Re: [PATCH v2 00/14] KVM/ARM Implementation

[haibin wang <wanghaibin202@xxxxxxxxx>]

Question: When I configure guestos  
with CONFIG_ARM_ARCH_TIMER,and use

rtsm_ve-cortex_a15x2.dtb(two cpus) to boot guestos, it can't boot success!

Which resource I used as follows:
kvm-arm:   git://github.com/virtualopensystems/linux-kvm-arm.git     branch :kvm-arm-v12-vgic-timers
qemu:        git://github.com/virtualopensystems/qemu.git                branch :kvm-arm
guide:     http://www.virtualopensystems.com/media/pdf/kvm-arm-guide.pdf
It is worth mentioning that i use the guest kernel which is same to host kernel,just like the guide said.
According to the guide,step by step,Qemu boot the guest with follow command line,that will be sucess!
 ./qemu-system-arm \
 -enable-kvm \
 -kernel zImage \
 -sd guest.cramfs \
 -dtb ./rtsm_ve-cortex_a15x1.dtb \
 -m 512 -M vexpress-a15 -cpu cortex-a15 -nographic \
 -append "console=ttyAMA0 mem=512M root=/dev/mmcblk0 rw"

But,I attempt to change the dtb file to rtsm_ve-cortex_a15x2.dtb which has two cpus ,there is 

something wrong happened! There is not any print info to help me find the problem.
 
I find that there is an end-less loop. depict as follow:

in the process of booting host :
      init_hyp_mode ==> kvm_timer_hyp_init , this function enable physical virtual timer ppi, and 

register the irq handler: kvm_arch_timer_handler
          
in the process of booting guest,  some reasons make guest exit :
      ||===>   VM_EXIT
      ||                 ||
      ||                 ||                  save content module:
      ||    |--------------------------------------------------------------------------------------------------------------|               
      ||    |   save_timer_state     (save hardware virtual timer,CNTV_CTL,                     |
      ||    |             ||               CNTV_CVAL regiests to vcpu and disenable                 |
      ||    |             ||               virtual timer)                                                               |
      ||    |             ||                                                                                                |
      ||    |  kvm_timer_sync_from_cpu    (software judge and decide to                       |
      ||    |             ||                            send virtual virtual timer ppi to guest or not)    |
      ||    |-------------||------------------------------------------------------------------------------------------------|
      ||                   ||
      ||                   ||
      ||             ioctl(KVM_RUN)    
      ||                   ||
      ||                   ||                 restore content module:
      ||    |-----------------------------------------------------------------------------------------------------------------------|               
      ||    |      restore_timer_state    (restore hardware virtual timer,CNTV_CTL,                     |
      ||    |             ||                CNTV_CVAL regiests from vcpu)                                           |
      ||    |             ||                                                                                                          |
      ||    |       GICD_ISPENDING        (because hardware virtual timer has been enabled, and |
      ||    |             ||               hostos has enabled the physical virtual timer ppi                     |
      ||    |             ||               so there will be a physical virtual timer ppi pending                 |
      ||    |             ||               for host)                                                                              |
      ||    |             ||      (host deal with the physical virtual timer ppi using                            |
      ||    |             ||       handler kvm_arch_timer_handler, the handler do nothing, so that    |
      ||    |             ||       run here next time, GICD_ISPENDING must have the physical         |
      ||    |             ||       virtual timer ppi pending)                                                              |
      ||    |             ||                                                                                                         |
      ||    |       HCR register setting                                                                                     |
      ||    |             ||                                                                                                         |   
      ||    |             ||                                                                                                         | 
      ||    |            ERET      (want to exectue guest instructions,deal with  virtual virtual          |
      ||    |             ||        timer ppi and read/write virtual timer registers to set next              |
      ||    |             ||        event timer,and only guest use and change virtual timer                 |
      ||    |             ||        registers to set next event,but GICD_ISPENDING has pending irq ,  |
      ||    |             ||         so next  will trap to hyp_irq,that is VM_EXIT)                                |
      ||    |-------------||------------------------------------------------------------------ -------------------------------------|
      ||                  ||
      ||<========||
 
Reasons are listed as follows:
(1) When storing the context, it will send(or not send) the virtual interrupt of virtual timer ppi to Guest
   according to kvm_timer_sync_from_cpu.
(2) The physical virtual timer ppi is enabled by hostos, however, when restoring content , there must be
   a physical virtual timer ppi pending, so that "eret" will trap to hyp_irq again immediately! this is the end-less loop.
 
The solution of the problem:
I think it is not necessary to enable the physical interrupt of virtual timer ppi in the function of 

kvm_timer_hyp_init when hostos initializing the hyp .
 
Signed-off-by: Wanghaibin<wanghaibin202@xxxxxxxxx>
---
 arch/arm/kvm/timer.c |   27 ---------------------------
 1 files changed, 0 insertions(+), 27 deletions(-)
diff --git a/arch/arm/kvm/timer.c b/arch/arm/kvm/timer.c
index a241298..a10ef3d 100644
--- a/arch/arm/kvm/timer.c
+++ b/arch/arm/kvm/timer.c
@@ -145,43 +145,16 @@ static const struct of_device_id arch_timer_of_match[] = {
 int kvm_timer_hyp_init(void)
 {
-       struct device_node *np;
-       unsigned int ppi;
-       int err;
        timecounter = arch_timer_get_timecounter();
        if (!timecounter)
                return -ENODEV;
-       np = of_find_matching_node(NULL, arch_timer_of_match);
-       if (!np) {
-               kvm_err("kvm_arch_timer: can't find DT node\n");
-               return -ENODEV;
-       }
-
-       ppi = irq_of_parse_and_map(np, 2);
-       if (!ppi) {
-               kvm_err("kvm_arch_timer: no virtual timer interrupt\n");
-               return -EINVAL;
-       }
-
-       err = request_percpu_irq(ppi, kvm_arch_timer_handler,
-                                "kvm guest timer", kvm_get_running_vcpus());
-       if (err) {
-               kvm_err("kvm_arch_timer: can't request interrupt %d (%d)\n",
-                       ppi, err);
-               return err;
-       }
-
        wqueue = create_singlethread_workqueue("kvm_arch_timer");
        if (!wqueue) {
-               free_percpu_irq(ppi, kvm_get_running_vcpus());
                return -ENOMEM;
        }
-       kvm_info("%s IRQ%d\n", np->name, ppi);
-       on_each_cpu(kvm_timer_init_interrupt, &ppi, 1);
-
        return 0;
 }

_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm