Re: [PATCH v21 02/29] LoongArch: KVM: Implement kvm module related interface
From: zhaotianrui
Date: Mon Sep 18 2023 - 23:13:40 EST
在 2023/9/19 上午10:57, Huacai Chen 写道:
On Tue, Sep 19, 2023 at 10:53 AM zhaotianrui <zhaotianrui@xxxxxxxxxxx> wrote:
在 2023/9/19 上午10:14, Huacai Chen 写道:
On Mon, Sep 18, 2023 at 4:39 PM bibo mao <maobibo@xxxxxxxxxxx> wrote:
在 2023/9/18 14:25, zhaotianrui 写道:
在 2023/9/18 下午12:03, Huacai Chen 写道:
On Mon, Sep 18, 2023 at 11:20 AM zhaotianrui <zhaotianrui@xxxxxxxxxxx> wrote:
在 2023/9/18 上午11:12, Huacai Chen 写道:
On Mon, Sep 18, 2023 at 11:08 AM zhaotianrui <zhaotianrui@xxxxxxxxxxx> wrote:
在 2023/9/18 上午9:45, Huacai Chen 写道:
On Mon, Sep 18, 2023 at 9:21 AM zhaotianrui <zhaotianrui@xxxxxxxxxxx> wrote:
在 2023/9/16 下午4:51, Huacai Chen 写道:
Hi, Tianrui,
On Fri, Sep 15, 2023 at 9:50 AM Tianrui Zhao <zhaotianrui@xxxxxxxxxxx> wrote:
Implement LoongArch kvm module init, module exit interface,
using kvm context to save the vpid info and vcpu world switch
interface pointer.
Reviewed-by: Bibo Mao <maobibo@xxxxxxxxxxx>
Signed-off-by: Tianrui Zhao <zhaotianrui@xxxxxxxxxxx>
---
arch/loongarch/kvm/main.c | 367 ++++++++++++++++++++++++++++++++++++++
1 file changed, 367 insertions(+)
create mode 100644 arch/loongarch/kvm/main.c
diff --git a/arch/loongarch/kvm/main.c b/arch/loongarch/kvm/main.c
new file mode 100644
index 0000000000..0deb9273d8
--- /dev/null
+++ b/arch/loongarch/kvm/main.c
@@ -0,0 +1,367 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2020-2023 Loongson Technology Corporation Limited
+ */
+
+#include <linux/err.h>
+#include <linux/module.h>
+#include <linux/kvm_host.h>
+#include <asm/cacheflush.h>
+#include <asm/cpufeature.h>
+#include <asm/kvm_csr.h>
+#include "trace.h"
+
+static struct kvm_context __percpu *vmcs;
+struct kvm_world_switch *kvm_loongarch_ops;
+unsigned long vpid_mask;
+static int gcsr_flag[CSR_MAX_NUMS];
+
+int get_gcsr_flag(int csr)
+{
+ if (csr < CSR_MAX_NUMS)
+ return gcsr_flag[csr];
+
+ return INVALID_GCSR;
+}
+
+static inline void set_gcsr_sw_flag(int csr)
+{
+ if (csr < CSR_MAX_NUMS)
+ gcsr_flag[csr] |= SW_GCSR;
+}
+
+static inline void set_gcsr_hw_flag(int csr)
+{
+ if (csr < CSR_MAX_NUMS)
+ gcsr_flag[csr] |= HW_GCSR;
+}
+
+/*
+ * The default value of gcsr_flag[CSR] is 0, and we use this
+ * function to set the flag to 1(SW_GCSR) or 2(HW_GCSR) if the
+ * gcsr is software or hardware. It will be used by get/set_gcsr,
+ * if gcsr_flag is HW we should use gcsrrd/gcsrwr to access it,
+ * else use sw csr to emulate it.
