* [RFC for-4.5 00/12] IOMMU support for ARM
@ 2014-02-07 17:42 Julien Grall
2014-02-07 17:43 ` [RFC for-4.5 01/12] xen/common: grant-table: only call IOMMU if paging mode translate is disabled Julien Grall
` (12 more replies)
0 siblings, 13 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:42 UTC (permalink / raw)
To: xen-devel; +Cc: stefano.stabellini, Julien Grall, tim, ian.campbell, patches
Hello,
This patch series add support for IOMMU on ARM. It have also added ARM SMMU
driver which is used for instance on Midway.
The IOMMU architecture for ARM is relying on the page table is shared between
the processor and each IOMMU.
The patch series is divided following:
- #1: fixing grant-table with IOMMU. Will be necessary for ARM later
- #2-#3: Make static some vtd functions
- #4-#5: Adding new device tree functions
- #6-#9: Prepare IOMMU code to add support for ARM
- #10-#11: Add IOMMU architecture for ARM
- #12: Add SMMU drivers
For now the 1:1 workaround is not removed because a same platform can have
DMA-capable device which are under an IOMMU and some not. It's a problem for
the swiotlb which needs to know if the device is procted or not when foreign
mapping is mapped in dom0.
When I talked with Stefano, 2 solutions came:
- Having a property in each "protected" device
- List in the hypervisor node the procted devices
I didn't yet decide which solution I will use.
Any comments, questions are welcomed.
Sincerely yours,
Julien Grall (12):
xen/common: grant-table: only call IOMMU if paging mode translate is
disabled
xen/passthrough: vtd: Don't export iommu_domain_teardown
xen/passthrough: vtd: Don't export iommu_set_pgd
xen/dts: Add dt_property_read_bool
xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle
xen/passthrough: rework dom0_pvh_reqs to use it also on ARM
xen/passthrough: iommu: Don't need to map dom0 page when the PT is
shared
xen/passthrough: iommu: Split generic IOMMU code
xen/passthrough: iommu: Introduce arch specific code
xen/passthrough: Introduce IOMMU ARM architure
MAINTAINERS: Add drivers/passthrough/arm
drivers/passthrough: arm: Add support for SMMU drivers
MAINTAINERS | 1 +
xen/arch/arm/Rules.mk | 1 +
xen/arch/arm/domain.c | 7 +
xen/arch/arm/domain_build.c | 2 +
xen/arch/arm/p2m.c | 4 +
xen/arch/arm/setup.c | 2 +
xen/arch/x86/domctl.c | 6 +-
xen/arch/x86/hvm/io.c | 2 +-
xen/arch/x86/tboot.c | 3 +-
xen/common/device_tree.c | 157 ++-
xen/common/grant_table.c | 7 +-
xen/drivers/passthrough/Makefile | 7 +-
xen/drivers/passthrough/amd/iommu_cmd.c | 2 +-
xen/drivers/passthrough/amd/iommu_guest.c | 8 +-
xen/drivers/passthrough/amd/iommu_map.c | 56 +-
xen/drivers/passthrough/amd/pci_amd_iommu.c | 53 +-
xen/drivers/passthrough/arm/Makefile | 2 +
xen/drivers/passthrough/arm/iommu.c | 65 +
xen/drivers/passthrough/arm/smmu.c | 1701 +++++++++++++++++++++++++++
xen/drivers/passthrough/iommu.c | 525 +--------
xen/drivers/passthrough/iommu_pci.c | 468 ++++++++
xen/drivers/passthrough/iommu_x86.c | 106 ++
xen/drivers/passthrough/vtd/iommu.c | 124 +-
xen/include/asm-arm/device.h | 3 +-
xen/include/asm-arm/domain.h | 2 +
xen/include/asm-arm/hvm/iommu.h | 10 +
xen/include/asm-arm/iommu.h | 36 +
xen/include/asm-x86/hvm/iommu.h | 29 +
xen/include/asm-x86/iommu.h | 50 +
xen/include/xen/device_tree.h | 75 ++
xen/include/xen/hvm/iommu.h | 27 +-
xen/include/xen/iommu.h | 51 +-
32 files changed, 2891 insertions(+), 701 deletions(-)
create mode 100644 xen/drivers/passthrough/arm/Makefile
create mode 100644 xen/drivers/passthrough/arm/iommu.c
create mode 100644 xen/drivers/passthrough/arm/smmu.c
create mode 100644 xen/drivers/passthrough/iommu_pci.c
create mode 100644 xen/drivers/passthrough/iommu_x86.c
create mode 100644 xen/include/asm-arm/hvm/iommu.h
create mode 100644 xen/include/asm-arm/iommu.h
create mode 100644 xen/include/asm-x86/iommu.h
--
1.7.10.4
^ permalink raw reply [flat|nested] 51+ messages in thread
* [RFC for-4.5 01/12] xen/common: grant-table: only call IOMMU if paging mode translate is disabled
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-10 8:02 ` Jan Beulich
2014-02-19 12:25 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 02/12] xen/passthrough: vtd: Don't export iommu_domain_teardown Julien Grall
` (11 subsequent siblings)
12 siblings, 2 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel
Cc: Keir Fraser, ian.campbell, patches, Julien Grall, tim,
stefano.stabellini
>From Xen point of view, ARM guests are PV guest with paging auto translate
enabled.
When IOMMU support will be added for ARM, mapping grant ref will always crash
Xen due to the BUG_ON in __gnttab_map_grant_ref.
On x86:
- PV guests always have paging mode translate disabled
- PVH and HVM guests have always paging mode translate enabled
It means that we can safely replace the check that the domain is a PV guests
by checking if the guest has paging mode translate enabled.
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Cc: Keir Fraser <keir@xen.org>
---
xen/common/grant_table.c | 7 ++-----
1 file changed, 2 insertions(+), 5 deletions(-)
diff --git a/xen/common/grant_table.c b/xen/common/grant_table.c
index 107b000..778bdb7 100644
--- a/xen/common/grant_table.c
+++ b/xen/common/grant_table.c
@@ -721,12 +721,10 @@ __gnttab_map_grant_ref(
double_gt_lock(lgt, rgt);
- if ( is_pv_domain(ld) && need_iommu(ld) )
+ if ( !paging_mode_translate(ld) && need_iommu(ld) )
{
unsigned int wrc, rdc;
int err = 0;
- /* Shouldn't happen, because you can't use iommu in a HVM domain. */
- BUG_ON(paging_mode_translate(ld));
/* We're not translated, so we know that gmfns and mfns are
the same things, so the IOMMU entry is always 1-to-1. */
mapcount(lgt, rd, frame, &wrc, &rdc);
@@ -931,11 +929,10 @@ __gnttab_unmap_common(
act->pin -= GNTPIN_hstw_inc;
}
- if ( is_pv_domain(ld) && need_iommu(ld) )
+ if ( !paging_mode_translate(ld) && need_iommu(ld) )
{
unsigned int wrc, rdc;
int err = 0;
- BUG_ON(paging_mode_translate(ld));
mapcount(lgt, rd, op->frame, &wrc, &rdc);
if ( (wrc + rdc) == 0 )
err = iommu_unmap_page(ld, op->frame);
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* [RFC for-4.5 02/12] xen/passthrough: vtd: Don't export iommu_domain_teardown
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
2014-02-07 17:43 ` [RFC for-4.5 01/12] xen/common: grant-table: only call IOMMU if paging mode translate is disabled Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-19 12:26 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 03/12] xen/passthrough: vtd: Don't export iommu_set_pgd Julien Grall
` (10 subsequent siblings)
12 siblings, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel
Cc: ian.campbell, patches, Julien Grall, tim, stefano.stabellini,
Xiantao Zhang
iommu_domain_teardown is only used internally in
xen/drivers/passthrough/vtd/iommu.c
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Cc: Xiantao Zhang <xiantao.zhang@intel.com>
---
xen/drivers/passthrough/vtd/iommu.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/xen/drivers/passthrough/vtd/iommu.c b/xen/drivers/passthrough/vtd/iommu.c
index 5f10034..a8d33fc 100644
--- a/xen/drivers/passthrough/vtd/iommu.c
+++ b/xen/drivers/passthrough/vtd/iommu.c
@@ -1701,7 +1701,7 @@ static int reassign_device_ownership(
return ret;
}
-void iommu_domain_teardown(struct domain *d)
+static void iommu_domain_teardown(struct domain *d)
{
struct hvm_iommu *hd = domain_hvm_iommu(d);
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* [RFC for-4.5 03/12] xen/passthrough: vtd: Don't export iommu_set_pgd
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
2014-02-07 17:43 ` [RFC for-4.5 01/12] xen/common: grant-table: only call IOMMU if paging mode translate is disabled Julien Grall
2014-02-07 17:43 ` [RFC for-4.5 02/12] xen/passthrough: vtd: Don't export iommu_domain_teardown Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-10 8:04 ` Jan Beulich
2014-02-19 12:27 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 04/12] xen/dts: Add dt_property_read_bool Julien Grall
` (9 subsequent siblings)
12 siblings, 2 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel
Cc: ian.campbell, patches, Julien Grall, tim, stefano.stabellini,
Jan Beulich, Xiantoa Zhang
iommu_set_pgd is only used internally in
xen/drivers/passthrough/vtd/iommu.c
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Cc: Xiantoa Zhang <xiantao.zhang@intel.com>
Cc: Jan Beulich <jbeulich@suse.com>
---
xen/drivers/passthrough/vtd/iommu.c | 2 +-
xen/include/xen/iommu.h | 1 -
2 files changed, 1 insertion(+), 2 deletions(-)
diff --git a/xen/drivers/passthrough/vtd/iommu.c b/xen/drivers/passthrough/vtd/iommu.c
index a8d33fc..d5ce5b7 100644
--- a/xen/drivers/passthrough/vtd/iommu.c
+++ b/xen/drivers/passthrough/vtd/iommu.c
@@ -1826,7 +1826,7 @@ static int vtd_ept_page_compatible(struct iommu *iommu)
/*
* set VT-d page table directory to EPT table if allowed
*/
-void iommu_set_pgd(struct domain *d)
+static void iommu_set_pgd(struct domain *d)
{
struct hvm_iommu *hd = domain_hvm_iommu(d);
mfn_t pgd_mfn;
diff --git a/xen/include/xen/iommu.h b/xen/include/xen/iommu.h
index 8bb0a1d..fcbc432 100644
--- a/xen/include/xen/iommu.h
+++ b/xen/include/xen/iommu.h
@@ -68,7 +68,6 @@ int iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,
unsigned int flags);
int iommu_unmap_page(struct domain *d, unsigned long gfn);
void iommu_pte_flush(struct domain *d, u64 gfn, u64 *pte, int order, int present);
-void iommu_set_pgd(struct domain *d);
void iommu_domain_teardown(struct domain *d);
void pt_pci_init(void);
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* [RFC for-4.5 04/12] xen/dts: Add dt_property_read_bool
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
` (2 preceding siblings ...)
2014-02-07 17:43 ` [RFC for-4.5 03/12] xen/passthrough: vtd: Don't export iommu_set_pgd Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-19 12:28 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 05/12] xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle Julien Grall
` (8 subsequent siblings)
12 siblings, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel; +Cc: stefano.stabellini, Julien Grall, tim, ian.campbell, patches
The function check if property exists in a specific node.
Signed-off-by: Julien Grall <julien.grall@linaro.org>
---
xen/common/device_tree.c | 6 ++----
xen/include/xen/device_tree.h | 21 +++++++++++++++++++++
2 files changed, 23 insertions(+), 4 deletions(-)
diff --git a/xen/common/device_tree.c b/xen/common/device_tree.c
index c66d1d5..ccdb7ff 100644
--- a/xen/common/device_tree.c
+++ b/xen/common/device_tree.c
@@ -512,10 +512,8 @@ static void __init *unflatten_dt_alloc(unsigned long *mem, unsigned long size,
}
/* Find a property with a given name for a given node and return it. */
-static const struct dt_property *
-dt_find_property(const struct dt_device_node *np,
- const char *name,
- u32 *lenp)
+const struct dt_property *dt_find_property(const struct dt_device_node *np,
+ const char *name, u32 *lenp)
{
const struct dt_property *pp;
diff --git a/xen/include/xen/device_tree.h b/xen/include/xen/device_tree.h
index 9a8c3de..7c075d9 100644
--- a/xen/include/xen/device_tree.h
+++ b/xen/include/xen/device_tree.h
@@ -15,6 +15,7 @@
#include <xen/init.h>
#include <xen/string.h>
#include <xen/types.h>
+#include <xen/stdbool.h>
#define DEVICE_TREE_MAX_DEPTH 16
@@ -347,6 +348,10 @@ struct dt_device_node *dt_find_compatible_node(struct dt_device_node *from,
const void *dt_get_property(const struct dt_device_node *np,
const char *name, u32 *lenp);
+const struct dt_property *dt_find_property(const struct dt_device_node *np,
+ const char *name, u32 *lenp);
+
+
/**
* dt_property_read_u32 - Helper to read a u32 property.
* @np: node to get the value
@@ -369,6 +374,22 @@ bool_t dt_property_read_u64(const struct dt_device_node *np,
const char *name, u64 *out_value);
/**
+ * dt_property_read_bool - Check if a property exists
+ * @np: node to get the value
+ * @name: name of the property
+ *
+ * Search for a property in a device node.
+ * Return true if the property exists false otherwise.
+ */
+static inline bool_t dt_property_read_bool(const struct dt_device_node *np,
+ const char *name)
+{
+ const struct dt_property *prop = dt_find_property(np, name, NULL);
+
+ return prop ? true : false;
+}
+
+/**
* dt_property_read_string - Find and read a string from a property
* @np: Device node from which the property value is to be read
* @propname: Name of the property to be searched
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* [RFC for-4.5 05/12] xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
` (3 preceding siblings ...)
2014-02-07 17:43 ` [RFC for-4.5 04/12] xen/dts: Add dt_property_read_bool Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-19 12:34 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM Julien Grall
` (7 subsequent siblings)
12 siblings, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel; +Cc: stefano.stabellini, Julien Grall, tim, ian.campbell, patches
Code adapted from linux drivers/of/base.c (commit ef42c58).
Signed-off-by: Julien Grall <julien.grall@linaro.org>
---
xen/common/device_tree.c | 151 ++++++++++++++++++++++++++++++++++++++++-
xen/include/xen/device_tree.h | 54 +++++++++++++++
2 files changed, 203 insertions(+), 2 deletions(-)
diff --git a/xen/common/device_tree.c b/xen/common/device_tree.c
index ccdb7ff..37a025a 100644
--- a/xen/common/device_tree.c
+++ b/xen/common/device_tree.c
@@ -1090,9 +1090,9 @@ int dt_device_get_address(const struct dt_device_node *dev, int index,
*
* Returns a node pointer.
*/
-static const struct dt_device_node *dt_find_node_by_phandle(dt_phandle handle)
+static struct dt_device_node *dt_find_node_by_phandle(dt_phandle handle)
{
- const struct dt_device_node *np;
+ struct dt_device_node *np;
dt_for_each_device_node(dt_host, np)
if ( np->phandle == handle )
@@ -1477,6 +1477,153 @@ bool_t dt_device_is_available(const struct dt_device_node *device)
return 0;
}
+static int __dt_parse_phandle_with_args(const struct dt_device_node *np,
+ const char *list_name,
+ const char *cells_name,
+ int cell_count, int index,
+ struct dt_phandle_args *out_args)
+{
+ const __be32 *list, *list_end;
+ int rc = 0, cur_index = 0;
+ u32 size, count = 0;
+ struct dt_device_node *node = NULL;
+ dt_phandle phandle;
+
+ /* Retrieve the phandle list property */
+ list = dt_get_property(np, list_name, &size);
+ if ( !list )
+ return -ENOENT;
+ list_end = list + size / sizeof(*list);
+
+ /* Loop over the phandles until all the requested entry is found */
+ while ( list < list_end )
+ {
+ rc = -EINVAL;
+ count = 0;
+
+ /*
+ * If phandle is 0, then it is an empty entry with no
+ * arguments. Skip forward to the next entry.
+ * */
+ phandle = be32_to_cpup(list++);
+ if ( phandle )
+ {
+ /*
+ * Find the provider node and parse the #*-cells
+ * property to determine the argument length.
+ *
+ * This is not needed if the cell count is hard-coded
+ * (i.e. cells_name not set, but cell_count is set),
+ * except when we're going to return the found node
+ * below.
+ */
+ if ( cells_name || cur_index == index )
+ {
+ node = dt_find_node_by_phandle(phandle);
+ if ( !node )
+ {
+ dt_printk(XENLOG_ERR "%s: could not find phandle\n",
+ np->full_name);
+ goto err;
+ }
+ }
+
+ if ( cells_name )
+ {
+ if ( !dt_property_read_u32(node, cells_name, &count) )
+ {
+ dt_printk("%s: could not get %s for %s\n",
+ np->full_name, cells_name, node->full_name);
+ goto err;
+ }
+ }
+ else
+ count = cell_count;
+
+ /*
+ * Make sure that the arguments actually fit in the
+ * remaining property data length
+ */
+ if ( list + count > list_end )
+ {
+ dt_printk(XENLOG_ERR "%s: arguments longer than property\n",
+ np->full_name);
+ goto err;
+ }
+ }
+
+ /*
+ * All of the error cases above bail out of the loop, so at
+ * this point, the parsing is successful. If the requested
+ * index matches, then fill the out_args structure and return,
+ * or return -ENOENT for an empty entry.
+ */
+ rc = -ENOENT;
+ if ( cur_index == index )
+ {
+ if (!phandle)
+ goto err;
+
+ if ( out_args )
+ {
+ int i;
+
+ WARN_ON(count > MAX_PHANDLE_ARGS);
+ if (count > MAX_PHANDLE_ARGS)
+ count = MAX_PHANDLE_ARGS;
+ out_args->np = node;
+ out_args->args_count = count;
+ for ( i = 0; i < count; i++ )
+ out_args->args[i] = be32_to_cpup(list++);
+ }
+
+ /* Found it! return success */
+ return 0;
+ }
+
+ node = NULL;
+ list += count;
+ cur_index++;
+ }
+
+ /*
+ * Returning result will be one of:
+ * -ENOENT : index is for empty phandle
+ * -EINVAL : parsing error on data
+ * [1..n] : Number of phandle (count mode; when index = -1)
+ */
+ rc = index < 0 ? cur_index : -ENOENT;
+err:
+ return rc;
+}
+
+struct dt_device_node *dt_parse_phandle(const struct dt_device_node *np,
+ const char *phandle_name, int index)
+{
+ struct dt_phandle_args args;
+
+ if (index < 0)
+ return NULL;
+
+ if (__dt_parse_phandle_with_args(np, phandle_name, NULL, 0,
+ index, &args))
+ return NULL;
+
+ return args.np;
+}
+
+
+int dt_parse_phandle_with_args(const struct dt_device_node *np,
+ const char *list_name,
+ const char *cells_name, int index,
+ struct dt_phandle_args *out_args)
+{
+ if ( index < 0 )
+ return -EINVAL;
+ return __dt_parse_phandle_with_args(np, list_name, cells_name, 0,
+ index, out_args);
+}
+
/**
* unflatten_dt_node - Alloc and populate a device_node from the flat tree
* @fdt: The parent device tree blob
diff --git a/xen/include/xen/device_tree.h b/xen/include/xen/device_tree.h
index 7c075d9..d429e60 100644
--- a/xen/include/xen/device_tree.h
+++ b/xen/include/xen/device_tree.h
@@ -112,6 +112,13 @@ struct dt_device_node {
};
+#define MAX_PHANDLE_ARGS 16
+struct dt_phandle_args {
+ struct dt_device_node *np;
+ int args_count;
+ uint32_t args[MAX_PHANDLE_ARGS];
+};
+
/**
* IRQ line type.
*
@@ -621,6 +628,53 @@ void dt_set_range(__be32 **cellp, const struct dt_device_node *np,
void dt_get_range(const __be32 **cellp, const struct dt_device_node *np,
u64 *address, u64 *size);
+/**
+ * dt_parse_phandle - Resolve a phandle property to a device_node pointer
+ * @np: Pointer to device node holding phandle property
+ * @phandle_name: Name of property holding a phandle value
+ * @index: For properties holding a table of phandles, this is the index into
+ * the table
+ *
+ * Returns the device_node pointer.
+ */
+struct dt_device_node *dt_parse_phandle(const struct dt_device_node *np,
+ const char *phandle_name,
+ int index);
+
+/**
+ * dt_parse_phandle_with_args() - Find a node pointed by phandle in a list
+ * @np: pointer to a device tree node containing a list
+ * @list_name: property name that contains a list
+ * @cells_name: property name that specifies phandles' arguments count
+ * @index: index of a phandle to parse out
+ * @out_args: optional pointer to output arguments structure (will be filled)
+ *
+ * This function is useful to parse lists of phandles and their arguments.
+ * Returns 0 on success and fills out_args, on error returns appropriate
+ * errno value.
+ *
+ * Example:
+ *
+ * phandle1: node1 {
+ * #list-cells = <2>;
+ * }
+ *
+ * phandle2: node2 {
+ * #list-cells = <1>;
+ * }
+ *
+ * node3 {
+ * list = <&phandle1 1 2 &phandle2 3>;
+ * }
+ *
+ * To get a device_node of the `node2' node you may call this:
+ * dt_parse_phandle_with_args(node3, "list", "#list-cells", 1, &args);
+ */
+int dt_parse_phandle_with_args(const struct dt_device_node *np,
+ const char *list_name,
+ const char *cells_name, int index,
+ struct dt_phandle_args *out_args);
+
#endif /* __XEN_DEVICE_TREE_H */
/*
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
` (4 preceding siblings ...)
2014-02-07 17:43 ` [RFC for-4.5 05/12] xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-10 8:07 ` Jan Beulich
2014-02-10 16:10 ` Julien Grall
2014-02-07 17:43 ` [RFC for-4.5 07/12] xen/passthrough: iommu: Don't need to map dom0 page when the PT is shared Julien Grall
` (6 subsequent siblings)
12 siblings, 2 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel
Cc: ian.campbell, patches, Julien Grall, tim, stefano.stabellini,
Jan Beulich, Xiantao Zhang
DOM0 on ARM will have the same requirements as DOM0 PVH when iommu is enabled.
Both PVH and ARM guest has paging mode translate enabled, so Xen can use it
to know if it needs to check the requirements.
Rename the function and remove "pvh" word in the commit message.
