Re: [Xen-devel] [PATCH V2 1/25] VIOMMU: Add vIOMMU helper functions to create, destroy and query capabilities

[Tian, Kevin]

> From: Lan, Tianyu
> Sent: Friday, August 18, 2017 8:22 AM
> 
> This patch is to introduct an abstract layer for arch vIOMMU
> implementation
> to deal with requests from dom0. Arch vIOMMU code needs to provide
> callback
> to perform create, destroy and query capabilities operation.
> 
> Signed-off-by: Lan Tianyu 
> ---
>  xen/arch/x86/Kconfig |   1 +
>  xen/arch/x86/setup.c |   1 +
>  xen/common/Kconfig   |   3 +
>  xen/common/Makefile  |   1 +
>  xen/common/domain.c  |   3 +
>  xen/common/viommu.c  | 165
> +++
>  xen/include/xen/sched.h  |   2 +
>  xen/include/xen/viommu.h |  71 
>  8 files changed, 247 insertions(+)
>  create mode 100644 xen/common/viommu.c
>  create mode 100644 xen/include/xen/viommu.h
> 
> diff --git a/xen/arch/x86/Kconfig b/xen/arch/x86/Kconfig
> index 30c2769..1f1de96 100644
> --- a/xen/arch/x86/Kconfig
> +++ b/xen/arch/x86/Kconfig
> @@ -23,6 +23,7 @@ config X86
>   select HAS_PDX
>   select NUMA
>   select VGA
> + select VIOMMU
> 
>  config ARCH_DEFCONFIG
>   string
> diff --git a/xen/arch/x86/setup.c b/xen/arch/x86/setup.c
> index db5df69..68f1631 100644
> --- a/xen/arch/x86/setup.c
> +++ b/xen/arch/x86/setup.c
> @@ -1513,6 +1513,7 @@ void __init noreturn __start_xen(unsigned long
> mbi_p)
>  early_msi_init();
> 
>  iommu_setup();/* setup iommu if available */
> +viommu_setup();

start_xen is about physical bits, why placing viommu stuff here?

> 
>  smp_prepare_cpus(max_cpus);
> 
> diff --git a/xen/common/Kconfig b/xen/common/Kconfig
> index dc8e876..2ad2c8d 100644
> --- a/xen/common/Kconfig
> +++ b/xen/common/Kconfig
> @@ -49,6 +49,9 @@ config HAS_CHECKPOLICY
>   string
>   option env="XEN_HAS_CHECKPOLICY"
> 
> +config VIOMMU
> + bool
> +
>  config KEXEC
>   bool "kexec support"
>   default y
> diff --git a/xen/common/Makefile b/xen/common/Makefile
> index 26c5a64..852553d 100644
> --- a/xen/common/Makefile
> +++ b/xen/common/Makefile
> @@ -56,6 +56,7 @@ obj-y += time.o
>  obj-y += timer.o
>  obj-y += trace.o
>  obj-y += version.o
> +obj-$(CONFIG_VIOMMU) += viommu.o
>  obj-y += virtual_region.o
>  obj-y += vm_event.o
>  obj-y += vmap.o
> diff --git a/xen/common/domain.c b/xen/common/domain.c
> index b22aacc..d1f9b10 100644
> --- a/xen/common/domain.c
> +++ b/xen/common/domain.c
> @@ -396,6 +396,9 @@ struct domain *domain_create(domid_t domid,
> unsigned int domcr_flags,
>  spin_unlock(&domlist_update_lock);
>  }
> 
> +if ( (err = viommu_init_domain(d)) != 0 )
> +goto fail;
> +

though viommu framework is common, it's better to have arch
specific code invoke above function based on their own 
requirement (e.g. if any dependency required to enforce)