+ */
+static void kvm_init_gcsr_flag(void)
+{
+ set_gcsr_hw_flag(LOONGARCH_CSR_CRMD);
+ set_gcsr_hw_flag(LOONGARCH_CSR_PRMD);
+ set_gcsr_hw_flag(LOONGARCH_CSR_EUEN);
+ set_gcsr_hw_flag(LOONGARCH_CSR_MISC);
+ set_gcsr_hw_flag(LOONGARCH_CSR_ECFG);
+ set_gcsr_hw_flag(LOONGARCH_CSR_ESTAT);
+ set_gcsr_hw_flag(LOONGARCH_CSR_ERA);
+ set_gcsr_hw_flag(LOONGARCH_CSR_BADV);
+ set_gcsr_hw_flag(LOONGARCH_CSR_BADI);
+ set_gcsr_hw_flag(LOONGARCH_CSR_EENTRY);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBIDX);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBEHI);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBELO0);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBELO1);
+ set_gcsr_hw_flag(LOONGARCH_CSR_ASID);
+ set_gcsr_hw_flag(LOONGARCH_CSR_PGDL);
+ set_gcsr_hw_flag(LOONGARCH_CSR_PGDH);
+ set_gcsr_hw_flag(LOONGARCH_CSR_PWCTL0);
+ set_gcsr_hw_flag(LOONGARCH_CSR_PWCTL1);
+ set_gcsr_hw_flag(LOONGARCH_CSR_STLBPGSIZE);
+ set_gcsr_hw_flag(LOONGARCH_CSR_RVACFG);
+ set_gcsr_hw_flag(LOONGARCH_CSR_CPUID);
+ set_gcsr_hw_flag(LOONGARCH_CSR_PRCFG1);
+ set_gcsr_hw_flag(LOONGARCH_CSR_PRCFG2);
+ set_gcsr_hw_flag(LOONGARCH_CSR_PRCFG3);
+ set_gcsr_hw_flag(LOONGARCH_CSR_KS0);
+ set_gcsr_hw_flag(LOONGARCH_CSR_KS1);
+ set_gcsr_hw_flag(LOONGARCH_CSR_KS2);
+ set_gcsr_hw_flag(LOONGARCH_CSR_KS3);
+ set_gcsr_hw_flag(LOONGARCH_CSR_KS4);
+ set_gcsr_hw_flag(LOONGARCH_CSR_KS5);
+ set_gcsr_hw_flag(LOONGARCH_CSR_KS6);
+ set_gcsr_hw_flag(LOONGARCH_CSR_KS7);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TMID);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TCFG);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TVAL);
+ set_gcsr_hw_flag(LOONGARCH_CSR_CNTC);
+ set_gcsr_hw_flag(LOONGARCH_CSR_LLBCTL);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBRENTRY);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBRBADV);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBRERA);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBRSAVE);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBRELO0);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBRELO1);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBREHI);
+ set_gcsr_hw_flag(LOONGARCH_CSR_TLBRPRMD);
+ set_gcsr_hw_flag(LOONGARCH_CSR_DMWIN0);
+ set_gcsr_hw_flag(LOONGARCH_CSR_DMWIN1);
+ set_gcsr_hw_flag(LOONGARCH_CSR_DMWIN2);
+ set_gcsr_hw_flag(LOONGARCH_CSR_DMWIN3);
+ set_gcsr_hw_flag(LOONGARCH_CSR_MWPS);
+ set_gcsr_hw_flag(LOONGARCH_CSR_FWPS);
+
+ set_gcsr_sw_flag(LOONGARCH_CSR_IMPCTL1);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IMPCTL2);
+ set_gcsr_sw_flag(LOONGARCH_CSR_MERRCTL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_MERRINFO1);
+ set_gcsr_sw_flag(LOONGARCH_CSR_MERRINFO2);
+ set_gcsr_sw_flag(LOONGARCH_CSR_MERRENTRY);
+ set_gcsr_sw_flag(LOONGARCH_CSR_MERRERA);
+ set_gcsr_sw_flag(LOONGARCH_CSR_MERRSAVE);
+ set_gcsr_sw_flag(LOONGARCH_CSR_CTAG);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DEBUG);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DERA);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DESAVE);
+ set_gcsr_sw_flag(LOONGARCH_CSR_PRCFG1);
+ set_gcsr_sw_flag(LOONGARCH_CSR_PRCFG2);
+ set_gcsr_sw_flag(LOONGARCH_CSR_PRCFG3);
+ set_gcsr_sw_flag(LOONGARCH_CSR_PGD);
+ set_gcsr_sw_flag(LOONGARCH_CSR_TINTCLR);
+
+ set_gcsr_sw_flag(LOONGARCH_CSR_FWPS);
+ set_gcsr_sw_flag(LOONGARCH_CSR_FWPC);
+ set_gcsr_sw_flag(LOONGARCH_CSR_MWPS);
+ set_gcsr_sw_flag(LOONGARCH_CSR_MWPC);
FWPS and MWPS are both HW CSR and SW CSR?
Huacai
The FWPC and MWPC should be SW GCSR, FWPS and MWPS should be HW GCSR, it
is my mistake.
But in user manual vol 3, section 1.5, FWPC/FWPS/MWPC/MWPS are all HW
GCSR, while DBxxxx and IBxxxx are SW GCSR.
Ok, It is my misunderstanding, as the FWPC and MWPC can control guest
debug register numbers, but I know they are all HW GCSR when I look up
the manual again.
So these lines can be removed?
set_gcsr_sw_flag(LOONGARCH_CSR_FWPS);
set_gcsr_sw_flag(LOONGARCH_CSR_FWPC);
set_gcsr_sw_flag(LOONGARCH_CSR_MWPS);
set_gcsr_sw_flag(LOONGARCH_CSR_MWPC);
And add FWPC/MWPC to hw list?