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Cc: Xiantao Zhang <xiantao.zhang@intel.com>
Cc: Jan Beulich <jbeulich@suse.com>
---
xen/drivers/passthrough/iommu.c | 14 +++++++++-----
1 file changed, 9 insertions(+), 5 deletions(-)
diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
index 19b0e23..26a5d91 100644
--- a/xen/drivers/passthrough/iommu.c
+++ b/xen/drivers/passthrough/iommu.c
@@ -130,13 +130,18 @@ int iommu_domain_init(struct domain *d)
return hd->platform_ops->init(d);
}
-static __init void check_dom0_pvh_reqs(struct domain *d)
+static __init void check_dom0_reqs(struct domain *d)
{
+ if ( !paging_mode_translate(d) )
+ return;
+
if ( !iommu_enabled )
- panic("Presently, iommu must be enabled for pvh dom0\n");
+ panic("Presently, iommu must be enabled to use dom0 with translate "
+ "paging mode\n");
if ( iommu_passthrough )
- panic("For pvh dom0, dom0-passthrough must not be enabled\n");
+ panic("Dom0 uses translate paging mode, dom0-passthrough must not be "
+ "enabled\n");
iommu_dom0_strict = 1;
}
@@ -145,8 +150,7 @@ void __init iommu_dom0_init(struct domain *d)
{
struct hvm_iommu *hd = domain_hvm_iommu(d);
- if ( is_pvh_domain(d) )
- check_dom0_pvh_reqs(d);
+ check_dom0_reqs(d);
if ( !iommu_enabled )
return;
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* [RFC for-4.5 07/12] xen/passthrough: iommu: Don't need to map dom0 page when the PT is shared
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
` (5 preceding siblings ...)
2014-02-07 17:43 ` [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-10 8:11 ` Jan Beulich
2014-02-19 12:37 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 08/12] xen/passthrough: iommu: Split generic IOMMU code Julien Grall
` (5 subsequent siblings)
12 siblings, 2 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel
Cc: ian.campbell, patches, Julien Grall, tim, stefano.stabellini,
Jan Beulich, Xiantao Zhang
Currently iommu_init_dom0 is browsing the page list and call map_page callback
on each page.
On both AMD and VTD drivers, the function will directly return if the page
table is shared with the processor. So Xen can safely avoid to run through
the page list.
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Cc: Xiantao Zhang <xiantao.zhang@intel.com>
Cc: Jan Beulich <jbeulich@suse.com>
---
xen/drivers/passthrough/iommu.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
index 26a5d91..0a26956 100644
--- a/xen/drivers/passthrough/iommu.c
+++ b/xen/drivers/passthrough/iommu.c
@@ -157,7 +157,7 @@ void __init iommu_dom0_init(struct domain *d)
register_keyhandler('o', &iommu_p2m_table);
d->need_iommu = !!iommu_dom0_strict;
- if ( need_iommu(d) )
+ if ( need_iommu(d) && !iommu_use_hap_pt(d) )
{
struct page_info *page;
unsigned int i = 0;
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* [RFC for-4.5 08/12] xen/passthrough: iommu: Split generic IOMMU code
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
` (6 preceding siblings ...)
2014-02-07 17:43 ` [RFC for-4.5 07/12] xen/passthrough: iommu: Don't need to map dom0 page when the PT is shared Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-10 8:22 ` Jan Beulich
2014-02-19 12:39 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 09/12] xen/passthrough: iommu: Introduce arch specific code Julien Grall
` (4 subsequent siblings)
12 siblings, 2 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel
Cc: ian.campbell, patches, Julien Grall, tim, stefano.stabellini,
Jan Beulich, Xiantao Zhang
The generic IOMMU framework code (xen/drivers/passthrough/iommu.c) contains
functions specific to x86 and PCI.
Split the framework in 3 distincts files:
- iommu.c: contains generic functions shared between x86 and ARM
(when it will be supported)
- iommu_pci.c: contains specific functions for PCI passthrough
- iommu_x86.c: contains specific functions for x86
iommu_pci.c will be only compiled when PCI is supported by the architecture
(eg. HAS_PCI is defined).
This patch is mostly code movement in new files.
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Cc: Xiantao Zhang <xiantao.zhang@intel.com>
Cc: Jan Beulich <jbeulich@suse.com>
---
xen/drivers/passthrough/Makefile | 6 +-
xen/drivers/passthrough/iommu.c | 473 +----------------------------------
xen/drivers/passthrough/iommu_pci.c | 468 ++++++++++++++++++++++++++++++++++
xen/drivers/passthrough/iommu_x86.c | 65 +++++
xen/drivers/passthrough/vtd/iommu.c | 42 ++--
xen/include/asm-x86/iommu.h | 46 ++++
xen/include/xen/hvm/iommu.h | 1 +
xen/include/xen/iommu.h | 42 ++--
8 files changed, 625 insertions(+), 518 deletions(-)
create mode 100644 xen/drivers/passthrough/iommu_pci.c
create mode 100644 xen/drivers/passthrough/iommu_x86.c
create mode 100644 xen/include/asm-x86/iommu.h
diff --git a/xen/drivers/passthrough/Makefile b/xen/drivers/passthrough/Makefile
index 7c40fa5..51e0a0d 100644
--- a/xen/drivers/passthrough/Makefile
+++ b/xen/drivers/passthrough/Makefile
@@ -3,5 +3,7 @@ subdir-$(x86) += amd
subdir-$(x86_64) += x86
obj-y += iommu.o
-obj-y += io.o
-obj-y += pci.o
+obj-$(x86) += iommu_x86.o
+obj-$(HAS_PCI) += iommu_pci.o
+obj-$(x86) += io.o
+obj-$(HAS_PCI) += pci.o
diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
index 0a26956..d733878 100644
--- a/xen/drivers/passthrough/iommu.c
+++ b/xen/drivers/passthrough/iommu.c
@@ -24,7 +24,6 @@
#include <xsm/xsm.h>
static void parse_iommu_param(char *s);
-static int iommu_populate_page_table(struct domain *d);
static void iommu_dump_p2m_table(unsigned char key);
/*
@@ -180,86 +179,7 @@ void __init iommu_dom0_init(struct domain *d)
return hd->platform_ops->dom0_init(d);
}
-int iommu_add_device(struct pci_dev *pdev)
-{
- struct hvm_iommu *hd;
- int rc;
- u8 devfn;
-
- if ( !pdev->domain )
- return -EINVAL;
-
- ASSERT(spin_is_locked(&pcidevs_lock));
-
- hd = domain_hvm_iommu(pdev->domain);
- if ( !iommu_enabled || !hd->platform_ops )
- return 0;
-
- rc = hd->platform_ops->add_device(pdev->devfn, pdev);
- if ( rc || !pdev->phantom_stride )
- return rc;
-
- for ( devfn = pdev->devfn ; ; )
- {
- devfn += pdev->phantom_stride;
- if ( PCI_SLOT(devfn) != PCI_SLOT(pdev->devfn) )
- return 0;
- rc = hd->platform_ops->add_device(devfn, pdev);
- if ( rc )
- printk(XENLOG_WARNING "IOMMU: add %04x:%02x:%02x.%u failed (%d)\n",
- pdev->seg, pdev->bus, PCI_SLOT(devfn), PCI_FUNC(devfn), rc);
- }
-}
-
-int iommu_enable_device(struct pci_dev *pdev)
-{
- struct hvm_iommu *hd;
-
- if ( !pdev->domain )
- return -EINVAL;
-
- ASSERT(spin_is_locked(&pcidevs_lock));
-
- hd = domain_hvm_iommu(pdev->domain);
- if ( !iommu_enabled || !hd->platform_ops ||
- !hd->platform_ops->enable_device )
- return 0;
-
- return hd->platform_ops->enable_device(pdev);
-}
-
-int iommu_remove_device(struct pci_dev *pdev)
-{
- struct hvm_iommu *hd;
- u8 devfn;
-
- if ( !pdev->domain )
- return -EINVAL;
-
- hd = domain_hvm_iommu(pdev->domain);
- if ( !iommu_enabled || !hd->platform_ops )
- return 0;
-
- for ( devfn = pdev->devfn ; pdev->phantom_stride; )
- {
- int rc;
-
- devfn += pdev->phantom_stride;
- if ( PCI_SLOT(devfn) != PCI_SLOT(pdev->devfn) )
- break;
- rc = hd->platform_ops->remove_device(devfn, pdev);
- if ( !rc )
- continue;
-
- printk(XENLOG_ERR "IOMMU: remove %04x:%02x:%02x.%u failed (%d)\n",
- pdev->seg, pdev->bus, PCI_SLOT(devfn), PCI_FUNC(devfn), rc);
- return rc;
- }
-
- return hd->platform_ops->remove_device(pdev->devfn, pdev);
-}
-
-static void iommu_teardown(struct domain *d)
+void iommu_teardown(struct domain *d)
{
const struct hvm_iommu *hd = domain_hvm_iommu(d);
@@ -268,151 +188,6 @@ static void iommu_teardown(struct domain *d)
tasklet_schedule(&iommu_pt_cleanup_tasklet);
}
-/*
- * If the device isn't owned by dom0, it means it already
- * has been assigned to other domain, or it doesn't exist.
- */
-static int device_assigned(u16 seg, u8 bus, u8 devfn)
-{
- struct pci_dev *pdev;
-
- spin_lock(&pcidevs_lock);
- pdev = pci_get_pdev_by_domain(dom0, seg, bus, devfn);
- spin_unlock(&pcidevs_lock);
-
- return pdev ? 0 : -EBUSY;
-}
-
-static int assign_device(struct domain *d, u16 seg, u8 bus, u8 devfn)
-{
- struct hvm_iommu *hd = domain_hvm_iommu(d);
- struct pci_dev *pdev;
- int rc = 0;
-
- if ( !iommu_enabled || !hd->platform_ops )
- return 0;
-
- /* Prevent device assign if mem paging or mem sharing have been
- * enabled for this domain */
- if ( unlikely(!need_iommu(d) &&
- (d->arch.hvm_domain.mem_sharing_enabled ||
- d->mem_event->paging.ring_page)) )
- return -EXDEV;
-
- if ( !spin_trylock(&pcidevs_lock) )
- return -ERESTART;
-
- if ( need_iommu(d) <= 0 )
- {
- if ( !iommu_use_hap_pt(d) )
- {
- rc = iommu_populate_page_table(d);
- if ( rc )
- {
- spin_unlock(&pcidevs_lock);
- return rc;
- }
- }
- d->need_iommu = 1;
- }
-
- pdev = pci_get_pdev_by_domain(dom0, seg, bus, devfn);
- if ( !pdev )
- {
- rc = pci_get_pdev(seg, bus, devfn) ? -EBUSY : -ENODEV;
- goto done;
- }
-
- pdev->fault.count = 0;
-
- if ( (rc = hd->platform_ops->assign_device(d, devfn, pdev)) )
- goto done;
-
- for ( ; pdev->phantom_stride; rc = 0 )
- {
- devfn += pdev->phantom_stride;
- if ( PCI_SLOT(devfn) != PCI_SLOT(pdev->devfn) )
- break;
- rc = hd->platform_ops->assign_device(d, devfn, pdev);
- if ( rc )
- printk(XENLOG_G_WARNING "d%d: assign %04x:%02x:%02x.%u failed (%d)\n",
- d->domain_id, seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn),
- rc);
- }
-
- done:
- if ( !has_arch_pdevs(d) && need_iommu(d) )
- iommu_teardown(d);
- spin_unlock(&pcidevs_lock);
-
- return rc;
-}
-
-static int iommu_populate_page_table(struct domain *d)
-{
- struct hvm_iommu *hd = domain_hvm_iommu(d);
- struct page_info *page;
- int rc = 0, n = 0;
-
- d->need_iommu = -1;
-
- this_cpu(iommu_dont_flush_iotlb) = 1;
- spin_lock(&d->page_alloc_lock);
-
- if ( unlikely(d->is_dying) )
- rc = -ESRCH;
-
- while ( !rc && (page = page_list_remove_head(&d->page_list)) )
- {
- if ( is_hvm_domain(d) ||
- (page->u.inuse.type_info & PGT_type_mask) == PGT_writable_page )
- {
- BUG_ON(SHARED_M2P(mfn_to_gmfn(d, page_to_mfn(page))));
- rc = hd->platform_ops->map_page(
- d, mfn_to_gmfn(d, page_to_mfn(page)), page_to_mfn(page),
- IOMMUF_readable|IOMMUF_writable);
- if ( rc )
- {
- page_list_add(page, &d->page_list);
- break;
- }
- }
- page_list_add_tail(page, &d->arch.relmem_list);
- if ( !(++n & 0xff) && !page_list_empty(&d->page_list) &&
- hypercall_preempt_check() )
- rc = -ERESTART;
- }
-
- if ( !rc )
- {
- /*
- * The expectation here is that generally there are many normal pages
- * on relmem_list (the ones we put there) and only few being in an
- * offline/broken state. The latter ones are always at the head of the
- * list. Hence we first move the whole list, and then move back the
- * first few entries.
- */
- page_list_move(&d->page_list, &d->arch.relmem_list);
- while ( (page = page_list_first(&d->page_list)) != NULL &&
- (page->count_info & (PGC_state|PGC_broken)) )
- {
- page_list_del(page, &d->page_list);
- page_list_add_tail(page, &d->arch.relmem_list);
- }
- }
-
- spin_unlock(&d->page_alloc_lock);
- this_cpu(iommu_dont_flush_iotlb) = 0;
-
- if ( !rc )
- iommu_iotlb_flush_all(d);
- else if ( rc != -ERESTART )
- iommu_teardown(d);
-
- return rc;
-}
-
-
void iommu_domain_destroy(struct domain *d)
{
struct hvm_iommu *hd = domain_hvm_iommu(d);
@@ -499,53 +274,6 @@ void iommu_iotlb_flush_all(struct domain *d)
hd->platform_ops->iotlb_flush_all(d);
}
-/* caller should hold the pcidevs_lock */
-int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn)
-{
- struct hvm_iommu *hd = domain_hvm_iommu(d);
- struct pci_dev *pdev = NULL;
- int ret = 0;
-
- if ( !iommu_enabled || !hd->platform_ops )
- return -EINVAL;
-
- ASSERT(spin_is_locked(&pcidevs_lock));
- pdev = pci_get_pdev_by_domain(d, seg, bus, devfn);
- if ( !pdev )
- return -ENODEV;
-
- while ( pdev->phantom_stride )
- {
- devfn += pdev->phantom_stride;
- if ( PCI_SLOT(devfn) != PCI_SLOT(pdev->devfn) )
- break;
- ret = hd->platform_ops->reassign_device(d, dom0, devfn, pdev);
- if ( !ret )
- continue;
-
- printk(XENLOG_G_ERR "d%d: deassign %04x:%02x:%02x.%u failed (%d)\n",
- d->domain_id, seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn), ret);
- return ret;
- }
-
- devfn = pdev->devfn;
- ret = hd->platform_ops->reassign_device(d, dom0, devfn, pdev);
- if ( ret )
- {
- dprintk(XENLOG_G_ERR,
- "d%d: deassign device (%04x:%02x:%02x.%u) failed\n",
- d->domain_id, seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
- return ret;
- }
-
- pdev->fault.count = 0;
-
- if ( !has_arch_pdevs(d) && need_iommu(d) )
- iommu_teardown(d);
-
- return ret;
-}
-
int __init iommu_setup(void)
{
int rc = -ENODEV;
@@ -586,86 +314,6 @@ int __init iommu_setup(void)
return rc;
}
-static int iommu_get_device_group(
- struct domain *d, u16 seg, u8 bus, u8 devfn,
- XEN_GUEST_HANDLE_64(uint32) buf, int max_sdevs)
-{
- struct hvm_iommu *hd = domain_hvm_iommu(d);
- struct pci_dev *pdev;
- int group_id, sdev_id;
- u32 bdf;
- int i = 0;
- const struct iommu_ops *ops = hd->platform_ops;
-
- if ( !iommu_enabled || !ops || !ops->get_device_group_id )
- return 0;
-
- group_id = ops->get_device_group_id(seg, bus, devfn);
-
- spin_lock(&pcidevs_lock);
- for_each_pdev( d, pdev )
- {
- if ( (pdev->seg != seg) ||
- ((pdev->bus == bus) && (pdev->devfn == devfn)) )
- continue;
-
- if ( xsm_get_device_group(XSM_HOOK, (seg << 16) | (pdev->bus << 8) | pdev->devfn) )
- continue;
-
- sdev_id = ops->get_device_group_id(seg, pdev->bus, pdev->devfn);
- if ( (sdev_id == group_id) && (i < max_sdevs) )
- {
- bdf = 0;
- bdf |= (pdev->bus & 0xff) << 16;
- bdf |= (pdev->devfn & 0xff) << 8;
-
- if ( unlikely(copy_to_guest_offset(buf, i, &bdf, 1)) )
- {
- spin_unlock(&pcidevs_lock);
- return -1;
- }
- i++;
- }
- }
- spin_unlock(&pcidevs_lock);
-
- return i;
-}
-
-void iommu_update_ire_from_apic(
- unsigned int apic, unsigned int reg, unsigned int value)
-{
- const struct iommu_ops *ops = iommu_get_ops();
- ops->update_ire_from_apic(apic, reg, value);
-}
-
-int iommu_update_ire_from_msi(
- struct msi_desc *msi_desc, struct msi_msg *msg)
-{
- const struct iommu_ops *ops = iommu_get_ops();
- return iommu_intremap ? ops->update_ire_from_msi(msi_desc, msg) : 0;
-}
-
-void iommu_read_msi_from_ire(
- struct msi_desc *msi_desc, struct msi_msg *msg)
-{
- const struct iommu_ops *ops = iommu_get_ops();
- if ( iommu_intremap )
- ops->read_msi_from_ire(msi_desc, msg);
-}
-
-unsigned int iommu_read_apic_from_ire(unsigned int apic, unsigned int reg)
-{
- const struct iommu_ops *ops = iommu_get_ops();
- return ops->read_apic_from_ire(apic, reg);
-}
-
-int __init iommu_setup_hpet_msi(struct msi_desc *msi)
-{
- const struct iommu_ops *ops = iommu_get_ops();
- return ops->setup_hpet_msi ? ops->setup_hpet_msi(msi) : -ENODEV;
-}
-
void iommu_resume()
{
const struct iommu_ops *ops = iommu_get_ops();
@@ -696,125 +344,6 @@ void iommu_crash_shutdown(void)
iommu_enabled = iommu_intremap = 0;
}
-int iommu_do_domctl(
- struct xen_domctl *domctl, struct domain *d,
- XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)
-{
- u16 seg;
- u8 bus, devfn;
- int ret = 0;
-
- if ( !iommu_enabled )
- return -ENOSYS;
-
- switch ( domctl->cmd )
- {
- case XEN_DOMCTL_get_device_group:
- {
- u32 max_sdevs;
- XEN_GUEST_HANDLE_64(uint32) sdevs;
-
- ret = xsm_get_device_group(XSM_HOOK, domctl->u.get_device_group.machine_sbdf);
- if ( ret )
- break;
-
- seg = domctl->u.get_device_group.machine_sbdf >> 16;
- bus = (domctl->u.get_device_group.machine_sbdf >> 8) & 0xff;
- devfn = domctl->u.get_device_group.machine_sbdf & 0xff;
- max_sdevs = domctl->u.get_device_group.max_sdevs;
- sdevs = domctl->u.get_device_group.sdev_array;
-
- ret = iommu_get_device_group(d, seg, bus, devfn, sdevs, max_sdevs);
- if ( ret < 0 )
- {
- dprintk(XENLOG_ERR, "iommu_get_device_group() failed!\n");
- ret = -EFAULT;
- domctl->u.get_device_group.num_sdevs = 0;
- }
- else
- {
- domctl->u.get_device_group.num_sdevs = ret;
- ret = 0;
- }
- if ( __copy_field_to_guest(u_domctl, domctl, u.get_device_group) )
- ret = -EFAULT;
- }
- break;
-
- case XEN_DOMCTL_test_assign_device:
- ret = xsm_test_assign_device(XSM_HOOK, domctl->u.assign_device.machine_sbdf);
- if ( ret )
- break;
-
- seg = domctl->u.assign_device.machine_sbdf >> 16;
- bus = (domctl->u.assign_device.machine_sbdf >> 8) & 0xff;
- devfn = domctl->u.assign_device.machine_sbdf & 0xff;
-
- if ( device_assigned(seg, bus, devfn) )
- {
- printk(XENLOG_G_INFO
- "%04x:%02x:%02x.%u already assigned, or non-existent\n",
- seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
- ret = -EINVAL;
- }
- break;
-
- case XEN_DOMCTL_assign_device:
- if ( unlikely(d->is_dying) )
- {
- ret = -EINVAL;
- break;
- }
-
- ret = xsm_assign_device(XSM_HOOK, d, domctl->u.assign_device.machine_sbdf);
- if ( ret )
- break;
-
- seg = domctl->u.assign_device.machine_sbdf >> 16;
- bus = (domctl->u.assign_device.machine_sbdf >> 8) & 0xff;
- devfn = domctl->u.assign_device.machine_sbdf & 0xff;
-
- ret = device_assigned(seg, bus, devfn) ?:
- assign_device(d, seg, bus, devfn);
- if ( ret == -ERESTART )
- ret = hypercall_create_continuation(__HYPERVISOR_domctl,
- "h", u_domctl);
- else if ( ret )
- printk(XENLOG_G_ERR "XEN_DOMCTL_assign_device: "
- "assign %04x:%02x:%02x.%u to dom%d failed (%d)\n",
- seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn),
- d->domain_id, ret);
-
- break;
-
- case XEN_DOMCTL_deassign_device:
- ret = xsm_deassign_device(XSM_HOOK, d, domctl->u.assign_device.machine_sbdf);
- if ( ret )
- break;
-
- seg = domctl->u.assign_device.machine_sbdf >> 16;
- bus = (domctl->u.assign_device.machine_sbdf >> 8) & 0xff;
- devfn = domctl->u.assign_device.machine_sbdf & 0xff;
-
- spin_lock(&pcidevs_lock);
- ret = deassign_device(d, seg, bus, devfn);
- spin_unlock(&pcidevs_lock);
- if ( ret )
- printk(XENLOG_G_ERR
- "deassign %04x:%02x:%02x.%u from dom%d failed (%d)\n",
- seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn),
- d->domain_id, ret);
-
- break;
-
- default:
- ret = -ENOSYS;
- break;
- }
-
- return ret;
-}
-
static void iommu_dump_p2m_table(unsigned char key)
{
struct domain *d;
diff --git a/xen/drivers/passthrough/iommu_pci.c b/xen/drivers/passthrough/iommu_pci.c
new file mode 100644
index 0000000..5b9d937
--- /dev/null
+++ b/xen/drivers/passthrough/iommu_pci.c
@@ -0,0 +1,468 @@
+/*
+ * 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, write to the Free Software Foundation, Inc., 59 Temple
+ * Place - Suite 330, Boston, MA 02111-1307 USA.
