KVM: ia64: Add kvm sal/pal virtulization support
authorXiantao Zhang <xiantao.zhang@intel.com>
Tue, 1 Apr 2008 06:59:30 +0000 (14:59 +0800)
committerAvi Kivity <avi@qumranet.com>
Sun, 27 Apr 2008 09:01:12 +0000 (12:01 +0300)
Some sal/pal calls would be traped to kvm for virtulization
from guest firmware.

Signed-off-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
arch/ia64/kvm/kvm_fw.c [new file with mode: 0644]

diff --git a/arch/ia64/kvm/kvm_fw.c b/arch/ia64/kvm/kvm_fw.c
new file mode 100644 (file)
index 0000000..091f936
--- /dev/null
@@ -0,0 +1,500 @@
+/*
+ * PAL/SAL call delegation
+ *
+ * Copyright (c) 2004 Li Susie <susie.li@intel.com>
+ * Copyright (c) 2005 Yu Ke <ke.yu@intel.com>
+ * Copyright (c) 2007 Xiantao Zhang <xiantao.zhang@intel.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+ * Place - Suite 330, Boston, MA 02111-1307 USA.
+ */
+
+#include <linux/kvm_host.h>
+#include <linux/smp.h>
+
+#include "vti.h"
+#include "misc.h"
+
+#include <asm/pal.h>
+#include <asm/sal.h>
+#include <asm/tlb.h>
+
+/*
+ * Handy macros to make sure that the PAL return values start out
+ * as something meaningful.
+ */
+#define INIT_PAL_STATUS_UNIMPLEMENTED(x)               \
+       {                                               \
+               x.status = PAL_STATUS_UNIMPLEMENTED;    \
+               x.v0 = 0;                               \
+               x.v1 = 0;                               \
+               x.v2 = 0;                               \
+       }
+
+#define INIT_PAL_STATUS_SUCCESS(x)                     \
+       {                                               \
+               x.status = PAL_STATUS_SUCCESS;          \
+               x.v0 = 0;                               \
+               x.v1 = 0;                               \
+               x.v2 = 0;                               \
+    }
+
+static void kvm_get_pal_call_data(struct kvm_vcpu *vcpu,
+               u64 *gr28, u64 *gr29, u64 *gr30, u64 *gr31) {
+       struct exit_ctl_data *p;
+
+       if (vcpu) {
+               p = &vcpu->arch.exit_data;
+               if (p->exit_reason == EXIT_REASON_PAL_CALL) {
+                       *gr28 = p->u.pal_data.gr28;
+                       *gr29 = p->u.pal_data.gr29;
+                       *gr30 = p->u.pal_data.gr30;
+                       *gr31 = p->u.pal_data.gr31;
+                       return ;
+               }
+       }
+       printk(KERN_DEBUG"Failed to get vcpu pal data!!!\n");
+}
+
+static void set_pal_result(struct kvm_vcpu *vcpu,
+               struct ia64_pal_retval result) {
+
+       struct exit_ctl_data *p;
+
+       p = kvm_get_exit_data(vcpu);
+       if (p && p->exit_reason == EXIT_REASON_PAL_CALL) {
+               p->u.pal_data.ret = result;
+               return ;
+       }
+       INIT_PAL_STATUS_UNIMPLEMENTED(p->u.pal_data.ret);
+}
+
+static void set_sal_result(struct kvm_vcpu *vcpu,
+               struct sal_ret_values result) {
+       struct exit_ctl_data *p;
+
+       p = kvm_get_exit_data(vcpu);
+       if (p && p->exit_reason == EXIT_REASON_SAL_CALL) {
+               p->u.sal_data.ret = result;
+               return ;
+       }
+       printk(KERN_WARNING"Failed to set sal result!!\n");
+}
+
+struct cache_flush_args {
+       u64 cache_type;
+       u64 operation;
+       u64 progress;
+       long status;
+};
+
+cpumask_t cpu_cache_coherent_map;
+
+static void remote_pal_cache_flush(void *data)
+{
+       struct cache_flush_args *args = data;
+       long status;
+       u64 progress = args->progress;
+
+       status = ia64_pal_cache_flush(args->cache_type, args->operation,
+                                       &progress, NULL);
+       if (status != 0)
+       args->status = status;
+}
+
+static struct ia64_pal_retval pal_cache_flush(struct kvm_vcpu *vcpu)
+{
+       u64 gr28, gr29, gr30, gr31;
+       struct ia64_pal_retval result = {0, 0, 0, 0};
+       struct cache_flush_args args = {0, 0, 0, 0};
+       long psr;
+
+       gr28 = gr29 = gr30 = gr31 = 0;
+       kvm_get_pal_call_data(vcpu, &gr28, &gr29, &gr30, &gr31);
+
+       if (gr31 != 0)
+               printk(KERN_ERR"vcpu:%p called cache_flush error!\n", vcpu);
+
