Palacios Public Git Repository

To checkout Palacios execute

  git clone http://v3vee.org/palacios/palacios.web/palacios.git
This will give you the master branch. You probably want the devel branch or one of the release branches. To switch to the devel branch, simply execute
  cd palacios
  git checkout --track -b devel origin/devel
The other branches are similar.


disabled serial port debugging
[palacios.git] / palacios / src / devices / icc_bus.c
index f5de549..0a74eee 100644 (file)
 #include <palacios/vmm_sprintf.h>
 #include <palacios/vm_guest.h>
 #include <devices/icc_bus.h>
+#include <devices/apic_regs.h>
 
 #define MAX_APICS 256
 
+#ifndef CONFIG_DEBUG_ICC_BUS
+#undef PrintDebug
+#define PrintDebug(fmt, args...)
+#endif
+
+
+void v3_force_exit(void *p) {
+#ifdef CONFIG_DEBUG_ICC_BUS
+    struct guest_info *core=(struct guest_info *)p;
+#endif
+    PrintDebug("core %u: Forced to exit!\n",core->cpu_id);
+}
 
 struct ipi_thunk_data {
     struct vm_device * target;
     uint64_t val;
 };
 
-struct int_cmd_reg {
-    union {
-        uint64_t val;
-
-        struct {
-            uint32_t lo;
-            uint32_t hi;
-        } __attribute__((packed));
-
-        struct {
-            uint_t vec           : 8;
-            uint_t msg_type      : 3;
-            uint_t dst_mode      : 1;
-            uint_t del_status    : 1;
-            uint_t rsvd1         : 1;
-            uint_t lvl           : 1;
-            uint_t trig_mode     : 1;
-            uint_t rem_rd_status : 2;
-            uint_t dst_shorthand : 2;
-            uint64_t rsvd2       : 36;
-            uint32_t dst         : 8;
-        } __attribute__((packed));
-    } __attribute__((packed));
-} __attribute__((packed));
-
-
 
 
 struct apic_data {
@@ -69,6 +56,8 @@ struct apic_data {
 
 struct icc_bus_state {
     struct apic_data apics[MAX_APICS];
+    
+    uint32_t         ioapic_id;
 };
 
 static struct v3_device_ops dev_ops = {
@@ -78,60 +67,309 @@ static struct v3_device_ops dev_ops = {
     .stop = NULL,
 };
 