+ */
+
+#include <xen/sched.h>
+#include <xen/iommu.h>
+#include <xen/paging.h>
+#include <xen/guest_access.h>
+#include <xen/event.h>
+#include <xen/softirq.h>
+#include <xsm/xsm.h>
+
+static int iommu_populate_page_table(struct domain *d)
+{
+ struct hvm_iommu *hd = domain_hvm_iommu(d);
+ struct page_info *page;
+ int rc = 0, n = 0;
+
+ d->need_iommu = -1;
+
+ this_cpu(iommu_dont_flush_iotlb) = 1;
+ spin_lock(&d->page_alloc_lock);
+
+ if ( unlikely(d->is_dying) )
+ rc = -ESRCH;
+
+
+ while ( !rc && (page = page_list_remove_head(&d->page_list)) )
+ {
+ if ( is_hvm_domain(d) ||
+ (page->u.inuse.type_info & PGT_type_mask) == PGT_writable_page )
+ {
+ BUG_ON(SHARED_M2P(mfn_to_gmfn(d, page_to_mfn(page))));
+ rc = hd->platform_ops->map_page(
+ d, mfn_to_gmfn(d, page_to_mfn(page)), page_to_mfn(page),
+ IOMMUF_readable|IOMMUF_writable);
+ if ( rc )
+ {
+ page_list_add(page, &d->page_list);
+ break;
+ }
+ }
+ page_list_add_tail(page, &d->arch.relmem_list);
+ if ( !(++n & 0xff) && !page_list_empty(&d->page_list) &&
+ hypercall_preempt_check() )
+ rc = -ERESTART;
+ }
+
+ if ( !rc )
+ {
+ /*
+ * The expectation here is that generally there are many normal pages
+ * on relmem_list (the ones we put there) and only few being in an
+ * offline/broken state. The latter ones are always at the head of the
+ * list. Hence we first move the whole list, and then move back the
+ * first few entries.
+ */
+ page_list_move(&d->page_list, &d->arch.relmem_list);
+ while ( (page = page_list_first(&d->page_list)) != NULL &&
+ (page->count_info & (PGC_state|PGC_broken)) )
+ {
+ page_list_del(page, &d->page_list);
+ page_list_add_tail(page, &d->arch.relmem_list);
+ }
+ }
+
+ spin_unlock(&d->page_alloc_lock);
+ this_cpu(iommu_dont_flush_iotlb) = 0;
+
+ if ( !rc )
+ iommu_iotlb_flush_all(d);
+ else if ( rc != -ERESTART )
+ iommu_teardown(d);
+
+ return rc;
+}
+
+int iommu_add_device(struct pci_dev *pdev)
+{
+ struct hvm_iommu *hd;
+ int rc;
+ u8 devfn;
+
+ if ( !pdev->domain )
+ return -EINVAL;
+
+ ASSERT(spin_is_locked(&pcidevs_lock));
+
+ hd = domain_hvm_iommu(pdev->domain);
+ if ( !iommu_enabled || !hd->platform_ops )
+ return 0;
+
+ rc = hd->platform_ops->add_device(pdev->devfn, pdev);
+ if ( rc || !pdev->phantom_stride )
+ return rc;
+
+ for ( devfn = pdev->devfn ; ; )
+ {
+ devfn += pdev->phantom_stride;
+ if ( PCI_SLOT(devfn) != PCI_SLOT(pdev->devfn) )
+ return 0;
+ rc = hd->platform_ops->add_device(devfn, pdev);
+ if ( rc )
+ printk(XENLOG_WARNING "IOMMU: add %04x:%02x:%02x.%u failed (%d)\n",
+ pdev->seg, pdev->bus, PCI_SLOT(devfn), PCI_FUNC(devfn), rc);
+ }
+}
+
+int iommu_enable_device(struct pci_dev *pdev)
+{
+ struct hvm_iommu *hd;
+
+ if ( !pdev->domain )
+ return -EINVAL;
+
+ ASSERT(spin_is_locked(&pcidevs_lock));
+
+ hd = domain_hvm_iommu(pdev->domain);
+ if ( !iommu_enabled || !hd->platform_ops ||
+ !hd->platform_ops->enable_device )
+ return 0;
+
+ return hd->platform_ops->enable_device(pdev);
+}
+
+int iommu_remove_device(struct pci_dev *pdev)
+{
+ struct hvm_iommu *hd;
+ u8 devfn;
+
+ if ( !pdev->domain )
+ return -EINVAL;
+
+ hd = domain_hvm_iommu(pdev->domain);
+ if ( !iommu_enabled || !hd->platform_ops )
+ return 0;
+
+ for ( devfn = pdev->devfn ; pdev->phantom_stride; )
+ {
+ int rc;
+
+ devfn += pdev->phantom_stride;
+ if ( PCI_SLOT(devfn) != PCI_SLOT(pdev->devfn) )
+ break;
+ rc = hd->platform_ops->remove_device(devfn, pdev);
+ if ( !rc )
+ continue;
+
+ printk(XENLOG_ERR "IOMMU: remove %04x:%02x:%02x.%u failed (%d)\n",
+ pdev->seg, pdev->bus, PCI_SLOT(devfn), PCI_FUNC(devfn), rc);
+ return rc;
+ }
+
+ return hd->platform_ops->remove_device(pdev->devfn, pdev);
+}
+
+/*
+ * If the device isn't owned by dom0, it means it already
+ * has been assigned to other domain, or it doesn't exist.
+ */
+static int device_assigned(u16 seg, u8 bus, u8 devfn)
+{
+ struct pci_dev *pdev = NULL;
+
+ spin_lock(&pcidevs_lock);
+ pdev = pci_get_pdev_by_domain(dom0, seg, bus, devfn);
+ spin_unlock(&pcidevs_lock);
+
+ return pdev ? 0 : -EBUSY;
+}
+
+static int assign_device(struct domain *d, u16 seg, u8 bus, u8 devfn)
+{
+ struct hvm_iommu *hd = domain_hvm_iommu(d);
+ struct pci_dev *pdev;
+ int rc = 0;
+
+ if ( !iommu_enabled || !hd->platform_ops )
+ return 0;
+
+ /* Prevent device assign if mem paging or mem sharing have been
+ * enabled for this domain */
+ if ( unlikely(!need_iommu(d) &&
+ (mem_sharing_enabled(d) ||
+ d->mem_event->paging.ring_page)) )
+ return -EXDEV;
+
+ if ( !spin_trylock(&pcidevs_lock) )
+ return -ERESTART;
+
+ if ( need_iommu(d) <= 0 )
+ {
+ if ( !iommu_use_hap_pt(d) )
+ {
+ rc = iommu_populate_page_table(d);
+ if ( rc )
+ {
+ spin_unlock(&pcidevs_lock);
+ return rc;
+ }
+ }
+ d->need_iommu = 1;
+ }
+
+ pdev = pci_get_pdev_by_domain(dom0, seg, bus, devfn);
+ if ( !pdev )
+ {
+ rc = pci_get_pdev(seg, bus, devfn) ? -EBUSY : -ENODEV;
+ goto done;
+ }
+
+ pdev->fault.count = 0;
+
+ if ( (rc = hd->platform_ops->assign_device(d, devfn, pdev)) )
+ goto done;
+
+ for ( ; pdev->phantom_stride; rc = 0 )
+ {
+ devfn += pdev->phantom_stride;
+ if ( PCI_SLOT(devfn) != PCI_SLOT(pdev->devfn) )
+ break;
+ rc = hd->platform_ops->assign_device(d, devfn, pdev);
+ if ( rc )
+ printk(XENLOG_G_WARNING "d%d: assign %04x:%02x:%02x.%u failed (%d)\n",
+ d->domain_id, seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn),
+ rc);
+ }
+
+ done:
+ if ( !has_arch_pdevs(d) && need_iommu(d) )
+ iommu_teardown(d);
+ spin_unlock(&pcidevs_lock);
+
+ return rc;
+}
+
+/* caller should hold the pcidevs_lock */
+int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn)
+{
+ struct hvm_iommu *hd = domain_hvm_iommu(d);
+ struct pci_dev *pdev = NULL;
+ int ret = 0;
+
+ if ( !iommu_enabled || !hd->platform_ops )
+ return -EINVAL;
+
+ ASSERT(spin_is_locked(&pcidevs_lock));
+ pdev = pci_get_pdev_by_domain(d, seg, bus, devfn);
+ if ( !pdev )
+ return -ENODEV;
+
+ while ( pdev->phantom_stride )
+ {
+ devfn += pdev->phantom_stride;
+ if ( PCI_SLOT(devfn) != PCI_SLOT(pdev->devfn) )
+ break;
+ ret = hd->platform_ops->reassign_device(d, dom0, devfn, pdev);
+ if ( !ret )
+ continue;
+
+ printk(XENLOG_G_ERR "d%d: deassign %04x:%02x:%02x.%u failed (%d)\n",
+ d->domain_id, seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn), ret);
+ return ret;
+ }
+
+ devfn = pdev->devfn;
+ ret = hd->platform_ops->reassign_device(d, dom0, devfn, pdev);
+ if ( ret )
+ {
+ dprintk(XENLOG_G_ERR,
+ "d%d: deassign device (%04x:%02x:%02x.%u) failed\n",
+ d->domain_id, seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
+ return ret;
+ }
+
+ pdev->fault.count = 0;
+
+ if ( !has_arch_pdevs(d) && need_iommu(d) )
+ iommu_teardown(d);
+
+ return ret;
+}
+
+static int iommu_get_device_group(
+ struct domain *d, u16 seg, u8 bus, u8 devfn,
+ XEN_GUEST_HANDLE_64(uint32) buf, int max_sdevs)
+{
+ struct hvm_iommu *hd = domain_hvm_iommu(d);
+ struct pci_dev *pdev;
+ int group_id, sdev_id;
+ u32 bdf;
+ int i = 0;
+ const struct iommu_ops *ops = hd->platform_ops;
+
+ if ( !iommu_enabled || !ops || !ops->get_device_group_id )
+ return 0;
+
+ group_id = ops->get_device_group_id(seg, bus, devfn);
+
+ spin_lock(&pcidevs_lock);
+ for_each_pdev( d, pdev )
+ {
+ if ( (pdev->seg != seg) ||
+ ((pdev->bus == bus) && (pdev->devfn == devfn)) )
+ continue;
+
+ if ( xsm_get_device_group(XSM_HOOK, (seg << 16) | (pdev->bus << 8) | pdev->devfn) )
+ continue;
+
+ sdev_id = ops->get_device_group_id(seg, pdev->bus, pdev->devfn);
+ if ( (sdev_id == group_id) && (i < max_sdevs) )
+ {
+ bdf = 0;
+ bdf |= (pdev->bus & 0xff) << 16;
+ bdf |= (pdev->devfn & 0xff) << 8;
+
+ if ( unlikely(copy_to_guest_offset(buf, i, &bdf, 1)) )
+ {
+ spin_unlock(&pcidevs_lock);
+ return -1;
+ }
+ i++;
+ }
+ }
+
+ spin_unlock(&pcidevs_lock);
+
+ return i;
+}
+
+int iommu_do_domctl(
+ struct xen_domctl *domctl, struct domain *d,
+ XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)
+{
+ u16 seg;
+ u8 bus, devfn;
+ int ret = 0;
+
+ if ( !iommu_enabled )
+ return -ENOSYS;
+
+ switch ( domctl->cmd )
+ {
+ case XEN_DOMCTL_get_device_group:
+ {
+ u32 max_sdevs;
+ XEN_GUEST_HANDLE_64(uint32) sdevs;
+
+ ret = xsm_get_device_group(XSM_HOOK, domctl->u.get_device_group.machine_sbdf);
+ if ( ret )
+ break;
+
+ seg = domctl->u.get_device_group.machine_sbdf >> 16;
+ bus = (domctl->u.get_device_group.machine_sbdf >> 8) & 0xff;
+ devfn = domctl->u.get_device_group.machine_sbdf & 0xff;
+ max_sdevs = domctl->u.get_device_group.max_sdevs;
+ sdevs = domctl->u.get_device_group.sdev_array;
+
+ ret = iommu_get_device_group(d, seg, bus, devfn, sdevs, max_sdevs);
+ if ( ret < 0 )
+ {
+ dprintk(XENLOG_ERR, "iommu_get_device_group() failed!\n");
+ ret = -EFAULT;
+ domctl->u.get_device_group.num_sdevs = 0;
+ }
+ else
+ {
+ domctl->u.get_device_group.num_sdevs = ret;
+ ret = 0;
+ }
+ if ( __copy_field_to_guest(u_domctl, domctl, u.get_device_group) )
+ ret = -EFAULT;
+ }
+ break;
+
+ case XEN_DOMCTL_test_assign_device:
+ ret = xsm_test_assign_device(XSM_HOOK, domctl->u.assign_device.machine_sbdf);
+ if ( ret )
+ break;
+
+ seg = domctl->u.assign_device.machine_sbdf >> 16;
+ bus = (domctl->u.assign_device.machine_sbdf >> 8) & 0xff;
+ devfn = domctl->u.assign_device.machine_sbdf & 0xff;
+
+ if ( device_assigned(seg, bus, devfn) )
+ {
+ printk(XENLOG_G_INFO
+ "%04x:%02x:%02x.%u already assigned, or non-existent\n",
+ seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
+ ret = -EINVAL;
+ }
+ break;
+
+ case XEN_DOMCTL_assign_device:
+ if ( unlikely(d->is_dying) )
+ {
+ ret = -EINVAL;
+ break;
+ }
+
+ ret = xsm_assign_device(XSM_HOOK, d, domctl->u.assign_device.machine_sbdf);
+ if ( ret )
+ break;
+
+ seg = domctl->u.assign_device.machine_sbdf >> 16;
+ bus = (domctl->u.assign_device.machine_sbdf >> 8) & 0xff;
+ devfn = domctl->u.assign_device.machine_sbdf & 0xff;
+
+ ret = device_assigned(seg, bus, devfn) ?:
+ assign_device(d, seg, bus, devfn);
+ if ( ret == -ERESTART )
+ ret = hypercall_create_continuation(__HYPERVISOR_domctl,
+ "h", u_domctl);
+ else if ( ret )
+ printk(XENLOG_G_ERR "XEN_DOMCTL_assign_device: "
+ "assign %04x:%02x:%02x.%u to dom%d failed (%d)\n",
+ seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn),
+ d->domain_id, ret);
+
+ break;
+
+ case XEN_DOMCTL_deassign_device:
+ ret = xsm_deassign_device(XSM_HOOK, d, domctl->u.assign_device.machine_sbdf);
+ if ( ret )
+ break;
+
+ seg = domctl->u.assign_device.machine_sbdf >> 16;
+ bus = (domctl->u.assign_device.machine_sbdf >> 8) & 0xff;
+ devfn = domctl->u.assign_device.machine_sbdf & 0xff;
+
+ spin_lock(&pcidevs_lock);
+ ret = deassign_device(d, seg, bus, devfn);
+ spin_unlock(&pcidevs_lock);
+ if ( ret )
+ printk(XENLOG_G_ERR
+ "deassign %04x:%02x:%02x.%u from dom%d failed (%d)\n",
+ seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn),
+ d->domain_id, ret);
+
+ break;
+
+ default:
+ ret = -ENOSYS;
+ break;
+ }
+
+ return ret;
+}
+
+/*
+ * Local variables:
+ * mode: C
+ * c-file-style: "BSD"
+ * c-basic-offset: 4
+ * indent-tabs-mode: nil
+ * End:
+ */
diff --git a/xen/drivers/passthrough/iommu_x86.c b/xen/drivers/passthrough/iommu_x86.c
new file mode 100644
index 0000000..bd3c23b
--- /dev/null
+++ b/xen/drivers/passthrough/iommu_x86.c
@@ -0,0 +1,65 @@
+/*
+ * 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, write to the Free Software Foundation, Inc., 59 Temple
+ * Place - Suite 330, Boston, MA 02111-1307 USA.
+ */
+
+#include <xen/sched.h>
+#include <xen/iommu.h>
+#include <xen/paging.h>
+#include <xen/guest_access.h>
+#include <xen/event.h>
+#include <xen/softirq.h>
+#include <xsm/xsm.h>
+
+void iommu_update_ire_from_apic(
+ unsigned int apic, unsigned int reg, unsigned int value)
+{
+ const struct iommu_ops *ops = iommu_get_ops();
+ ops->update_ire_from_apic(apic, reg, value);
+}
+
+int iommu_update_ire_from_msi(
+ struct msi_desc *msi_desc, struct msi_msg *msg)
+{
+ const struct iommu_ops *ops = iommu_get_ops();
+ return iommu_intremap ? ops->update_ire_from_msi(msi_desc, msg) : 0;
+}
+
+void iommu_read_msi_from_ire(
+ struct msi_desc *msi_desc, struct msi_msg *msg)
+{
+ const struct iommu_ops *ops = iommu_get_ops();
+ if ( iommu_intremap )
+ ops->read_msi_from_ire(msi_desc, msg);
+}
+
+unsigned int iommu_read_apic_from_ire(unsigned int apic, unsigned int reg)
+{
+ const struct iommu_ops *ops = iommu_get_ops();
+ return ops->read_apic_from_ire(apic, reg);
+}
+
+int __init iommu_setup_hpet_msi(struct msi_desc *msi)
+{
+ const struct iommu_ops *ops = iommu_get_ops();
+ return ops->setup_hpet_msi ? ops->setup_hpet_msi(msi) : -ENODEV;
+}
+
+/*
+ * Local variables:
+ * mode: C
+ * c-file-style: "BSD"
+ * c-basic-offset: 4
+ * indent-tabs-mode: nil
+ * End:
+ */
diff --git a/xen/drivers/passthrough/vtd/iommu.c b/xen/drivers/passthrough/vtd/iommu.c
index d5ce5b7..faa794b 100644
--- a/xen/drivers/passthrough/vtd/iommu.c
+++ b/xen/drivers/passthrough/vtd/iommu.c
@@ -1784,31 +1784,31 @@ static int intel_iommu_unmap_page(struct domain *d, unsigned long gfn)
void iommu_pte_flush(struct domain *d, u64 gfn, u64 *pte,
int order, int present)
-{
- struct acpi_drhd_unit *drhd;
- struct iommu *iommu = NULL;
- struct hvm_iommu *hd = domain_hvm_iommu(d);
- int flush_dev_iotlb;
- int iommu_domid;
+ {
+ struct acpi_drhd_unit *drhd;
+ struct iommu *iommu = NULL;
+ struct hvm_iommu *hd = domain_hvm_iommu(d);
+ int flush_dev_iotlb;
+ int iommu_domid;
- iommu_flush_cache_entry(pte, sizeof(struct dma_pte));
+ iommu_flush_cache_entry(pte, sizeof(struct dma_pte));
- for_each_drhd_unit ( drhd )
- {
- iommu = drhd->iommu;
- if ( !test_bit(iommu->index, &hd->iommu_bitmap) )
- continue;
+ for_each_drhd_unit ( drhd )
+ {
+ iommu = drhd->iommu;
+ if ( !test_bit(iommu->index, &hd->iommu_bitmap) )
+ continue;
- flush_dev_iotlb = find_ats_dev_drhd(iommu) ? 1 : 0;
- iommu_domid= domain_iommu_domid(d, iommu);
- if ( iommu_domid == -1 )
- continue;
- if ( iommu_flush_iotlb_psi(iommu, iommu_domid,
- (paddr_t)gfn << PAGE_SHIFT_4K,
- order, !present, flush_dev_iotlb) )
- iommu_flush_write_buffer(iommu);
+ flush_dev_iotlb = find_ats_dev_drhd(iommu) ? 1 : 0;
+ iommu_domid= domain_iommu_domid(d, iommu);
+ if ( iommu_domid == -1 )
+ continue;
+ if ( iommu_flush_iotlb_psi(iommu, iommu_domid,
+ (paddr_t)gfn << PAGE_SHIFT_4K,
+ order, !present, flush_dev_iotlb) )
+ iommu_flush_write_buffer(iommu);
+ }
}
-}
static int vtd_ept_page_compatible(struct iommu *iommu)
{
diff --git a/xen/include/asm-x86/iommu.h b/xen/include/asm-x86/iommu.h
new file mode 100644
index 0000000..34c1896
--- /dev/null
+++ b/xen/include/asm-x86/iommu.h
@@ -0,0 +1,46 @@
+/*
+ * 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, write to the Free Software Foundation, Inc., 59 Temple
+ * Place - Suite 330, Boston, MA 02111-1307 USA.