+       /* Always call Host Pal in int=1 */
+       gr30 &= ~PAL_CACHE_FLUSH_CHK_INTRS;
+       args.cache_type = gr29;
+       args.operation = gr30;
+       smp_call_function(remote_pal_cache_flush,
+                               (void *)&args, 1, 1);
+       if (args.status != 0)
+               printk(KERN_ERR"pal_cache_flush error!,"
+                               "status:0x%lx\n", args.status);
+       /*
+        * Call Host PAL cache flush
+        * Clear psr.ic when call PAL_CACHE_FLUSH
+        */
+       local_irq_save(psr);
+       result.status = ia64_pal_cache_flush(gr29, gr30, &result.v1,
+                                               &result.v0);
+       local_irq_restore(psr);
+       if (result.status != 0)
+               printk(KERN_ERR"vcpu:%p crashed due to cache_flush err:%ld"
+                               "in1:%lx,in2:%lx\n",
+                               vcpu, result.status, gr29, gr30);
+
+#if 0
+       if (gr29 == PAL_CACHE_TYPE_COHERENT) {
+               cpus_setall(vcpu->arch.cache_coherent_map);
+               cpu_clear(vcpu->cpu, vcpu->arch.cache_coherent_map);
+               cpus_setall(cpu_cache_coherent_map);
+               cpu_clear(vcpu->cpu, cpu_cache_coherent_map);
+       }
+#endif
+       return result;
+}
+
+struct ia64_pal_retval pal_cache_summary(struct kvm_vcpu *vcpu)
+{
+
+       struct ia64_pal_retval result;
+
+       PAL_CALL(result, PAL_CACHE_SUMMARY, 0, 0, 0);
+       return result;
+}
+
+static struct ia64_pal_retval pal_freq_base(struct kvm_vcpu *vcpu)
+{
+
+       struct ia64_pal_retval result;
+
+       PAL_CALL(result, PAL_FREQ_BASE, 0, 0, 0);
+
+       /*
+        * PAL_FREQ_BASE may not be implemented in some platforms,
+        * call SAL instead.
+        */
+       if (result.v0 == 0) {
+               result.status = ia64_sal_freq_base(SAL_FREQ_BASE_PLATFORM,
+                                                       &result.v0,
+                                                       &result.v1);
+               result.v2 = 0;
+       }
+
+       return result;
+}
+
+static struct ia64_pal_retval pal_freq_ratios(struct kvm_vcpu *vcpu)
+{
+
+       struct ia64_pal_retval result;
+
+       PAL_CALL(result, PAL_FREQ_RATIOS, 0, 0, 0);
+       return result;
+}
+
+static struct ia64_pal_retval pal_logical_to_physica(struct kvm_vcpu *vcpu)
+{
+       struct ia64_pal_retval result;
+
+       INIT_PAL_STATUS_UNIMPLEMENTED(result);
+       return result;
+}
+
+static struct ia64_pal_retval pal_platform_addr(struct kvm_vcpu *vcpu)
+{
+
+       struct ia64_pal_retval result;
+
+       INIT_PAL_STATUS_SUCCESS(result);
+       return result;
+}
+
+static struct ia64_pal_retval pal_proc_get_features(struct kvm_vcpu *vcpu)
+{
+
+       struct ia64_pal_retval result = {0, 0, 0, 0};
+       long in0, in1, in2, in3;
+
+       kvm_get_pal_call_data(vcpu, &in0, &in1, &in2, &in3);
+       result.status = ia64_pal_proc_get_features(&result.v0, &result.v1,
+                       &result.v2, in2);
+
+       return result;
+}
+
+static struct ia64_pal_retval pal_cache_info(struct kvm_vcpu *vcpu)
+{
+
+       pal_cache_config_info_t ci;
+       long status;
+       unsigned long in0, in1, in2, in3, r9, r10;
+
+       kvm_get_pal_call_data(vcpu, &in0, &in1, &in2, &in3);
+       status = ia64_pal_cache_config_info(in1, in2, &ci);
+       r9 = ci.pcci_info_1.pcci1_data;
+       r10 = ci.pcci_info_2.pcci2_data;
+       return ((struct ia64_pal_retval){status, r9, r10, 0});
+}
+
+#define GUEST_IMPL_VA_MSB      59
+#define GUEST_RID_BITS         18
+
+static struct ia64_pal_retval pal_vm_summary(struct kvm_vcpu *vcpu)
+{
+
+       pal_vm_info_1_u_t vminfo1;
+       pal_vm_info_2_u_t vminfo2;
+       struct ia64_pal_retval result;
+
+       PAL_CALL(result, PAL_VM_SUMMARY, 0, 0, 0);
+       if (!result.status) {
+               vminfo1.pvi1_val = result.v0;
+               vminfo1.pal_vm_info_1_s.max_itr_entry = 8;
+               vminfo1.pal_vm_info_1_s.max_dtr_entry = 8;
+               result.v0 = vminfo1.pvi1_val;
+               vminfo2.pal_vm_info_2_s.impl_va_msb = GUEST_IMPL_VA_MSB;
+               vminfo2.pal_vm_info_2_s.rid_size = GUEST_RID_BITS;
