Patchwork [v3] Fix X86 CPU topology in KVM mode

login
register
mail settings
Submitter Bharata B Rao
Date Nov. 2, 2011, 8:46 a.m.
Message ID <1320223568-20322-1-git-send-email-bharata.rao@gmail.com>
Download mbox | patch
Permalink /patch/123240/
State New
Headers show

Comments

Bharata B Rao - Nov. 2, 2011, 8:46 a.m.
From: Bharata B Rao <bharata.rao@gmail.com>

apic id returned to guest kernel in ebx for cpuid(function=1) depends on
CPUX86State->cpuid_apic_id which gets populated after the cpuid information
is cached in the host kernel. This results in broken CPU topology in guest.

Fix this by setting cpuid_apic_id before cpuid information is passed to
the host kernel. This is done by moving the setting of cpuid_apic_id
to cpu_x86_init() where it will work for both KVM as well as TCG modes.

Signed-off-by: Bharata B Rao <bharata.rao@gmail.com>
---
This is the next post of the fix that addresses Jan's last comment from
my v2 post (http://lists.gnu.org/archive/html/qemu-devel/2011-09/msg01049.html)

Compile tested for all targets.
Tested that it works for x86_64-softmmu with KVM and TCG modes
As with the earlier version, I haven't tested i486.

A quick recap: Without this fix, the APIC IDs will not be set correctly
for vCPUs. There are two effects of this in KVM mode:

1 The final topology the guest comes up with will not be correct for
options like -smp sockets=2,cores=4,threads=2.

2. The guest kernel crashes for options like
-smp sockets=2,threads=4,cores=2 -numa node,nodeid=0,cpus=0-7
-numa node,nodeid=1,cpus=8-15

[    0.256180] divide error: 0000 [#1] SMP 
[    0.256970] last sysfs file: 
[    0.256970] CPU 1 
[    0.256970] Modules linked in:
[    0.256970] 
[    0.256970] Pid: 2, comm: kthreadd Not tainted 2.6.38.6-26.rc1.fc15.x86_64 #1 Bochs Bochs
[    0.256970] RIP: 0010:[<ffffffff8104f4d4>]  [<ffffffff8104f4d4>] select_task_rq_fair+0x44a/0x571
[    0.256970] RSP: 0000:ffff880095a97c60  EFLAGS: 00010006
[    0.256970] RAX: 0000000000100000 RBX: ffff880095a8cfc0 RCX: 0000000000000000
[    0.256970] RDX: 0000000000000000 RSI: 0000000000000100 RDI: 0000000000000000
[    0.256970] RBP: ffff880095a97d10 R08: 0000000000000100 R09: ffff880095a8cff8
[    0.256970] R10: 0000000000013840 R11: 0000000000800711 R12: 00000000ffffffff
[    0.256970] R13: ffff88009fc4f810 R14: 0000000000000001 R15: 0000000000000000
[    0.256970] FS:  0000000000000000(0000) GS:ffff88009fc40000(0000) knlGS:0000000000000000
[    0.256970] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[    0.256970] CR2: 00000000ffffffff CR3: 0000000001a03000 CR4: 00000000000006e0
[    0.256970] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[    0.256970] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[    0.256970] Process kthreadd (pid: 2, threadinfo ffff880095a96000, task ffff880095a99720)
[    0.256970] Stack:
[    0.256970]  ffffffff81475873 ffffffff81a02140 ffff880095a97ce0 ffffffff8106c5a3
[    0.256970]  ffff880095a8cfd8 0000000100000007 0000000000013840 0000000000013840
[    0.256970]  ffff880095a8cfd8 0000007d00000001 ffff880000000000 ffff8800954081e8
[    0.256970] Call Trace:
[    0.256970]  [<ffffffff81475873>] ? _raw_spin_lock_irq+0x1c/0x1e
[    0.256970]  [<ffffffff8106c5a3>] ? alloc_pid+0x2e6/0x335
[    0.256970]  [<ffffffff81048960>] select_task_rq+0x16/0x46
[    0.256970]  [<ffffffff8104e29a>] wake_up_new_task+0x3a/0xde
[    0.256970]  [<ffffffff810546ce>] do_fork+0x1f1/0x2bf
[    0.256970]  [<ffffffff8100804e>] ? load_TLS+0x10/0x14
[    0.256970]  [<ffffffff81008714>] ? __switch_to+0xc6/0x220
[    0.256970]  [<ffffffff81010c1a>] kernel_thread+0x75/0x77
[    0.256970]  [<ffffffff8106eacf>] ? kthread+0x0/0x8c
[    0.256970]  [<ffffffff8100a9e0>] ? kernel_thread_helper+0x0/0x10
[    0.256970]  [<ffffffff8106ee93>] kthreadd+0xe7/0x124
[    0.256970]  [<ffffffff8100a9e4>] kernel_thread_helper+0x4/0x10
[    0.256970]  [<ffffffff8106edac>] ? kthreadd+0x0/0x124
[    0.256970]  [<ffffffff8100a9e0>] ? kernel_thread_helper+0x0/0x10

