Re: [Xen-devel] [PATCH] [RFC V2] xen/arm: Restrict "p2m_ipa_bits" according to the IOMMU requirements
Hi,
On 9/11/19 5:34 PM, Oleksandr wrote:
On 10.09.19 21:55, Julien Grall wrote:
Hi,
Hi Julien
On 9/10/19 5:24 PM, Oleksandr wrote:
On 10.09.19 18:11, Julien Grall wrote:
Hi Oleksandr,
Hi, Julien
On 8/23/19 8:34 PM, Oleksandr Tyshchenko wrote:
From: Oleksandr Tyshchenko
There is a strict requirement for the IOMMU which wants to share
the P2M table with the CPU. The IOMMU's Stage-2 input size must be
equal
to the P2M IPA size. It is not a problem when the IOMMU can support
all values the CPU supports. In that case, the IOMMU driver would just
use any "p2m_ipa_bits" value as is. But, there are cases when not.
In order to make P2M sharing possible on the platforms which
IPMMUs have a limitation in maximum Stage-2 input size introduce
the following logic.
First initialize the IOMMU subsystem and gather requirements regarding
the maximum IPA bits supported by each IOMMU device to figure out
the minimum value among them. In the P2M code, take into the account
the IOMMU requirements and choose suitable "pa_range" according
to the restricted "p2m_ipa_bits".
As I pointed in the previous version, all the code you modify is
arm64 specific. For arm32, the number of IPA bits is
hardcoded. So if you modify p2m_ipa_bits, you would end up to
misconfigure VTCR.
In other words, for Arm32, you need to check p2m_ipa_bits is at
least 40-bits before overriding it.
But, all modifications with p2m_ipa_bits are done before
setup_virt_paging(), where, actually, the p2m_ipa_bits is hard-coded
to 40 bits. How can we end up misconfiguring VTCR for ARM32? Or I
really missed something?
Sorry if I wasn't cleared, I meant the VTCR for the IOMMU. You would
end up to configure with a value that is bigger than what it can support.
I am ok if you don't restrict the p2m_ipa_bits and just fail. The
point is to notify the user ASAP rather than allowing to continue.
This would make the behavior similar to the current implementation
(although the error would be different).
So, in IOMMU driver we should check if IOMMU is able to support at least
40-bit IPA before trying to restrict. If yes, then go ahead, but if no,
then just fail. Correct?
There are no need to do this in the IOMU drivers. And I actually don't
such check in the drivers.
This is the similar problem to half initialized IOMMU. If the user
requests IOMMU and doesn't work, then we don't want to continue an
panic. Such check can be done directly in the function setup_virt_paging().
+{
+ /*
+ * Calculate the minimum of the maximum IPA bits that any IOMMU
+ * can support.
+ */
+ if ( iommu_ipa_bits < p2m_ipa_bits )
+ p2m_ipa_bits = iommu_ipa_bits;
+}
+
/* VTCR value to be configured by all CPUs. Set only once by the
boot CPU */
static uint32_t __read_mostly vtcr;
@@ -1966,10 +1977,28 @@ void __init setup_virt_paging(void)
[7] = { 0 } /* Invalid */
};
- unsigned int cpu;
+ unsigned int i, cpu;
unsigned int pa_range = 0x10; /* Larger than any possible
value */
bool vmid_8_bit = false;
+ if ( iommu_enabled )
Could we make this IOMMU-agnostic? The main reason to convert from
p2m_ipa_bits to pa_range is to cater the rest of the code.
But we could rework the code to do the computation with p2m_ipa_bits
and then look-up for the pa_range.
I am afraid, I don't completely understand your idea of making this
IOMMU-agnostic and what I should do...
Roughly what you are doing today is:
if ( iommu_enabled )
pa_range = find_pa_range_from_p2m_bits().
for_each_cpu()
if ( cpu.pa_range < pa_range )
pa_range = cpu.pa_range
What you could do is:
Thank you for the clarification. I think I understand your idea.
But ...
for_ech_cpu()
if ( p2m_ipa_bits < pa_range_info[cpu.pa_range].pabits )
Probably you meant ">" here?