+#ifdef CONFIG_DEBUG_ICC_BUS
+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
+
+
+static int deliver(uint32_t src_apic, struct apic_data *dest_apic, struct int_cmd_reg *icr, struct icc_bus_state * state, uint32_t extirq) {
+
+    switch (icr->del_mode) {                                           
+
+       case 0:  //fixed
+       case 1: // lowest priority
+       case 7: // ExtInt
+           PrintDebug("icc_bus: delivering IRQ to core %u\n",dest_apic->core->cpu_id); 
+           dest_apic->ops->raise_intr(dest_apic->core, 
+                                      icr->del_mode!=7 ? icr->vec : extirq,
+                                      dest_apic->priv_data); 
+           if (src_apic!=state->ioapic_id && dest_apic->core->cpu_id != 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("icc_bus: non-local core, forcing it to exit\n"); 
+               V3_Call_On_CPU(dest_apic->core->cpu_id,v3_force_exit,(void*)(dest_apic->core));
+               // TODO: do what the print says
+           }                                                   
+           break;                                                      
+           
+       case 2:   //SMI                 
+           PrintError("icc_bus: SMI delivery is unsupported\n");       
+           return -1;                                          
+           break;                                                      
+           
+       case 3:  //reserved                                             
+           PrintError("icc_bus: Reserved delivery mode 3 is unsupported\n"); 
+           return -1;                                          
+           break;                                                      
+
+       case 4:  //NMI                                  
+           PrintError("icc_bus: NMI delivery is unsupported\n"); 
+           return -1;                                          
+           break;                                                      
+
+       case 5: { //INIT
+           struct guest_info *core = dest_apic->core;
+
+           PrintDebug("icc_bus: INIT delivery to core %u\n",core->cpu_id);
+
+           // TODO: any APIC reset on dest core (shouldn't be needed, but not sure...)
+
+           // Sanity check
+           if (core->cpu_mode != INIT) { 
+               PrintError("icc_bus: Warning: core %u is not in INIT state, ignored\n",core->cpu_id);
+               // Only a warning, since INIT INIT SIPI is common
+               break;
+           }
+
+           // We transition the target core to SIPI state
+           core->cpu_mode = 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("icc_bus: INIT delivery done\n");
+
+       }
+           break;                                                      
+
+       case 6: { //SIPI
+           struct guest_info *core = dest_apic->core;
+
+           // Sanity check
+           if (core->cpu_mode!=SIPI) { 
+               PrintError("icc_bus: core %u is not in SIPI state, ignored!\n",core->cpu_id);
+               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
+           //
+           core->rip = 0;
+           core->segments.cs.selector = icr->vec << 8;
+           core->segments.cs.limit = 0xffff;
+           core->segments.cs.base = icr->vec << 12;
+
+           PrintDebug("icc_bus: SIPI delivery (0x%x -> 0x%x:0x0) to core %u\n",
+                      icr->vec, core->segments.cs.selector, core->cpu_id);
+           // Maybe need to adjust the APIC?
+           
+           // We transition the target core to SIPI state
+           core->cpu_mode = REAL;  // note: locking should not be needed here
+
+           // As with INIT, we should not need to do anything else
+
+           PrintDebug("icc_bus: SIPI delivery done\n");
+
+       }
+           break;                                                      
+    }
 
+    return 0;
+} 
 
 
-int v3_icc_send_irq(struct vm_device * icc_bus, uint8_t apic_num, uint32_t irq_num) {
-    struct icc_bus_state * state = (struct icc_bus_state *)icc_bus->private_data;
-    struct apic_data * apic = &(state->apics[apic_num]);    
-
+//
+// icr_data contains interrupt vector *except* for ext_int
+// in which case it is given via irq
+//
 
-    struct int_cmd_reg icr;
-    icr.lo = irq_num;
+int v3_icc_send_ipi(struct vm_device * icc_bus, uint32_t src_apic, uint64_t icr_data, 
+                   uint32_t dfr_data, uint32_t extirq) {
 
+    PrintDebug("icc_bus: icc_bus=%p, src_apic=%u, icr_data=%llx, extirq=%u\n",icc_bus,src_apic,icr_data,extirq);
 
-    char * type = NULL;
-    char * dest = NULL;
-    char foo[8];
+    struct int_cmd_reg *icr = (struct int_cmd_reg *)&icr_data;
+    struct dst_fmt_reg *dfr = (struct dst_fmt_reg*)&dfr_data;
 
-    switch (icr.dst_shorthand) {
-       case 0x0:
-           sprintf(foo, "%d", icr.dst);
-           dest = foo;
-           break;
-       case 0x1:
-           dest = "(self)";
-           break;
-       case 0x2:
-           dest = "(broadcast inclusive)";
-           break;
-       case 0x3:
-           dest = "(broadcast)";
-           break;
+    struct icc_bus_state * state = (struct icc_bus_state *)icc_bus->private_data;
+    struct apic_data * dest_apic = NULL;
+    PrintDebug("icc_bus: icc_bus=%p, src_apic=%u, icr_data=%llx, extirq=%u\n", 
+              icc_bus, src_apic, icr_data, extirq);
+
+    // initial sanity checks
+    if ((src_apic >= MAX_APICS) || 
+       ((state->apics[src_apic].present == 0) && 
+        (src_apic != state->ioapic_id))) { 
+       PrintError("icc_bus: Apparently sending from unregistered apic id=%u\n",src_apic);
+       return -1;
     }
 