Anthony, could you please consider this for 1.0 ?

 hw/pc.c              |    1 -
 target-i386/helper.c |    1 +
 2 files changed, 1 insertions(+), 1 deletions(-)
Jan Kiszka - Nov. 2, 2011, 8:50 a.m.
On 2011-11-02 09:46, bharata.rao@gmail.com wrote:
> From: Bharata B Rao <bharata.rao@gmail.com>
> 
> apic id returned to guest kernel in ebx for cpuid(function=1) depends on
> CPUX86State->cpuid_apic_id which gets populated after the cpuid information
> is cached in the host kernel. This results in broken CPU topology in guest.
> 
> Fix this by setting cpuid_apic_id before cpuid information is passed to
> the host kernel. This is done by moving the setting of cpuid_apic_id
> to cpu_x86_init() where it will work for both KVM as well as TCG modes.
> 
> Signed-off-by: Bharata B Rao <bharata.rao@gmail.com>
> ---
> This is the next post of the fix that addresses Jan's last comment from
> my v2 post (http://lists.gnu.org/archive/html/qemu-devel/2011-09/msg01049.html)
> 
> Compile tested for all targets.
> Tested that it works for x86_64-softmmu with KVM and TCG modes
> As with the earlier version, I haven't tested i486.
> 
> A quick recap: Without this fix, the APIC IDs will not be set correctly
> for vCPUs. There are two effects of this in KVM mode:
> 
> 1 The final topology the guest comes up with will not be correct for
> options like -smp sockets=2,cores=4,threads=2.
> 
> 2. The guest kernel crashes for options like
> -smp sockets=2,threads=4,cores=2 -numa node,nodeid=0,cpus=0-7
> -numa node,nodeid=1,cpus=8-15
> 
> [    0.256180] divide error: 0000 [#1] SMP 
> [    0.256970] last sysfs file: 
> [    0.256970] CPU 1 
> [    0.256970] Modules linked in:
> [    0.256970] 
> [    0.256970] Pid: 2, comm: kthreadd Not tainted 2.6.38.6-26.rc1.fc15.x86_64 #1 Bochs Bochs
> [    0.256970] RIP: 0010:[<ffffffff8104f4d4>]  [<ffffffff8104f4d4>] select_task_rq_fair+0x44a/0x571
> [    0.256970] RSP: 0000:ffff880095a97c60  EFLAGS: 00010006
> [    0.256970] RAX: 0000000000100000 RBX: ffff880095a8cfc0 RCX: 0000000000000000
> [    0.256970] RDX: 0000000000000000 RSI: 0000000000000100 RDI: 0000000000000000
> [    0.256970] RBP: ffff880095a97d10 R08: 0000000000000100 R09: ffff880095a8cff8
> [    0.256970] R10: 0000000000013840 R11: 0000000000800711 R12: 00000000ffffffff
> [    0.256970] R13: ffff88009fc4f810 R14: 0000000000000001 R15: 0000000000000000
> [    0.256970] FS:  0000000000000000(0000) GS:ffff88009fc40000(0000) knlGS:0000000000000000
> [    0.256970] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
> [    0.256970] CR2: 00000000ffffffff CR3: 0000000001a03000 CR4: 00000000000006e0
> [    0.256970] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> [    0.256970] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
> [    0.256970] Process kthreadd (pid: 2, threadinfo ffff880095a96000, task ffff880095a99720)
> [    0.256970] Stack:
> [    0.256970]  ffffffff81475873 ffffffff81a02140 ffff880095a97ce0 ffffffff8106c5a3
> [    0.256970]  ffff880095a8cfd8 0000000100000007 0000000000013840 0000000000013840
> [    0.256970]  ffff880095a8cfd8 0000007d00000001 ffff880000000000 ffff8800954081e8
> [    0.256970] Call Trace:
> [    0.256970]  [<ffffffff81475873>] ? _raw_spin_lock_irq+0x1c/0x1e
> [    0.256970]  [<ffffffff8106c5a3>] ? alloc_pid+0x2e6/0x335
> [    0.256970]  [<ffffffff81048960>] select_task_rq+0x16/0x46
> [    0.256970]  [<ffffffff8104e29a>] wake_up_new_task+0x3a/0xde
> [    0.256970]  [<ffffffff810546ce>] do_fork+0x1f1/0x2bf
> [    0.256970]  [<ffffffff8100804e>] ? load_TLS+0x10/0x14
> [    0.256970]  [<ffffffff81008714>] ? __switch_to+0xc6/0x220
> [    0.256970]  [<ffffffff81010c1a>] kernel_thread+0x75/0x77
> [    0.256970]  [<ffffffff8106eacf>] ? kthread+0x0/0x8c
> [    0.256970]  [<ffffffff8100a9e0>] ? kernel_thread_helper+0x0/0x10
> [    0.256970]  [<ffffffff8106ee93>] kthreadd+0xe7/0x124
> [    0.256970]  [<ffffffff8100a9e4>] kernel_thread_helper+0x4/0x10
> [    0.256970]  [<ffffffff8106edac>] ? kthreadd+0x0/0x124
> [    0.256970]  [<ffffffff8100a9e0>] ? kernel_thread_helper+0x0/0x10
> 
> Anthony, could you please consider this for 1.0 ?
> 
>  hw/pc.c              |    1 -
>  target-i386/helper.c |    1 +
>  2 files changed, 1 insertions(+), 1 deletions(-)
> 
> 
> diff --git a/hw/pc.c b/hw/pc.c
> index eb4c2d8..25bfa85 100644
> --- a/hw/pc.c
> +++ b/hw/pc.c
> @@ -941,7 +941,6 @@ static CPUState *pc_new_cpu(const char *cpu_model)
>          exit(1);
>      }
>      if ((env->cpuid_features & CPUID_APIC) || smp_cpus > 1) {
> -        env->cpuid_apic_id = env->cpu_index;
>          env->apic_state = apic_init(env, env->cpuid_apic_id);
>      }
>      qemu_register_reset(pc_cpu_reset, env);
> diff --git a/target-i386/helper.c b/target-i386/helper.c
> index 5df40d4..9302823 100644
> --- a/target-i386/helper.c
> +++ b/target-i386/helper.c
> @@ -1256,6 +1256,7 @@ CPUX86State *cpu_x86_init(const char *cpu_model)
>          cpu_x86_close(env);
>          return NULL;
>      }
> +    env->cpuid_apic_id = env->cpu_index;
>      mce_init(env);
>  
>      qemu_init_vcpu(env);