Yes.
p2m_ipa_bits = pa_range_info[cpu.pa_range].pabits;
pa_range = find_pa_range_from_p2m_bits();
/* Check validity */
Cheers,
--
Julien Grall
___
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH] [RFC V2] xen/arm: Restrict "p2m_ipa_bits" according to the IOMMU requirements
On 10.09.19 21:55, Julien Grall wrote:
Hi,
Hi Julien
On 9/10/19 5:24 PM, Oleksandr wrote:
On 10.09.19 18:11, Julien Grall wrote:
Hi Oleksandr,
Hi, Julien
On 8/23/19 8:34 PM, Oleksandr Tyshchenko wrote:
From: Oleksandr Tyshchenko
There is a strict requirement for the IOMMU which wants to share
the P2M table with the CPU. The IOMMU's Stage-2 input size must be
equal
to the P2M IPA size. It is not a problem when the IOMMU can support
all values the CPU supports. In that case, the IOMMU driver would just
use any "p2m_ipa_bits" value as is. But, there are cases when not.
In order to make P2M sharing possible on the platforms which
IPMMUs have a limitation in maximum Stage-2 input size introduce
the following logic.
First initialize the IOMMU subsystem and gather requirements regarding
the maximum IPA bits supported by each IOMMU device to figure out
the minimum value among them. In the P2M code, take into the account
the IOMMU requirements and choose suitable "pa_range" according
to the restricted "p2m_ipa_bits".
As I pointed in the previous version, all the code you modify is
arm64 specific. For arm32, the number of IPA bits is
hardcoded. So if you modify p2m_ipa_bits, you would end up to
misconfigure VTCR.
In other words, for Arm32, you need to check p2m_ipa_bits is at
least 40-bits before overriding it.
But, all modifications with p2m_ipa_bits are done before
setup_virt_paging(), where, actually, the p2m_ipa_bits is hard-coded
to 40 bits. How can we end up misconfiguring VTCR for ARM32? Or I
really missed something?
Sorry if I wasn't cleared, I meant the VTCR for the IOMMU. You would
end up to configure with a value that is bigger than what it can support.
I am ok if you don't restrict the p2m_ipa_bits and just fail. The
point is to notify the user ASAP rather than allowing to continue.
This would make the behavior similar to the current implementation
(although the error would be different).
So, in IOMMU driver we should check if IOMMU is able to support at least
40-bit IPA before trying to restrict. If yes, then go ahead, but if no,
then just fail. Correct?
+{
+ /*
+ * Calculate the minimum of the maximum IPA bits that any IOMMU
+ * can support.
+ */
+ if ( iommu_ipa_bits < p2m_ipa_bits )
+ p2m_ipa_bits = iommu_ipa_bits;
+}
+
/* VTCR value to be configured by all CPUs. Set only once by the
boot CPU */
static uint32_t __read_mostly vtcr;
@@ -1966,10 +1977,28 @@ void __init setup_virt_paging(void)
[7] = { 0 } /* Invalid */
};
- unsigned int cpu;
+ unsigned int i, cpu;
unsigned int pa_range = 0x10; /* Larger than any possible
value */
bool vmid_8_bit = false;
+ if ( iommu_enabled )
Could we make this IOMMU-agnostic? The main reason to convert from
p2m_ipa_bits to pa_range is to cater the rest of the code.
But we could rework the code to do the computation with p2m_ipa_bits
and then look-up for the pa_range.
I am afraid, I don't completely understand your idea of making this
IOMMU-agnostic and what I should do...
Roughly what you are doing today is:
if ( iommu_enabled )
pa_range = find_pa_range_from_p2m_bits().
for_each_cpu()
if ( cpu.pa_range < pa_range )
pa_range = cpu.pa_range
What you could do is:
Thank you for the clarification. I think I understand your idea.
But ...
for_ech_cpu()
if ( p2m_ipa_bits < pa_range_info[cpu.pa_range].pabits )
Probably you meant ">" here?
p2m_ipa_bits = pa_range_info[cpu.pa_range].pabits;
pa_range = find_pa_range_from_p2m_bits();
/* Check validity */
--
Regards,
Oleksandr Tyshchenko
___
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH] [RFC V2] xen/arm: Restrict "p2m_ipa_bits" according to the IOMMU requirements
Hi,
On 9/10/19 5:24 PM, Oleksandr wrote:
On 10.09.19 18:11, Julien Grall wrote:
Hi Oleksandr,
Hi, Julien
On 8/23/19 8:34 PM, Oleksandr Tyshchenko wrote:
From: Oleksandr Tyshchenko
There is a strict requirement for the IOMMU which wants to share
the P2M table with the CPU. The IOMMU's Stage-2 input size must be equal
to the P2M IPA size. It is not a problem when the IOMMU can support
all values the CPU supports. In that case, the IOMMU driver would just
use any "p2m_ipa_bits" value as is. But, there are cases when not.
In order to make P2M sharing possible on the platforms which
IPMMUs have a limitation in maximum Stage-2 input size introduce
the following logic.