+*/
+#ifndef __ARCH_X86_IOMMU_H__
+#define __ARCH_X86_IOMMU_H__
+
+#define MAX_IOMMUS 32
+
+#include <asm/msi.h>
+
+void iommu_update_ire_from_apic(unsigned int apic, unsigned int reg, unsigned int value);
+int iommu_update_ire_from_msi(struct msi_desc *msi_desc, struct msi_msg *msg);
+void iommu_read_msi_from_ire(struct msi_desc *msi_desc, struct msi_msg *msg);
+unsigned int iommu_read_apic_from_ire(unsigned int apic, unsigned int reg);
+int iommu_setup_hpet_msi(struct msi_desc *);
+
+void iommu_share_p2m_table(struct domain *d);
+
+/* While VT-d specific, this must get declared in a generic header. */
+int adjust_vtd_irq_affinities(void);
+void iommu_pte_flush(struct domain *d, u64 gfn, u64 *pte, int order, int present);
+int iommu_supports_eim(void);
+int iommu_enable_x2apic_IR(void);
+void iommu_disable_x2apic_IR(void);
+void iommu_set_dom0_mapping(struct domain *d);
+
+#endif /* !__ARCH_X86_IOMMU_H__ */
+/*
+ * Local variables:
+ * mode: C
+ * c-file-style: "BSD"
+ * c-basic-offset: 4
+ * indent-tabs-mode: nil
+ * End:
+ */
diff --git a/xen/include/xen/hvm/iommu.h b/xen/include/xen/hvm/iommu.h
index 26539e0..2abb4e3 100644
--- a/xen/include/xen/hvm/iommu.h
+++ b/xen/include/xen/hvm/iommu.h
@@ -21,6 +21,7 @@
#define __XEN_HVM_IOMMU_H__
#include <xen/iommu.h>
+#include <asm/hvm/iommu.h>
struct g2m_ioport {
struct list_head list;
diff --git a/xen/include/xen/iommu.h b/xen/include/xen/iommu.h
index fcbc432..60df9d6 100644
--- a/xen/include/xen/iommu.h
+++ b/xen/include/xen/iommu.h
@@ -25,6 +25,7 @@
#include <xen/pci.h>
#include <public/hvm/ioreq.h>
#include <public/domctl.h>
+#include <asm/iommu.h>
extern bool_t iommu_enable, iommu_enabled;
extern bool_t force_iommu, iommu_verbose;
@@ -39,17 +40,12 @@ extern bool_t amd_iommu_perdev_intremap;
#define domain_hvm_iommu(d) (&d->arch.hvm_domain.hvm_iommu)
-#define MAX_IOMMUS 32
-
#define PAGE_SHIFT_4K (12)
#define PAGE_SIZE_4K (1UL << PAGE_SHIFT_4K)
#define PAGE_MASK_4K (((u64)-1) << PAGE_SHIFT_4K)
#define PAGE_ALIGN_4K(addr) (((addr) + PAGE_SIZE_4K - 1) & PAGE_MASK_4K)
int iommu_setup(void);
-int iommu_supports_eim(void);
-int iommu_enable_x2apic_IR(void);
-void iommu_disable_x2apic_IR(void);
int iommu_add_device(struct pci_dev *pdev);
int iommu_enable_device(struct pci_dev *pdev);
@@ -59,6 +55,9 @@ void iommu_dom0_init(struct domain *d);
void iommu_domain_destroy(struct domain *d);
int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn);
+/* Function used internally, use iommu_domain_destroy */
+void iommu_teardown(struct domain *d);
+
/* iommu_map_page() takes flags to direct the mapping operation. */
#define _IOMMUF_readable 0
#define IOMMUF_readable (1u<<_IOMMUF_readable)
@@ -67,9 +66,8 @@ int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn);
int iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,
unsigned int flags);
int iommu_unmap_page(struct domain *d, unsigned long gfn);
-void iommu_pte_flush(struct domain *d, u64 gfn, u64 *pte, int order, int present);
-void iommu_domain_teardown(struct domain *d);
+#ifdef HAS_PCI
void pt_pci_init(void);
struct pirq;
@@ -84,62 +82,60 @@ void free_hvm_irq_dpci(struct hvm_irq_dpci *dpci);
bool_t pt_irq_need_timer(uint32_t flags);
#define PT_IRQ_TIME_OUT MILLISECS(8)
+#endif /* HAS_PCI */
+#ifdef CONFIG_X86
struct msi_desc;
struct msi_msg;
+#endif /* CONFIG_X86 */
+
struct page_info;
struct iommu_ops {
int (*init)(struct domain *d);
void (*dom0_init)(struct domain *d);
+#ifdef HAS_PCI
int (*add_device)(u8 devfn, struct pci_dev *);
int (*enable_device)(struct pci_dev *pdev);
int (*remove_device)(u8 devfn, struct pci_dev *);
int (*assign_device)(struct domain *, u8 devfn, struct pci_dev *);
+ int (*reassign_device)(struct domain *s, struct domain *t,
+ u8 devfn, struct pci_dev *);
+ int (*get_device_group_id)(u16 seg, u8 bus, u8 devfn);
+#endif /* HAS_PCI */
void (*teardown)(struct domain *d);
int (*map_page)(struct domain *d, unsigned long gfn, unsigned long mfn,
unsigned int flags);
int (*unmap_page)(struct domain *d, unsigned long gfn);
void (*free_page_table)(struct page_info *);
- int (*reassign_device)(struct domain *s, struct domain *t,
- u8 devfn, struct pci_dev *);
- int (*get_device_group_id)(u16 seg, u8 bus, u8 devfn);
+#ifdef CONFIG_X86
void (*update_ire_from_apic)(unsigned int apic, unsigned int reg, unsigned int value);
int (*update_ire_from_msi)(struct msi_desc *msi_desc, struct msi_msg *msg);
void (*read_msi_from_ire)(struct msi_desc *msi_desc, struct msi_msg *msg);
unsigned int (*read_apic_from_ire)(unsigned int apic, unsigned int reg);
int (*setup_hpet_msi)(struct msi_desc *);
+ void (*share_p2m)(struct domain *d);
+#endif /* CONFIG_X86 */
void (*suspend)(void);
void (*resume)(void);
- void (*share_p2m)(struct domain *d);
void (*crash_shutdown)(void);
void (*iotlb_flush)(struct domain *d, unsigned long gfn, unsigned int page_count);
void (*iotlb_flush_all)(struct domain *d);
void (*dump_p2m_table)(struct domain *d);
};
-void iommu_update_ire_from_apic(unsigned int apic, unsigned int reg, unsigned int value);
-int iommu_update_ire_from_msi(struct msi_desc *msi_desc, struct msi_msg *msg);
-void iommu_read_msi_from_ire(struct msi_desc *msi_desc, struct msi_msg *msg);
-unsigned int iommu_read_apic_from_ire(unsigned int apic, unsigned int reg);
-int iommu_setup_hpet_msi(struct msi_desc *);
-
void iommu_suspend(void);
void iommu_resume(void);
void iommu_crash_shutdown(void);
-void iommu_set_dom0_mapping(struct domain *d);
-void iommu_share_p2m_table(struct domain *d);
-
+#if HAS_PCI
int iommu_do_domctl(struct xen_domctl *, struct domain *d,
XEN_GUEST_HANDLE_PARAM(xen_domctl_t));
+#endif
void iommu_iotlb_flush(struct domain *d, unsigned long gfn, unsigned int page_count);
void iommu_iotlb_flush_all(struct domain *d);
-/* While VT-d specific, this must get declared in a generic header. */
-int adjust_vtd_irq_affinities(void);
-
/*
* The purpose of the iommu_dont_flush_iotlb optional cpu flag is to
* avoid unecessary iotlb_flush in the low level IOMMU code.
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* [RFC for-4.5 09/12] xen/passthrough: iommu: Introduce arch specific code
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
` (7 preceding siblings ...)
2014-02-07 17:43 ` [RFC for-4.5 08/12] xen/passthrough: iommu: Split generic IOMMU code Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-19 12:40 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 10/12] xen/passthrough: Introduce IOMMU ARM architure Julien Grall
` (3 subsequent siblings)
12 siblings, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel
Cc: Joseph Cihula, Keir Fraser, ian.campbell, patches, Shane Wang,
Julien Grall, tim, stefano.stabellini, Jan Beulich,
Suravee Suthikulpanit, Gang Wei, Xiantao Zhang
Currently the structure hvm_iommu (xen/include/xen/hvm/iommu.h) contains
x86 specific fields.
This patch creates:
- arch_hvm_iommu structure which will contain architecture depend
fields
- arch_iommu_domain_{init,destroy} function to execute arch
specific during domain creation/destruction
Also move iommu_use_hap_pt and domain_hvm_iommu in asm-x86/iommu.h.
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Cc: Keir Fraser <keir@xen.org>
Cc: Jan Beulich <jbeulich@suse.com>
Cc: Joseph Cihula <joseph.cihula@intel.com>
Cc: Gang Wei <gang.wei@intel.com>
Cc: Shane Wang <shane.wang@intel.com>
Cc: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Cc: Xiantao Zhang <xiantao.zhang@intel.com>
---
xen/arch/x86/domctl.c | 6 +-
xen/arch/x86/hvm/io.c | 2 +-
xen/arch/x86/tboot.c | 3 +-
xen/drivers/passthrough/amd/iommu_cmd.c | 2 +-
xen/drivers/passthrough/amd/iommu_guest.c | 8 +--
xen/drivers/passthrough/amd/iommu_map.c | 56 +++++++++----------
xen/drivers/passthrough/amd/pci_amd_iommu.c | 53 +++++++++---------
xen/drivers/passthrough/iommu.c | 36 +++---------
xen/drivers/passthrough/iommu_x86.c | 41 ++++++++++++++
xen/drivers/passthrough/vtd/iommu.c | 80 +++++++++++++--------------
xen/include/asm-x86/hvm/iommu.h | 29 ++++++++++
xen/include/asm-x86/iommu.h | 4 ++
xen/include/xen/hvm/iommu.h | 26 +--------
xen/include/xen/iommu.h | 8 +--
14 files changed, 191 insertions(+), 163 deletions(-)
diff --git a/xen/arch/x86/domctl.c b/xen/arch/x86/domctl.c
index 26635ff..e55d9d5 100644
--- a/xen/arch/x86/domctl.c
+++ b/xen/arch/x86/domctl.c
@@ -745,7 +745,7 @@ long arch_do_domctl(
"ioport_map:add: dom%d gport=%x mport=%x nr=%x\n",
d->domain_id, fgp, fmp, np);
- list_for_each_entry(g2m_ioport, &hd->g2m_ioport_list, list)
+ list_for_each_entry(g2m_ioport, &hd->arch.g2m_ioport_list, list)
if (g2m_ioport->mport == fmp )
{
g2m_ioport->gport = fgp;
@@ -764,7 +764,7 @@ long arch_do_domctl(
g2m_ioport->gport = fgp;
g2m_ioport->mport = fmp;
g2m_ioport->np = np;
- list_add_tail(&g2m_ioport->list, &hd->g2m_ioport_list);
+ list_add_tail(&g2m_ioport->list, &hd->arch.g2m_ioport_list);
}
if ( !ret )
ret = ioports_permit_access(d, fmp, fmp + np - 1);
@@ -779,7 +779,7 @@ long arch_do_domctl(
printk(XENLOG_G_INFO
"ioport_map:remove: dom%d gport=%x mport=%x nr=%x\n",
d->domain_id, fgp, fmp, np);
- list_for_each_entry(g2m_ioport, &hd->g2m_ioport_list, list)
+ list_for_each_entry(g2m_ioport, &hd->arch.g2m_ioport_list, list)
if ( g2m_ioport->mport == fmp )
{
list_del(&g2m_ioport->list);
diff --git a/xen/arch/x86/hvm/io.c b/xen/arch/x86/hvm/io.c
index bf6309d..ddb03f8 100644
--- a/xen/arch/x86/hvm/io.c
+++ b/xen/arch/x86/hvm/io.c
@@ -451,7 +451,7 @@ int dpci_ioport_intercept(ioreq_t *p)
unsigned int s = 0, e = 0;
int rc;
- list_for_each_entry( g2m_ioport, &hd->g2m_ioport_list, list )
+ list_for_each_entry( g2m_ioport, &hd->arch.g2m_ioport_list, list )
{
s = g2m_ioport->gport;
e = s + g2m_ioport->np;
diff --git a/xen/arch/x86/tboot.c b/xen/arch/x86/tboot.c
index ccde4a0..c40fe12 100644
--- a/xen/arch/x86/tboot.c
+++ b/xen/arch/x86/tboot.c
@@ -230,7 +230,8 @@ static void tboot_gen_domain_integrity(const uint8_t key[TB_KEY_SIZE],
if ( !is_idle_domain(d) )
{
struct hvm_iommu *hd = domain_hvm_iommu(d);
- update_iommu_mac(&ctx, hd->pgd_maddr, agaw_to_level(hd->agaw));
+ update_iommu_mac(&ctx, hd->arch.pgd_maddr,
+ agaw_to_level(hd->arch.agaw));
}
}
diff --git a/xen/drivers/passthrough/amd/iommu_cmd.c b/xen/drivers/passthrough/amd/iommu_cmd.c
index d27bd3c..f39bd9d 100644
--- a/xen/drivers/passthrough/amd/iommu_cmd.c
+++ b/xen/drivers/passthrough/amd/iommu_cmd.c
@@ -355,7 +355,7 @@ static void _amd_iommu_flush_pages(struct domain *d,
unsigned long flags;
struct amd_iommu *iommu;
struct hvm_iommu *hd = domain_hvm_iommu(d);
- unsigned int dom_id = hd->domain_id;
+ unsigned int dom_id = hd->arch.domain_id;
/* send INVALIDATE_IOMMU_PAGES command */
for_each_amd_iommu ( iommu )
diff --git a/xen/drivers/passthrough/amd/iommu_guest.c b/xen/drivers/passthrough/amd/iommu_guest.c
index 477de20..bd31bb5 100644
--- a/xen/drivers/passthrough/amd/iommu_guest.c
+++ b/xen/drivers/passthrough/amd/iommu_guest.c
@@ -60,12 +60,12 @@ static uint16_t guest_bdf(struct domain *d, uint16_t machine_bdf)
static inline struct guest_iommu *domain_iommu(struct domain *d)
{
- return domain_hvm_iommu(d)->g_iommu;
+ return domain_hvm_iommu(d)->arch.g_iommu;
}
static inline struct guest_iommu *vcpu_iommu(struct vcpu *v)
{
- return domain_hvm_iommu(v->domain)->g_iommu;
+ return domain_hvm_iommu(v->domain)->arch.g_iommu;
}
static void guest_iommu_enable(struct guest_iommu *iommu)
@@ -886,7 +886,7 @@ int guest_iommu_init(struct domain* d)
guest_iommu_reg_init(iommu);
iommu->domain = d;
- hd->g_iommu = iommu;
+ hd->arch.g_iommu = iommu;
tasklet_init(&iommu->cmd_buffer_tasklet,
guest_iommu_process_command, (unsigned long)d);
@@ -907,7 +907,7 @@ void guest_iommu_destroy(struct domain *d)
tasklet_kill(&iommu->cmd_buffer_tasklet);
xfree(iommu);
- domain_hvm_iommu(d)->g_iommu = NULL;
+ domain_hvm_iommu(d)->arch.g_iommu = NULL;
}
static int guest_iommu_mmio_range(struct vcpu *v, unsigned long addr)
diff --git a/xen/drivers/passthrough/amd/iommu_map.c b/xen/drivers/passthrough/amd/iommu_map.c
index 1294561..be34e90 100644
--- a/xen/drivers/passthrough/amd/iommu_map.c
+++ b/xen/drivers/passthrough/amd/iommu_map.c
@@ -344,7 +344,7 @@ static int iommu_update_pde_count(struct domain *d, unsigned long pt_mfn,
struct hvm_iommu *hd = domain_hvm_iommu(d);
bool_t ok = 0;
- ASSERT( spin_is_locked(&hd->mapping_lock) && pt_mfn );
+ ASSERT( spin_is_locked(&hd->arch.mapping_lock) && pt_mfn );
next_level = merge_level - 1;
@@ -398,7 +398,7 @@ static int iommu_merge_pages(struct domain *d, unsigned long pt_mfn,
unsigned long first_mfn;
struct hvm_iommu *hd = domain_hvm_iommu(d);
- ASSERT( spin_is_locked(&hd->mapping_lock) && pt_mfn );
+ ASSERT( spin_is_locked(&hd->arch.mapping_lock) && pt_mfn );
table = map_domain_page(pt_mfn);
pde = table + pfn_to_pde_idx(gfn, merge_level);
@@ -448,8 +448,8 @@ static int iommu_pde_from_gfn(struct domain *d, unsigned long pfn,
struct page_info *table;
struct hvm_iommu *hd = domain_hvm_iommu(d);
- table = hd->root_table;
- level = hd->paging_mode;
+ table = hd->arch.root_table;
+ level = hd->arch.paging_mode;
BUG_ON( table == NULL || level < IOMMU_PAGING_MODE_LEVEL_1 ||
level > IOMMU_PAGING_MODE_LEVEL_6 );
@@ -557,11 +557,11 @@ static int update_paging_mode(struct domain *d, unsigned long gfn)
unsigned long old_root_mfn;
struct hvm_iommu *hd = domain_hvm_iommu(d);
- level = hd->paging_mode;
- old_root = hd->root_table;
+ level = hd->arch.paging_mode;
+ old_root = hd->arch.root_table;
offset = gfn >> (PTE_PER_TABLE_SHIFT * (level - 1));
- ASSERT(spin_is_locked(&hd->mapping_lock) && is_hvm_domain(d));
+ ASSERT(spin_is_locked(&hd->arch.mapping_lock) && is_hvm_domain(d));
while ( offset >= PTE_PER_TABLE_SIZE )
{
@@ -587,8 +587,8 @@ static int update_paging_mode(struct domain *d, unsigned long gfn)
if ( new_root != NULL )
{
- hd->paging_mode = level;
- hd->root_table = new_root;
+ hd->arch.paging_mode = level;
+ hd->arch.root_table = new_root;
if ( !spin_is_locked(&pcidevs_lock) )
AMD_IOMMU_DEBUG("%s Try to access pdev_list "
@@ -613,9 +613,9 @@ static int update_paging_mode(struct domain *d, unsigned long gfn)
/* valid = 0 only works for dom0 passthrough mode */
amd_iommu_set_root_page_table((u32 *)device_entry,
- page_to_maddr(hd->root_table),
- hd->domain_id,
- hd->paging_mode, 1);
+ page_to_maddr(hd->arch.root_table),
+ hd->arch.domain_id,
+ hd->arch.paging_mode, 1);
amd_iommu_flush_device(iommu, req_id);
bdf += pdev->phantom_stride;
@@ -638,14 +638,14 @@ int amd_iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,
unsigned long pt_mfn[7];
unsigned int merge_level;
- BUG_ON( !hd->root_table );
+ BUG_ON( !hd->arch.root_table );
if ( iommu_use_hap_pt(d) )
return 0;
memset(pt_mfn, 0, sizeof(pt_mfn));
- spin_lock(&hd->mapping_lock);
+ spin_lock(&hd->arch.mapping_lock);
/* Since HVM domain is initialized with 2 level IO page table,
* we might need a deeper page table for lager gfn now */
@@ -653,7 +653,7 @@ int amd_iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,
{
if ( update_paging_mode(d, gfn) )
{
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
AMD_IOMMU_DEBUG("Update page mode failed gfn = %lx\n", gfn);
domain_crash(d);
return -EFAULT;
@@ -662,7 +662,7 @@ int amd_iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,
if ( iommu_pde_from_gfn(d, gfn, pt_mfn) || (pt_mfn[1] == 0) )
{
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
AMD_IOMMU_DEBUG("Invalid IO pagetable entry gfn = %lx\n", gfn);
domain_crash(d);
return -EFAULT;
@@ -684,7 +684,7 @@ int amd_iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,
amd_iommu_flush_pages(d, gfn, 0);
for ( merge_level = IOMMU_PAGING_MODE_LEVEL_2;
- merge_level <= hd->paging_mode; merge_level++ )
+ merge_level <= hd->arch.paging_mode; merge_level++ )
{
if ( pt_mfn[merge_level] == 0 )
break;
@@ -697,7 +697,7 @@ int amd_iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,
if ( iommu_merge_pages(d, pt_mfn[merge_level], gfn,
flags, merge_level) )
{
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
AMD_IOMMU_DEBUG("Merge iommu page failed at level %d, "
"gfn = %lx mfn = %lx\n", merge_level, gfn, mfn);
domain_crash(d);
@@ -706,7 +706,7 @@ int amd_iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,
}
out:
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
return 0;
}
@@ -715,14 +715,14 @@ int amd_iommu_unmap_page(struct domain *d, unsigned long gfn)
unsigned long pt_mfn[7];
struct hvm_iommu *hd = domain_hvm_iommu(d);
- BUG_ON( !hd->root_table );
+ BUG_ON( !hd->arch.root_table );
if ( iommu_use_hap_pt(d) )
return 0;
memset(pt_mfn, 0, sizeof(pt_mfn));
- spin_lock(&hd->mapping_lock);
+ spin_lock(&hd->arch.mapping_lock);
/* Since HVM domain is initialized with 2 level IO page table,
* we might need a deeper page table for lager gfn now */
@@ -730,7 +730,7 @@ int amd_iommu_unmap_page(struct domain *d, unsigned long gfn)
{
if ( update_paging_mode(d, gfn) )
{
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
AMD_IOMMU_DEBUG("Update page mode failed gfn = %lx\n", gfn);
domain_crash(d);
return -EFAULT;
@@ -739,7 +739,7 @@ int amd_iommu_unmap_page(struct domain *d, unsigned long gfn)
if ( iommu_pde_from_gfn(d, gfn, pt_mfn) || (pt_mfn[1] == 0) )
{
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
AMD_IOMMU_DEBUG("Invalid IO pagetable entry gfn = %lx\n", gfn);
domain_crash(d);
return -EFAULT;
@@ -747,7 +747,7 @@ int amd_iommu_unmap_page(struct domain *d, unsigned long gfn)
/* mark PTE as 'page not present' */
clear_iommu_pte_present(pt_mfn[1], gfn);
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
amd_iommu_flush_pages(d, gfn, 0);
@@ -792,13 +792,13 @@ void amd_iommu_share_p2m(struct domain *d)
pgd_mfn = pagetable_get_mfn(p2m_get_pagetable(p2m_get_hostp2m(d)));
p2m_table = mfn_to_page(mfn_x(pgd_mfn));
- if ( hd->root_table != p2m_table )
+ if ( hd->arch.root_table != p2m_table )
{
- free_amd_iommu_pgtable(hd->root_table);
- hd->root_table = p2m_table;
+ free_amd_iommu_pgtable(hd->arch.root_table);
+ hd->arch.root_table = p2m_table;
/* When sharing p2m with iommu, paging mode = 4 */
- hd->paging_mode = IOMMU_PAGING_MODE_LEVEL_4;
+ hd->arch.paging_mode = IOMMU_PAGING_MODE_LEVEL_4;
AMD_IOMMU_DEBUG("Share p2m table with iommu: p2m table = %#lx\n",
mfn_x(pgd_mfn));
}
diff --git a/xen/drivers/passthrough/amd/pci_amd_iommu.c b/xen/drivers/passthrough/amd/pci_amd_iommu.c
index c26aabc..0c3cd3e 100644
--- a/xen/drivers/passthrough/amd/pci_amd_iommu.c
+++ b/xen/drivers/passthrough/amd/pci_amd_iommu.c
@@ -120,7 +120,8 @@ static void amd_iommu_setup_domain_device(
struct hvm_iommu *hd = domain_hvm_iommu(domain);
- BUG_ON( !hd->root_table || !hd->paging_mode || !iommu->dev_table.buffer );
+ BUG_ON( !hd->arch.root_table || !hd->arch.paging_mode ||
+ !iommu->dev_table.buffer );
if ( iommu_passthrough && (domain->domain_id == 0) )
valid = 0;
@@ -138,8 +139,8 @@ static void amd_iommu_setup_domain_device(
{
/* bind DTE to domain page-tables */
amd_iommu_set_root_page_table(
- (u32 *)dte, page_to_maddr(hd->root_table), hd->domain_id,
- hd->paging_mode, valid);
+ (u32 *)dte, page_to_maddr(hd->arch.root_table), hd->arch.domain_id,
+ hd->arch.paging_mode, valid);
if ( pci_ats_device(iommu->seg, bus, pdev->devfn) &&
iommu_has_cap(iommu, PCI_CAP_IOTLB_SHIFT) )
@@ -151,8 +152,8 @@ static void amd_iommu_setup_domain_device(
"root table = %#"PRIx64", "
"domain = %d, paging mode = %d\n",
req_id, pdev->type,
- page_to_maddr(hd->root_table),
- hd->domain_id, hd->paging_mode);
+ page_to_maddr(hd->arch.root_table),
+ hd->arch.domain_id, hd->arch.paging_mode);
}
spin_unlock_irqrestore(&iommu->lock, flags);
@@ -225,17 +226,17 @@ int __init amd_iov_detect(void)
static int allocate_domain_resources(struct hvm_iommu *hd)
{
/* allocate root table */
- spin_lock(&hd->mapping_lock);
- if ( !hd->root_table )
+ spin_lock(&hd->arch.mapping_lock);
+ if ( !hd->arch.root_table )
{
- hd->root_table = alloc_amd_iommu_pgtable();
- if ( !hd->root_table )
+ hd->arch.root_table = alloc_amd_iommu_pgtable();
+ if ( !hd->arch.root_table )
{
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
return -ENOMEM;
}
}
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
return 0;
}
@@ -262,18 +263,18 @@ static int amd_iommu_domain_init(struct domain *d)
/* allocate page directroy */
if ( allocate_domain_resources(hd) != 0 )
{
- if ( hd->root_table )
- free_domheap_page(hd->root_table);
+ if ( hd->arch.root_table )
+ free_domheap_page(hd->arch.root_table);
return -ENOMEM;
}
/* For pv and dom0, stick with get_paging_mode(max_page)
* For HVM dom0, use 2 level page table at first */
- hd->paging_mode = is_hvm_domain(d) ?