>  return d;
> 
>   fail:
> diff --git a/xen/common/viommu.c b/xen/common/viommu.c
> new file mode 100644
> index 000..6874d9f
> --- /dev/null
> +++ b/xen/common/viommu.c
> @@ -0,0 +1,165 @@
> +/*
> + * common/viommu.c
> + *
> + * Copyright (c) 2017 Intel Corporation
> + * Author: Lan Tianyu 
> + *
> + * This program is free software; you can redistribute it and/or modify it
> + * under the terms and conditions of the GNU General Public License,
> + * version 2, as published by the Free Software Foundation.
> + *
> + * This program is distributed in the hope it will be useful, but WITHOUT
> + * ANY WARRANTY; without even the implied warranty of
> MERCHANTABILITY or
> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
> License for
> + * more details.
> + *
> + * You should have received a copy of the GNU General Public License
> along with
> + * this program; If not, see .
> + */
> +
> +#include 
> +#include 
> +#include 
> +#include 
> +
> +bool __read_mostly opt_viommu;
> +boolean_param("viommu", opt_viommu);
> +
> +static spinlock_t type_list_lock;
> +static struct list_head type_list;
> +
> +struct viommu_type {
> +u64 type;

please add some comment what 'type' stands for.

> +struct viommu_ops *ops;
> +struct list_head node;
> +};
> +
> +int viommu_init_domain(struct domain *d)
> +{
> +d->viommu.nr_viommu = 0;
> +return 0;
> +}
> +
> +static struct viommu_type *viommu_get_type(u64 type)
> +{
> +struct viommu_type *viommu_type = NULL;
> +
> +spin_lock(&type_list_lock);
> +list_for_each_entry( viommu_type, &type_list, node )
> +{
> +if ( viommu_type->type == type )
> +{
> +spin_unlock(&type_list_lock);
> +return viommu_type;
> +}
> +}
> +spin_unlock(&type_list_lock);
> +
> +return NULL;
> +}
> +
> +int viommu_register_type(u64 type, struct viommu_ops * ops)
> +{
> +struct viommu_type *viommu_type = NULL;
> +
> +if ( !viommu_enabled() )
> +return -EINVAL;

shouldn't above be domain specific check?

Thanks
Kevin

___
Xen-devel mailing list
Xen-devel@lists.xe

Re: [Xen-devel] [PATCH V2 1/25] VIOMMU: Add vIOMMU helper functions to create, destroy and query capabilities

[Roger Pau Monné]

On Wed, Aug 23, 2017 at 03:10:48PM +0800, Lan Tianyu wrote:
> On 2017年08月22日 23:27, Roger Pau Monné wrote:
> > On Thu, Aug 17, 2017 at 08:22:16PM -0400, Lan Tianyu wrote:
> >> +int viommu_register_type(u64 type, struct viommu_ops * ops)
> >> +{
> >> +struct viommu_type *viommu_type = NULL;
> >> +
> >> +if ( !viommu_enabled() )
> >> +return -EINVAL;
> > 
> > ENODEV is maybe better here.
> 
> Will update.
> 
> > 
> >> +
> >> +if ( viommu_get_type(type) )
> >> +return -EEXIST;
> >> +
> >> +viommu_type = xzalloc(struct viommu_type);
> >> +if ( !viommu_type )
> >> +return -ENOMEM;
> >> +
> >> +viommu_type->type = type;
> >> +viommu_type->ops = ops;
> >> +
> >> +spin_lock(&type_list_lock);
> >> +list_add_tail(&viommu_type->node, &type_list);
> >> +spin_unlock(&type_list_lock);
> >> +
> >> +return 0;
> >> +}
> > 
> > Hm, I haven't seen the usage of this function, but from the looks of
> > it it seems like you want to use something similar to what's used by
> > the scheduler in order to register vIOMMU types.
> > 
> > See the logic around REGISTER_SCHEDULER in xen/sched-if.h.
> 
> Each vIOMMU devel model needs to call viommu_register_type() to register
> its vIOMMU type. Tool stack will pass viommu_type and vIOMMU abstract
> layer call vendor's vIOMMU callback to query vIOMMU capabilities, create
> and destroy() vIOMMU. We just need to maintain type list.
> REGISTER_SCHEDULER seems to heavy which reserve a scheduler array in the
> Xen data section.

I will let the maintainers comment, but I find it easier to use
something similar to REGISTER_SCHEDULER rather than have each IOMMU
type have it's initialization function hooked up in the init procedure
and calling viommu_register_type.