+               result.v1 = vminfo2.pvi2_val;
+       }
+
+       return result;
+}
+
+static struct ia64_pal_retval pal_vm_info(struct kvm_vcpu *vcpu)
+{
+       struct ia64_pal_retval result;
+
+       INIT_PAL_STATUS_UNIMPLEMENTED(result);
+
+       return result;
+}
+
+static  u64 kvm_get_pal_call_index(struct kvm_vcpu *vcpu)
+{
+       u64 index = 0;
+       struct exit_ctl_data *p;
+
+       p = kvm_get_exit_data(vcpu);
+       if (p && (p->exit_reason == EXIT_REASON_PAL_CALL))
+               index = p->u.pal_data.gr28;
+
+       return index;
+}
+
+int kvm_pal_emul(struct kvm_vcpu *vcpu, struct kvm_run *run)
+{
+
+       u64 gr28;
+       struct ia64_pal_retval result;
+       int ret = 1;
+
+       gr28 = kvm_get_pal_call_index(vcpu);
+       /*printk("pal_call index:%lx\n",gr28);*/
+       switch (gr28) {
+       case PAL_CACHE_FLUSH:
+               result = pal_cache_flush(vcpu);
+               break;
+       case PAL_CACHE_SUMMARY:
+               result = pal_cache_summary(vcpu);
+               break;
+       case PAL_HALT_LIGHT:
+       {
+               vcpu->arch.timer_pending = 1;
+               INIT_PAL_STATUS_SUCCESS(result);
+               if (kvm_highest_pending_irq(vcpu) == -1)
+                       ret = kvm_emulate_halt(vcpu);
+
+       }
+               break;
+
+       case PAL_FREQ_RATIOS:
+               result = pal_freq_ratios(vcpu);
+               break;
+
+       case PAL_FREQ_BASE:
+               result = pal_freq_base(vcpu);
+               break;
+
+       case PAL_LOGICAL_TO_PHYSICAL :
+               result = pal_logical_to_physica(vcpu);
+               break;
+
+       case PAL_VM_SUMMARY :
+               result = pal_vm_summary(vcpu);
+               break;
+
+       case PAL_VM_INFO :
+               result = pal_vm_info(vcpu);
+               break;
+       case PAL_PLATFORM_ADDR :
+               result = pal_platform_addr(vcpu);
+               break;
+       case PAL_CACHE_INFO:
+               result = pal_cache_info(vcpu);
+               break;
+       case PAL_PTCE_INFO:
+               INIT_PAL_STATUS_SUCCESS(result);
+               result.v1 = (1L << 32) | 1L;
+               break;
+       case PAL_VM_PAGE_SIZE:
+               result.status = ia64_pal_vm_page_size(&result.v0,
+                                                       &result.v1);
+               break;
+       case PAL_RSE_INFO:
+               result.status = ia64_pal_rse_info(&result.v0,
+                                       (pal_hints_u_t *)&result.v1);
+               break;
+       case PAL_PROC_GET_FEATURES:
+               result = pal_proc_get_features(vcpu);
+               break;
+       case PAL_DEBUG_INFO:
+               result.status = ia64_pal_debug_info(&result.v0,
+                                                       &result.v1);
+               break;
+       case PAL_VERSION:
+               result.status = ia64_pal_version(
+                               (pal_version_u_t *)&result.v0,
+                               (pal_version_u_t *)&result.v1);
+
+               break;
+       case PAL_FIXED_ADDR:
+               result.status = PAL_STATUS_SUCCESS;
+               result.v0 = vcpu->vcpu_id;
+               break;
+       default:
+               INIT_PAL_STATUS_UNIMPLEMENTED(result);
+               printk(KERN_WARNING"kvm: Unsupported pal call,"
+                                       " index:0x%lx\n", gr28);
+       }
+       set_pal_result(vcpu, result);
+       return ret;
+}
+
+static struct sal_ret_values sal_emulator(struct kvm *kvm,
+                               long index, unsigned long in1,
+                               unsigned long in2, unsigned long in3,
+                               unsigned long in4, unsigned long in5,
+                               unsigned long in6, unsigned long in7)
+{
+       unsigned long r9  = 0;
+       unsigned long r10 = 0;
+       long r11 = 0;
+       long status;
+
+       status = 0;
+       switch (index) {
+       case SAL_FREQ_BASE:
+               status = ia64_sal_freq_base(in1, &r9, &r10);
+               break;
+       case SAL_PCI_CONFIG_READ:
+               printk(KERN_WARNING"kvm: Not allowed to call here!"