+ hd->arch.paging_mode = is_hvm_domain(d) ?
IOMMU_PAGING_MODE_LEVEL_2 :
get_paging_mode(max_page);
- hd->domain_id = d->domain_id;
+ hd->arch.domain_id = d->domain_id;
guest_iommu_init(d);
@@ -333,8 +334,8 @@ void amd_iommu_disable_domain_device(struct domain *domain,
AMD_IOMMU_DEBUG("Disable: device id = %#x, "
"domain = %d, paging mode = %d\n",
- req_id, domain_hvm_iommu(domain)->domain_id,
- domain_hvm_iommu(domain)->paging_mode);
+ req_id, domain_hvm_iommu(domain)->arch.domain_id,
+ domain_hvm_iommu(domain)->arch.paging_mode);
}
spin_unlock_irqrestore(&iommu->lock, flags);
@@ -374,7 +375,7 @@ static int reassign_device(struct domain *source, struct domain *target,
/* IO page tables might be destroyed after pci-detach the last device
* In this case, we have to re-allocate root table for next pci-attach.*/
- if ( t->root_table == NULL )
+ if ( t->arch.root_table == NULL )
allocate_domain_resources(t);
amd_iommu_setup_domain_device(target, iommu, devfn, pdev);
@@ -456,13 +457,13 @@ static void deallocate_iommu_page_tables(struct domain *d)
if ( iommu_use_hap_pt(d) )
return;
- spin_lock(&hd->mapping_lock);
- if ( hd->root_table )
+ spin_lock(&hd->arch.mapping_lock);
+ if ( hd->arch.root_table )
{
- deallocate_next_page_table(hd->root_table, hd->paging_mode);
- hd->root_table = NULL;
+ deallocate_next_page_table(hd->arch.root_table, hd->arch.paging_mode);
+ hd->arch.root_table = NULL;
}
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
}
@@ -593,11 +594,11 @@ static void amd_dump_p2m_table(struct domain *d)
{
struct hvm_iommu *hd = domain_hvm_iommu(d);
- if ( !hd->root_table )
+ if ( !hd->arch.root_table )
return;
- printk("p2m table has %d levels\n", hd->paging_mode);
- amd_dump_p2m_table_level(hd->root_table, hd->paging_mode, 0, 0);
+ printk("p2m table has %d levels\n", hd->arch.paging_mode);
+ amd_dump_p2m_table_level(hd->arch.root_table, hd->arch.paging_mode, 0, 0);
}
const struct iommu_ops amd_iommu_ops = {
diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
index d733878..2346da9 100644
--- a/xen/drivers/passthrough/iommu.c
+++ b/xen/drivers/passthrough/iommu.c
@@ -117,10 +117,11 @@ static void __init parse_iommu_param(char *s)
int iommu_domain_init(struct domain *d)
{
struct hvm_iommu *hd = domain_hvm_iommu(d);
+ int ret = 0;
- spin_lock_init(&hd->mapping_lock);
- INIT_LIST_HEAD(&hd->g2m_ioport_list);
- INIT_LIST_HEAD(&hd->mapped_rmrrs);
+ ret = arch_iommu_domain_init(d);
+ if ( ret )
+ return ret;
if ( !iommu_enabled )
return 0;
@@ -190,10 +191,7 @@ void iommu_teardown(struct domain *d)
void iommu_domain_destroy(struct domain *d)
{
- struct hvm_iommu *hd = domain_hvm_iommu(d);
- struct list_head *ioport_list, *rmrr_list, *tmp;
- struct g2m_ioport *ioport;
- struct mapped_rmrr *mrmrr;
+ struct hvm_iommu *hd = domain_hvm_iommu(d);
if ( !iommu_enabled || !hd->platform_ops )
return;
@@ -201,20 +199,8 @@ void iommu_domain_destroy(struct domain *d)
if ( need_iommu(d) )
iommu_teardown(d);
- list_for_each_safe ( ioport_list, tmp, &hd->g2m_ioport_list )
- {
- ioport = list_entry(ioport_list, struct g2m_ioport, list);
- list_del(&ioport->list);
- xfree(ioport);
- }
-
- list_for_each_safe ( rmrr_list, tmp, &hd->mapped_rmrrs )
- {
- mrmrr = list_entry(rmrr_list, struct mapped_rmrr, list);
- list_del(&mrmrr->list);
- xfree(mrmrr);
- }
-}
+ arch_iommu_domain_destroy(d);
+ }
int iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,
unsigned int flags)
@@ -328,14 +314,6 @@ void iommu_suspend()
ops->suspend();
}
-void iommu_share_p2m_table(struct domain* d)
-{
- const struct iommu_ops *ops = iommu_get_ops();
-
- if ( iommu_enabled && is_hvm_domain(d) )
- ops->share_p2m(d);
-}
-
void iommu_crash_shutdown(void)
{
const struct iommu_ops *ops = iommu_get_ops();
diff --git a/xen/drivers/passthrough/iommu_x86.c b/xen/drivers/passthrough/iommu_x86.c
index bd3c23b..c137cef 100644
--- a/xen/drivers/passthrough/iommu_x86.c
+++ b/xen/drivers/passthrough/iommu_x86.c
@@ -55,6 +55,47 @@ int __init iommu_setup_hpet_msi(struct msi_desc *msi)
return ops->setup_hpet_msi ? ops->setup_hpet_msi(msi) : -ENODEV;
}
+void iommu_share_p2m_table(struct domain* d)
+{
+ const struct iommu_ops *ops = iommu_get_ops();
+
+ if ( iommu_enabled && is_hvm_domain(d) )
+ ops->share_p2m(d);
+}
+
+int arch_iommu_domain_init(struct domain *d)
+{
+ struct hvm_iommu *hd = domain_hvm_iommu(d);
+
+ spin_lock_init(&hd->arch.mapping_lock);
+ INIT_LIST_HEAD(&hd->arch.g2m_ioport_list);
+ INIT_LIST_HEAD(&hd->arch.mapped_rmrrs);
+
+ return 0;
+}
+
+void arch_iommu_domain_destroy(struct domain *d)
+{
+ struct hvm_iommu *hd = domain_hvm_iommu(d);
+ struct list_head *ioport_list, *rmrr_list, *tmp;
+ struct g2m_ioport *ioport;
+ struct mapped_rmrr *mrmrr;
+
+ list_for_each_safe ( ioport_list, tmp, &hd->arch.g2m_ioport_list )
+ {
+ ioport = list_entry(ioport_list, struct g2m_ioport, list);
+ list_del(&ioport->list);
+ xfree(ioport);
+ }
+
+ list_for_each_safe ( rmrr_list, tmp, &hd->arch.mapped_rmrrs )
+ {
+ mrmrr = list_entry(rmrr_list, struct mapped_rmrr, list);
+ list_del(&mrmrr->list);
+ xfree(mrmrr);
+ }
+}
+
/*
* Local variables:
* mode: C
diff --git a/xen/drivers/passthrough/vtd/iommu.c b/xen/drivers/passthrough/vtd/iommu.c
index faa794b..a7a5253 100644
--- a/xen/drivers/passthrough/vtd/iommu.c
+++ b/xen/drivers/passthrough/vtd/iommu.c
@@ -249,16 +249,16 @@ static u64 addr_to_dma_page_maddr(struct domain *domain, u64 addr, int alloc)
struct acpi_drhd_unit *drhd;
struct pci_dev *pdev;
struct hvm_iommu *hd = domain_hvm_iommu(domain);
- int addr_width = agaw_to_width(hd->agaw);
+ int addr_width = agaw_to_width(hd->arch.agaw);
struct dma_pte *parent, *pte = NULL;
- int level = agaw_to_level(hd->agaw);
+ int level = agaw_to_level(hd->arch.agaw);
int offset;
u64 pte_maddr = 0, maddr;
u64 *vaddr = NULL;
addr &= (((u64)1) << addr_width) - 1;
- ASSERT(spin_is_locked(&hd->mapping_lock));
- if ( hd->pgd_maddr == 0 )
+ ASSERT(spin_is_locked(&hd->arch.mapping_lock));
+ if ( hd->arch.pgd_maddr == 0 )
{
/*
* just get any passthrough device in the domainr - assume user
@@ -266,11 +266,11 @@ static u64 addr_to_dma_page_maddr(struct domain *domain, u64 addr, int alloc)
*/
pdev = pci_get_pdev_by_domain(domain, -1, -1, -1);
drhd = acpi_find_matched_drhd_unit(pdev);
- if ( !alloc || ((hd->pgd_maddr = alloc_pgtable_maddr(drhd, 1)) == 0) )
+ if ( !alloc || ((hd->arch.pgd_maddr = alloc_pgtable_maddr(drhd, 1)) == 0) )
goto out;
}
- parent = (struct dma_pte *)map_vtd_domain_page(hd->pgd_maddr);
+ parent = (struct dma_pte *)map_vtd_domain_page(hd->arch.pgd_maddr);
while ( level > 1 )
{
offset = address_level_offset(addr, level);
@@ -580,7 +580,7 @@ static void __intel_iommu_iotlb_flush(struct domain *d, unsigned long gfn,
{
iommu = drhd->iommu;
- if ( !test_bit(iommu->index, &hd->iommu_bitmap) )
+ if ( !test_bit(iommu->index, &hd->arch.iommu_bitmap) )
continue;
flush_dev_iotlb = find_ats_dev_drhd(iommu) ? 1 : 0;
@@ -622,12 +622,12 @@ static void dma_pte_clear_one(struct domain *domain, u64 addr)
u64 pg_maddr;
struct mapped_rmrr *mrmrr;
- spin_lock(&hd->mapping_lock);
+ spin_lock(&hd->arch.mapping_lock);
/* get last level pte */
pg_maddr = addr_to_dma_page_maddr(domain, addr, 0);
if ( pg_maddr == 0 )
{
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
return;
}
@@ -636,13 +636,13 @@ static void dma_pte_clear_one(struct domain *domain, u64 addr)
if ( !dma_pte_present(*pte) )
{
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
unmap_vtd_domain_page(page);
return;
}
dma_clear_pte(*pte);
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
iommu_flush_cache_entry(pte, sizeof(struct dma_pte));
if ( !this_cpu(iommu_dont_flush_iotlb) )
@@ -653,8 +653,8 @@ static void dma_pte_clear_one(struct domain *domain, u64 addr)
/* if the cleared address is between mapped RMRR region,
* remove the mapped RMRR
*/
- spin_lock(&hd->mapping_lock);
- list_for_each_entry ( mrmrr, &hd->mapped_rmrrs, list )
+ spin_lock(&hd->arch.mapping_lock);
+ list_for_each_entry ( mrmrr, &hd->arch.mapped_rmrrs, list )
{
if ( addr >= mrmrr->base && addr <= mrmrr->end )
{
@@ -663,7 +663,7 @@ static void dma_pte_clear_one(struct domain *domain, u64 addr)
break;
}
}
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
}
static void iommu_free_pagetable(u64 pt_maddr, int level)
@@ -1248,7 +1248,7 @@ static int intel_iommu_domain_init(struct domain *d)
{
struct hvm_iommu *hd = domain_hvm_iommu(d);
- hd->agaw = width_to_agaw(DEFAULT_DOMAIN_ADDRESS_WIDTH);
+ hd->arch.agaw = width_to_agaw(DEFAULT_DOMAIN_ADDRESS_WIDTH);
return 0;
}
@@ -1345,16 +1345,16 @@ int domain_context_mapping_one(
}
else
{
- spin_lock(&hd->mapping_lock);
+ spin_lock(&hd->arch.mapping_lock);
/* Ensure we have pagetables allocated down to leaf PTE. */
- if ( hd->pgd_maddr == 0 )
+ if ( hd->arch.pgd_maddr == 0 )
{
addr_to_dma_page_maddr(domain, 0, 1);
- if ( hd->pgd_maddr == 0 )
+ if ( hd->arch.pgd_maddr == 0 )
{
nomem:
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
spin_unlock(&iommu->lock);
unmap_vtd_domain_page(context_entries);
return -ENOMEM;
@@ -1362,7 +1362,7 @@ int domain_context_mapping_one(
}
/* Skip top levels of page tables for 2- and 3-level DRHDs. */
- pgd_maddr = hd->pgd_maddr;
+ pgd_maddr = hd->arch.pgd_maddr;
for ( agaw = level_to_agaw(4);
agaw != level_to_agaw(iommu->nr_pt_levels);
agaw-- )
@@ -1380,7 +1380,7 @@ int domain_context_mapping_one(
else
context_set_translation_type(*context, CONTEXT_TT_MULTI_LEVEL);
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
}
if ( context_set_domain_id(context, domain, iommu) )
@@ -1406,7 +1406,7 @@ int domain_context_mapping_one(
iommu_flush_iotlb_dsi(iommu, 0, 1, flush_dev_iotlb);
}
- set_bit(iommu->index, &hd->iommu_bitmap);
+ set_bit(iommu->index, &hd->arch.iommu_bitmap);
unmap_vtd_domain_page(context_entries);
@@ -1652,7 +1652,7 @@ static int domain_context_unmap(
struct hvm_iommu *hd = domain_hvm_iommu(domain);
int iommu_domid;
- clear_bit(iommu->index, &hd->iommu_bitmap);
+ clear_bit(iommu->index, &hd->arch.iommu_bitmap);
iommu_domid = domain_iommu_domid(domain, iommu);
if ( iommu_domid == -1 )
@@ -1711,10 +1711,10 @@ static void iommu_domain_teardown(struct domain *d)
if ( iommu_use_hap_pt(d) )
return;
- spin_lock(&hd->mapping_lock);
- iommu_free_pagetable(hd->pgd_maddr, agaw_to_level(hd->agaw));
- hd->pgd_maddr = 0;
- spin_unlock(&hd->mapping_lock);
+ spin_lock(&hd->arch.mapping_lock);
+ iommu_free_pagetable(hd->arch.pgd_maddr, agaw_to_level(hd->arch.agaw));
+ hd->arch.pgd_maddr = 0;
+ spin_unlock(&hd->arch.mapping_lock);
}
static int intel_iommu_map_page(
@@ -1733,12 +1733,12 @@ static int intel_iommu_map_page(
if ( iommu_passthrough && (d->domain_id == 0) )
return 0;
- spin_lock(&hd->mapping_lock);
+ spin_lock(&hd->arch.mapping_lock);
pg_maddr = addr_to_dma_page_maddr(d, (paddr_t)gfn << PAGE_SHIFT_4K, 1);
if ( pg_maddr == 0 )
{
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
return -ENOMEM;
}
page = (struct dma_pte *)map_vtd_domain_page(pg_maddr);
@@ -1755,14 +1755,14 @@ static int intel_iommu_map_page(
if ( old.val == new.val )
{
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
unmap_vtd_domain_page(page);
return 0;
}
*pte = new;
iommu_flush_cache_entry(pte, sizeof(struct dma_pte));
- spin_unlock(&hd->mapping_lock);
+ spin_unlock(&hd->arch.mapping_lock);
unmap_vtd_domain_page(page);
if ( !this_cpu(iommu_dont_flush_iotlb) )
@@ -1796,7 +1796,7 @@ void iommu_pte_flush(struct domain *d, u64 gfn, u64 *pte,
for_each_drhd_unit ( drhd )
{
iommu = drhd->iommu;
- if ( !test_bit(iommu->index, &hd->iommu_bitmap) )
+ if ( !test_bit(iommu->index, &hd->arch.iommu_bitmap) )
continue;
flush_dev_iotlb = find_ats_dev_drhd(iommu) ? 1 : 0;
@@ -1837,7 +1837,7 @@ static void iommu_set_pgd(struct domain *d)
return;
pgd_mfn = pagetable_get_mfn(p2m_get_pagetable(p2m_get_hostp2m(d)));
- hd->pgd_maddr = pagetable_get_paddr(pagetable_from_mfn(pgd_mfn));
+ hd->arch.pgd_maddr = pagetable_get_paddr(pagetable_from_mfn(pgd_mfn));
}
static int rmrr_identity_mapping(struct domain *d,
@@ -1852,10 +1852,10 @@ static int rmrr_identity_mapping(struct domain *d,
ASSERT(rmrr->base_address < rmrr->end_address);
/*
- * No need to acquire hd->mapping_lock, as the only theoretical race is
+ * No need to acquire hd->arch.mapping_lock, as the only theoretical race is
* with the insertion below (impossible due to holding pcidevs_lock).
*/
- list_for_each_entry( mrmrr, &hd->mapped_rmrrs, list )
+ list_for_each_entry( mrmrr, &hd->arch.mapped_rmrrs, list )
{
if ( mrmrr->base == rmrr->base_address &&
mrmrr->end == rmrr->end_address )
@@ -1880,9 +1880,9 @@ static int rmrr_identity_mapping(struct domain *d,
return -ENOMEM;
mrmrr->base = rmrr->base_address;
mrmrr->end = rmrr->end_address;
- spin_lock(&hd->mapping_lock);
- list_add_tail(&mrmrr->list, &hd->mapped_rmrrs);
- spin_unlock(&hd->mapping_lock);
+ spin_lock(&hd->arch.mapping_lock);
+ list_add_tail(&mrmrr->list, &hd->arch.mapped_rmrrs);
+ spin_unlock(&hd->arch.mapping_lock);
return 0;
}
@@ -2427,8 +2427,8 @@ static void vtd_dump_p2m_table(struct domain *d)
return;
hd = domain_hvm_iommu(d);
- printk("p2m table has %d levels\n", agaw_to_level(hd->agaw));
- vtd_dump_p2m_table_level(hd->pgd_maddr, agaw_to_level(hd->agaw), 0, 0);
+ printk("p2m table has %d levels\n", agaw_to_level(hd->arch.agaw));
+ vtd_dump_p2m_table_level(hd->arch.pgd_maddr, agaw_to_level(hd->arch.agaw), 0, 0);
}
const struct iommu_ops intel_iommu_ops = {
diff --git a/xen/include/asm-x86/hvm/iommu.h b/xen/include/asm-x86/hvm/iommu.h
index d488edf..a3f83d0 100644
--- a/xen/include/asm-x86/hvm/iommu.h
+++ b/xen/include/asm-x86/hvm/iommu.h
@@ -39,4 +39,33 @@ static inline int iommu_hardware_setup(void)
return 0;
}
+struct g2m_ioport {
+ struct list_head list;
+ unsigned int gport;
+ unsigned int mport;
+ unsigned int np;
+};
+
+struct mapped_rmrr {
+ struct list_head list;
+ u64 base;
+ u64 end;
+};
+
+struct arch_hvm_iommu
+{
+ u64 pgd_maddr; /* io page directory machine address */
+ int agaw; /* adjusted guest address width, 0 is level 2 30-bit */
+ u64 iommu_bitmap; /* bitmap of iommu(s) that the domain uses */
+ /* amd iommu support */
+ int domain_id;
+ int paging_mode;
+ struct page_info *root_table;
+ struct guest_iommu *g_iommu;
+
+ struct list_head g2m_ioport_list; /* guest to machine ioport mapping */
+ struct list_head mapped_rmrrs;
+ spinlock_t mapping_lock; /* io page table lock */
+};
+
#endif /* __ASM_X86_HVM_IOMMU_H__ */
diff --git a/xen/include/asm-x86/iommu.h b/xen/include/asm-x86/iommu.h
index 34c1896..021cd80 100644
--- a/xen/include/asm-x86/iommu.h
+++ b/xen/include/asm-x86/iommu.h
@@ -19,6 +19,10 @@
#include <asm/msi.h>
+/* Does this domain have a P2M table we can use as its IOMMU pagetable? */
+#define iommu_use_hap_pt(d) (hap_enabled(d) && iommu_hap_pt_share)
+#define domain_hvm_iommu(d) (&d->arch.hvm_domain.hvm_iommu)
+
void iommu_update_ire_from_apic(unsigned int apic, unsigned int reg, unsigned int value);
int iommu_update_ire_from_msi(struct msi_desc *msi_desc, struct msi_msg *msg);
void iommu_read_msi_from_ire(struct msi_desc *msi_desc, struct msi_msg *msg);
diff --git a/xen/include/xen/hvm/iommu.h b/xen/include/xen/hvm/iommu.h
index 2abb4e3..f8f8a93 100644
--- a/xen/include/xen/hvm/iommu.h
+++ b/xen/include/xen/hvm/iommu.h
@@ -23,32 +23,8 @@
#include <xen/iommu.h>
#include <asm/hvm/iommu.h>
-struct g2m_ioport {
- struct list_head list;
- unsigned int gport;
- unsigned int mport;
- unsigned int np;
-};
-
-struct mapped_rmrr {
- struct list_head list;
- u64 base;
- u64 end;
-};
-
struct hvm_iommu {
- u64 pgd_maddr; /* io page directory machine address */
- spinlock_t mapping_lock; /* io page table lock */
- int agaw; /* adjusted guest address width, 0 is level 2 30-bit */
- struct list_head g2m_ioport_list; /* guest to machine ioport mapping */
- u64 iommu_bitmap; /* bitmap of iommu(s) that the domain uses */
- struct list_head mapped_rmrrs;
-
- /* amd iommu support */
- int domain_id;
- int paging_mode;
- struct page_info *root_table;
- struct guest_iommu *g_iommu;
+ struct arch_hvm_iommu arch;
/* iommu_ops */
const struct iommu_ops *platform_ops;
diff --git a/xen/include/xen/iommu.h b/xen/include/xen/iommu.h
index 60df9d6..9a69b76 100644
--- a/xen/include/xen/iommu.h
+++ b/xen/include/xen/iommu.h
@@ -35,11 +35,6 @@ extern bool_t iommu_hap_pt_share;
extern bool_t iommu_debug;
extern bool_t amd_iommu_perdev_intremap;
-/* Does this domain have a P2M table we can use as its IOMMU pagetable? */
-#define iommu_use_hap_pt(d) (hap_enabled(d) && iommu_hap_pt_share)
-
-#define domain_hvm_iommu(d) (&d->arch.hvm_domain.hvm_iommu)
-
#define PAGE_SHIFT_4K (12)
#define PAGE_SIZE_4K (1UL << PAGE_SHIFT_4K)
#define PAGE_MASK_4K (((u64)-1) << PAGE_SHIFT_4K)
@@ -55,6 +50,9 @@ void iommu_dom0_init(struct domain *d);
void iommu_domain_destroy(struct domain *d);
int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn);
+void arch_iommu_domain_destroy(struct domain *d);
+int arch_iommu_domain_init(struct domain *d);
+
/* Function used internally, use iommu_domain_destroy */
void iommu_teardown(struct domain *d);
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* [RFC for-4.5 10/12] xen/passthrough: Introduce IOMMU ARM architure
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
` (8 preceding siblings ...)