From an abstract point of view I don't really see how this is
different from a scheduler for example, the IOMMU types are fixed and
compiled into Xen, and they need to be initialized/registered at boot
time.

> >> +static inline bool viommu_enabled(void) { return opt_viommu; }
> > 
> > I think those static inline functions should also follow the coding
> > standard.
> 
> Yes, I am not sure what wrong here. Could you elaborate? Thanks.

IMHO they should be:

static inline bool viommu_enabled(void)
{
return opt_viommu;
}

Roger.

___
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

Re: [Xen-devel] [PATCH V2 1/25] VIOMMU: Add vIOMMU helper functions to create, destroy and query capabilities

[Lan Tianyu]

On 2017年08月22日 23:27, Roger Pau Monné wrote:
> On Thu, Aug 17, 2017 at 08:22:16PM -0400, Lan Tianyu wrote:
>> This patch is to introduct an abstract layer for arch vIOMMU implementation
>> to deal with requests from dom0. Arch vIOMMU code needs to provide callback
>> to perform create, destroy and query capabilities operation.
>>
>> Signed-off-by: Lan Tianyu 
>> ---
>>  xen/arch/x86/Kconfig |   1 +
>>  xen/arch/x86/setup.c |   1 +
>>  xen/common/Kconfig   |   3 +
>>  xen/common/Makefile  |   1 +
>>  xen/common/domain.c  |   3 +
>>  xen/common/viommu.c  | 165 
>> +++
>>  xen/include/xen/sched.h  |   2 +
>>  xen/include/xen/viommu.h |  71 
>>  8 files changed, 247 insertions(+)
>>  create mode 100644 xen/common/viommu.c
>>  create mode 100644 xen/include/xen/viommu.h
>>
>> diff --git a/xen/arch/x86/Kconfig b/xen/arch/x86/Kconfig
>> index 30c2769..1f1de96 100644
>> --- a/xen/arch/x86/Kconfig
>> +++ b/xen/arch/x86/Kconfig
>> @@ -23,6 +23,7 @@ config X86
>>  select HAS_PDX
>>  select NUMA
>>  select VGA
>> +select VIOMMU
>>  
>>  config ARCH_DEFCONFIG
>>  string
>> diff --git a/xen/arch/x86/setup.c b/xen/arch/x86/setup.c
>> index db5df69..68f1631 100644
>> --- a/xen/arch/x86/setup.c
>> +++ b/xen/arch/x86/setup.c
>> @@ -1513,6 +1513,7 @@ void __init noreturn __start_xen(unsigned long mbi_p)
>>  early_msi_init();
>>  
>>  iommu_setup();/* setup iommu if available */
>> +viommu_setup();
>>  
>>  smp_prepare_cpus(max_cpus);
>>  
>> diff --git a/xen/common/Kconfig b/xen/common/Kconfig
>> index dc8e876..2ad2c8d 100644
>> --- a/xen/common/Kconfig
>> +++ b/xen/common/Kconfig
>> @@ -49,6 +49,9 @@ config HAS_CHECKPOLICY
>>  string
>>  option env="XEN_HAS_CHECKPOLICY"
>>  
>> +config VIOMMU
>> +bool
>> +
>>  config KEXEC
>>  bool "kexec support"
>>  default y
>> diff --git a/xen/common/Makefile b/xen/common/Makefile
>> index 26c5a64..852553d 100644
>> --- a/xen/common/Makefile
>> +++ b/xen/common/Makefile
>> @@ -56,6 +56,7 @@ obj-y += time.o
>>  obj-y += timer.o
>>  obj-y += trace.o
>>  obj-y += version.o
>> +obj-$(CONFIG_VIOMMU) += viommu.o
>>  obj-y += virtual_region.o
>>  obj-y += vm_event.o
>>  obj-y += vmap.o
>> diff --git a/xen/common/domain.c b/xen/common/domain.c
>> index b22aacc..d1f9b10 100644
>> --- a/xen/common/domain.c
>> +++ b/xen/common/domain.c
>> @@ -396,6 +396,9 @@ struct domain *domain_create(domid_t domid, unsigned int 
>> domcr_flags,
>>  spin_unlock(&domlist_update_lock);
>>  }
>>  
>> +if ( (err = viommu_init_domain(d)) != 0 )
>> +goto fail;
>> +
>>  return d;
>>  
>>   fail:
>> diff --git a/xen/common/viommu.c b/xen/common/viommu.c
>> new file mode 100644
>> index 000..6874d9f
>> --- /dev/null
>> +++ b/xen/common/viommu.c
>> @@ -0,0 +1,165 @@
>> +/*
>> + * common/viommu.c
>> + * 
>> + * Copyright (c) 2017 Intel Corporation
>> + * Author: Lan Tianyu  
>> + *
>> + * This program is free software; you can redistribute it and/or modify it
>> + * under the terms and conditions of the GNU General Public License,
>> + * version 2, as published by the Free Software Foundation.
>> + *
>> + * This program is distributed in the hope it will be useful, but WITHOUT
>> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
>> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
>> + * more details.
>> + *
>> + * You should have received a copy of the GNU General Public License along 
>> with
>> + * this program; If not, see .
>> + */
>> +
>> +#include 
>> +#include 
>> +#include 
>> +#include 
>> +
>> +bool __read_mostly opt_viommu;
>> +boolean_param("viommu", opt_viommu);
>> +
>> +static spinlock_t type_list_lock;
> 
> static DEFINE_SPINLOCK(type_list_lock);
> 
>> +static struct list_head type_list;
> 
> static LIST_HEAD(type_list);
> 
>> +
>> +struct viommu_type {
>> +u64 type;
>> +struct viommu_ops *ops;
>> +struct list_head node;
>> +};
>> +
>> +int viommu_init_domain(struct domain *d)
>> +{
>> +d->viommu.nr_viommu = 0;
>> +return 0;
>> +}
> 
> If you don't use the viommu_info struct you can also get rid of this.
> The array entries will point to NULL which can be used to signal not
> initialized.