Acked-by: Jan Kiszka <jan.kiszka@siemens.com>
Anthony Liguori - Nov. 2, 2011, 12:56 p.m.
On 11/02/2011 03:46 AM, bharata.rao@gmail.com wrote:
> From: Bharata B Rao<bharata.rao@gmail.com>
>
> apic id returned to guest kernel in ebx for cpuid(function=1) depends on
> CPUX86State->cpuid_apic_id which gets populated after the cpuid information
> is cached in the host kernel. This results in broken CPU topology in guest.
>
> Fix this by setting cpuid_apic_id before cpuid information is passed to
> the host kernel. This is done by moving the setting of cpuid_apic_id
> to cpu_x86_init() where it will work for both KVM as well as TCG modes.
>
> Signed-off-by: Bharata B Rao<bharata.rao@gmail.com>

Applied.  Thanks.

Regards,

Anthony Liguori

> ---
> This is the next post of the fix that addresses Jan's last comment from
> my v2 post (http://lists.gnu.org/archive/html/qemu-devel/2011-09/msg01049.html)
>
> Compile tested for all targets.
> Tested that it works for x86_64-softmmu with KVM and TCG modes
> As with the earlier version, I haven't tested i486.
>
> A quick recap: Without this fix, the APIC IDs will not be set correctly
> for vCPUs. There are two effects of this in KVM mode:
>
> 1 The final topology the guest comes up with will not be correct for
> options like -smp sockets=2,cores=4,threads=2.
>
> 2. The guest kernel crashes for options like
> -smp sockets=2,threads=4,cores=2 -numa node,nodeid=0,cpus=0-7
> -numa node,nodeid=1,cpus=8-15
>
> [    0.256180] divide error: 0000 [#1] SMP
> [    0.256970] last sysfs file:
> [    0.256970] CPU 1
> [    0.256970] Modules linked in:
> [    0.256970]
> [    0.256970] Pid: 2, comm: kthreadd Not tainted 2.6.38.6-26.rc1.fc15.x86_64 #1 Bochs Bochs
> [    0.256970] RIP: 0010:[<ffffffff8104f4d4>]  [<ffffffff8104f4d4>] select_task_rq_fair+0x44a/0x571
> [    0.256970] RSP: 0000:ffff880095a97c60  EFLAGS: 00010006
> [    0.256970] RAX: 0000000000100000 RBX: ffff880095a8cfc0 RCX: 0000000000000000
> [    0.256970] RDX: 0000000000000000 RSI: 0000000000000100 RDI: 0000000000000000
> [    0.256970] RBP: ffff880095a97d10 R08: 0000000000000100 R09: ffff880095a8cff8
> [    0.256970] R10: 0000000000013840 R11: 0000000000800711 R12: 00000000ffffffff
> [    0.256970] R13: ffff88009fc4f810 R14: 0000000000000001 R15: 0000000000000000
> [    0.256970] FS:  0000000000000000(0000) GS:ffff88009fc40000(0000) knlGS:0000000000000000
> [    0.256970] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
> [    0.256970] CR2: 00000000ffffffff CR3: 0000000001a03000 CR4: 00000000000006e0
> [    0.256970] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> [    0.256970] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
> [    0.256970] Process kthreadd (pid: 2, threadinfo ffff880095a96000, task ffff880095a99720)
> [    0.256970] Stack:
> [    0.256970]  ffffffff81475873 ffffffff81a02140 ffff880095a97ce0 ffffffff8106c5a3
