state * dst_apic, uint8_t mda) { + + + if (dst_apic->dst_fmt.model == 0xf) { + + if (mda == 0xff) { + // always deliver broadcast + return 1; + } + + return should_deliver_flat_ipi(dst_core, dst_apic, mda); + } else if (dst_apic->dst_fmt.model == 0x0) { + + if (mda == 0xff) { + // always deliver broadcast + return 1; + } + + return should_deliver_cluster_ipi(dst_core, dst_apic, mda); + } else { + PrintError("apic %u core %u: invalid destination format register value 0x%x for logical mode delivery.\n", + dst_apic->lapic_id.val, dst_core->cpu_id, dst_apic->dst_fmt.model); return -1; } } -static int apic_read(addr_t guest_addr, void * dst, uint_t length, void * priv_data) { - struct vm_device * dev = (struct vm_device *)priv_data; - struct apic_state * apic = (struct apic_state *)dev->private_data; +static int deliver_ipi(struct apic_state * src_apic, + struct apic_state * dst_apic, + uint32_t vector, uint8_t del_mode) { + + struct guest_info * dst_core = dst_apic->core; + + switch (del_mode) { + + case 0: //fixed + case 1: // lowest priority + PrintDebug("delivering IRQ %d to core %u\n", vector, dst_core->cpu_id); + + activate_apic_irq(dst_apic, vector); + + if (dst_apic != src_apic) { + // Assume core # is same as logical processor for now + // TODO FIX THIS FIX THIS + // THERE SHOULD BE: guestapicid->virtualapicid map, + // cpu_id->logical processor map + // host maitains logical proc->phsysical proc + PrintDebug(" non-local core, forcing it to exit\n"); + + v3_interrupt_cpu(dst_core->vm_info, dst_core->cpu_id, 0); + } + + break; + case 5: { //INIT + + PrintDebug(" INIT delivery to core %u\n", dst_core->cpu_id); + + // TODO: any APIC reset on dest core (shouldn't be needed, but not sure...) + + // Sanity check + if (dst_apic->ipi_state != INIT_ST) { + PrintError(" Warning: core %u is not in INIT state (mode = %d), ignored\n", + dst_core->cpu_id, dst_apic->ipi_state); + // Only a warning, since INIT INIT SIPI is common + break; + } + + // We transition the target core to SIPI state + dst_apic->ipi_state = SIPI; // note: locking should not be needed here + + // That should be it since the target core should be + // waiting in host on this transition + // either it's on another core or on a different preemptive thread + // in both cases, it will quickly notice this transition + // in particular, we should not need to force an exit here + + PrintDebug(" INIT delivery done\n"); + + break; + } + case 6: { //SIPI + + // Sanity check + if (dst_apic->ipi_state != SIPI) { + PrintError(" core %u is not in SIPI state (mode = %d), ignored!\n", + dst_core->cpu_id, dst_apic->ipi_state); + break; + } + + // Write the RIP, CS, and descriptor + // assume the rest is already good to go + // + // vector VV -> rip at 0 + // CS = VV00 + // This means we start executing at linear address VV000 + // + // So the selector needs to be VV00 + // and the base needs to be VV000 + // + dst_core->rip = 0; + dst_core->segments.cs.selector = vector << 8; + dst_core->segments.cs.limit = 0xffff; + dst_core->segments.cs.base = vector << 12; + + PrintDebug(" SIPI delivery (0x%x -> 0x%x:0x0) to core %u\n", + vector, dst_core->segments.cs.selector, dst_core->cpu_id); + // Maybe need to adjust the APIC? + + // We transition the target core to SIPI state + dst_core->core_run_state = CORE_RUNNING; // note: locking should not be needed here + dst_apic->ipi_state = STARTED; + + // As with INIT, we should not need to do anything else + + PrintDebug(" SIPI delivery done\n"); + + break; + } + case 2: // SMI + case 3: // reserved + case 4: // NMI + case 7: // ExtInt + default: + PrintError("IPI %d delivery is unsupported\n", del_mode); + return -1; + } + + return 0; + +} + + +static int route_ipi(struct apic_dev_state * apic_dev, + struct apic_state * src_apic, + struct int_cmd_reg * icr) { + struct apic_state * dest_apic = NULL; + + PrintDebug("route_ipi: src_apic=%p, icr_data=%p\n", + src_apic, (void *)(addr_t)icr->val); + + + if ((icr->dst_mode == 0) && (icr->dst >= apic_dev->num_apics)) { + PrintError("route_ipi: Attempted send to unregistered apic id=%u\n", + icr->dst); + return -1; + } + + dest_apic = &(apic_dev->apics[icr->dst]); + + + PrintDebug("route_ipi: IPI %s %u from apic %p to %s %s %u (icr=0x%llx)\n", + deliverymode_str[icr->del_mode], + icr->vec, + src_apic, + (icr->dst_mode == 0) ? "(physical)" : "(logical)", + shorthand_str[icr->dst_shorthand], + icr->dst, + icr->val); + + switch (icr->dst_shorthand) { + + case 0: // no shorthand + if (icr->dst_mode == 0) { + // physical delivery + + if (deliver_ipi(src_apic, dest_apic, + icr->vec, icr->del_mode) == -1) { + PrintError("Error: Could not deliver IPI\n"); + return -1; + } + + } else { + // logical delivery + int i; + uint8_t mda = icr->dst; + + for (i = 0; i < apic_dev->num_apics; i++) { + dest_apic = &(apic_dev->apics[i]); + int del_flag = should_deliver_ipi(dest_apic->core, dest_apic, mda); + + if (del_flag == -1) { + PrintError("Error checking delivery mode\n"); + return -1; + } else if (del_flag == 1) { + if (deliver_ipi(src_apic, dest_apic, + icr->vec, icr->del_mode) == -1) { + PrintError("Error: Could not deliver IPI\n"); + return -1; + } + } + } + } + + break; + + case 1: // self + + if (src_apic == NULL) { + PrintError("Sending IPI to self from generic IPI sender\n"); + break; + } + + if (icr->dst_mode == 0) { + if (deliver_ipi(src_apic, src_apic, icr->vec, icr->del_mode) == -1) { + PrintError("Could not deliver IPI\n"); + return -1; + } + } else { + // logical delivery + PrintError("use of logical delivery in self is not yet supported.\n"); + return -1; + } + break; + + case 2: + case 3: { // all and all-but-me + // assuming that logical verus physical doesn't matter + // although it is odd that both are used + int i; + + for (i = 0; i < apic_dev->num_apics; i++) { + dest_apic = &(apic_dev->apics[i]); + + if ((dest_apic != src_apic) || (icr->dst_shorthand == 2)) { + if (deliver_ipi(src_apic, dest_apic, icr->vec, icr->del_mode) == -1) { + PrintError("Error: Could not deliver IPI\n"); + return -1; + } + } + } + + break; + } + default: + PrintError("Error routing IPI, invalid Mode (%d)\n", icr->dst_shorthand); + return -1; + } + + + return 0; +} + + + +static int apic_read(struct guest_info * core, addr_t guest_addr, void * dst, uint_t length, void * priv_data) { + struct apic_dev_state * apic_dev = (struct apic_dev_state *)(priv_data); + struct apic_state * apic = &(apic_dev->apics[core->cpu_id]); addr_t reg_addr = guest_addr - apic->base_addr; struct apic_msr * msr = (struct apic_msr *)&(apic->base_addr_msr.value); uint32_t val = 0; - PrintDebug("Read apic address space (%p)\n", - (void *)guest_addr); + PrintDebug("apic %u: core %u: at %p: Read apic address space (%p)\n", + apic->lapic_id.val, core->cpu_id, apic, (void *)guest_addr); if (msr->apic_enable == 0) { - PrintError("Write to APIC address space with disabled APIC\n"); + PrintError("apic %u: core %u: Read from APIC address space with disabled APIC, apic msr=0x%llx\n", + apic->lapic_id.val, core->cpu_id, apic->base_addr_msr.value); + return -1; } @@ -569,15 +1000,26 @@ static int apic_read(addr_t guest_addr, void * dst, uint_t length, void * priv_d // Unhandled Registers case EXT_INT_LOC_VEC_TBL_OFFSET0: + val = apic->ext_intr_vec_tbl[0].val; + break; case EXT_INT_LOC_VEC_TBL_OFFSET1: + val = apic->ext_intr_vec_tbl[1].val; + break; case EXT_INT_LOC_VEC_TBL_OFFSET2: + val = apic->ext_intr_vec_tbl[2].val; + break; case EXT_INT_LOC_VEC_TBL_OFFSET3: + val = apic->ext_intr_vec_tbl[3].val; + break; + + case EXT_APIC_FEATURE_OFFSET: case EXT_APIC_CMD_OFFSET: case SEOI_OFFSET: default: - PrintError("Read from Unhandled APIC Register: %x\n", (uint32_t)reg_addr); + PrintError("apic %u: core %u: Read from Unhandled APIC Register: %x (getting zero)\n", + apic->lapic_id.val, core->cpu_id, (uint32_t)reg_addr); return -1; } @@ -599,34 +1041,44 @@ static int apic_read(addr_t guest_addr, void * dst, uint_t length, void * priv_d *val_ptr = val; } else { - PrintError("Invalid apic read length (%d)\n", length); + PrintError("apic %u: core %u: Invalid apic read length (%d)\n", + apic->lapic_id.val, core->cpu_id, length); return -1; } - PrintDebug("Read finished (val=%x)\n", *(uint32_t *)dst); + PrintDebug("apic %u: core %u: Read finished (val=%x)\n", + apic->lapic_id.val, core->cpu_id, *(uint32_t *)dst); return length; } -static int apic_write(addr_t guest_addr, void * src, uint_t length, void * priv_data) { - struct vm_device * dev = (struct vm_device *)priv_data; - struct apic_state * apic = (struct apic_state *)dev->private_data; +/** + * + */ +static int apic_write(struct guest_info * core, addr_t guest_addr, void * src, uint_t length, void * priv_data) { + struct apic_dev_state * apic_dev = (struct apic_dev_state *)(priv_data); + struct apic_state * apic = &(apic_dev->apics[core->cpu_id]); addr_t reg_addr = guest_addr - apic->base_addr; struct apic_msr * msr = (struct apic_msr *)&(apic->base_addr_msr.value); uint32_t op_val = *(uint32_t *)src; - PrintDebug("Write to apic address space (%p) (val=%x)\n", - (void *)guest_addr, *(uint32_t *)src); + PrintDebug("apic %u: core %u: at %p and priv_data is at %p\n", + apic->lapic_id.val, core->cpu_id, apic, priv_data); + + PrintDebug("apic %u: core %u: write to address space (%p) (val=%x)\n", + apic->lapic_id.val, core->cpu_id, (void *)guest_addr, *(uint32_t *)src); if (msr->apic_enable == 0) { - PrintError("Write to APIC address space with disabled APIC\n"); + PrintError("apic %u: core %u: Write to APIC address space with disabled APIC, apic msr=0x%llx\n", + apic->lapic_id.val, core->cpu_id, apic->base_addr_msr.value); return -1; } if (length != 4) { - PrintError("Invalid apic write length (%d)\n", length); + PrintError("apic %u: core %u: Invalid apic write length (%d)\n", + apic->lapic_id.val, length, core->cpu_id); return -1; } @@ -660,22 +1112,26 @@ static int apic_write(addr_t guest_addr, void * src, uint_t length, void * priv_ case TRIG_OFFSET7: case PPR_OFFSET: case EXT_APIC_FEATURE_OFFSET: -#if 1 - PrintError("Attempting to write to read only register %p (ignored)\n", (void *)reg_addr); -#else - PrintError("Attempting to write to