[PATCH] hvf: Avoid mapping regions < PAGE_SIZE as ram

[PATCH] hvf: Avoid mapping regions < PAGE_SIZE as ram

[Alexander Graf]

HVF has generic memory listener code that adds all RAM regions as HVF RAM
regions. However, HVF can only handle page aligned, page granule regions.

So let's ignore regions that are not page aligned and sized. They will be
trapped as MMIO instead.

Signed-off-by: Alexander Graf <agraf@csgraf.de>
---
 accel/hvf/hvf-accel-ops.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/accel/hvf/hvf-accel-ops.c b/accel/hvf/hvf-accel-ops.c
index 6bf319d34c..090155853a 100644
--- a/accel/hvf/hvf-accel-ops.c
+++ b/accel/hvf/hvf-accel-ops.c
@@ -135,6 +135,12 @@ static void hvf_set_phys_mem(MemoryRegionSection *section, bool add)
         }
     }
 
+    if (int128_get64(section->size) & (qemu_real_host_page_size - 1) ||
+        section->offset_within_address_space & (qemu_real_host_page_size - 1)) {
+        /* Not page aligned, so we can not map as RAM */
+        add = false;
+    }
+
     mem = hvf_find_overlap_slot(
             section->offset_within_address_space,
             int128_get64(section->size));
-- 
2.30.1 (Apple Git-130)

Re: [PATCH] hvf: Avoid mapping regions < PAGE_SIZE as ram

[Philippe Mathieu-Daudé]

Hi Alex,

On 10/25/21 10:25, Alexander Graf wrote:
> HVF has generic memory listener code that adds all RAM regions as HVF RAM
> regions. However, HVF can only handle page aligned, page granule regions.
> 
> So let's ignore regions that are not page aligned and sized. They will be
> trapped as MMIO instead.
> 
> Signed-off-by: Alexander Graf <agraf@csgraf.de>
> ---
>  accel/hvf/hvf-accel-ops.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/accel/hvf/hvf-accel-ops.c b/accel/hvf/hvf-accel-ops.c
> index 6bf319d34c..090155853a 100644
> --- a/accel/hvf/hvf-accel-ops.c
> +++ b/accel/hvf/hvf-accel-ops.c
> @@ -135,6 +135,12 @@ static void hvf_set_phys_mem(MemoryRegionSection *section, bool add)
>          }
>      }
>  
> +    if (int128_get64(section->size) & (qemu_real_host_page_size - 1) ||
> +        section->offset_within_address_space & (qemu_real_host_page_size - 1)) {

Could we use QEMU_IS_ALIGNED() instead?

> +        /* Not page aligned, so we can not map as RAM */
> +        add = false;
> +    }
> +
>      mem = hvf_find_overlap_slot(
>              section->offset_within_address_space,
>              int128_get64(section->size));
> 

Re: [PATCH] hvf: Avoid mapping regions < PAGE_SIZE as ram

[Paolo Bonzini]

On 25/10/21 10:25, Alexander Graf wrote:
> HVF has generic memory listener code that adds all RAM regions as HVF RAM
> regions. However, HVF can only handle page aligned, page granule regions.
> 
> So let's ignore regions that are not page aligned and sized. They will be
> trapped as MMIO instead.
> 
> Signed-off-by: Alexander Graf <agraf@csgraf.de>
> ---
>   accel/hvf/hvf-accel-ops.c | 6 ++++++
>   1 file changed, 6 insertions(+)
> 
> diff --git a/accel/hvf/hvf-accel-ops.c b/accel/hvf/hvf-accel-ops.c
> index 6bf319d34c..090155853a 100644
> --- a/accel/hvf/hvf-accel-ops.c
> +++ b/accel/hvf/hvf-accel-ops.c
> @@ -135,6 +135,12 @@ static void hvf_set_phys_mem(MemoryRegionSection *section, bool add)
>           }
>       }
>   
> +    if (int128_get64(section->size) & (qemu_real_host_page_size - 1) ||
> +        section->offset_within_address_space & (qemu_real_host_page_size - 1)) {
> +        /* Not page aligned, so we can not map as RAM */
> +        add = false;
> +    }
> +
>       mem = hvf_find_overlap_slot(
>               section->offset_within_address_space,
>               int128_get64(section->size));
> 

Queued, thanks.

Paolo

Re: [PATCH] hvf: Avoid mapping regions < PAGE_SIZE as ram

[Alexander Graf]

On 25.10.21 19:11, Paolo Bonzini wrote:
> On 25/10/21 10:25, Alexander Graf wrote:
>> HVF has generic memory listener code that adds all RAM regions as HVF 
>> RAM
>> regions. However, HVF can only handle page aligned, page granule 
>> regions.
>>
>> So let's ignore regions that are not page aligned and sized. They 
>> will be
>> trapped as MMIO instead.
>>
>> Signed-off-by: Alexander Graf <agraf@csgraf.de>
>> ---
>>   accel/hvf/hvf-accel-ops.c | 6 ++++++
>>   1 file changed, 6 insertions(+)
>>
>> diff --git a/accel/hvf/hvf-accel-ops.c b/accel/hvf/hvf-accel-ops.c
>> index 6bf319d34c..090155853a 100644
>> --- a/accel/hvf/hvf-accel-ops.c
>> +++ b/accel/hvf/hvf-accel-ops.c
>> @@ -135,6 +135,12 @@ static void hvf_set_phys_mem(MemoryRegionSection 
>> *section, bool add)
>>           }
>>       }
>>   +    if (int128_get64(section->size) & (qemu_real_host_page_size - 
>> 1) ||
>> +        section->offset_within_address_space & 
>> (qemu_real_host_page_size - 1)) {
>> +        /* Not page aligned, so we can not map as RAM */
>> +        add = false;
>> +    }
>> +
>>       mem = hvf_find_overlap_slot(
>>               section->offset_within_address_space,
>>               int128_get64(section->size));
>>
>
> Queued, thanks.


You probably want v2 instead :)

Alex

Re: [PATCH] hvf: Avoid mapping regions < PAGE_SIZE as ram

[Paolo Bonzini]

On 25/10/21 21:10, Alexander Graf wrote:
>>>           }
>>>       }
>>>   +    if (int128_get64(section->size) & (qemu_real_host_page_size - 
>>> 1) ||
>>> +        section->offset_within_address_space & 
>>> (qemu_real_host_page_size - 1)) {
>>> +        /* Not page aligned, so we can not map as RAM */
>>> +        add = false;
>>> +    }
>>> +
>>>       mem = hvf_find_overlap_slot(
>>>               section->offset_within_address_space,
>>>               int128_get64(section->size));
>>>
>>
>> Queued, thanks.
> 
> 
> You probably want v2 instead :)

That's actually what I had applied. :)

Paolo