+                       " SAL_PCI_CONFIG_READ\n");
+               break;
+       case SAL_PCI_CONFIG_WRITE:
+               printk(KERN_WARNING"kvm: Not allowed to call here!"
+                       " SAL_PCI_CONFIG_WRITE\n");
+               break;
+       case SAL_SET_VECTORS:
+               if (in1 == SAL_VECTOR_OS_BOOT_RENDEZ) {
+                       if (in4 != 0 || in5 != 0 || in6 != 0 || in7 != 0) {
+                               status = -2;
+                       } else {
+                               kvm->arch.rdv_sal_data.boot_ip = in2;
+                               kvm->arch.rdv_sal_data.boot_gp = in3;
+                       }
+                       printk("Rendvous called! iip:%lx\n\n", in2);
+               } else
+                       printk(KERN_WARNING"kvm: CALLED SAL_SET_VECTORS %lu."
+                                                       "ignored...\n", in1);
+               break;
+       case SAL_GET_STATE_INFO:
+               /* No more info.  */
+               status = -5;
+               r9 = 0;
+               break;
+       case SAL_GET_STATE_INFO_SIZE:
+               /* Return a dummy size.  */
+               status = 0;
+               r9 = 128;
+               break;
+       case SAL_CLEAR_STATE_INFO:
+               /* Noop.  */
+               break;
+       case SAL_MC_RENDEZ:
+               printk(KERN_WARNING
+                       "kvm: called SAL_MC_RENDEZ. ignored...\n");
+               break;
+       case SAL_MC_SET_PARAMS:
+               printk(KERN_WARNING
+                       "kvm: called  SAL_MC_SET_PARAMS.ignored!\n");
+               break;
+       case SAL_CACHE_FLUSH:
+               if (1) {
+                       /*Flush using SAL.
+                       This method is faster but has a side
+                       effect on other vcpu running on
+                       this cpu.  */
+                       status = ia64_sal_cache_flush(in1);
+               } else {
+                       /*Maybe need to implement the method
+                       without side effect!*/
+                       status = 0;
+               }
+               break;
+       case SAL_CACHE_INIT:
+               printk(KERN_WARNING
+                       "kvm: called SAL_CACHE_INIT.  ignored...\n");
+               break;
+       case SAL_UPDATE_PAL:
+               printk(KERN_WARNING
+                       "kvm: CALLED SAL_UPDATE_PAL.  ignored...\n");
+               break;
+       default:
+               printk(KERN_WARNING"kvm: called SAL_CALL with unknown index."
+                                               " index:%ld\n", index);
+               status = -1;
+               break;
+       }
+       return ((struct sal_ret_values) {status, r9, r10, r11});
+}
+
+static void kvm_get_sal_call_data(struct kvm_vcpu *vcpu, u64 *in0, u64 *in1,
+               u64 *in2, u64 *in3, u64 *in4, u64 *in5, u64 *in6, u64 *in7){
+
+       struct exit_ctl_data *p;
+
+       p = kvm_get_exit_data(vcpu);
+
+       if (p) {
+               if (p->exit_reason == EXIT_REASON_SAL_CALL) {
+                       *in0 = p->u.sal_data.in0;
+                       *in1 = p->u.sal_data.in1;
+                       *in2 = p->u.sal_data.in2;
+                       *in3 = p->u.sal_data.in3;
+                       *in4 = p->u.sal_data.in4;
+                       *in5 = p->u.sal_data.in5;
+                       *in6 = p->u.sal_data.in6;
+                       *in7 = p->u.sal_data.in7;
+                       return ;
+               }
+       }
+       *in0 = 0;
+}
+
+void kvm_sal_emul(struct kvm_vcpu *vcpu)
+{
+
+       struct sal_ret_values result;
+       u64 index, in1, in2, in3, in4, in5, in6, in7;
+
+       kvm_get_sal_call_data(vcpu, &index, &in1, &in2,
+                       &in3, &in4, &in5, &in6, &in7);
+       result = sal_emulator(vcpu->kvm, index, in1, in2, in3,
+                                       in4, in5, in6, in7);
+       set_sal_result(vcpu, result);
+}