2014-02-07 17:43 ` [RFC for-4.5 09/12] xen/passthrough: iommu: Introduce arch specific code Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-19 12:49 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 11/12] MAINTAINERS: Add drivers/passthrough/arm Julien Grall
` (2 subsequent siblings)
12 siblings, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel
Cc: ian.campbell, patches, Julien Grall, tim, stefano.stabellini,
Jan Beulich, Xiantao Zhang
This patch contains the architecture to use IOMMU on ARM. There is no
IOMMU drivers on this patch.
The code will run through the device tree and will initialize every IOMMUs.
It's possible to have multiple IOMMUs on the same platform, but they should
be handled with the same drivers. For now, there is no support for using
multiple iommu drivers at runtime.
Each new IOMMU drivers should contain:
static const char * const myiommu_dt_compat[] __initcontst =
{
/* list of device compatible with the drivers. Will be matched with
* the "compatible" property on the device tree
*/
NULL,
};
DT_DEVICE_START(myiommu, "MY IOMMU", DEVICE_IOMMU)
.compatible = myiommu_compatible,
.init = myiommu_init,
DT_DEVICE_END
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Cc: Xiantao Zhang <xiantao.zhang@intel.com>
Cc: Jan Beulich <jbeulich@suse.com>
---
xen/arch/arm/Rules.mk | 1 +
xen/arch/arm/domain.c | 7 ++++
xen/arch/arm/domain_build.c | 2 ++
xen/arch/arm/p2m.c | 4 +++
xen/arch/arm/setup.c | 2 ++
xen/drivers/passthrough/Makefile | 1 +
xen/drivers/passthrough/arm/Makefile | 1 +
xen/drivers/passthrough/arm/iommu.c | 65 ++++++++++++++++++++++++++++++++++
xen/include/asm-arm/device.h | 3 +-
xen/include/asm-arm/domain.h | 2 ++
xen/include/asm-arm/hvm/iommu.h | 10 ++++++
xen/include/asm-arm/iommu.h | 36 +++++++++++++++++++
12 files changed, 133 insertions(+), 1 deletion(-)
create mode 100644 xen/drivers/passthrough/arm/Makefile
create mode 100644 xen/drivers/passthrough/arm/iommu.c
create mode 100644 xen/include/asm-arm/hvm/iommu.h
create mode 100644 xen/include/asm-arm/iommu.h
diff --git a/xen/arch/arm/Rules.mk b/xen/arch/arm/Rules.mk
index 57f2eb1..1703551 100644
--- a/xen/arch/arm/Rules.mk
+++ b/xen/arch/arm/Rules.mk
@@ -9,6 +9,7 @@
HAS_DEVICE_TREE := y
HAS_VIDEO := y
HAS_ARM_HDLCD := y
+HAS_PASSTHROUGH := y
CFLAGS += -I$(BASEDIR)/include
diff --git a/xen/arch/arm/domain.c b/xen/arch/arm/domain.c
index c279a27..b3c0dda 100644
--- a/xen/arch/arm/domain.c
+++ b/xen/arch/arm/domain.c
@@ -557,6 +557,9 @@ int arch_domain_create(struct domain *d, unsigned int domcr_flags)
if ( (d->domain_id == 0) && (rc = domain_vuart_init(d)) )
goto fail;
+ if ( (rc = iommu_domain_init(d)) != 0 )
+ goto fail;
+
return 0;
fail:
@@ -568,6 +571,10 @@ fail:
void arch_domain_destroy(struct domain *d)
{
+ /* IOMMU page table is shared with P2M, always call
+ * iommu_domain_destroy() before p2m_teardown().
+ */
+ iommu_domain_destroy(d);
p2m_teardown(d);
domain_vgic_free(d);
domain_vuart_free(d);
diff --git a/xen/arch/arm/domain_build.c b/xen/arch/arm/domain_build.c
index 6d9a801..1f845d7 100644
--- a/xen/arch/arm/domain_build.c
+++ b/xen/arch/arm/domain_build.c
@@ -1091,6 +1091,8 @@ int construct_dom0(struct domain *d)
}
}
+ iommu_dom0_init(d);
+
return 0;
}
diff --git a/xen/arch/arm/p2m.c b/xen/arch/arm/p2m.c
index 85ca330..57304d0 100644
--- a/xen/arch/arm/p2m.c
+++ b/xen/arch/arm/p2m.c
@@ -401,12 +401,16 @@ static int create_p2m_entries(struct domain *d,
if ( flush )
{
+ unsigned long sgfn = paddr_to_pfn(start_gpaddr);
+ unsigned long egfn = paddr_to_pfn(end_gpaddr);
+
/* At the beginning of the function, Xen is updating VTTBR
* with the domain where the mappings are created. In this
* case it's only necessary to flush TLBs on every CPUs with
* the current VMID (our domain).
*/
flush_tlb();
+ iommu_iotlb_flush(d, sgfn, egfn - sgfn);
}
if ( op == ALLOCATE || op == INSERT )
diff --git a/xen/arch/arm/setup.c b/xen/arch/arm/setup.c
index 1f6d713..5a687d1 100644
--- a/xen/arch/arm/setup.c
+++ b/xen/arch/arm/setup.c
@@ -725,6 +725,8 @@ void __init start_xen(unsigned long boot_phys_offset,
local_irq_enable();
local_abort_enable();
+ iommu_setup(); /* setup iommu if available */
+
smp_prepare_cpus(cpus);
initialize_keytable();
diff --git a/xen/drivers/passthrough/Makefile b/xen/drivers/passthrough/Makefile
index 51e0a0d..aded1ea 100644
--- a/xen/drivers/passthrough/Makefile
+++ b/xen/drivers/passthrough/Makefile
@@ -1,6 +1,7 @@
subdir-$(x86) += vtd
subdir-$(x86) += amd
subdir-$(x86_64) += x86
+subdir-$(arm) += arm
obj-y += iommu.o
obj-$(x86) += iommu_x86.o
diff --git a/xen/drivers/passthrough/arm/Makefile b/xen/drivers/passthrough/arm/Makefile
new file mode 100644
index 0000000..0484b79
--- /dev/null
+++ b/xen/drivers/passthrough/arm/Makefile
@@ -0,0 +1 @@
+obj-y += iommu.o
diff --git a/xen/drivers/passthrough/arm/iommu.c b/xen/drivers/passthrough/arm/iommu.c
new file mode 100644
index 0000000..7cf36cd
--- /dev/null
+++ b/xen/drivers/passthrough/arm/iommu.c
@@ -0,0 +1,65 @@
+/*
+ * xen/drivers/passthrough/arm/iommu.c
+ *
+ * Generic IOMMU framework via the device tree
+ *
+ * Julien Grall <julien.grall@linaro.org>
+ * Copyright (c) 2013 Linaro Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that 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.
+ */
+
+#include <xen/lib.h>
+#include <xen/iommu.h>
+#include <xen/device_tree.h>
+#include <asm/device.h>
+
+static const struct iommu_ops *iommu_ops;
+
+const struct iommu_ops *iommu_get_ops(void)
+{
+ return iommu_ops;
+}
+
+void __init iommu_set_ops(const struct iommu_ops *ops)
+{
+ BUG_ON(ops == NULL);
+
+ if ( iommu_ops && iommu_ops != ops )
+ printk("WARNING: IOMMU ops already set to a different value\n");
+
+ iommu_ops = ops;
+}
+
+int __init iommu_hardware_setup(void)
+{
+ struct dt_device_node *np;
+ int rc;
+ unsigned int num_iommus = 0;
+
+ dt_for_each_device_node(dt_host, np)
+ {
+ rc = device_init(np, DEVICE_IOMMU, NULL);
+ if ( !rc )
+ num_iommus++;
+ }
+
+ return ( num_iommus > 0 ) ? 0 : -ENODEV;
+}
+
+int arch_iommu_domain_init(struct domain *d)
+{
+ return 0;
+}
+
+void arch_iommu_domain_destroy(struct domain *d)
+{
+}
diff --git a/xen/include/asm-arm/device.h b/xen/include/asm-arm/device.h
index 9e47ca6..ed04344 100644
--- a/xen/include/asm-arm/device.h
+++ b/xen/include/asm-arm/device.h
@@ -6,7 +6,8 @@
enum device_type
{
- DEVICE_SERIAL
+ DEVICE_SERIAL,
+ DEVICE_IOMMU,
};
struct device_desc {
diff --git a/xen/include/asm-arm/domain.h b/xen/include/asm-arm/domain.h
index af8c64b..15d814a 100644
--- a/xen/include/asm-arm/domain.h
+++ b/xen/include/asm-arm/domain.h
@@ -9,6 +9,7 @@
#include <asm/vfp.h>
#include <public/hvm/params.h>
#include <xen/serial.h>
+#include <xen/hvm/iommu.h>
/* Represents state corresponding to a block of 32 interrupts */
struct vgic_irq_rank {
@@ -72,6 +73,7 @@ struct pending_irq
struct hvm_domain
{
uint64_t params[HVM_NR_PARAMS];
+ struct hvm_iommu hvm_iommu;
} __cacheline_aligned;
#ifdef CONFIG_ARM_64
diff --git a/xen/include/asm-arm/hvm/iommu.h b/xen/include/asm-arm/hvm/iommu.h
new file mode 100644
index 0000000..461c8cf
--- /dev/null
+++ b/xen/include/asm-arm/hvm/iommu.h
@@ -0,0 +1,10 @@
+#ifndef __ASM_ARM_HVM_IOMMU_H_
+#define __ASM_ARM_HVM_IOMMU_H_
+
+struct arch_hvm_iommu
+{
+ /* Private information for the IOMMU drivers */
+ void *priv;
+};
+
+#endif /* __ASM_ARM_HVM_IOMMU_H_ */
diff --git a/xen/include/asm-arm/iommu.h b/xen/include/asm-arm/iommu.h
new file mode 100644
index 0000000..81eec83
--- /dev/null
+++ b/xen/include/asm-arm/iommu.h
@@ -0,0 +1,36 @@
+/*
+ * 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, write to the Free Software Foundation, Inc., 59 Temple
+ * Place - Suite 330, Boston, MA 02111-1307 USA.
+*/
+#ifndef __ARCH_ARM_IOMMU_H__
+#define __ARCH_ARM_IOMMU_H__
+
+/* Always share P2M Table between the CPU and the IOMMU */
+#define iommu_use_hap_pt(d) (1)
+#define domain_hvm_iommu(d) (&d->arch.hvm_domain.hvm_iommu)
+
+const struct iommu_ops *iommu_get_ops(void);
+void __init iommu_set_ops(const struct iommu_ops *ops);
+
+int __init iommu_hardware_setup(void);
+
+#endif /* __ARCH_ARM_IOMMU_H__ */
+
+/*
+ * Local variables:
+ * mode: C
+ * c-file-style: "BSD"
+ * c-basic-offset: 4
+ * indent-tabs-mode: nil
+ * End:
+ */
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* [RFC for-4.5 11/12] MAINTAINERS: Add drivers/passthrough/arm
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
` (9 preceding siblings ...)
2014-02-07 17:43 ` [RFC for-4.5 10/12] xen/passthrough: Introduce IOMMU ARM architure Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-19 12:49 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers Julien Grall
2014-02-07 17:51 ` [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
12 siblings, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel
Cc: Keir Fraser, ian.campbell, patches, Julien Grall, tim,
stefano.stabellini
Add the ARM IOMMU directory to "ARM ARCHITECTURE" part
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Cc: Keir Fraser <keir@xen.org>
---
MAINTAINERS | 1 +
1 file changed, 1 insertion(+)
diff --git a/MAINTAINERS b/MAINTAINERS
index 7757cdd..ad6c8a9 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -130,6 +130,7 @@ S: Supported
L: xen-devel@lists.xen.org
F: xen/arch/arm/
F: xen/include/asm-arm/
+F: xen/drivers/passthrough/arm
CPU POOLS
M: Juergen Gross <juergen.gross@ts.fujitsu.com>
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
` (10 preceding siblings ...)
2014-02-07 17:43 ` [RFC for-4.5 11/12] MAINTAINERS: Add drivers/passthrough/arm Julien Grall
@ 2014-02-07 17:43 ` Julien Grall
2014-02-19 14:15 ` Ian Campbell
2014-02-07 17:51 ` [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
12 siblings, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:43 UTC (permalink / raw)
To: xen-devel
Cc: ian.campbell, patches, Julien Grall, tim, stefano.stabellini,
Jan Beulich, Xiantao Zhang
This patch add support for ARM architected SMMU driver. It's based on the
linux drivers (drivers/iommu/arm-smmu) commit 89ac23cd.
Signed-off-by: Julien Grall<julien.grall@linaro.org>
Cc: Xiantao Zhang <xiantao.zhang@intel.com>
Cc: Jan Beulich <jbeulich@suse.com>
---
xen/drivers/passthrough/arm/Makefile | 1 +
xen/drivers/passthrough/arm/smmu.c | 1701 ++++++++++++++++++++++++++++++++++
2 files changed, 1702 insertions(+)
create mode 100644 xen/drivers/passthrough/arm/smmu.c
diff --git a/xen/drivers/passthrough/arm/Makefile b/xen/drivers/passthrough/arm/Makefile
index 0484b79..f4cd26e 100644
--- a/xen/drivers/passthrough/arm/Makefile
+++ b/xen/drivers/passthrough/arm/Makefile
@@ -1 +1,2 @@
obj-y += iommu.o
+obj-y += smmu.o
diff --git a/xen/drivers/passthrough/arm/smmu.c b/xen/drivers/passthrough/arm/smmu.c
new file mode 100644
index 0000000..9bf2aa3
--- /dev/null
+++ b/xen/drivers/passthrough/arm/smmu.c
@@ -0,0 +1,1701 @@
+/*
+ * IOMMU API for ARM architected SMMU implementations.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that 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, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Based on Linux drivers/iommu/arm-smmu.c (commit 89a23cd)
+ * Copyright (C) 2013 ARM Limited
+ *
+ * Author: Will Deacon <will.deacon@arm.com>
+ *
+ * Xen modification:
+ * Julien Grall <julien.grall@linaro.org>
+ * Copyright (C) 2014 Linaro Limited.
+ *
+ * This driver currently supports:
+ * - SMMUv1 and v2 implementations (didn't try v2 SMMU)
+ * - Stream-matching and stream-indexing
+ * - v7/v8 long-descriptor format
+ * - Non-secure access to the SMMU
+ * - 4k pages, p2m shared with the processor
+ * - Up to 40-bit addressing
+ * - Context fault reporting
+ */
+
+#include <xen/config.h>
+#include <xen/delay.h>
+#include <xen/errno.h>
+#include <xen/irq.h>
+#include <xen/lib.h>
+#include <xen/list.h>
+#include <xen/mm.h>
+#include <xen/rbtree.h>
+#include <xen/sched.h>
+#include <asm/atomic.h>
+#include <asm/device.h>
+#include <asm/io.h>
+#include <asm/platform.h>
+
+#define SZ_4K (1 << 12)
+#define SZ_64K (1 << 16)
+
+/* Driver options */
+#define SMMU_OPT_SECURE_CONFIG_ACCESS (1 << 0)
+
+/* Maximum number of stream IDs assigned to a single device */
+#define MAX_MASTER_STREAMIDS MAX_PHANDLE_ARGS
+
+/* Maximum stream ID */
+#define SMMU_MAX_STREAMIDS (SZ_64K - 1)
+
+/* Maximum number of context banks per SMMU */
+#define SMMU_MAX_CBS 128
+
+/* Maximum number of mapping groups per SMMU */
+#define SMMU_MAX_SMRS 128
+
+/* SMMU global address space */
+#define SMMU_GR0(smmu) ((smmu)->base)
+#define SMMU_GR1(smmu) ((smmu)->base + (smmu)->pagesize)
+
+/*
+ * SMMU global address space with conditional offset to access secure aliases of
+ * non-secure registers (e.g. nsCR0: 0x400, nsGFSR: 0x448, nsGFSYNR0: 0x450)
+ */
+#define SMMU_GR0_NS(smmu) \
+ ((smmu)->base + \
+ ((smmu->options & SMMU_OPT_SECURE_CONFIG_ACCESS) \
+ ? 0x400 : 0))
+
+/* Page table bits */
+#define SMMU_PTE_PAGE (((pteval_t)3) << 0)
+#define SMMU_PTE_CONT (((pteval_t)1) << 52)
+#define SMMU_PTE_AF (((pteval_t)1) << 10)
+#define SMMU_PTE_SH_NS (((pteval_t)0) << 8)
+#define SMMU_PTE_SH_OS (((pteval_t)2) << 8)
+#define SMMU_PTE_SH_IS (((pteval_t)3) << 8)
+
+#if PAGE_SIZE == SZ_4K
+#define SMMU_PTE_CONT_ENTRIES 16
+#elif PAGE_SIZE == SZ_64K
+#define SMMU_PTE_CONT_ENTRIES 32
+#else
+#define SMMU_PTE_CONT_ENTRIES 1
+#endif
+
+#define SMMU_PTE_CONT_SIZE (PAGE_SIZE * SMMU_PTE_CONT_ENTRIES)
+#define SMMU_PTE_CONT_MASK (~(SMMU_PTE_CONT_SIZE - 1))
+#define SMMU_PTE_HWTABLE_SIZE (PTRS_PER_PTE * sizeof(pte_t))
+
+/* Stage-1 PTE */
+#define SMMU_PTE_AP_UNPRIV (((pteval_t)1) << 6)
+#define SMMU_PTE_AP_RDONLY (((pteval_t)2) << 6)
+#define SMMU_PTE_ATTRINDX_SHIFT 2
+#define SMMU_PTE_nG (((pteval_t)1) << 11)
+
+/* Stage-2 PTE */
+#define SMMU_PTE_HAP_FAULT (((pteval_t)0) << 6)
+#define SMMU_PTE_HAP_READ (((pteval_t)1) << 6)
+#define SMMU_PTE_HAP_WRITE (((pteval_t)2) << 6)
+#define SMMU_PTE_MEMATTR_OIWB (((pteval_t)0xf) << 2)
+#define SMMU_PTE_MEMATTR_NC (((pteval_t)0x5) << 2)
+#define SMMU_PTE_MEMATTR_DEV (((pteval_t)0x1) << 2)
+
+/* Configuration registers */
+#define SMMU_GR0_sCR0 0x0
+#define SMMU_sCR0_CLIENTPD (1 << 0)
+#define SMMU_sCR0_GFRE (1 << 1)
+#define SMMU_sCR0_GFIE (1 << 2)
+#define SMMU_sCR0_GCFGFRE (1 << 4)
+#define SMMU_sCR0_GCFGFIE (1 << 5)
+#define SMMU_sCR0_USFCFG (1 << 10)
+#define SMMU_sCR0_VMIDPNE (1 << 11)
+#define SMMU_sCR0_PTM (1 << 12)
+#define SMMU_sCR0_FB (1 << 13)
+#define SMMU_sCR0_BSU_SHIFT 14
+#define SMMU_sCR0_BSU_MASK 0x3
+
+/* Identification registers */
+#define SMMU_GR0_ID0 0x20
+#define SMMU_GR0_ID1 0x24
+#define SMMU_GR0_ID2 0x28
+#define SMMU_GR0_ID3 0x2c
+#define SMMU_GR0_ID4 0x30
+#define SMMU_GR0_ID5 0x34
+#define SMMU_GR0_ID6 0x38
+#define SMMU_GR0_ID7 0x3c
+#define SMMU_GR0_sGFSR 0x48
+#define SMMU_GR0_sGFSYNR0 0x50
+#define SMMU_GR0_sGFSYNR1 0x54
+#define SMMU_GR0_sGFSYNR2 0x58
+#define SMMU_GR0_PIDR0 0xfe0
+#define SMMU_GR0_PIDR1 0xfe4
+#define SMMU_GR0_PIDR2 0xfe8
+
+#define SMMU_ID0_S1TS (1 << 30)
+#define SMMU_ID0_S2TS (1 << 29)
+#define SMMU_ID0_NTS (1 << 28)
+#define SMMU_ID0_SMS (1 << 27)
+#define SMMU_ID0_PTFS_SHIFT 24
+#define SMMU_ID0_PTFS_MASK 0x2
+#define SMMU_ID0_PTFS_V8_ONLY 0x2
+#define SMMU_ID0_CTTW (1 << 14)
+#define SMMU_ID0_NUMIRPT_SHIFT 16
+#define SMMU_ID0_NUMIRPT_MASK 0xff
+#define SMMU_ID0_NUMSMRG_SHIFT 0
+#define SMMU_ID0_NUMSMRG_MASK 0xff
+
+#define SMMU_ID1_PAGESIZE (1 << 31)