Yes, just check the memory  for struct domain will be set to all zero
after allocation.

> 
>> +
>> +static struct viommu_type *viommu_get_type(u64 type)
>> +{
>> +struct viommu_type *viommu_type = NULL;
>> +
>> +spin_lock(&type_list_lock);
>> +list_for_each_entry( viommu_type, &type_list, node )
>> +{
>> +if ( viommu_type->type == type )
>> +{
>> +spin_unlock(&type_list_lock);
>> +return viommu_type;
>> +}
>> +}
>> +spin_unlock(&type_list_lock);
>> +
>> +return NULL;
>> +}
>> +
>> +int viommu_register_type(u64 type, struct viommu_ops * ops)
>> +{
>> +struct viommu_t

Re: [Xen-devel] [PATCH V2 1/25] VIOMMU: Add vIOMMU helper functions to create, destroy and query capabilities

[Roger Pau Monné]

On Thu, Aug 17, 2017 at 08:22:16PM -0400, Lan Tianyu wrote:
> This patch is to introduct an abstract layer for arch vIOMMU implementation
> to deal with requests from dom0. Arch vIOMMU code needs to provide callback
> to perform create, destroy and query capabilities operation.
> 
> Signed-off-by: Lan Tianyu 
> ---
>  xen/arch/x86/Kconfig |   1 +
>  xen/arch/x86/setup.c |   1 +
>  xen/common/Kconfig   |   3 +
>  xen/common/Makefile  |   1 +
>  xen/common/domain.c  |   3 +
>  xen/common/viommu.c  | 165 
> +++
>  xen/include/xen/sched.h  |   2 +
>  xen/include/xen/viommu.h |  71 
>  8 files changed, 247 insertions(+)
>  create mode 100644 xen/common/viommu.c
>  create mode 100644 xen/include/xen/viommu.h
> 
> diff --git a/xen/arch/x86/Kconfig b/xen/arch/x86/Kconfig
> index 30c2769..1f1de96 100644
> --- a/xen/arch/x86/Kconfig
> +++ b/xen/arch/x86/Kconfig
> @@ -23,6 +23,7 @@ config X86
>   select HAS_PDX
>   select NUMA
>   select VGA
> + select VIOMMU
>  
>  config ARCH_DEFCONFIG
>   string
> diff --git a/xen/arch/x86/setup.c b/xen/arch/x86/setup.c
> index db5df69..68f1631 100644
> --- a/xen/arch/x86/setup.c
> +++ b/xen/arch/x86/setup.c
> @@ -1513,6 +1513,7 @@ void __init noreturn __start_xen(unsigned long mbi_p)
>  early_msi_init();
>  
>  iommu_setup();/* setup iommu if available */
> +viommu_setup();
>  
>  smp_prepare_cpus(max_cpus);
>  
> diff --git a/xen/common/Kconfig b/xen/common/Kconfig
> index dc8e876..2ad2c8d 100644
> --- a/xen/common/Kconfig
> +++ b/xen/common/Kconfig
> @@ -49,6 +49,9 @@ config HAS_CHECKPOLICY
>   string
>   option env="XEN_HAS_CHECKPOLICY"
>  
> +config VIOMMU
> + bool
> +
>  config KEXEC
>   bool "kexec support"
>   default y
> diff --git a/xen/common/Makefile b/xen/common/Makefile
> index 26c5a64..852553d 100644
> --- a/xen/common/Makefile
> +++ b/xen/common/Makefile