> [    0.256970]  ffff880095a8cfd8 0000000100000007 0000000000013840 0000000000013840
> [    0.256970]  ffff880095a8cfd8 0000007d00000001 ffff880000000000 ffff8800954081e8
> [    0.256970] Call Trace:
> [    0.256970]  [<ffffffff81475873>] ? _raw_spin_lock_irq+0x1c/0x1e
> [    0.256970]  [<ffffffff8106c5a3>] ? alloc_pid+0x2e6/0x335
> [    0.256970]  [<ffffffff81048960>] select_task_rq+0x16/0x46
> [    0.256970]  [<ffffffff8104e29a>] wake_up_new_task+0x3a/0xde
> [    0.256970]  [<ffffffff810546ce>] do_fork+0x1f1/0x2bf
> [    0.256970]  [<ffffffff8100804e>] ? load_TLS+0x10/0x14
> [    0.256970]  [<ffffffff81008714>] ? __switch_to+0xc6/0x220
> [    0.256970]  [<ffffffff81010c1a>] kernel_thread+0x75/0x77
> [    0.256970]  [<ffffffff8106eacf>] ? kthread+0x0/0x8c
> [    0.256970]  [<ffffffff8100a9e0>] ? kernel_thread_helper+0x0/0x10
> [    0.256970]  [<ffffffff8106ee93>] kthreadd+0xe7/0x124
> [    0.256970]  [<ffffffff8100a9e4>] kernel_thread_helper+0x4/0x10
> [    0.256970]  [<ffffffff8106edac>] ? kthreadd+0x0/0x124
> [    0.256970]  [<ffffffff8100a9e0>] ? kernel_thread_helper+0x0/0x10
>
> Anthony, could you please consider this for 1.0 ?
>
>   hw/pc.c              |    1 -
>   target-i386/helper.c |    1 +
>   2 files changed, 1 insertions(+), 1 deletions(-)
>
>
> diff --git a/hw/pc.c b/hw/pc.c
> index eb4c2d8..25bfa85 100644
> --- a/hw/pc.c
> +++ b/hw/pc.c
> @@ -941,7 +941,6 @@ static CPUState *pc_new_cpu(const char *cpu_model)
>           exit(1);
>       }
>       if ((env->cpuid_features&  CPUID_APIC) || smp_cpus>  1) {
> -        env->cpuid_apic_id = env->cpu_index;
>           env->apic_state = apic_init(env, env->cpuid_apic_id);
>       }
>       qemu_register_reset(pc_cpu_reset, env);
> diff --git a/target-i386/helper.c b/target-i386/helper.c
> index 5df40d4..9302823 100644
> --- a/target-i386/helper.c
> +++ b/target-i386/helper.c
> @@ -1256,6 +1256,7 @@ CPUX86State *cpu_x86_init(const char *cpu_model)
>           cpu_x86_close(env);
>           return NULL;
>       }
> +    env->cpuid_apic_id = env->cpu_index;
>       mce_init(env);
>
>       qemu_init_vcpu(env);
>
>

Patch

diff --git a/hw/pc.c b/hw/pc.c
index eb4c2d8..25bfa85 100644
--- a/hw/pc.c
+++ b/hw/pc.c
@@ -941,7 +941,6 @@  static CPUState *pc_new_cpu(const char *cpu_model)
         exit(1);
     }
     if ((env->cpuid_features & CPUID_APIC) || smp_cpus > 1) {
-        env->cpuid_apic_id = env->cpu_index;
         env->apic_state = apic_init(env, env->cpuid_apic_id);
     }
     qemu_register_reset(pc_cpu_reset, env);
diff --git a/target-i386/helper.c b/target-i386/helper.c
index 5df40d4..9302823 100644
--- a/target-i386/helper.c
+++ b/target-i386/helper.c
@@ -1256,6 +1256,7 @@  CPUX86State *cpu_x86_init(const char *cpu_model)
         cpu_x86_close(env);
         return NULL;
     }
+    env->cpuid_apic_id = env->cpu_index;
     mce_init(env);
 
     qemu_init_vcpu(env);