First initialize the IOMMU subsystem and gather requirements regarding
the maximum IPA bits supported by each IOMMU device to figure out
the minimum value among them. In the P2M code, take into the account
the IOMMU requirements and choose suitable "pa_range" according
to the restricted "p2m_ipa_bits".
As I pointed in the previous version, all the code you modify is arm64
specific. For arm32, the number of IPA bits is
hardcoded. So if you modify p2m_ipa_bits, you would end up to
misconfigure VTCR.
In other words, for Arm32, you need to check p2m_ipa_bits is at least
40-bits before overriding it.
But, all modifications with p2m_ipa_bits are done before
setup_virt_paging(), where, actually, the p2m_ipa_bits is hard-coded to
40 bits. How can we end up misconfiguring VTCR for ARM32? Or I really
missed something?
Sorry if I wasn't cleared, I meant the VTCR for the IOMMU. You would end
up to configure with a value that is bigger than what it can support.
I am ok if you don't restrict the p2m_ipa_bits and just fail. The point
is to notify the user ASAP rather than allowing to continue.
This would make the behavior similar to the current implementation
(although the error would be different).
[...]
+{
+ /*
+ * Calculate the minimum of the maximum IPA bits that any IOMMU
+ * can support.
+ */
+ if ( iommu_ipa_bits < p2m_ipa_bits )
+ p2m_ipa_bits = iommu_ipa_bits;
+}
+
/* VTCR value to be configured by all CPUs. Set only once by the
boot CPU */
static uint32_t __read_mostly vtcr;
@@ -1966,10 +1977,28 @@ void __init setup_virt_paging(void)
[7] = { 0 } /* Invalid */
};
- unsigned int cpu;
+ unsigned int i, cpu;
unsigned int pa_range = 0x10; /* Larger than any possible value */
bool vmid_8_bit = false;
+ if ( iommu_enabled )
Could we make this IOMMU-agnostic? The main reason to convert from
p2m_ipa_bits to pa_range is to cater the rest of the code.
But we could rework the code to do the computation with p2m_ipa_bits
and then look-up for the pa_range.
I am afraid, I don't completely understand your idea of making this
IOMMU-agnostic and what I should do...
Roughly what you are doing today is:
if ( iommu_enabled )
pa_range = find_pa_range_from_p2m_bits().
for_each_cpu()
if ( cpu.pa_range < pa_range )
pa_range = cpu.pa_range
What you could do is:
for_ech_cpu()
if ( p2m_ipa_bits < pa_range_info[cpu.pa_range].pabits )
p2m_ipa_bits = pa_range_info[cpu.pa_range].pabits;
pa_range = find_pa_range_from_p2m_bits();
/* Check validity */
...
Cheers,
--
Julien Grall
___
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH] [RFC V2] xen/arm: Restrict "p2m_ipa_bits" according to the IOMMU requirements
On 10.09.19 18:11, Julien Grall wrote:
Hi Oleksandr,
Hi, Julien
On 8/23/19 8:34 PM, Oleksandr Tyshchenko wrote:
From: Oleksandr Tyshchenko
There is a strict requirement for the IOMMU which wants to share
the P2M table with the CPU. The IOMMU's Stage-2 input size must be equal
to the P2M IPA size. It is not a problem when the IOMMU can support
all values the CPU supports. In that case, the IOMMU driver would just
use any "p2m_ipa_bits" value as is. But, there are cases when not.
In order to make P2M sharing possible on the platforms which
IPMMUs have a limitation in maximum Stage-2 input size introduce
the following logic.
First initialize the IOMMU subsystem and gather requirements regarding
the maximum IPA bits supported by each IOMMU device to figure out
the minimum value among them. In the P2M code, take into the account
the IOMMU requirements and choose suitable "pa_range" according
to the restricted "p2m_ipa_bits".
As I pointed in the previous version, all the code you modify is arm64
specific. For arm32, the number of IPA bits is
hardcoded. So if you modify p2m_ipa_bits, you would end up to
misconfigure VTCR.
In other words, for Arm32, you need to check p2m_ipa_bits is at least
40-bits before overriding it.
But, all modifications with p2m_ipa_bits are done before
setup_virt_paging(), where, actually, the p2m_ipa_bits is hard-coded to
40 bits. How can we end up misconfiguring VTCR for ARM32? Or I really
missed something?