> @@ -56,6 +56,7 @@ obj-y += time.o
>  obj-y += timer.o
>  obj-y += trace.o
>  obj-y += version.o
> +obj-$(CONFIG_VIOMMU) += viommu.o
>  obj-y += virtual_region.o
>  obj-y += vm_event.o
>  obj-y += vmap.o
> diff --git a/xen/common/domain.c b/xen/common/domain.c
> index b22aacc..d1f9b10 100644
> --- a/xen/common/domain.c
> +++ b/xen/common/domain.c
> @@ -396,6 +396,9 @@ struct domain *domain_create(domid_t domid, unsigned int 
> domcr_flags,
>  spin_unlock(&domlist_update_lock);
>  }
>  
> +if ( (err = viommu_init_domain(d)) != 0 )
> +goto fail;
> +
>  return d;
>  
>   fail:
> diff --git a/xen/common/viommu.c b/xen/common/viommu.c
> new file mode 100644
> index 000..6874d9f
> --- /dev/null
> +++ b/xen/common/viommu.c
> @@ -0,0 +1,165 @@
> +/*
> + * common/viommu.c
> + * 
> + * Copyright (c) 2017 Intel Corporation
> + * Author: Lan Tianyu  
> + *
> + * This program is free software; you can redistribute it and/or modify it
> + * under the terms and conditions of the GNU General Public License,
> + * version 2, as published by the Free Software Foundation.
> + *
> + * This program is distributed in the hope it will be useful, but WITHOUT
> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
> + * more details.
> + *
> + * You should have received a copy of the GNU General Public License along 
> with
> + * this program; If not, see .
> + */
> +
> +#include 
> +#include 
> +#include 
> +#include 
> +
> +bool __read_mostly opt_viommu;
> +boolean_param("viommu", opt_viommu);
> +
> +static spinlock_t type_list_lock;

static DEFINE_SPINLOCK(type_list_lock);

> +static struct list_head type_list;

static LIST_HEAD(type_list);

> +
> +struct viommu_type {
> +u64 type;
> +struct viommu_ops *ops;
> +struct list_head node;
> +};
> +
> +int viommu_init_domain(struct domain *d)
> +{
> +d->viommu.nr_viommu = 0;
> +return 0;
> +}

If you don't use the viommu_info struct you can also get rid of this.
The array entries will point to NULL which can be used to signal not
initialized.

> +
> +static struct viommu_type *viommu_get_type(u64 type)
> +{
> +struct viommu_type *viommu_type = NULL;
> +
> +spin_lock(&type_list_lock);
> +list_for_each_entry( viommu_type, &type_list, node )
> +{
> +if ( viommu_type->type == type )
> +{
> +spin_unlock(&type_list_lock);
> +return viommu_type;
> +}
> +}
> +spin_unlock(&type_list_lock);
> +
> +return NULL;
> +}
> +
> +int viommu_register_type(u64 type, struct viommu_ops * ops)
> +{
> +struct viommu_type *viommu_type = NULL;
> +
> +if ( !viommu_enabled() )
> +return -EINVAL;

ENODEV is maybe better here.