Can you discuss more with our hw engineers? I found in section 1.8:
PGD, TINTCLR, PRCFG1~3 are all HW GCSR.
If guest can access hw CSR directly and do not cause a exception, it means hw GCSR. on the other hand, if cause a exception to return to KVM to emulate it by software, it means SW GCSR. And I re-check the PGD, TINTCLR, PRCFG1~3 CSR, they should be hw GCSR.
CSR_PGD is logical read only register instead, its content is the same
CSR_PGDL if highest bit of BADV is 0, else its content is the same with
CSR_PGDH.
TINTCLR is write only register, write 1 to clear timer interrupt and read
value is 0.
So CSR_PGD and TINTCLR need not save and restore during vcpu switching or
vm migration, it is not necessary to set CSR_PGD/TINTCLR from qemu user space.
PRCFG1~3 is read only registers, it is decided by hardware and can not be
set by kvm, it is not necessary to save or retore. However there maybe brings
problems for vm migration on diferent hardwares if PRCFG1~3 is different.
During my tests, reading FWPC/FWPS/MWPC/MWPS causes guest exit, but
user manual vol 3 says they are HW GCSR, who can tell me why?
It will cause guest exiting when guest visit these registers above for
the first time, and KVM could config FWPC, MWPC to enable the debug
registers for guest, so that it will not cause exiting at next time. And
we will add this handling later when we want to use this function in guest.
Then what type should we mark them? HW GCSR or SW GCSR?
Huacai
It should be marked by SW GCSR at the initial time in this function, and
when guest visit it, we should change the flag to HW GCSR.
Thanks
Tianrui Zhao
Thanks
Tianrui Zhao
Huacai
Regards
Bibo Mao
Thanks
Tianrui Zhao
Huacai
Huacai
Yes, It is.
Thanks
Tianrui Zhao
Thanks
Tianrui Zhao
Huacai
Thanks
Tianrui Zhao
+
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB0ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB0MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB0CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB0ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB1ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB1MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB1CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB1ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB2ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB2MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB2CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB2ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB3ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB3MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB3CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB3ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB4ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB4MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB4CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB4ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB5ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB5MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB5CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB5ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB6ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB6MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB6CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB6ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB7ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB7MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB7CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_DB7ASID);
+
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB0ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB0MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB0CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB0ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB1ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB1MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB1CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB1ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB2ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB2MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB2CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB2ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB3ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB3MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB3CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB3ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB4ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB4MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB4CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB4ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB5ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB5MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB5CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB5ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB6ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB6MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB6CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB6ASID);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB7ADDR);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB7MASK);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB7CTRL);
+ set_gcsr_sw_flag(LOONGARCH_CSR_IB7ASID);
+
+ set_gcsr_sw_flag(LOONGARCH_CSR_PERFCTRL0);
+ set_gcsr_sw_flag(LOONGARCH_CSR_PERFCNTR0);
+ set_gcsr_sw_flag(LOONGARCH_CSR_PERFCTRL1);
+ set_gcsr_sw_flag(LOONGARCH_CSR_PERFCNTR1);
+ set_gcsr_sw_flag(LOONGARCH_CSR_PERFCTRL2);
+ set_gcsr_sw_flag(LOONGARCH_CSR_PERFCNTR2);
+ set_gcsr_sw_flag(LOONGARCH_CSR_PERFCTRL3);
+ set_gcsr_sw_flag(LOONGARCH_CSR_PERFCNTR3);
+}
+
+static void kvm_update_vpid(struct kvm_vcpu *vcpu, int cpu)
+{
+ struct kvm_context *context;
+ unsigned long vpid;
+
+ context = per_cpu_ptr(vcpu->kvm->arch.vmcs, cpu);
+ vpid = context->vpid_cache + 1;
+ if (!(vpid & vpid_mask)) {
+ /* finish round of 64 bit loop */
+ if (unlikely(!vpid))
+ vpid = vpid_mask + 1;
+
+ /* vpid 0 reserved for root */
+ ++vpid;
+
+ /* start new vpid cycle */
+ kvm_flush_tlb_all();
+ }
+
+ context->vpid_cache = vpid;
+ vcpu->arch.vpid = vpid;
+}
+
+void kvm_check_vpid(struct kvm_vcpu *vcpu)
+{
+ struct kvm_context *context;
+ bool migrated;
+ unsigned long ver, old, vpid;
+ int cpu;
+
+ cpu = smp_processor_id();
+ /*
+ * Are we entering guest context on a different CPU to last time?
+ * If so, the vCPU's guest TLB state on this CPU may be stale.