read only register %p (error)\n", (void *)reg_addr); - return -1; -#endif + + PrintError("apic %u: core %u: Attempting to write to read only register %p (error)\n", + apic->lapic_id.val, core->cpu_id, (void *)reg_addr); + // return -1; + break; // Data registers case APIC_ID_OFFSET: + PrintDebug("apic %u: core %u: my id is being changed to %u\n", + apic->lapic_id.val, core->cpu_id, op_val); + apic->lapic_id.val = op_val; break; case TPR_OFFSET: apic->task_prio.val = op_val; break; case LDR_OFFSET: + PrintDebug("apic %u: core %u: setting log_dst.val to 0x%x\n", + apic->lapic_id.val, core->cpu_id, op_val); apic->log_dst.val = op_val; break; case DFR_OFFSET: @@ -742,30 +1198,57 @@ static int apic_write(addr_t guest_addr, void * src, uint_t length, void * priv_ case IER_OFFSET7: *(uint32_t *)(apic->int_en_reg + 28) = op_val; break; - - // Action Registers - case INT_CMD_LO_OFFSET: - case INT_CMD_HI_OFFSET: - case EOI_OFFSET: - { - // do eoi - apic_do_eoi(apic); - break; - } - // Unhandled Registers case EXT_INT_LOC_VEC_TBL_OFFSET0: + apic->ext_intr_vec_tbl[0].val = op_val; + break; case EXT_INT_LOC_VEC_TBL_OFFSET1: + apic->ext_intr_vec_tbl[1].val = op_val; + break; case EXT_INT_LOC_VEC_TBL_OFFSET2: + apic->ext_intr_vec_tbl[2].val = op_val; + break; case EXT_INT_LOC_VEC_TBL_OFFSET3: + apic->ext_intr_vec_tbl[3].val = op_val; + break; + + + // Action Registers + case EOI_OFFSET: + // do eoi + apic_do_eoi(apic); + break; + + case INT_CMD_LO_OFFSET: + apic->int_cmd.lo = op_val; + + PrintDebug("apic %u: core %u: sending cmd 0x%llx to apic %u\n", + apic->lapic_id.val, core->cpu_id, + apic->int_cmd.val, apic->int_cmd.dst); + + if (route_ipi(apic_dev, apic, &(apic->int_cmd)) == -1) { + PrintError("IPI Routing failure\n"); + return -1; + } + + break; + + case INT_CMD_HI_OFFSET: + apic->int_cmd.hi = op_val; + break; + + + // Unhandled Registers case EXT_APIC_CMD_OFFSET: case SEOI_OFFSET: default: - PrintError("Write to Unhandled APIC Register: %x\n", (uint32_t)reg_addr); + PrintError("apic %u: core %u: Write to Unhandled APIC Register: %x (ignored)\n", + apic->lapic_id.val, core->cpu_id, (uint32_t)reg_addr); + return -1; } - PrintDebug("Write finished\n"); + PrintDebug("apic %u: core %u: Write finished\n", apic->lapic_id.val, core->cpu_id); return length; } @@ -774,90 +1257,109 @@ static int apic_write(addr_t guest_addr, void * src, uint_t length, void * priv_ /* Interrupt Controller Functions */ -static int apic_intr_pending(void * private_data) { - struct vm_device * dev = (struct vm_device *)private_data; - struct apic_state * apic = (struct apic_state *)dev->private_data; - int i = 0; +// returns 1 if an interrupt is pending, 0 otherwise +static int apic_intr_pending(struct guest_info * core, void * private_data) { + struct apic_dev_state * apic_dev = (struct apic_dev_state *)(private_data); + struct apic_state * apic = &(apic_dev->apics[core->cpu_id]); + int req_irq = get_highest_irr(apic); + int svc_irq = get_highest_isr(apic); + // PrintDebug("apic %u: core %u: req_irq=%d, svc_irq=%d\n",apic->lapic_id.val,info->cpu_id,req_irq,svc_irq); - // just scan the request register looking for any set bit - // we should probably just do this with uint64 casts - for (i = 0; i < 32; i++) { - if (apic->int_req_reg[i] & 0xff) { - return 1; - } + if ((req_irq >= 0) && + (req_irq > svc_irq)) { + return 1; } + return 0; } -static int