-    switch (icr.msg_type) {
-       case 0x0:
-           type = "";
-           break;
-       case 0x4:
-           type = "(NMI)";
-           break;
-       case 0x5:
-           type = "(INIT)";
-           break;
-       case 0x6:
-           type = "(Startup)";
-           break;
+
+    if ((icr->dst_mode == 0) && (state->apics[icr->dst].present == 0)) { 
+       PrintError("icc_bus: Attempted send to unregistered apic id=%u\n", icr->dst);
+       return -1;
     }
 
+    dest_apic =  &(state->apics[icr->dst]);
 
-    PrintDebug("Sending IPI of type %s and destination type %s from LAPIC %u to LAPIC %u.\n", 
-              type, dest, V3_Get_CPU(), apic_num);
 
-    apic->ops->raise_intr(apic->core, irq_num & 0xff, apic->priv_data);
+    PrintDebug("icc_bus: IPI %s %u from %s %u to %s %s %u (icr=0x%llx, dfr=0x%x) (extirq=%u)\n",
+              deliverymode_str[icr->del_mode], icr->vec, 
+              src_apic==state->ioapic_id ? "ioapic" : "apic",
+              src_apic,               
+              icr->dst_mode==0 ? "(physical)" : "(logical)", 
+              shorthand_str[icr->dst_shorthand], icr->dst,icr->val, dfr->val,
+              extirq);
 
-    //V3_Call_On_CPU(apic_num,  icc_force_exit, (void *)(uint64_t)(val & 0xff));
+    /*
+
+    if (icr->dst==state->ioapic_id) { 
+       PrintError("icc_bus: Attempted send to ioapic ignored\n");
+       return -1;
+    }
+    */
+
+
+
+    switch (icr->dst_shorthand) {
+
+       case 0:  // no shorthand
+
+           if (icr->dst_mode==0) { 
+               // physical delivery
+               struct apic_data * dest_apic =  &(state->apics[icr->dst]);
+               if (deliver(src_apic,dest_apic,icr,state,extirq)) { 
+                   return -1;
+               }
+           } else {
+               // logical delivery
+               uint8_t mda = icr->dst; // message destination address, not physical address
+               
+               if (dfr->model==0xf) { 
+                   // flat model
+                   // this means we deliver the IPI each destination APIC where
+                   // mda of sender & ldr of receiver is nonzero
+                   // mda=0xff means broadcast to all
+                   //
+                   int i;
+                   for (i=0;i<MAX_APICS;i++) { 
+                       struct apic_data *dest_apic=&(state->apics[i]);
+                       if (dest_apic->present &&
+                           dest_apic->ops->should_deliver_flat(dest_apic->core,
+                                                               mda,
+                                                               dest_apic->priv_data)) {
+                           if (deliver(src_apic,dest_apic,icr,state,extirq)) { 
+                               return -1;
+                           }
+                       }
+                   }
+               } else if (dfr->model==0x0) {
+                   // cluster model
+                   //
+                   // there are two variants of this
+                   //
+                   // 1. (ancient P5/P6) All apics are on one bus
+                   //    mda[31:28] is the target cluster, 
+                   //    mda[27:24] has one bit for each apic in the cluster
+                   //    mda[31:28] of sending apic == ldr[31:28] of dest apic means
+                   //      the dest apic is part of the cluster
+                   //      then mda[27:24] & ldr[27:24] nonzero means to deliver
+                   //    also, mda=0xff still means broadcast 
+                   //    So, basically, you have 15 clusters of 4 apics each + broadcast
+                   //
+                   // 2. (current) hierarchical cluster model
+                   //    This is some hwat unclearly documented in volume 3, 9-32
+                   //    basically, you have a hierarchy of clusters that where
+                   //    each cluster has 4 agents (APICs?) and a cluster manager.
+                   //    The cluster manager is not an apic, though, and outside of
+                   //    scope of documents.  Again, you have 15 clusters of 4 apics
+                   //    each + broadcast.   My impression is that this is identical 
+                   //    to variant 1 for our purposes. 
+                   //
+                   //
+                   // if we are in lowest priorty mode, we should just pick one
+                   // according to the arbitrarion prioty register
+                   int i;
+                   for (i=0;i<MAX_APICS;i++) { 
+                       struct apic_data *dest_apic=&(state->apics[i]);
+                       if (dest_apic->present &&
+                           dest_apic->ops->should_deliver_cluster(dest_apic->core,
+                                                                  mda,
+                                                                  dest_apic->priv_data)) {
+                           if (deliver(src_apic,dest_apic,icr,state,extirq)) { 
+                               return -1;
+                           }
+                       }
+                   }
+               } else {
+                   PrintError("icc_bus: unknown logical delivery model 0x%x\n", dfr->model);
+                   return -1;
+               }
+
+           }
+           
+           break;
+           
+       case 1:  // self
+
+           if (icr->dst_mode==0) { 
+               // physical delivery
+               if (icr->dst==state->ioapic_id) { 
+                   PrintError("icc_bus: ioapic attempting to send to itself\n");
+                   return -1;
+               }
+               struct apic_data *dest_apic=&(state->apics[src_apic]);
+               if (deliver(src_apic,dest_apic,icr,state,extirq)) { 
+                   return -1;
+               }
+           } else {
+               // logical delivery
+               PrintError("icc_bus: 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<MAX_APICS;i++) { 
+               struct apic_data *dest_apic=&(state->apics[i]);
+               if (dest_apic->present && (i!=src_apic || icr->dst_shorthand==2)) { 
+                   if (deliver(src_apic,dest_apic,icr,state,extirq)) { 
+                       return -1;
+                   }
+               }
+           }
+       }
+           break;
+
+       default:
+           return -1;
+    }
+    
 