+#define SMMU_ID1_NUMPAGENDXB_SHIFT 28
+#define SMMU_ID1_NUMPAGENDXB_MASK 7
+#define SMMU_ID1_NUMS2CB_SHIFT 16
+#define SMMU_ID1_NUMS2CB_MASK 0xff
+#define SMMU_ID1_NUMCB_SHIFT 0
+#define SMMU_ID1_NUMCB_MASK 0xff
+
+#define SMMU_ID2_OAS_SHIFT 4
+#define SMMU_ID2_OAS_MASK 0xf
+#define SMMU_ID2_IAS_SHIFT 0
+#define SMMU_ID2_IAS_MASK 0xf
+#define SMMU_ID2_UBS_SHIFT 8
+#define SMMU_ID2_UBS_MASK 0xf
+#define SMMU_ID2_PTFS_4K (1 << 12)
+#define SMMU_ID2_PTFS_16K (1 << 13)
+#define SMMU_ID2_PTFS_64K (1 << 14)
+
+#define SMMU_PIDR2_ARCH_SHIFT 4
+#define SMMU_PIDR2_ARCH_MASK 0xf
+
+/* Global TLB invalidation */
+#define SMMU_GR0_STLBIALL 0x60
+#define SMMU_GR0_TLBIVMID 0x64
+#define SMMU_GR0_TLBIALLNSNH 0x68
+#define SMMU_GR0_TLBIALLH 0x6c
+#define SMMU_GR0_sTLBGSYNC 0x70
+#define SMMU_GR0_sTLBGSTATUS 0x74
+#define SMMU_sTLBGSTATUS_GSACTIVE (1 << 0)
+#define SMMU_TLB_LOOP_TIMEOUT 1000000 /* 1s! */
+
+/* Stream mapping registers */
+#define SMMU_GR0_SMR(n) (0x800 + ((n) << 2))
+#define SMMU_SMR_VALID (1 << 31)
+#define SMMU_SMR_MASK_SHIFT 16
+#define SMMU_SMR_MASK_MASK 0x7fff
+#define SMMU_SMR_ID_SHIFT 0
+#define SMMU_SMR_ID_MASK 0x7fff
+
+#define SMMU_GR0_S2CR(n) (0xc00 + ((n) << 2))
+#define SMMU_S2CR_CBNDX_SHIFT 0
+#define SMMU_S2CR_CBNDX_MASK 0xff
+#define SMMU_S2CR_TYPE_SHIFT 16
+#define SMMU_S2CR_TYPE_MASK 0x3
+#define SMMU_S2CR_TYPE_TRANS (0 << SMMU_S2CR_TYPE_SHIFT)
+#define SMMU_S2CR_TYPE_BYPASS (1 << SMMU_S2CR_TYPE_SHIFT)
+#define SMMU_S2CR_TYPE_FAULT (2 << SMMU_S2CR_TYPE_SHIFT)
+
+/* Context bank attribute registers */
+#define SMMU_GR1_CBAR(n) (0x0 + ((n) << 2))
+#define SMMU_CBAR_VMID_SHIFT 0
+#define SMMU_CBAR_VMID_MASK 0xff
+#define SMMU_CBAR_S1_MEMATTR_SHIFT 12
+#define SMMU_CBAR_S1_MEMATTR_MASK 0xf
+#define SMMU_CBAR_S1_MEMATTR_WB 0xf
+#define SMMU_CBAR_TYPE_SHIFT 16
+#define SMMU_CBAR_TYPE_MASK 0x3
+#define SMMU_CBAR_TYPE_S2_TRANS (0 << SMMU_CBAR_TYPE_SHIFT)
+#define SMMU_CBAR_TYPE_S1_TRANS_S2_BYPASS (1 << SMMU_CBAR_TYPE_SHIFT)
+#define SMMU_CBAR_TYPE_S1_TRANS_S2_FAULT (2 << SMMU_CBAR_TYPE_SHIFT)
+#define SMMU_CBAR_TYPE_S1_TRANS_S2_TRANS (3 << SMMU_CBAR_TYPE_SHIFT)
+#define SMMU_CBAR_IRPTNDX_SHIFT 24
+#define SMMU_CBAR_IRPTNDX_MASK 0xff
+
+#define SMMU_GR1_CBA2R(n) (0x800 + ((n) << 2))
+#define SMMU_CBA2R_RW64_32BIT (0 << 0)
+#define SMMU_CBA2R_RW64_64BIT (1 << 0)
+
+/* Translation context bank */
+#define SMMU_CB_BASE(smmu) ((smmu)->base + ((smmu)->size >> 1))
+#define SMMU_CB(smmu, n) ((n) * (smmu)->pagesize)
+
+#define SMMU_CB_SCTLR 0x0
+#define SMMU_CB_RESUME 0x8
+#define SMMU_CB_TCR2 0x10
+#define SMMU_CB_TTBR0_LO 0x20
+#define SMMU_CB_TTBR0_HI 0x24
+#define SMMU_CB_TCR 0x30
+#define SMMU_CB_S1_MAIR0 0x38
+#define SMMU_CB_FSR 0x58
+#define SMMU_CB_FAR_LO 0x60
+#define SMMU_CB_FAR_HI 0x64
+#define SMMU_CB_FSYNR0 0x68
+#define SMMU_CB_S1_TLBIASID 0x610
+
+#define SMMU_SCTLR_S1_ASIDPNE (1 << 12)
+#define SMMU_SCTLR_CFCFG (1 << 7)
+#define SMMU_SCTLR_CFIE (1 << 6)
+#define SMMU_SCTLR_CFRE (1 << 5)
+#define SMMU_SCTLR_E (1 << 4)
+#define SMMU_SCTLR_AFE (1 << 2)
+#define SMMU_SCTLR_TRE (1 << 1)
+#define SMMU_SCTLR_M (1 << 0)
+#define SMMU_SCTLR_EAE_SBOP (SMMU_SCTLR_AFE | SMMU_SCTLR_TRE)
+
+#define SMMU_RESUME_RETRY (0 << 0)
+#define SMMU_RESUME_TERMINATE (1 << 0)
+
+#define SMMU_TCR_EAE (1 << 31)
+
+#define SMMU_TCR_PASIZE_SHIFT 16
+#define SMMU_TCR_PASIZE_MASK 0x7
+
+#define SMMU_TCR_TG0_4K (0 << 14)
+#define SMMU_TCR_TG0_64K (1 << 14)
+
+#define SMMU_TCR_SH0_SHIFT 12
+#define SMMU_TCR_SH0_MASK 0x3
+#define SMMU_TCR_SH_NS 0
+#define SMMU_TCR_SH_OS 2
+#define SMMU_TCR_SH_IS 3
+
+#define SMMU_TCR_ORGN0_SHIFT 10
+#define SMMU_TCR_IRGN0_SHIFT 8
+#define SMMU_TCR_RGN_MASK 0x3
+#define SMMU_TCR_RGN_NC 0
+#define SMMU_TCR_RGN_WBWA 1
+#define SMMU_TCR_RGN_WT 2
+#define SMMU_TCR_RGN_WB 3
+
+#define SMMU_TCR_SL0_SHIFT 6
+#define SMMU_TCR_SL0_MASK 0x3
+#define SMMU_TCR_SL0_LVL_2 0
+#define SMMU_TCR_SL0_LVL_1 1
+
+#define SMMU_TCR_T1SZ_SHIFT 16
+#define SMMU_TCR_T0SZ_SHIFT 0
+#define SMMU_TCR_SZ_MASK 0xf
+
+#define SMMU_TCR2_SEP_SHIFT 15
+#define SMMU_TCR2_SEP_MASK 0x7
+
+#define SMMU_TCR2_PASIZE_SHIFT 0
+#define SMMU_TCR2_PASIZE_MASK 0x7
+
+/* Common definitions for PASize and SEP fields */
+#define SMMU_TCR2_ADDR_32 0
+#define SMMU_TCR2_ADDR_36 1
+#define SMMU_TCR2_ADDR_40 2
+#define SMMU_TCR2_ADDR_42 3
+#define SMMU_TCR2_ADDR_44 4
+#define SMMU_TCR2_ADDR_48 5
+
+#define SMMU_TTBRn_HI_ASID_SHIFT 16
+
+#define SMMU_MAIR_ATTR_SHIFT(n) ((n) << 3)
+#define SMMU_MAIR_ATTR_MASK 0xff
+#define SMMU_MAIR_ATTR_DEVICE 0x04
+#define SMMU_MAIR_ATTR_NC 0x44
+#define SMMU_MAIR_ATTR_WBRWA 0xff
+#define SMMU_MAIR_ATTR_IDX_NC 0
+#define SMMU_MAIR_ATTR_IDX_CACHE 1
+#define SMMU_MAIR_ATTR_IDX_DEV 2
+
+#define SMMU_FSR_MULTI (1 << 31)
+#define SMMU_FSR_SS (1 << 30)
+#define SMMU_FSR_UUT (1 << 8)
+#define SMMU_FSR_ASF (1 << 7)
+#define SMMU_FSR_TLBLKF (1 << 6)
+#define SMMU_FSR_TLBMCF (1 << 5)
+#define SMMU_FSR_EF (1 << 4)
+#define SMMU_FSR_PF (1 << 3)
+#define SMMU_FSR_AFF (1 << 2)
+#define SMMU_FSR_TF (1 << 1)
+
+#define SMMU_FSR_IGN (SMMU_FSR_AFF | SMMU_FSR_ASF | \
+ SMMU_FSR_TLBMCF | SMMU_FSR_TLBLKF)
+#define SMMU_FSR_FAULT (SMMU_FSR_MULTI | SMMU_FSR_SS | \
+ SMMU_FSR_UUT | SMMU_FSR_EF | \
+ SMMU_FSR_PF | SMMU_FSR_TF | \
+ SMMU_FSR_IGN)
+
+#define SMMU_FSYNR0_WNR (1 << 4)
+
+#define smmu_print(dev, lvl, fmt, ...) \
+ printk(lvl "smmu: %s: " fmt, dt_node_full_name(dev->node), ## __VA_ARGS__)
+
+#define smmu_err(dev, fmt, ...) smmu_print(dev, XENLOG_ERR, fmt, ## __VA_ARGS__)
+
+#define smmu_dbg(dev, fmt, ...) \
+ smmu_print(dev, XENLOG_DEBUG, fmt, ## __VA_ARGS__)
+
+#define smmu_info(dev, fmt, ...) \
+ smmu_print(dev, XENLOG_INFO, fmt, ## __VA_ARGS__)
+
+#define smmu_warn(dev, fmt, ...) \
+ smmu_print(dev, XENLOG_WARNING, fmt, ## __VA_ARGS__)
+
+struct arm_smmu_device {
+ const struct dt_device_node *node;
+
+ void __iomem *base;
+ unsigned long size;
+ unsigned long pagesize;
+
+#define SMMU_FEAT_COHERENT_WALK (1 << 0)
+#define SMMU_FEAT_STREAM_MATCH (1 << 1)
+#define SMMU_FEAT_TRANS_S1 (1 << 2)
+#define SMMU_FEAT_TRANS_S2 (1 << 3)
+#define SMMU_FEAT_TRANS_NESTED (1 << 4)
+ u32 features;
+ u32 options;
+ int version;
+
+ u32 num_context_banks;
+ u32 num_s2_context_banks;
+ DECLARE_BITMAP(context_map, SMMU_MAX_CBS);
+ atomic_t irptndx;
+
+ u32 num_mapping_groups;
+ DECLARE_BITMAP(smr_map, SMMU_MAX_SMRS);
+
+ unsigned long input_size;
+ unsigned long s1_output_size;
+ unsigned long s2_output_size;
+
+ u32 num_global_irqs;
+ u32 num_context_irqs;
+ struct dt_irq *irqs;
+
+ u32 smr_mask_mask;
+ u32 smr_id_mask;
+
+ unsigned long *sids;
+
+ struct list_head list;
+ struct rb_root masters;
+};
+
+struct arm_smmu_smr {
+ u8 idx;
+ u16 mask;
+ u16 id;
+};
+
+#define INVALID_IRPTNDX 0xff
+
+#define SMMU_CB_ASID(cfg) ((cfg)->cbndx)
+#define SMMU_CB_VMID(cfg) ((cfg)->cbndx + 1)
+
+struct arm_smmu_domain_cfg {
+ struct arm_smmu_device *smmu;
+ u8 cbndx;
+ u8 irptndx;
+ u32 cbar;
+ /* Domain associated to this device */
+ struct domain *domain;
+ /* List of master which use this structure */
+ struct list_head masters;
+
+ /* Used to link domain context for a same domain */
+ struct list_head list;
+};
+
+struct arm_smmu_master {
+ const struct dt_device_node *dt_node;
+
+ /*
+ * The following is specific to the master's position in the
+ * SMMU chain.
+ */
+ struct rb_node node;
+ u32 num_streamids;
+ u16 streamids[MAX_MASTER_STREAMIDS];
+ int num_s2crs;
+
+ struct arm_smmu_smr *smrs;
+ struct arm_smmu_domain_cfg *cfg;
+
+ /* Used to link masters in a same domain context */
+ struct list_head list;
+};
+
+static LIST_HEAD(arm_smmu_devices);
+
+struct arm_smmu_domain {
+ spinlock_t lock;
+ struct list_head contexts;
+};
+
+struct arm_smmu_option_prop {
+ u32 opt;
+ const char *prop;
+};
+
+static const struct arm_smmu_option_prop arm_smmu_options [] __initconst =
+{
+ { SMMU_OPT_SECURE_CONFIG_ACCESS, "calxeda,smmu-secure-config-access" },
+ { 0, NULL},
+};
+
+static void __init check_driver_options(struct arm_smmu_device *smmu)
+{
+ int i = 0;
+
+ do {
+ if ( dt_property_read_bool(smmu->node, arm_smmu_options[i].prop) )
+ {
+ smmu->options |= arm_smmu_options[i].opt;
+ smmu_dbg(smmu, "option %s\n", arm_smmu_options[i].prop);
+ }
+ } while ( arm_smmu_options[++i].opt );
+}
+
+static void arm_smmu_context_fault(int irq, void *data,
+ struct cpu_user_regs *regs)
+{
+ u32 fsr, far, fsynr;
+ unsigned long iova;
+ struct arm_smmu_domain_cfg *cfg = data;
+ struct arm_smmu_device *smmu = cfg->smmu;
+ void __iomem *cb_base;
+
+ cb_base = SMMU_CB_BASE(smmu) + SMMU_CB(smmu, cfg->cbndx);
+ fsr = readl_relaxed(cb_base + SMMU_CB_FSR);
+
+ if ( !(fsr & SMMU_FSR_FAULT) )
+ return;
+
+ if ( fsr & SMMU_FSR_IGN )
+ smmu_err(smmu, "Unexpected context fault (fsr 0x%u)\n", fsr);
+
+ fsynr = readl_relaxed(cb_base + SMMU_CB_FSYNR0);
+ far = readl_relaxed(cb_base + SMMU_CB_FAR_LO);
+ iova = far;
+#ifdef CONFIG_ARM_64
+ far = readl_relaxed(cb_base + SMMU_CB_FAR_HI);
+ iova |= ((unsigned long)far << 32);
+#endif
+
+ smmu_err(smmu,
+ "Unhandled context fault: iova=0x%08lx, fsynr=0x%x, cb=%d\n",
+ iova, fsynr, cfg->cbndx);
+
+ /* Clear the faulting FSR */
+ writel(fsr, cb_base + SMMU_CB_FSR);
+
+ /* Terminate any stalled transactions */
+ if ( fsr & SMMU_FSR_SS )
+ writel_relaxed(SMMU_RESUME_TERMINATE, cb_base + SMMU_CB_RESUME);
+}
+
+static void arm_smmu_global_fault(int irq, void *data,
+ struct cpu_user_regs *regs)
+{
+ u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
+ struct arm_smmu_device *smmu = data;
+ void __iomem *gr0_base = SMMU_GR0_NS(smmu);
+
+ gfsr = readl_relaxed(gr0_base + SMMU_GR0_sGFSR);
+ gfsynr0 = readl_relaxed(gr0_base + SMMU_GR0_sGFSYNR0);
+ gfsynr1 = readl_relaxed(gr0_base + SMMU_GR0_sGFSYNR1);
+ gfsynr2 = readl_relaxed(gr0_base + SMMU_GR0_sGFSYNR2);
+
+ if ( !gfsr )
+ return;
+
+ smmu_err(smmu, "Unexpected global fault, this could be serious\n");
+ smmu_err(smmu,
+ "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
+ gfsr, gfsynr0, gfsynr1, gfsynr2);
+ writel(gfsr, gr0_base + SMMU_GR0_sGFSR);
+}
+
+static struct arm_smmu_master *
+find_smmu_master(struct arm_smmu_device *smmu,
+ const struct dt_device_node *dev_node)
+{
+ struct rb_node *node = smmu->masters.rb_node;
+
+ while ( node )
+ {
+ struct arm_smmu_master *master;
+
+ master = container_of(node, struct arm_smmu_master, node);
+
+ if ( dev_node < master->dt_node )
+ node = node->rb_left;
+ else if ( dev_node > master->dt_node )
+ node = node->rb_right;
+ else
+ return master;
+ }
+
+ return NULL;
+}
+
+static __init int insert_smmu_master(struct arm_smmu_device *smmu,
+ struct arm_smmu_master *master)
+{
+ struct rb_node **new, *parent;
+
+ new = &smmu->masters.rb_node;
+ parent = NULL;
+ while ( *new )
+ {
+ struct arm_smmu_master *this;
+
+ this = container_of(*new, struct arm_smmu_master, node);
+
+ parent = *new;
+ if ( master->dt_node < this->dt_node )
+ new = &((*new)->rb_left);
+ else if (master->dt_node > this->dt_node)
+ new = &((*new)->rb_right);
+ else
+ return -EEXIST;
+ }
+
+ rb_link_node(&master->node, parent, new);
+ rb_insert_color(&master->node, &smmu->masters);
+ return 0;
+}
+
+static __init int register_smmu_master(struct arm_smmu_device *smmu,
+ struct dt_phandle_args *masterspec)
+{
+ int i, sid;
+ struct arm_smmu_master *master;
+ int rc = 0;
+
+ smmu_dbg(smmu, "Try to add master %s\n", masterspec->np->name);
+
+ master = find_smmu_master(smmu, masterspec->np);
+ if ( master )
+ {
+ smmu_err(smmu,
+ "rejecting multiple registrations for master device %s\n",
+ masterspec->np->name);
+ return -EBUSY;
+ }
+
+ if ( masterspec->args_count > MAX_MASTER_STREAMIDS )
+ {
+ smmu_err(smmu,
+ "reached maximum number (%d) of stream IDs for master device %s\n",
+ MAX_MASTER_STREAMIDS, masterspec->np->name);
+ return -ENOSPC;
+ }
+
+ master = xzalloc(struct arm_smmu_master);
+ if ( !master )
+ return -ENOMEM;
+
+ INIT_LIST_HEAD(&master->list);
+ master->dt_node = masterspec->np;
+ master->num_streamids = masterspec->args_count;
+
+ for ( i = 0; i < master->num_streamids; ++i )
+ {
+ sid = masterspec->args[i];
+ if ( test_and_set_bit(sid, smmu->sids) )
+ {
+ smmu_err(smmu, "duplicate stream ID (%d)\n", sid);
+ xfree(master);
+ return -EEXIST;
+ }
+ master->streamids[i] = masterspec->args[i];
+ }
+
+ rc = insert_smmu_master(smmu, master);
+ /* Insertion should never fail */
+ ASSERT(rc == 0);
+
+ return 0;
+}
+
+static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end)
+{
+ int idx;
+
+ do
+ {
+ idx = find_next_zero_bit(map, end, start);
+ if ( idx == end )
+ return -ENOSPC;
+ } while ( test_and_set_bit(idx, map) );
+
+ return idx;
+}
+
+static void __arm_smmu_free_bitmap(unsigned long *map, int idx)
+{
+ clear_bit(idx, map);
+}
+
+static void arm_smmu_tlb_sync(struct arm_smmu_device *smmu)
+{
+ int count = 0;
+ void __iomem *gr0_base = SMMU_GR0(smmu);
+
+ writel_relaxed(0, gr0_base + SMMU_GR0_sTLBGSYNC);
+ while ( readl_relaxed(gr0_base + SMMU_GR0_sTLBGSTATUS) &
+ SMMU_sTLBGSTATUS_GSACTIVE )
+ {
+ cpu_relax();
+ if ( ++count == SMMU_TLB_LOOP_TIMEOUT )
+ {
+ smmu_err(smmu, "TLB sync timed out -- SMMU may be deadlocked\n");
+ return;
+ }
+ udelay(1);
+ }
+}
+
+static void arm_smmu_tlb_inv_context(struct arm_smmu_domain_cfg *cfg)
+{
+ struct arm_smmu_device *smmu = cfg->smmu;
+ void __iomem *base = SMMU_GR0(smmu);
+
+ writel_relaxed(SMMU_CB_VMID(cfg),
+ base + SMMU_GR0_TLBIVMID);
+
+ arm_smmu_tlb_sync(smmu);
+}
+
+static void arm_smmu_iotlb_flush_all(struct domain *d)
+{
+ struct arm_smmu_domain *smmu_domain = domain_hvm_iommu(d)->arch.priv;
+ struct arm_smmu_domain_cfg *cfg;
+
+ spin_lock(&smmu_domain->lock);
+ list_for_each_entry(cfg, &smmu_domain->contexts, list)
+ arm_smmu_tlb_inv_context(cfg);
+ spin_unlock(&smmu_domain->lock);
+}
+
+static void arm_smmu_iotlb_flush(struct domain *d, unsigned long gfn,
+ unsigned int page_count)
+{
+ /* ARM SMMU v1 doesn't have flush by VMA and VMID */
+ arm_smmu_iotlb_flush_all(d);
+}
+
+static int determine_smr_mask(struct arm_smmu_device *smmu,
+ struct arm_smmu_master *master,
+ struct arm_smmu_smr *smr, int start, int order)
+{
+ u16 i, zero_bits_mask, one_bits_mask, const_mask;
+ int nr;
+
+ nr = 1 << order;
+
+ if ( nr == 1 )
+ {
+ /* no mask, use streamid to match and be done with it */
+ smr->mask = 0;
+ smr->id = master->streamids[start];
+ return 0;
+ }
+
+ zero_bits_mask = 0;
+ one_bits_mask = 0xffff;
+ for ( i = start; i < start + nr; i++)
+ {
+ zero_bits_mask |= master->streamids[i]; /* const 0 bits */
+ one_bits_mask &= master->streamids[i]; /* const 1 bits */
+ }
+ zero_bits_mask = ~zero_bits_mask;
+
+ /* bits having constant values (either 0 or 1) */
+ const_mask = zero_bits_mask | one_bits_mask;
+
+ i = hweight16(~const_mask);
+ if ( (1 << i) == nr )
+ {
+ smr->mask = ~const_mask;
+ smr->id = one_bits_mask;
+ }
+ else
+ /* no usable mask for this set of streamids */
+ return 1;
+
+ if ( ((smr->mask & smmu->smr_mask_mask) != smr->mask) ||
+ ((smr->id & smmu->smr_id_mask) != smr->id) )
+ /* insufficient number of mask/id bits */
+ return 1;
+
+ return 0;
+}
+
+static int determine_smr_mapping(struct arm_smmu_device *smmu,
+ struct arm_smmu_master *master,
+ struct arm_smmu_smr *smrs, int max_smrs)
+{
+ int nr_sid, nr, i, bit, start;
+
+ /*
+ * This function is called only once -- when a master is added
+ * to a domain. If master->num_s2crs != 0 then this master
+ * was already added to a domain.