apic_get_intr_number(void * private_data) { - struct vm_device * dev = (struct vm_device *)private_data; - struct apic_state * apic = (struct apic_state *)dev->private_data; - int i = 0, j = 0; - +static int apic_get_intr_number(struct guest_info * core, void * private_data) { + struct apic_dev_state * apic_dev = (struct apic_dev_state *)(private_data); + struct apic_state * apic = &(apic_dev->apics[core->cpu_id]); + int req_irq = get_highest_irr(apic); + int svc_irq = get_highest_isr(apic); - // We iterate backwards to find the highest priority - for (i = 31; i >= 0; i--) { - uchar_t req_major = apic->int_req_reg[i]; - - if (req_major & 0xff) { - for (j = 7; j >= 0; j--) { - if ((req_major >> j) == 0x1) { - return (i * 8) + j; - } - } - } + if (svc_irq == -1) { + return req_irq; + } else if (svc_irq < req_irq) { + return req_irq; } return -1; } -static int apic_raise_intr(void * private_data, int irq) { -#ifdef CRAY_XT - // The Seastar is connected directly to the LAPIC via LINT0 on the ICC bus - if (irq == 238) { - struct vm_device * dev = (struct vm_device *)private_data; - struct apic_state * apic = (struct apic_state *)dev->private_data; +int v3_apic_send_ipi(struct v3_vm_info * vm, struct v3_gen_ipi * ipi, void * dev_data) { + struct apic_dev_state * apic_dev = (struct apic_dev_state *)dev_data; + struct int_cmd_reg tmp_icr; - return activate_apic_irq(apic, irq); - } -#endif - return 0; -} + // zero out all the fields + tmp_icr.val = 0; -static int apic_lower_intr(void * private_data, int irq) { - return 0; + + tmp_icr.vec = ipi->vector; + tmp_icr.del_mode = ipi->mode; + tmp_icr.dst_mode = ipi->logical; + tmp_icr.trig_mode = ipi->trigger_mode; + tmp_icr.dst_shorthand = ipi->dst_shorthand; + tmp_icr.dst = ipi->dst; + + + return route_ipi(apic_dev, NULL, &tmp_icr); } -static int apic_begin_irq(void * private_data, int irq) { - struct vm_device * dev = (struct vm_device *)private_data; - struct apic_state * apic = (struct apic_state *)dev->private_data; - int major_offset = (irq & ~0x00000007) >> 3; - int minor_offset = irq & 0x00000007; - uchar_t * req_location = apic->int_req_reg + major_offset; - uchar_t * svc_location = apic->int_svc_reg + major_offset; - uchar_t flag = 0x01 << minor_offset; - *svc_location |= flag; - *req_location &= ~flag; +int v3_apic_raise_intr(struct v3_vm_info * vm, uint32_t irq, uint32_t dst, void * dev_data) { + struct apic_dev_state * apic_dev = (struct apic_dev_state *)(dev_data); + struct apic_state * apic = &(apic_dev->apics[dst]); + + PrintDebug("apic %u core ?: raising interrupt IRQ %u (dst = %u).\n", apic->lapic_id.val, irq, dst); + + activate_apic_irq(apic, irq); + + if (V3_Get_CPU() != dst) { + v3_interrupt_cpu(vm, dst, 0); + } return 0; } -int v3_apic_raise_intr(struct vm_device * apic_dev, int intr_num) { - struct apic_state * apic = (struct apic_state *)apic_dev->private_data; - return activate_apic_irq(apic, intr_num); +static int apic_begin_irq(struct guest_info * core, void * private_data, int irq) { + struct apic_dev_state * apic_dev = (struct apic_dev_state *)(private_data); + struct apic_state * apic = &(apic_dev->apics[core->cpu_id]); + int major_offset = (irq & ~0x00000007) >> 3; + int minor_offset = irq & 0x00000007; + uint8_t * req_location = apic->int_req_reg + major_offset; + uint8_t * svc_location = apic->int_svc_reg + major_offset; + uint8_t flag = 0x01 << minor_offset; + + if (*req_location & flag) { + // we will only pay attention to a begin irq if we + // know that we initiated it! + *svc_location |= flag; + *req_location &= ~flag; + } else { + // do nothing... + //PrintDebug("apic %u: core %u: begin irq for %d ignored since I don't own it\n", + // apic->lapic_id.val, core->cpu_id, irq); + } + + return 0; } + /* Timer Functions */ -static void apic_update_time(ullong_t cpu_cycles, ullong_t cpu_freq, void * priv_data) { - struct vm_device * dev = (struct vm_device *)priv_data; - struct apic_state * apic = (struct apic_state *)dev->private_data; +static void apic_update_time(struct guest_info * core, + uint64_t cpu_cycles, uint64_t cpu_freq, + void * priv_data) { + struct apic_dev_state * apic_dev = (struct apic_dev_state *)(priv_data); + struct apic_state * apic = &(apic_dev->apics[core->cpu_id]); + // The 32 bit GCC runtime is a pile of shit #ifdef __V3_64BIT__ uint64_t tmr_ticks = 0; @@ -865,7 +1367,7 @@ static void apic_update_time(ullong_t cpu_cycles, ullong_t cpu_freq, void * priv uint32_t tmr_ticks = 0; #endif - uchar_t tmr_div = *(uchar_t *)&(apic->tmr_div_cfg.val); + uint8_t tmr_div = *(uint8_t *)&(apic->tmr_div_cfg.val); uint_t shift_num = 0; @@ -875,7 +1377,7 @@ static void apic_update_time(ullong_t cpu_cycles, ullong_t cpu_freq, void * priv if ((apic->tmr_init_cnt == 0) || ( (apic->tmr_vec_tbl.tmr_mode == APIC_TMR_ONESHOT) && (apic->tmr_cur_cnt == 0))) { - //PrintDebug("APIC timer not yet initialized\n"); + //PrintDebug("apic %u: core %u: APIC timer not yet initialized\n",apic->lapic_id.val,info->cpu_id); return; } @@ -906,12 +1408,13 @@ static void apic_update_time(ullong_t cpu_cycles, ullong_t cpu_freq, void * priv shift_num = 7; break; default: - PrintError("Invalid Timer Divider configuration\n"); + PrintError("apic %u: core %u: Invalid Timer Divider configuration\n", + apic->lapic_id.val, core->cpu_id); return; } tmr_ticks = cpu_cycles >> shift_num; - // PrintDebug("Timer Ticks: %p\n", (void *)tmr_ticks); + // PrintDebug("Timer Ticks: %p\n", (void *)tmr_ticks); if (tmr_ticks < apic->tmr_cur_cnt) { apic->tmr_cur_cnt -= tmr_ticks; @@ -920,10 +1423,19 @@ static void apic_update_time(ullong_t cpu_cycles, ullong_t cpu_freq, void * priv apic->tmr_cur_cnt = 0; // raise irq - PrintDebug("Raising APIC Timer interrupt (periodic=%d) (icnt=%d) (div=%d)\n", + PrintDebug("apic %u: core %u: Raising APIC Timer interrupt (periodic=%d) (icnt=%d) (div=%d)\n", + apic->lapic_id.val, core->cpu_id, apic->tmr_vec_tbl.tmr_mode, apic->tmr_init_cnt, shift_num); + + if (apic_intr_pending(core, priv_data)) { + PrintDebug("apic %u: core %u: Overriding pending IRQ %d\n", + apic->lapic_id.val, core->cpu_id, + apic_get_intr_number(core, priv_data)); + } + if (activate_internal_irq(apic, APIC_TMR_INT) == -1) { - PrintError("Could not raise Timer interrupt\n"); + PrintError("apic %u: core %u: Could not raise Timer interrupt\n", + apic->lapic_id.val, core->cpu_id); } if (apic->tmr_vec_tbl.tmr_mode == APIC_TMR_PERIODIC) { @@ -936,61 +1448,114 @@ static void apic_update_time(ullong_t cpu_cycles, ullong_t cpu_freq, void * priv } - static struct intr_ctrl_ops intr_ops = { .intr_pending = apic_intr_pending, .get_intr_number = apic_get_intr_number, - .raise_intr = apic_raise_intr, .begin_irq = apic_begin_irq, - .lower_intr = apic_lower_intr, }; -static struct vm_timer_ops timer_ops = { - .update_time = apic_update_time, +static struct v3_timer_ops timer_ops = { + .update_timer = apic_update_time, }; -static int apic_init(struct vm_device * dev) { - struct guest_info * info = dev->vm; - struct apic_state * apic = (struct apic_state *)(dev->private_data); - v3_register_intr_controller(dev->vm, &intr_ops, dev); - v3_add_timer(dev->vm, &timer_ops, dev); - init_apic_state(apic); +static int apic_free(struct vm_device * dev) { + struct apic_dev_state * apic_dev = (struct apic_dev_state *)dev->private_data; + int i = 0; - v3_hook_msr(info, BASE_ADDR_MSR, read_apic_msr, write_apic_msr, dev); + for (i = 0; i < dev->vm->num_cores; i++) { + struct apic_state * apic = &(apic_dev->apics[i]); + struct guest_info * core = &(dev->vm->cores[i]); + - v3_hook_full_mem(info, apic->base_addr, apic->base_addr + PAGE_SIZE_4KB, apic_read, apic_write, dev); + // unregister intr controller - return 0; -} + if (apic->timer) { + v3_remove_timer(core, apic->timer); + } -static int apic_deinit(struct vm_device * dev) { - struct guest_info * info = dev->vm; + // unhook memory - v3_unhook_msr(info, BASE_ADDR_MSR); + } + v3_unhook_msr(dev->vm, BASE_ADDR_MSR); + + V3_Free(apic_dev); return 0; } -static struct vm_device_ops dev_ops = { - .init = apic_init, - .deinit = apic_deinit, - .reset = NULL, - .start = NULL, - .stop = NULL, +static struct v3_device_ops dev_ops = { + .free = apic_free, }; -struct vm_device * v3_create_apic() { - PrintDebug("Creating APIC\n"); - struct apic_state * apic = (struct apic_state *)V3_Malloc(sizeof(struct apic_state)); - struct vm_device * device = v3_create_device("APIC", &dev_ops, apic); - - return device; + +static int apic_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) { + char * dev_id = v3_cfg_val(cfg, "ID"); + struct apic_dev_state * apic_dev = NULL; + int i = 0; + + PrintDebug("apic: creating an APIC for each core\n"); + + apic_dev = (struct apic_dev_state *)V3_Malloc(sizeof(struct apic_dev_state) + + sizeof(struct apic_state) * vm->num_cores); + + apic_dev->num_apics = vm->num_cores; + + struct vm_device * dev = v3_allocate_device(dev_id, &dev_ops, apic_dev); + + if (v3_attach_device(vm, dev) == -1) { + PrintError("apic: Could not attach device %s\n", dev_id); + V3_Free(apic_dev); + return -1; + } + + + for (i = 0; i < vm->num_cores; i++) { + struct apic_state * apic = &(apic_dev->apics[i]); + struct guest_info * core = &(vm->cores[i]); + + apic->core = core; + + init_apic_state(apic, i); + + v3_register_intr_controller(core, &intr_ops, apic_dev); + + apic->timer = v3_add_timer(core, &timer_ops, apic_dev); + + if (apic->timer == NULL) { + PrintError("APIC: Failed to attach timer to core %d\n", i); + v3_detach_device(dev); + return -1; + } + + v3_hook_full_mem(vm, core->cpu_id, apic->base_addr, apic->base_addr + PAGE_SIZE_4KB, apic_read, apic_write, apic_dev); + + PrintDebug("apic %u: (setup device): done, my id is %u\n", i, apic->lapic_id.val); + } + +#ifdef CONFIG_DEBUG_APIC + for (i = 0; i < vm->num_cores; i++) { + struct apic_state * apic = &(apic_dev->apics[i]); + PrintDebug("apic: sanity check: apic %u (at %p) has id %u and msr value %llx\n", + i, apic, apic->lapic_id.val, apic->base_addr_msr.value); + } +#endif + + + PrintDebug("apic: priv_data is at %p\n", apic_dev); + + v3_hook_msr(vm, BASE_ADDR_MSR, read_apic_msr, write_apic_msr, apic_dev); + + return 0; } + + + +device_register("LAPIC", apic_init)