     return 0;
 }
@@ -146,7 +384,7 @@ int v3_icc_register_apic(struct guest_info  * core, struct vm_device * icc_bus,
     struct apic_data * apic = &(icc->apics[apic_num]);
 
     if (apic->present == 1) {
-       PrintError("Attempt to re-register apic %u\n", apic_num);
+       PrintError("icc_bus: Attempt to re-register apic %u\n", apic_num);
        return -1;
     }
     
@@ -155,26 +393,42 @@ int v3_icc_register_apic(struct guest_info  * core, struct vm_device * icc_bus,
     apic->core = core;
     apic->ops = ops;
    
-    PrintDebug("Registered apic%u\n", apic_num);
+    PrintDebug("icc_bus: Registered apic %u\n", apic_num);
 
     return 0;
 }
 
 
+int v3_icc_register_ioapic(struct v3_vm_info *vm, struct vm_device * icc_bus, uint8_t apic_num)
+{
+    struct icc_bus_state * icc = (struct icc_bus_state *)icc_bus->private_data;
+
+    if (icc->ioapic_id) { 
+       PrintError("icc_bus: Attempt to register a second ioapic!\n");
+       return -1;
+    }
+
+    icc->ioapic_id=apic_num;
+
+    PrintDebug("icc_bus: Registered ioapic %u\n", apic_num);
+    
+
+    return 0;
+}
 
 
 static int icc_bus_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
-    PrintDebug("Creating ICC_BUS\n");
+    PrintDebug("icc_bus: Creating ICC_BUS\n");
 
-    char * name = v3_cfg_val(cfg, "name");
+    char * dev_id = v3_cfg_val(cfg, "ID");
 
     struct icc_bus_state * icc_bus = (struct icc_bus_state *)V3_Malloc(sizeof(struct icc_bus_state));
     memset(icc_bus, 0, sizeof(struct icc_bus_state));
 
-    struct vm_device * dev = v3_allocate_device(name, &dev_ops, icc_bus);
+    struct vm_device * dev = v3_allocate_device(dev_id, &dev_ops, icc_bus);
 
     if (v3_attach_device(vm, dev) == -1) {
-        PrintError("Could not attach device %s\n", name);
+        PrintError("icc_bus: Could not attach device %s\n", dev_id);
         return -1;
     }