> +
> +if ( viommu_get_type(type) )
> +return -EEXIST;
> +
> +viommu_type = xzalloc(struct viommu_type);
> +if ( !viommu_type )
> +return -ENOMEM;
> +
> +

Re: [Xen-devel] [PATCH V2 1/25] VIOMMU: Add vIOMMU helper functions to create, destroy and query capabilities

[Wei Liu]

On Thu, Aug 17, 2017 at 08:22:16PM -0400, Lan Tianyu wrote:
> diff --git a/xen/common/domain.c b/xen/common/domain.c
> index b22aacc..d1f9b10 100644
> --- a/xen/common/domain.c
> +++ b/xen/common/domain.c
> @@ -396,6 +396,9 @@ struct domain *domain_create(domid_t domid, unsigned int 
> domcr_flags,
>  spin_unlock(&domlist_update_lock);
>  }
>  
> +if ( (err = viommu_init_domain(d)) != 0 )
> +goto fail;
> +

Where is the code to destroy viommu during domain destruction?

I suppose you will need a viommu_destroy_domain and call it somewhere in
complete_domain_destroy.

> +
> +#include 
> +#include 
> +#include 
> +#include 
> +
> +bool __read_mostly opt_viommu;
> +boolean_param("viommu", opt_viommu);

Missing patch to xen command line option doc.

> +
> +static spinlock_t type_list_lock;
> +static struct list_head type_list;
> +
> +struct viommu_type {
> +u64 type;

uintXX_t here and in all other places please.

[...]
> +
> +static int viommu_create(struct domain *d, u64 type, u64 base_address,
> +u64 length, u64 caps)
> +{
> +struct viommu_info *info = &d->viommu;
> +struct viommu *viommu;
> +struct viommu_type *viommu_type = NULL;
> +int rc;
> +
> +viommu_type = viommu_get_type(type);
> +if ( !viommu_type )
> +return -EINVAL;
> +
> +if ( info->nr_viommu >= NR_VIOMMU_PER_DOMAIN

E2BIG for this?

> +|| !viommu_type->ops || !viommu_type->ops->create )
> +return -EINVAL;
> +
> +viommu = xzalloc(struct viommu);
> +if ( !viommu )
> +return -ENOMEM;
> +
[...]
> diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h
> index 6673b27..98a965a 100644
> --- a/xen/include/xen/sched.h
> +++ b/xen/include/xen/sched.h
> @@ -21,6 +21,7 @@
>  #include 
>  #include 
>  #include 
> +#include 
>  #include 
>  #include 
>  #include 
> @@ -477,6 +478,7 @@ struct domain
>  /* vNUMA topology accesses are protected by rwlock. */
>  rwlock_t vnuma_rwlock;
>  struct vnuma_info *vnuma;

Please add a new line here.

> +struct viommu_info viommu;
>  
>  /* Common monitor options */
>  struct {
> diff --git a/xen/include/xen/viommu.h b/xen/include/xen/viommu.h
> new file mode 100644
> index 000..506ea54
> --- /dev/null
> +++ b/xen/include/xen/viommu.h
> @@ -0,0 +1,71 @@
> +/*
> + * include/xen/viommu.h
> + *
> + * Copyright (c) 2017, Intel Corporation
> + * Author: Lan Tianyu  
> + *
> + * This program is free software; you can redistribute it and/or modify it
> + * under the terms and conditions of the GNU General Public License,
> + * version 2, as published by the Free Software Foundation.
> + *
> + * This program is distributed in the hope it will be useful, but WITHOUT
> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
> + * more details.
> + *
> + * You should have received a copy of the GNU General Public License along 
> with
> + * this program; If not, see .
> + *
> + */
> +#ifndef __XEN_VIOMMU_H__
> +#define __XEN_VIOMMU_H__
> +
> +#define NR_VIOMMU_PER_DOMAIN 1
> +
> +struct viommu;
> +
> +struct viommu_ops {
> +u64 (*query_caps)(struct domain *d);
> +int (*create)(struct domain *d, struct viommu *viommu);
> +int (*destroy)(struct viommu *viommu);
> +};
> +
> +struct viommu {
> +u64 base_address;
> +u64 length;
> +u64 caps;
> +u32 viommu_id;
> +const struct viommu_ops *ops;
> +void *priv;
> +};
> +
> +struct viommu_info {
> +u32 nr_viommu;

