Re: [Qemu-devel] [PATCH] spapr: Correct RAM size calculation for HPT resizing

[David Gibson]

On Tue, Oct 10, 2017 at 05:21:53PM +0200, Laurent Vivier wrote:
> On 10/10/2017 16:44, Greg Kurz wrote:
> > On Wed, 11 Oct 2017 00:21:59 +1100
> > David Gibson  wrote:
> > 
> >> In order to prevent the guest from forcing the allocation of large amounts
> >> of qemu memory (or host kernel memory, in the case of KVM HV), we limit
> >> the size of Hashed Page Table (HPT) it is allowed to allocated, based on
> >> its RAM size.
> >>
> >> However, the current calculation is not correct: it only adds up the size
> >> of plugged memory, ignoring the base memory size.  This patch corrects it.
> >>
> >> While we're there, use get_plugged_memory_size() instead of directly
> >> calling pc_existing_dimms_capacity().  The only difference is that it
> >> will abort on failure, which is right: a failure here indicates something
> >> wrong within qemu.
> >>
> >> Signed-off-by: David Gibson 
> >> ---
> >>  hw/ppc/spapr_hcall.c | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/hw/ppc/spapr_hcall.c b/hw/ppc/spapr_hcall.c
> >> index 8d72bb7c1c..06af1b15c0 100644
> >> --- a/hw/ppc/spapr_hcall.c
> >> +++ b/hw/ppc/spapr_hcall.c
> >> @@ -494,7 +494,7 @@ static target_ulong h_resize_hpt_prepare(PowerPCCPU 
> >> *cpu,
> >>  return H_PARAMETER;
> >>  }
> >>  
> >> -current_ram_size = pc_existing_dimms_capacity(_fatal);
> >> +current_ram_size = ram_size + get_plugged_memory_size();
> > 
> > current_ram_size is initialized earlier in this function:
> > 
> > uint64_t current_ram_size = MACHINE(spapr)->ram_size;
> > 
> > which is is initialized to ram_size in vl.c. Why not doing:
> > 
> > current_ram_size += get_plugged_memory_size();
> > 
> > ?
> 
> I agree, it seems like the original intend of the first patch...

Yes, I think so.  However, splitting the calculation like that
demonstrably misread someone reading the code (i.e. me), so I'm going
to just ditch the initializerin the new spin.

-- 
David Gibson| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
| _way_ _around_!
http://www.ozlabs.org/~dgibson


signature.asc
Description: PGP signature

Re: [Qemu-devel] [PATCH] spapr: Correct RAM size calculation for HPT resizing

[Laurent Vivier]

On 10/10/2017 16:44, Greg Kurz wrote:
> On Wed, 11 Oct 2017 00:21:59 +1100
> David Gibson  wrote:
> 
>> In order to prevent the guest from forcing the allocation of large amounts
>> of qemu memory (or host kernel memory, in the case of KVM HV), we limit
>> the size of Hashed Page Table (HPT) it is allowed to allocated, based on
>> its RAM size.
>>
>> However, the current calculation is not correct: it only adds up the size
>> of plugged memory, ignoring the base memory size.  This patch corrects it.
>>
>> While we're there, use get_plugged_memory_size() instead of directly
>> calling pc_existing_dimms_capacity().  The only difference is that it
>> will abort on failure, which is right: a failure here indicates something
>> wrong within qemu.
>>
>> Signed-off-by: David Gibson 
>> ---
>>  hw/ppc/spapr_hcall.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/hw/ppc/spapr_hcall.c b/hw/ppc/spapr_hcall.c
>> index 8d72bb7c1c..06af1b15c0 100644
>> --- a/hw/ppc/spapr_hcall.c
>> +++ b/hw/ppc/spapr_hcall.c
>> @@ -494,7 +494,7 @@ static target_ulong h_resize_hpt_prepare(PowerPCCPU *cpu,
>>  return H_PARAMETER;
>>  }
>>  
>> -current_ram_size = pc_existing_dimms_capacity(_fatal);
>> +current_ram_size = ram_size + get_plugged_memory_size();
> 
> current_ram_size is initialized earlier in this function:
> 
> uint64_t current_ram_size = MACHINE(spapr)->ram_size;
> 
> which is is initialized to ram_size in vl.c. Why not doing:
> 
> current_ram_size += get_plugged_memory_size();
> 
> ?

I agree, it seems like the original intend of the first patch...