Signed-off-by: Oleksandr Tyshchenko
CC: Julien Grall
---
Still RFC:
1. Patch assumes that IPMMU support is already in.
2. Not checked for the SMMU.
Changes since RFC V1 [1]:
- Don't update p2m_ipa_bits by the IOMMU drivers directly,
introduce p2m_restrict_ipa_bits()
- Clarify patch subject/description
- Add more comments to code
- Check for equivalent "pabits" in setup_virt_paging()
- Remove ASSERTs from the SMMU and IPMMU drivers
[1]
https://lists.xenproject.org/archives/html/xen-devel/2019-08/msg02078.html
---
xen/arch/arm/p2m.c | 33
++--
xen/arch/arm/setup.c | 11 +--
xen/drivers/passthrough/arm/ipmmu-vmsa.c | 19 --
xen/drivers/passthrough/arm/smmu.c | 16
xen/include/asm-arm/p2m.h | 8
5 files changed, 60 insertions(+), 27 deletions(-)
diff --git a/xen/arch/arm/p2m.c b/xen/arch/arm/p2m.c
index 2374e92..f742d9c 100644
--- a/xen/arch/arm/p2m.c
+++ b/xen/arch/arm/p2m.c
@@ -34,7 +34,8 @@ static unsigned int __read_mostly max_vmid =
MAX_VMID_8_BIT;
#define P2M_ROOT_PAGES (1<
I think it would be worth explaining that this is required so the
number of P2M bits can be restricted by external entity (e.g IOMMU).
ok, will add explanation
+unsigned int __read_mostly p2m_ipa_bits = 64;
/* Helpers to lookup the properties of each level */
static const paddr_t level_masks[] =
@@ -1912,6 +1913,16 @@ struct page_info *get_page_from_gva(struct
vcpu *v, vaddr_t va,
return page;
}
+void __init p2m_restrict_ipa_bits(unsigned int iommu_ipa_bits)
The name of the function is quite generic as most of the code in it.
So can we avoid use the term IOMMU in it?
yes, will do
+{
+ /*
+ * Calculate the minimum of the maximum IPA bits that any IOMMU
+ * can support.
+ */
+ if ( iommu_ipa_bits < p2m_ipa_bits )
+ p2m_ipa_bits = iommu_ipa_bits;
+}
+
/* VTCR value to be configured by all CPUs. Set only once by the
boot CPU */
static uint32_t __read_mostly vtcr;
@@ -1966,10 +1977,28 @@ void __init setup_virt_paging(void)
[7] = { 0 } /* Invalid */
};
- unsigned int cpu;
+ unsigned int i, cpu;
unsigned int pa_range = 0x10; /* Larger than any possible value */
bool vmid_8_bit = false;
+ if ( iommu_enabled )
Could we make this IOMMU-agnostic? The main reason to convert from
p2m_ipa_bits to pa_range is to cater the rest of the code.
But we could rework the code to do the computation with p2m_ipa_bits
and then look-up for the pa_range.
I am afraid, I don't completely understand your idea of making this
IOMMU-agnostic and what I should do...
+ {
+ /*
+ * Choose suitable "pa_range" according to the IOMMU
requirements
+ * (restricted "p2m_ipa_bits" value).
+ * As P2M table is always configured with IPA bits == PA bits,
+ * check for equivalent "pabits" and store it's index.
+ */
+ for ( i = 0; i < ARRAY_SIZE(pa_range_info); i++ )
+ {
+ if ( p2m_ipa_bits == pa_range_info[i].pabits )
+ {
+ pa_range = i;
+ break;
+ }
+ }
+ }
+
for_each_online_cpu ( cpu )
{
const struct cpuinfo_arm *info = _data[cpu];
diff --git a/xen/arch/arm/setup.c b/xen/arch/arm/setup.c
index 51a6677..413f3e6 100644
--- a/xen/arch/arm/setup.c
+++
Re: [Xen-devel] [PATCH] [RFC V2] xen/arm: Restrict "p2m_ipa_bits" according to the IOMMU requirements
Hi Oleksandr,
On 8/23/19 8:34 PM, Oleksandr Tyshchenko wrote:
From: Oleksandr Tyshchenko
There is a strict requirement for the IOMMU which wants to share
the P2M table with the CPU. The IOMMU's Stage-2 input size must be equal
to the P2M IPA size. It is not a problem when the IOMMU can support
all values the CPU supports. In that case, the IOMMU driver would just
use any "p2m_ipa_bits" value as is. But, there are cases when not.
In order to make P2M sharing possible on the platforms which
IPMMUs have a limitation in maximum Stage-2 input size introduce
the following logic.
First initialize the IOMMU subsystem and gather requirements regarding
the maximum IPA bits supported by each IOMMU device to figure out
the minimum value among them. In the P2M code, take into the account
the IOMMU requirements and choose suitable "pa_range" according
to the restricted "p2m_ipa_bits".