unsigned int

> +struct viommu *viommu[NR_VIOMMU_PER_DOMAIN]; /* viommu array*/
> +};
> +
> +#ifdef CONFIG_VIOMMU
> +extern bool_t opt_viommu;

bool

___
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

Re: [Xen-devel] [PATCH V2 1/25] VIOMMU: Add vIOMMU helper functions to create, destroy and query capabilities

[Lan Tianyu]

On 2017年08月18日 16:41, Jan Beulich wrote:
 On 18.08.17 at 02:22,  wrote:
>> This patch is to introduct an abstract layer for arch vIOMMU implementation
>> to deal with requests from dom0. Arch vIOMMU code needs to provide callback
>> to perform create, destroy and query capabilities operation.
>>
>> Signed-off-by: Lan Tianyu 
> 
> What is this? It looks to be a standalone patch when there was
> a V2 series a couple of days ago with 1/25 titled "DOMCTL:
> Introduce new DOMCTL commands for vIOMMU support". I'm
> confused.
> 
> Jan
> 
Hi Jan:
Sorry. Missed this patch in this patchset and this should be the first
one. I will send v3 with some fixs.
-- 
Best regards
Tianyu Lan

___
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

Re: [Xen-devel] [PATCH V2 1/25] VIOMMU: Add vIOMMU helper functions to create, destroy and query capabilities

[Jan Beulich]

>>> On 18.08.17 at 02:22,  wrote:
> This patch is to introduct an abstract layer for arch vIOMMU implementation
> to deal with requests from dom0. Arch vIOMMU code needs to provide callback
> to perform create, destroy and query capabilities operation.
> 
> Signed-off-by: Lan Tianyu 

What is this? It looks to be a standalone patch when there was
a V2 series a couple of days ago with 1/25 titled "DOMCTL:
Introduce new DOMCTL commands for vIOMMU support". I'm
confused.

Jan


___
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

[Xen-devel] [PATCH V2 1/25] VIOMMU: Add vIOMMU helper functions to create, destroy and query capabilities

[Lan Tianyu]

This patch is to introduct an abstract layer for arch vIOMMU implementation
to deal with requests from dom0. Arch vIOMMU code needs to provide callback
to perform create, destroy and query capabilities operation.

Signed-off-by: Lan Tianyu 
---
 xen/arch/x86/Kconfig |   1 +
 xen/arch/x86/setup.c |   1 +
 xen/common/Kconfig   |   3 +
 xen/common/Makefile  |   1 +
 xen/common/domain.c  |   3 +
 xen/common/viommu.c  | 165 +++
 xen/include/xen/sched.h  |   2 +
 xen/include/xen/viommu.h |  71 
 8 files changed, 247 insertions(+)
 create mode 100644 xen/common/viommu.c
 create mode 100644 xen/include/xen/viommu.h

diff --git a/xen/arch/x86/Kconfig b/xen/arch/x86/Kconfig
index 30c2769..1f1de96 100644
--- a/xen/arch/x86/Kconfig
+++ b/xen/arch/x86/Kconfig
@@ -23,6 +23,7 @@ config X86
select HAS_PDX
select NUMA
select VGA
+   select VIOMMU
 
 config ARCH_DEFCONFIG
string
diff --git a/xen/arch/x86/setup.c b/xen/arch/x86/setup.c
index db5df69..68f1631 100644
--- a/xen/arch/x86/setup.c
+++ b/xen/arch/x86/setup.c
@@ -1513,6 +1513,7 @@ void __init noreturn __start_xen(unsigned long mbi_p)
 early_msi_init();
 
 iommu_setup();/* setup iommu if available */
+viommu_setup();
 
 smp_prepare_cpus(max_cpus);
 
diff --git a/xen/common/Kconfig b/xen/common/Kconfig
index dc8e876..2ad2c8d 100644
--- a/xen/common/Kconfig
+++ b/xen/common/Kconfig
@@ -49,6 +49,9 @@ config HAS_CHECKPOLICY
string
option env="XEN_HAS_CHECKPOLICY"
 