Thanks,
Laurent

Re: [Qemu-devel] [PATCH] spapr: Correct RAM size calculation for HPT resizing

[Greg Kurz]

On Wed, 11 Oct 2017 00:21:59 +1100
David Gibson  wrote:

> In order to prevent the guest from forcing the allocation of large amounts
> of qemu memory (or host kernel memory, in the case of KVM HV), we limit
> the size of Hashed Page Table (HPT) it is allowed to allocated, based on
> its RAM size.
> 
> However, the current calculation is not correct: it only adds up the size
> of plugged memory, ignoring the base memory size.  This patch corrects it.
> 
> While we're there, use get_plugged_memory_size() instead of directly
> calling pc_existing_dimms_capacity().  The only difference is that it
> will abort on failure, which is right: a failure here indicates something
> wrong within qemu.
> 
> Signed-off-by: David Gibson 
> ---
>  hw/ppc/spapr_hcall.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/hw/ppc/spapr_hcall.c b/hw/ppc/spapr_hcall.c
> index 8d72bb7c1c..06af1b15c0 100644
> --- a/hw/ppc/spapr_hcall.c
> +++ b/hw/ppc/spapr_hcall.c
> @@ -494,7 +494,7 @@ static target_ulong h_resize_hpt_prepare(PowerPCCPU *cpu,
>  return H_PARAMETER;
>  }
>  
> -current_ram_size = pc_existing_dimms_capacity(_fatal);
> +current_ram_size = ram_size + get_plugged_memory_size();

current_ram_size is initialized earlier in this function:

uint64_t current_ram_size = MACHINE(spapr)->ram_size;

which is is initialized to ram_size in vl.c. Why not doing:

current_ram_size += get_plugged_memory_size();

?

>  
>  /* We only allow the guest to allocate an HPT one order above what
>   * we'd normally give them (to stop a small guest claiming a huge

Re: [Qemu-devel] [PATCH] spapr: Correct RAM size calculation for HPT resizing

[Laurent Vivier]

On 10/10/2017 15:21, David Gibson wrote:
> In order to prevent the guest from forcing the allocation of large amounts
> of qemu memory (or host kernel memory, in the case of KVM HV), we limit
> the size of Hashed Page Table (HPT) it is allowed to allocated, based on
> its RAM size.
> 
> However, the current calculation is not correct: it only adds up the size
> of plugged memory, ignoring the base memory size.  This patch corrects it.
> 
> While we're there, use get_plugged_memory_size() instead of directly
> calling pc_existing_dimms_capacity().  The only difference is that it
> will abort on failure, which is right: a failure here indicates something
> wrong within qemu.
> 
> Signed-off-by: David Gibson 
> ---
>  hw/ppc/spapr_hcall.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/hw/ppc/spapr_hcall.c b/hw/ppc/spapr_hcall.c
> index 8d72bb7c1c..06af1b15c0 100644
> --- a/hw/ppc/spapr_hcall.c
> +++ b/hw/ppc/spapr_hcall.c
> @@ -494,7 +494,7 @@ static target_ulong h_resize_hpt_prepare(PowerPCCPU *cpu,
>  return H_PARAMETER;
>  }
>  
> -current_ram_size = pc_existing_dimms_capacity(_fatal);
> +current_ram_size = ram_size + get_plugged_memory_size();
>  
>  /* We only allow the guest to allocate an HPT one order above what
>   * we'd normally give them (to stop a small guest claiming a huge
> 

According to the content of qmp_query_memory_size_summary(), it's the
good way to compute the memory size...

Reviewed-by: Laurent Vivier 

[Qemu-devel] [PATCH] spapr: Correct RAM size calculation for HPT resizing

[David Gibson]

In order to prevent the guest from forcing the allocation of large amounts
of qemu memory (or host kernel memory, in the case of KVM HV), we limit
the size of Hashed Page Table (HPT) it is allowed to allocated, based on
its RAM size.

However, the current calculation is not correct: it only adds up the size
of plugged memory, ignoring the base memory size.  This patch corrects it.

While we're there, use get_plugged_memory_size() instead of directly
calling pc_existing_dimms_capacity().  The only difference is that it
will abort on failure, which is right: a failure here indicates something
wrong within qemu.

Signed-off-by: David Gibson 
---
 hw/ppc/spapr_hcall.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/hw/ppc/spapr_hcall.c b/hw/ppc/spapr_hcall.c
index 8d72bb7c1c..06af1b15c0 100644
--- a/hw/ppc/spapr_hcall.c
+++ b/hw/ppc/spapr_hcall.c
@@ -494,7 +494,7 @@ static target_ulong h_resize_hpt_prepare(PowerPCCPU *cpu,
 return H_PARAMETER;
 }
 
-current_ram_size = pc_existing_dimms_capacity(_fatal);
+current_ram_size = ram_size + get_plugged_memory_size();
 
 /* We only allow the guest to allocate an HPT one order above what
  * we'd normally give them (to stop a small guest claiming a huge
-- 
2.13.6