Hi Greg,
On 1/26/24 17:08, Greg KH wrote:
> On Fri, Jan 26, 2024 at 04:20:16PM -0800, Dongli Zhang wrote:
>> Hi Maxim and Paolo,
>>
>> This is the linux-stable backport request regarding the below patch.
>
> For what tree(s)?
It is linux-5.15.y as in the Subject of the patch.
However, more versions require this bugfix, e.g., 6.1 or 5.4.
I have a backport for 5.4 as well.
I just send the version on top of 5.15 for suggestion, or there
is already a backport available.
>
> And you forgot to sign off on the patch :(
I have a signed-off after the commit message. There are some conflicts:
e.g., the smram buffer offsets used in function calls.
I have added the commit messages to explain the conflicts between
Paolo's signed-off and my own signed-off.
BTW, I have created a kvm selftest program to reproduce this issue. Although I
cannot reproduce on baremetal (perhaps it is too fast), I can always reproduce
on a KVM running on top of a VM.
$ ./smm_interrupt_window
Create thread for vcpu=0
Create thread for vcpu=1
Waiting for 2-second for test to start ...
vcpu=0: stage = 1
vcpu=1: stage = 2
Start the test!
==== Test Assertion Failure ====
x86_64/mytest.c:96: exit_reason == (2)
pid=5541 tid=5544 errno=0 - Success
1 0x0000000000401dd3: vcpu_worker at mytest.c:96
2 0x0000000000417cc9: start_thread at libpthread.o:?
3 0x0000000000470d32: __clone at ??:?
Wanted KVM exit reason: 2 (IO), got: 9 (FAIL_ENTRY)
There are below in the dmesg.
[ 165.292990] VMCS 0000000088f567e4, last attempted VM-entry on CPU 14
... ...
[ 165.304272] RFLAGS=0x00000002 DR7 = 0x0000000000000400
... ...
[ 165.329264] Interruptibility = 00000009 ActivityState = 00000000
// SPDX-License-Identifier: GPL-2.0
/*
* Reproduce the issue fixed by the commit fb28875fd7da ("KVM: x86: smm:
* preserve interrupt shadow in SMRAM").
*
* The vCPU#0 sends SMI to vCPU#1 that is running sti to trap into the
* interrupt window.
*
* Adapted from smm_test.c
*/
#include <pthread.h>
#include "kvm_util.h"
#include "processor.h"
#include "vmx.h"
#define SMRAM_SIZE 65536
#define SMRAM_MEMSLOT ((1 << 16) | 1)
#define SMRAM_PAGES (SMRAM_SIZE / PAGE_SIZE)
#define SMRAM_GPA 0x1000000
#define SMRAM_STAGE 0xfe
#define STR(x) #x
#define XSTR(s) STR(s)
#define SYNC_PORT 0xe
#define NR_VCPUS 2
uint8_t smi_handler[] = {
0xb0, SMRAM_STAGE, /* mov $SMRAM_STAGE, %al */
0x0f, 0xaa, /* rsm */
};
static inline void sync_with_host(uint64_t phase)
{
asm volatile("in $" XSTR(SYNC_PORT)", %%al \n"
: "+a" (phase));
}
static void guest_code(int cpu)
{
uint64_t apicbase = rdmsr(MSR_IA32_APICBASE);
int i;
wrmsr(MSR_IA32_APICBASE, apicbase | X2APIC_ENABLE);
if (cpu == 0) {
sync_with_host(1);
/*
* vCPU#0 keeps cli/nop/sti
*/
while(1) {
asm volatile("cli");
asm volatile("nop");
asm volatile("nop");
asm volatile("nop");
asm volatile("sti");
asm volatile("nop");
asm volatile("nop");
asm volatile("nop");
}
}
if (cpu == 1) {
sync_with_host(2);
/*
* vCPU#1 keeps sending SMI to vCPU#0
*/
while(1) {
x2apic_write_reg(APIC_ICR, APIC_INT_ASSERT | APIC_DM_SMI);
for (i = 0; i < 1000000; i++)
asm volatile("nop");
}
}
}
static void *vcpu_worker(void *data)
{
struct kvm_vcpu *vcpu = data;
int stage_reported;
struct kvm_regs regs;
pr_info("Create thread for vcpu=%u\n", vcpu->id);
if (vcpu->id == 0) {
vcpu_run(vcpu);
TEST_ASSERT_KVM_EXIT_REASON(vcpu, KVM_EXIT_IO);
memset(®s, 0, sizeof(regs));
vcpu_regs_get(vcpu, ®s);
stage_reported = regs.rax & 0xff;
pr_info("vcpu=%u: stage = %d\n", vcpu->id, stage_reported);
vcpu_run(vcpu);
TEST_ASSERT_KVM_EXIT_REASON(vcpu, KVM_EXIT_IO);
memset(®s, 0, sizeof(regs));
vcpu_regs_get(vcpu, ®s);
stage_reported = regs.rax & 0xff;
pr_info("vcpu=%u: stage = %d\n", vcpu->id, stage_reported);
}
if (vcpu->id == 1) {
pr_info("Waiting for 2-second for test to start ...\n");
sleep(2);
vcpu_run(vcpu);
TEST_ASSERT_KVM_EXIT_REASON(vcpu, KVM_EXIT_IO);
memset(®s, 0, sizeof(regs));
vcpu_regs_get(vcpu, ®s);
stage_reported = regs.rax & 0xff;
pr_info("vcpu=%u: stage = %d\n", vcpu->id, stage_reported);
pr_info("Start the test!\n");
vcpu_run(vcpu);
}
return NULL;
}
int main(int argc, char **argv)
{
struct kvm_vcpu *vcpus[NR_VCPUS];
struct kvm_vm *vm;
pthread_t tids[NR_VCPUS];
vm = vm_create_with_vcpus(NR_VCPUS, guest_code, vcpus);
vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS, SMRAM_GPA,
SMRAM_MEMSLOT, SMRAM_PAGES, 0);
TEST_ASSERT(vm_phy_pages_alloc(vm, SMRAM_PAGES, SMRAM_GPA, SMRAM_MEMSLOT)
== SMRAM_GPA, "could not allocate guest physical addresses?");
memset(addr_gpa2hva(vm, SMRAM_GPA), 0x0, SMRAM_SIZE);
memcpy(addr_gpa2hva(vm, SMRAM_GPA) + 0x8000, smi_handler, sizeof(smi_handler));
vcpu_set_msr(vcpus[0], MSR_IA32_SMBASE, SMRAM_GPA);
vcpu_set_msr(vcpus[1], MSR_IA32_SMBASE, SMRAM_GPA);
vcpu_args_set(vcpus[0], 1, 0);
vcpu_args_set(vcpus[1], 1, 1);
pthread_create(&tids[0], NULL, vcpu_worker, vcpus[0]);
pthread_create(&tids[1], NULL, vcpu_worker, vcpus[1]);
pthread_join(tids[0], NULL);
pthread_join(tids[1], NULL);
return 0;
}
Dongli Zhang
>
> thanks,
>
> greg k-h