+config VIOMMU
+   bool
+
 config KEXEC
bool "kexec support"
default y
diff --git a/xen/common/Makefile b/xen/common/Makefile
index 26c5a64..852553d 100644
--- a/xen/common/Makefile
+++ b/xen/common/Makefile
@@ -56,6 +56,7 @@ obj-y += time.o
 obj-y += timer.o
 obj-y += trace.o
 obj-y += version.o
+obj-$(CONFIG_VIOMMU) += viommu.o
 obj-y += virtual_region.o
 obj-y += vm_event.o
 obj-y += vmap.o
diff --git a/xen/common/domain.c b/xen/common/domain.c
index b22aacc..d1f9b10 100644
--- a/xen/common/domain.c
+++ b/xen/common/domain.c
@@ -396,6 +396,9 @@ struct domain *domain_create(domid_t domid, unsigned int 
domcr_flags,
 spin_unlock(&domlist_update_lock);
 }
 
+if ( (err = viommu_init_domain(d)) != 0 )
+goto fail;
+
 return d;
 
  fail:
diff --git a/xen/common/viommu.c b/xen/common/viommu.c
new file mode 100644
index 000..6874d9f
--- /dev/null
+++ b/xen/common/viommu.c
@@ -0,0 +1,165 @@
+/*
+ * common/viommu.c
+ * 
+ * Copyright (c) 2017 Intel Corporation
+ * Author: Lan Tianyu  
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; If not, see .
+ */
+
+#include 
+#include 
+#include 
+#include 
+
+bool __read_mostly opt_viommu;
+boolean_param("viommu", opt_viommu);
+
+static spinlock_t type_list_lock;
+static struct list_head type_list;
+
+struct viommu_type {
+u64 type;
+struct viommu_ops *ops;
+struct list_head node;
+};
+
+int viommu_init_domain(struct domain *d)
+{
+d->viommu.nr_viommu = 0;
+return 0;
+}
+
+static struct viommu_type *viommu_get_type(u64 type)
+{
+struct viommu_type *viommu_type = NULL;
+
+spin_lock(&type_list_lock);
+list_for_each_entry( viommu_type, &type_list, node )
+{
+if ( viommu_type->type == type )
+{
+spin_unlock(&type_list_lock);
+return viommu_type;
+}
+}
+spin_unlock(&type_list_lock);
+
+return NULL;
+}
+
+int viommu_register_type(u64 type, struct viommu_ops * ops)
+{
+struct viommu_type *viommu_type = NULL;
+
+if ( !viommu_enabled() )
+return -EINVAL;
+
+if ( viommu_get_type(type) )
+return -EEXIST;
+
+viommu_type = xzalloc(struct viommu_type);
+if ( !viommu_type )
+return -ENOMEM;
+
+viommu_type->type = type;
+viommu_type->ops = ops;
+
+spin_lock(&type_list_lock);
+list_add_tail(&viommu_type->node, &type_list);
+spin_unlock(&type_list_lock);
+
+return 0;
+}
+
+static int viommu_create(struct domain *d, u64 type, u64 base_address,
+u64 length, u64 caps)
+{
+struct viommu_info *info = &d->viommu;
+struct viommu *viommu;
+struct viommu_type *viommu_type = NULL;
+int rc;
+
+viommu_type = viommu_get_type(type);
+if ( !viommu_type )
+return -EINVAL;
+
+if ( info->nr_viommu >= NR_VIOMMU_PER_DOMAIN
+|| !viommu_type->ops || !viommu_type->ops-