As I pointed in the previous version, all the code you modify is arm64
specific. For arm32, the number of IPA bits is
hardcoded. So if you modify p2m_ipa_bits, you would end up to
misconfigure VTCR.
In other words, for Arm32, you need to check p2m_ipa_bits is at least
40-bits before overriding it.
Signed-off-by: Oleksandr Tyshchenko
CC: Julien Grall
---
Still RFC:
1. Patch assumes that IPMMU support is already in.
2. Not checked for the SMMU.
Changes since RFC V1 [1]:
- Don't update p2m_ipa_bits by the IOMMU drivers directly,
introduce p2m_restrict_ipa_bits()
- Clarify patch subject/description
- Add more comments to code
- Check for equivalent "pabits" in setup_virt_paging()
- Remove ASSERTs from the SMMU and IPMMU drivers
[1] https://lists.xenproject.org/archives/html/xen-devel/2019-08/msg02078.html
---
xen/arch/arm/p2m.c | 33 ++--
xen/arch/arm/setup.c | 11 +--
xen/drivers/passthrough/arm/ipmmu-vmsa.c | 19 --
xen/drivers/passthrough/arm/smmu.c | 16
xen/include/asm-arm/p2m.h| 8
5 files changed, 60 insertions(+), 27 deletions(-)
diff --git a/xen/arch/arm/p2m.c b/xen/arch/arm/p2m.c
index 2374e92..f742d9c 100644
--- a/xen/arch/arm/p2m.c
+++ b/xen/arch/arm/p2m.c
@@ -34,7 +34,8 @@ static unsigned int __read_mostly max_vmid = MAX_VMID_8_BIT;
#define P2M_ROOT_PAGES(1<
-unsigned int __read_mostly p2m_ipa_bits;
+/* Larger than any possible value */
I think it would be worth explaining that this is required so the number
of P2M bits can be restricted by external entity (e.g IOMMU).
+unsigned int __read_mostly p2m_ipa_bits = 64;
/* Helpers to lookup the properties of each level */
static const paddr_t level_masks[] =
@@ -1912,6 +1913,16 @@ struct page_info *get_page_from_gva(struct vcpu *v,
vaddr_t va,
return page;
}
+void __init p2m_restrict_ipa_bits(unsigned int iommu_ipa_bits)
The name of the function is quite generic as most of the code in it. So
can we avoid use the term IOMMU in it?
+{
+/*
+ * Calculate the minimum of the maximum IPA bits that any IOMMU
+ * can support.
+ */
+if ( iommu_ipa_bits < p2m_ipa_bits )
+p2m_ipa_bits = iommu_ipa_bits;
+}
+
/* VTCR value to be configured by all CPUs. Set only once by the boot CPU */
static uint32_t __read_mostly vtcr;
@@ -1966,10 +1977,28 @@ void __init setup_virt_paging(void)
[7] = { 0 } /* Invalid */
};
-unsigned int cpu;
+unsigned int i, cpu;
unsigned int pa_range = 0x10; /* Larger than any possible value */
bool vmid_8_bit = false;
+if ( iommu_enabled )
Could we make this IOMMU-agnostic? The main reason to convert from
p2m_ipa_bits to pa_range is to cater the rest of the code.
But we could rework the code to do the computation with p2m_ipa_bits and
then look-up for the pa_range. In all honesty, I think we can completely
avoid pa_range but this is probably going to require more a bit more work.
+{
+/*
+ * Choose suitable "pa_range" according to the IOMMU requirements
+ * (restricted "p2m_ipa_bits" value).
+ * As P2M table is always configured with IPA bits == PA bits,
+ * check for equivalent "pabits" and store it's index.
+ */
+for ( i = 0; i < ARRAY_SIZE(pa_range_info); i++ )
+{
+if ( p2m_ipa_bits == pa_range_info[i].pabits )
+{
+pa_range = i;
+break;
+}
+}
+}
+
for_each_online_cpu ( cpu )
{
const struct cpuinfo_arm *info = _data[cpu];
diff --git a/xen/arch/arm/setup.c b/xen/arch/arm/setup.c
index 51a6677..413f3e6 100644
--- a/xen/arch/arm/setup.c
+++ b/xen/arch/arm/setup.c
@@ -936,12 +936,19 @@ void __init start_xen(unsigned long boot_phys_offset,
printk("Brought up %ld CPUs\n", (long)num_online_cpus());
/* TODO: smp_cpus_done(); */
-setup_virt_paging();
-
+/*
+ *