+ */
+ context = per_cpu_ptr(vcpu->kvm->arch.vmcs, cpu);
+ migrated = (vcpu->cpu != cpu);
+
+ /*
+ * Check if our vpid is of an older version
+ *
+ * We also discard the stored vpid if we've executed on
+ * another CPU, as the guest mappings may have changed without
+ * hypervisor knowledge.
+ */
+ ver = vcpu->arch.vpid & ~vpid_mask;
+ old = context->vpid_cache & ~vpid_mask;
+ if (migrated || (ver != old)) {
+ kvm_update_vpid(vcpu, cpu);
+ trace_kvm_vpid_change(vcpu, vcpu->arch.vpid);
+ vcpu->cpu = cpu;
+ }
+
+ /* Restore GSTAT(0x50).vpid */
+ vpid = (vcpu->arch.vpid & vpid_mask) << CSR_GSTAT_GID_SHIFT;
+ change_csr_gstat(vpid_mask << CSR_GSTAT_GID_SHIFT, vpid);
+}
+
+static int kvm_loongarch_env_init(void)
+{
+ struct kvm_context *context;
+ int cpu, order;
+ void *addr;
+
+ vmcs = alloc_percpu(struct kvm_context);
+ if (!vmcs) {
+ pr_err("kvm: failed to allocate percpu kvm_context\n");
+ return -ENOMEM;
+ }
+
+ kvm_loongarch_ops = kzalloc(sizeof(*kvm_loongarch_ops), GFP_KERNEL);
+ if (!kvm_loongarch_ops) {
+ free_percpu(vmcs);
+ vmcs = NULL;
+ return -ENOMEM;
+ }
+ /*
+ * There will be problem in world switch code if there
+ * is page fault reenter, since pgd register is shared
+ * between root kernel and kvm hypervisor. World switch
+ * entry need be unmapped area, cannot be tlb mapped area.
+ * In future if hw pagetable walking is supported, or there
+ * is separate pgd registers between root kernel and kvm
+ * hypervisor, copying about world switch code will not be used.
+ */
+
+ order = get_order(kvm_vector_size + kvm_enter_guest_size);
+ addr = (void *)__get_free_pages(GFP_KERNEL, order);
+ if (!addr) {
+ free_percpu(vmcs);
+ vmcs = NULL;
+ kfree(kvm_loongarch_ops);
+ kvm_loongarch_ops = NULL;
+ return -ENOMEM;
+ }
+
+ memcpy(addr, kvm_vector_entry, kvm_vector_size);
+ memcpy(addr + kvm_vector_size, kvm_enter_guest, kvm_enter_guest_size);
+ flush_icache_range((unsigned long)addr, (unsigned long)addr +
+ kvm_vector_size + kvm_enter_guest_size);
+ kvm_loongarch_ops->guest_eentry = addr;
+ kvm_loongarch_ops->enter_guest = addr + kvm_vector_size;
+ kvm_loongarch_ops->page_order = order;
+
+ vpid_mask = read_csr_gstat();
+ vpid_mask = (vpid_mask & CSR_GSTAT_GIDBIT) >> CSR_GSTAT_GIDBIT_SHIFT;
+ if (vpid_mask)
+ vpid_mask = GENMASK(vpid_mask - 1, 0);
+
+ for_each_possible_cpu(cpu) {
+ context = per_cpu_ptr(vmcs, cpu);
+ context->vpid_cache = vpid_mask + 1;
+ context->last_vcpu = NULL;
+ }
+
+ kvm_init_fault();
+ kvm_init_gcsr_flag();
+
+ return 0;
+}
+
+static void kvm_loongarch_env_exit(void)
+{
+ unsigned long addr;
+
+ if (vmcs)
+ free_percpu(vmcs);
+
+ if (kvm_loongarch_ops) {
+ if (kvm_loongarch_ops->guest_eentry) {
+ addr = (unsigned long)kvm_loongarch_ops->guest_eentry;
+ free_pages(addr, kvm_loongarch_ops->page_order);
+ }
+ kfree(kvm_loongarch_ops);
+ }
+}
+
+static int kvm_loongarch_init(void)
+{
+ int r;
+
+ if (!cpu_has_lvz) {
+ kvm_info("hardware virtualization not available\n");
+ return -ENODEV;
+ }
+ r = kvm_loongarch_env_init();
+ if (r)
+ return r;
+
+ return kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
+}
+
+static void kvm_loongarch_exit(void)
+{
+ kvm_exit();
+ kvm_loongarch_env_exit();
+}
+
+module_init(kvm_loongarch_init);
+module_exit(kvm_loongarch_exit);
+
+#ifdef MODULE
+static const struct cpu_feature loongarch_kvm_feature[] = {
+ { .feature = cpu_feature(LOONGARCH_LVZ) },
+ {},
+};
+MODULE_DEVICE_TABLE(cpu, loongarch_kvm_feature);
+#endif
--
2.39.1