+ */
+ BUG_ON(master->num_s2crs);
+
+ start = nr = 0;
+ nr_sid = master->num_streamids;
+ do
+ {
+ /*
+ * largest power-of-2 number of streamids for which to
+ * determine a usable mask/id pair for stream matching
+ */
+ bit = fls(nr_sid);
+ if (!bit)
+ return 0;
+
+ /*
+ * iterate over power-of-2 numbers to determine
+ * largest possible mask/id pair for stream matching
+ * of next 2**i streamids
+ */
+ for ( i = bit - 1; i >= 0; i-- )
+ {
+ if( !determine_smr_mask(smmu, master,
+ &smrs[master->num_s2crs],
+ start, i))
+ break;
+ }
+
+ if ( i < 0 )
+ goto out;
+
+ nr = 1 << i;
+ nr_sid -= nr;
+ start += nr;
+ master->num_s2crs++;
+ } while ( master->num_s2crs <= max_smrs );
+
+out:
+ if ( nr_sid )
+ {
+ /* not enough mapping groups available */
+ master->num_s2crs = 0;
+ return -ENOSPC;
+ }
+
+ return 0;
+}
+
+static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
+ struct arm_smmu_master *master)
+{
+ int i, max_smrs, ret;
+ struct arm_smmu_smr *smrs;
+ void __iomem *gr0_base = SMMU_GR0(smmu);
+
+ if ( !(smmu->features & SMMU_FEAT_STREAM_MATCH) )
+ return 0;
+
+ if ( master->smrs )
+ return -EEXIST;
+
+ max_smrs = min(smmu->num_mapping_groups, master->num_streamids);
+ smrs = xmalloc_array(struct arm_smmu_smr, max_smrs);
+ if ( !smrs )
+ {
+ smmu_err(smmu, "failed to allocated %d SMRs for master %s\n",
+ max_smrs, dt_node_name(master->dt_node));
+ return -ENOMEM;
+ }
+
+ ret = determine_smr_mapping(smmu, master, smrs, max_smrs);
+ if ( ret )
+ goto err_free_smrs;
+
+ /* Allocate the SMRs on the root SMMU */
+ for ( i = 0; i < master->num_s2crs; ++i )
+ {
+ int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
+ smmu->num_mapping_groups);
+ if ( idx < 0 )
+ {
+ smmu_err(smmu, "failed to allocate free SMR\n");
+ goto err_free_bitmap;
+ }
+ smrs[i].idx = idx;
+ }
+
+ /* It worked! Now, poke the actual hardware */
+ for ( i = 0; i < master->num_s2crs; ++i )
+ {
+ u32 reg = SMMU_SMR_VALID | smrs[i].id << SMMU_SMR_ID_SHIFT |
+ smrs[i].mask << SMMU_SMR_MASK_SHIFT;
+ smmu_dbg(smmu, "SMR%d: 0x%x\n", smrs[i].idx, reg);
+ writel_relaxed(reg, gr0_base + SMMU_GR0_SMR(smrs[i].idx));
+ }
+
+ master->smrs = smrs;
+ return 0;
+
+err_free_bitmap:
+ while (--i >= 0)
+ __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
+ master->num_s2crs = 0;
+err_free_smrs:
+ xfree(smrs);
+ return -ENOSPC;
+}
+
+/* Forward declaration */
+static void arm_smmu_destroy_domain_context(struct arm_smmu_domain_cfg *cfg);
+
+static int arm_smmu_domain_add_master(struct domain *d,
+ struct arm_smmu_domain_cfg *cfg,
+ struct arm_smmu_master *master)
+{
+ int i, ret;
+ struct arm_smmu_device *smmu = cfg->smmu;
+ void __iomem *gr0_base = SMMU_GR0(smmu);
+ struct arm_smmu_smr *smrs = master->smrs;
+
+ if ( master->cfg )
+ return -EBUSY;
+
+ ret = arm_smmu_master_configure_smrs(smmu, master);
+ if ( ret )
+ return ret;
+
+ /* Now we're at the root, time to point at our context bank */
+ if ( !master->num_s2crs )
+ master->num_s2crs = master->num_streamids;
+
+ for ( i = 0; i < master->num_s2crs; ++i )
+ {
+ u32 idx, s2cr;
+
+ idx = smrs ? smrs[i].idx : master->streamids[i];
+ s2cr = (SMMU_S2CR_TYPE_TRANS << SMMU_S2CR_TYPE_SHIFT) |
+ (cfg->cbndx << SMMU_S2CR_CBNDX_SHIFT);
+ smmu_dbg(smmu, "S2CR%d: 0x%x\n", idx, s2cr);
+ writel_relaxed(s2cr, gr0_base + SMMU_GR0_S2CR(idx));
+ }
+
+ master->cfg = cfg;
+ list_add(&master->list, &cfg->masters);
+
+ return 0;
+}
+
+static void arm_smmu_domain_remove_master(struct arm_smmu_master *master)
+{
+ int i;
+ struct arm_smmu_domain_cfg *cfg = master->cfg;
+ struct arm_smmu_device *smmu = cfg->smmu;
+ void __iomem *gr0_base = SMMU_GR0(smmu);
+ struct arm_smmu_smr *smrs = master->smrs;
+
+ /*
+ * We *must* clear the S2CR first, because freeing the SMR means
+ * that it can be reallocated immediately
+ */
+ for ( i = 0; i < master->num_streamids; ++i )
+ {
+ u16 sid = master->streamids[i];
+ writel_relaxed(SMMU_S2CR_TYPE_FAULT,
+ gr0_base + SMMU_GR0_S2CR(sid));
+ }
+
+ /* Invalidate the SMRs before freeing back to the allocator */
+ for (i = 0; i < master->num_streamids; ++i) {
+ u8 idx = smrs[i].idx;
+ writel_relaxed(~SMMU_SMR_VALID, gr0_base + SMMU_GR0_SMR(idx));
+ __arm_smmu_free_bitmap(smmu->smr_map, idx);
+ }
+
+ master->smrs = NULL;
+ xfree(smrs);
+
+ master->cfg = NULL;
+ list_del(&master->list);
+ INIT_LIST_HEAD(&master->list);
+}
+
+static void arm_smmu_init_context_bank(struct arm_smmu_domain_cfg *cfg)
+{
+ u32 reg;
+ struct arm_smmu_device *smmu = cfg->smmu;
+ void __iomem *cb_base, *gr0_base, *gr1_base;
+ paddr_t p2maddr;
+
+ ASSERT(cfg->domain != NULL);
+ p2maddr = page_to_maddr(cfg->domain->arch.p2m.first_level);
+
+ gr0_base = SMMU_GR0(smmu);
+ gr1_base = SMMU_GR1(smmu);
+ cb_base = SMMU_CB_BASE(smmu) + SMMU_CB(smmu, cfg->cbndx);
+
+ /* CBAR */
+ reg = cfg->cbar;
+ if ( smmu->version == 1 )
+ reg |= cfg->irptndx << SMMU_CBAR_IRPTNDX_SHIFT;
+
+ reg |= SMMU_CB_VMID(cfg) << SMMU_CBAR_VMID_SHIFT;
+ writel_relaxed(reg, gr1_base + SMMU_GR1_CBAR(cfg->cbndx));
+
+ if ( smmu->version > 1 )
+ {
+ /* CBA2R */
+#ifdef CONFIG_ARM_64
+ reg = SMMU_CBA2R_RW64_64BIT;
+#else
+ reg = SMMU_CBA2R_RW64_32BIT;
+#endif
+ writel_relaxed(reg, gr1_base + SMMU_GR1_CBA2R(cfg->cbndx));
+ }
+
+ /* TTBR0 */
+ reg = (p2maddr & ((1ULL << 32) - 1));
+ writel_relaxed(reg, cb_base + SMMU_CB_TTBR0_LO);
+ reg = (p2maddr >> 32);
+ writel_relaxed(reg, cb_base + SMMU_CB_TTBR0_HI);
+
+ /*
+ * TCR
+ * We use long descriptor, with inner-shareable WBWA tables in TTBR0.
+ */
+ if ( smmu->version > 1 )
+ {
+ /* 4K Page Table */
+ if ( PAGE_SIZE == SZ_4K )
+ reg = SMMU_TCR_TG0_4K;
+ else
+ reg = SMMU_TCR_TG0_64K;
+
+ switch ( smmu->s2_output_size ) {
+ case 32:
+ reg |= (SMMU_TCR2_ADDR_32 << SMMU_TCR_PASIZE_SHIFT);
+ break;
+ case 36:
+ reg |= (SMMU_TCR2_ADDR_36 << SMMU_TCR_PASIZE_SHIFT);
+ break;
+ case 40:
+ reg |= (SMMU_TCR2_ADDR_40 << SMMU_TCR_PASIZE_SHIFT);
+ break;
+ case 42:
+ reg |= (SMMU_TCR2_ADDR_42 << SMMU_TCR_PASIZE_SHIFT);
+ break;
+ case 44:
+ reg |= (SMMU_TCR2_ADDR_44 << SMMU_TCR_PASIZE_SHIFT);
+ break;
+ case 48:
+ reg |= (SMMU_TCR2_ADDR_48 << SMMU_TCR_PASIZE_SHIFT);
+ break;
+ }
+ }
+ else
+ reg = 0;
+
+ /* The attribute to walk the page table should be the same as VTCR_EL2 */
+ reg |= SMMU_TCR_EAE |
+ (SMMU_TCR_SH_NS << SMMU_TCR_SH0_SHIFT) |
+ (SMMU_TCR_RGN_WBWA << SMMU_TCR_ORGN0_SHIFT) |
+ (SMMU_TCR_RGN_WBWA << SMMU_TCR_IRGN0_SHIFT) |
+ (SMMU_TCR_SL0_LVL_1 << SMMU_TCR_SL0_SHIFT);
+ writel_relaxed(reg, cb_base + SMMU_CB_TCR);
+
+ /* SCTLR */
+ reg = SMMU_SCTLR_CFCFG |
+ SMMU_SCTLR_CFIE |
+ SMMU_SCTLR_CFRE |
+ SMMU_SCTLR_M |
+ SMMU_SCTLR_EAE_SBOP;
+
+ writel_relaxed(reg, cb_base + SMMU_CB_SCTLR);
+}
+
+static struct arm_smmu_domain_cfg *
+arm_smmu_alloc_domain_context(struct domain *d,
+ struct arm_smmu_device *smmu)
+{
+ const struct dt_irq *irq;
+ int ret, start;
+ struct arm_smmu_domain_cfg *cfg;
+ struct arm_smmu_domain *smmu_domain = domain_hvm_iommu(d)->arch.priv;
+
+ ASSERT(spin_is_locked(&smmu_domain->lock));
+
+ cfg = xzalloc(struct arm_smmu_domain_cfg);
+ if ( !cfg )
+ return NULL;
+
+ /* Master already initialized to another domain ... */
+ if ( cfg->domain != NULL )
+ goto out_free_mem;
+
+ cfg->cbar = SMMU_CBAR_TYPE_S2_TRANS;
+ start = 0;
+
+ ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
+ smmu->num_context_banks);
+ if ( ret < 0 )
+ goto out_free_mem;
+
+ cfg->cbndx = ret;
+ if ( smmu->version == 1 )
+ {
+ cfg->irptndx = atomic_inc_return(&smmu->irptndx);
+ cfg->irptndx %= smmu->num_context_irqs;
+ }
+ else
+ cfg->irptndx = cfg->cbndx;
+
+ irq = &smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
+ ret = request_dt_irq(irq, arm_smmu_context_fault,
+ "arm-smmu-context-fault", cfg);
+ if ( ret )
+ {
+ smmu_err(smmu, "failed to request context IRQ %d (%u)\n",
+ cfg->irptndx, irq->irq);
+ cfg->irptndx = INVALID_IRPTNDX;
+ goto out_free_context;
+ }
+
+ cfg->domain = d;
+ cfg->smmu = smmu;
+
+ arm_smmu_init_context_bank(cfg);
+ list_add(&cfg->list, &smmu_domain->contexts);
+ INIT_LIST_HEAD(&cfg->masters);
+
+ return cfg;
+
+out_free_context:
+ __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
+out_free_mem:
+ xfree(cfg);
+
+ return NULL;
+}
+
+static void arm_smmu_destroy_domain_context(struct arm_smmu_domain_cfg *cfg)
+{
+ struct domain *d = cfg->domain;
+ struct arm_smmu_domain *smmu_domain = domain_hvm_iommu(d)->arch.priv;
+ struct arm_smmu_device *smmu = cfg->smmu;
+ void __iomem *cb_base;
+ const struct dt_irq *irq;
+
+ ASSERT(spin_is_locked(&smmu_domain->lock));
+ BUG_ON(!list_empty(&cfg->masters));
+
+ /* Disable the context bank and nuke the TLB before freeing it */
+ cb_base = SMMU_CB_BASE(smmu) + SMMU_CB(smmu, cfg->cbndx);
+ writel_relaxed(0, cb_base + SMMU_CB_SCTLR);
+ arm_smmu_tlb_inv_context(cfg);
+
+ if ( cfg->irptndx != INVALID_IRPTNDX )
+ {
+ irq = &smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
+ release_dt_irq(irq, cfg);
+ }
+
+ __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
+}
+
+static struct arm_smmu_device *
+arm_smmu_find_smmu_by_dev(const struct dt_device_node *dev)
+{
+ struct arm_smmu_device *smmu;
+ struct arm_smmu_master *master = NULL;
+
+ list_for_each_entry( smmu, &arm_smmu_devices, list )
+ {
+ master = find_smmu_master(smmu, dev);
+ if ( master )
+ break;
+ }
+
+ if ( !master )
+ return NULL;
+
+ return smmu;
+}
+
+static int arm_smmu_attach_dev(struct domain *d,
+ const struct dt_device_node *dev)
+{
+ struct arm_smmu_device *smmu = arm_smmu_find_smmu_by_dev(dev);
+ struct arm_smmu_master *master;
+ struct arm_smmu_domain *smmu_domain = domain_hvm_iommu(d)->arch.priv;
+ struct arm_smmu_domain_cfg *cfg = NULL;
+ struct arm_smmu_domain_cfg *curr;
+ int ret;
+
+ printk(XENLOG_DEBUG "arm-smmu: attach %s to domain %d\n",
+ dt_node_full_name(dev), d->domain_id);
+
+ if ( !smmu )
+ {
+ printk(XENLOG_ERR "%s: cannot attach to SMMU, is it on the same bus?\n",
+ dt_node_full_name(dev));
+ return -ENODEV;
+ }
+
+ master = find_smmu_master(smmu, dev);
+ BUG_ON(master == NULL);
+
+ /* Check if the device is already assigned to someone */
+ if ( master->cfg )
+ return -EBUSY;
+
+ spin_lock(&smmu_domain->lock);
+ list_for_each_entry( curr, &smmu_domain->contexts, list )
+ {
+ if ( curr->smmu == smmu )
+ {
+ cfg = curr;
+ break;
+ }
+ }
+
+ if ( !cfg )
+ {
+ cfg = arm_smmu_alloc_domain_context(d, smmu);
+ if ( !cfg )
+ {
+ smmu_err(smmu, "unable to allocate context for domain %u\n",
+ d->domain_id);
+ spin_unlock(&smmu_domain->lock);
+ return -ENOMEM;
+ }
+ }
+ spin_unlock(&smmu_domain->lock);
+
+ ret = arm_smmu_domain_add_master(d, cfg, master);
+ if ( ret )
+ {
+ spin_lock(&smmu_domain->lock);
+ if ( list_empty(&cfg->masters) )
+ arm_smmu_destroy_domain_context(cfg);
+ spin_unlock(&smmu_domain->lock);
+ }
+
+ return ret;
+}
+
+static __init int arm_smmu_id_size_to_bits(int size)
+{
+ switch ( size )
+ {
+ case 0:
+ return 32;
+ case 1:
+ return 36;
+ case 2:
+ return 40;
+ case 3:
+ return 42;
+ case 4:
+ return 44;
+ case 5:
+ default:
+ return 48;
+ }
+}
+
+static __init int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
+{
+ unsigned long size;
+ void __iomem *gr0_base = SMMU_GR0(smmu);
+ u32 id;
+
+ smmu_info(smmu, "probing hardware configuration...\n");
+
+ /*
+ * Primecell ID
+ */
+ id = readl_relaxed(gr0_base + SMMU_GR0_PIDR2);
+ smmu->version = ((id >> SMMU_PIDR2_ARCH_SHIFT) & SMMU_PIDR2_ARCH_MASK) + 1;
+ smmu_info(smmu, "SMMUv%d with:\n", smmu->version);
+
+ /* ID0 */
+ id = readl_relaxed(gr0_base + SMMU_GR0_ID0);
+#ifndef CONFIG_ARM_64
+ if ( ((id >> SMMU_ID0_PTFS_SHIFT) & SMMU_ID0_PTFS_MASK) ==
+ SMMU_ID0_PTFS_V8_ONLY )
+ {
+ smmu_err(smmu, "\tno v7 descriptor support!\n");
+ return -ENODEV;
+ }
+#endif
+ if ( id & SMMU_ID0_S1TS )
+ {
+ smmu->features |= SMMU_FEAT_TRANS_S1;
+ smmu_info(smmu, "\tstage 1 translation\n");
+ }
+
+ if ( id & SMMU_ID0_S2TS )
+ {
+ smmu->features |= SMMU_FEAT_TRANS_S2;
+ smmu_info(smmu, "\tstage 2 translation\n");
+ }
+
+ if ( id & SMMU_ID0_NTS )
+ {
+ smmu->features |= SMMU_FEAT_TRANS_NESTED;
+ smmu_info(smmu, "\tnested translation\n");
+ }
+
+ if ( !(smmu->features &
+ (SMMU_FEAT_TRANS_S1 | SMMU_FEAT_TRANS_S2 |
+ SMMU_FEAT_TRANS_NESTED)) )
+ {
+ smmu_err(smmu, "\tno translation support!\n");
+ return -ENODEV;
+ }
+
+ /* We need at least support for Stage 2 */
+ if ( !(smmu->features & SMMU_FEAT_TRANS_S2) )
+ {
+ smmu_err(smmu, "\tno stage 2 translation!\n");
+ return -ENODEV;
+ }
+
+ if ( id & SMMU_ID0_CTTW )
+ {
+ smmu->features |= SMMU_FEAT_COHERENT_WALK;
+ smmu_info(smmu, "\tcoherent table walk\n");
+ }
+
+ if ( id & SMMU_ID0_SMS )
+ {
+ u32 smr, sid, mask;
+
+ smmu->features |= SMMU_FEAT_STREAM_MATCH;
+ smmu->num_mapping_groups = (id >> SMMU_ID0_NUMSMRG_SHIFT) &
+ SMMU_ID0_NUMSMRG_MASK;
+ if ( smmu->num_mapping_groups == 0 )
+ {
+ smmu_err(smmu,
+ "stream-matching supported, but no SMRs present!\n");
+ return -ENODEV;
+ }
+
+ smr = SMMU_SMR_MASK_MASK << SMMU_SMR_MASK_SHIFT;
+ smr |= (SMMU_SMR_ID_MASK << SMMU_SMR_ID_SHIFT);
+ writel_relaxed(smr, gr0_base + SMMU_GR0_SMR(0));
+ smr = readl_relaxed(gr0_base + SMMU_GR0_SMR(0));
+
+ mask = (smr >> SMMU_SMR_MASK_SHIFT) & SMMU_SMR_MASK_MASK;
+ sid = (smr >> SMMU_SMR_ID_SHIFT) & SMMU_SMR_ID_MASK;
+ if ( (mask & sid) != sid )
+ {
+ smmu_err(smmu,
+ "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
+ mask, sid);
+ return -ENODEV;
+ }
+ smmu->smr_mask_mask = mask;
+ smmu->smr_id_mask = sid;
+
+ smmu_info(smmu,
+ "\tstream matching with %u register groups, mask 0x%x\n",
+ smmu->num_mapping_groups, mask);
+ }
+
+ /* ID1 */
+ id = readl_relaxed(gr0_base + SMMU_GR0_ID1);
+ smmu->pagesize = (id & SMMU_ID1_PAGESIZE) ? SZ_64K : SZ_4K;
+
+ /* Check for size mismatch of SMMU address space from mapped region */
+ size = 1 << (((id >> SMMU_ID1_NUMPAGENDXB_SHIFT) &
+ SMMU_ID1_NUMPAGENDXB_MASK) + 1);
+ size *= (smmu->pagesize << 1);
+ if ( smmu->size != size )
+ smmu_warn(smmu, "SMMU address space size (0x%lx) differs "
+ "from mapped region size (0x%lx)!\n", size, smmu->size);
+
+ smmu->num_s2_context_banks = (id >> SMMU_ID1_NUMS2CB_SHIFT) &
+ SMMU_ID1_NUMS2CB_MASK;
+ smmu->num_context_banks = (id >> SMMU_ID1_NUMCB_SHIFT) &
+ SMMU_ID1_NUMCB_MASK;
+ if ( smmu->num_s2_context_banks > smmu->num_context_banks )
+ {
+ smmu_err(smmu, "impossible number of S2 context banks!\n");
+ return -ENODEV;
+ }
+ smmu_info(smmu, "\t%u context banks (%u stage-2 only)\n",
+ smmu->num_context_banks, smmu->num_s2_context_banks);
+
+ /* ID2 */
+ id = readl_relaxed(gr0_base + SMMU_GR0_ID2);
+ size = arm_smmu_id_size_to_bits((id >> SMMU_ID2_IAS_SHIFT) &
+ SMMU_ID2_IAS_MASK);
+
+ /*
+ * Stage-1 output limited by stage-2 input size due to VTCR_EL2
+ * setup (see setup_virt_paging)
+ */
+ /* Current maximum output size of 40 bits */
+ smmu->s1_output_size = min(40UL, size);
+
+ /* The stage-2 output mask is also applied for bypass */
+ size = arm_smmu_id_size_to_bits((id >> SMMU_ID2_OAS_SHIFT) &
+ SMMU_ID2_OAS_MASK);
+ smmu->s2_output_size = min((unsigned long)PADDR_BITS, size);
+
+ if ( smmu->version == 1 )
+ smmu->input_size = 32;
+ else
+ {
+#ifdef CONFIG_ARM_64
+ size = (id >> SMMU_ID2_UBS_SHIFT) & SMMU_ID2_UBS_MASK;
+ size = min(39, arm_smmu_id_size_to_bits(size));
+#else
+ size = 32;
+#endif
+ smmu->input_size = size;
+
+ if ( (PAGE_SIZE == SZ_4K && !(id & SMMU_ID2_PTFS_4K) ) ||
+ (PAGE_SIZE == SZ_64K && !(id & SMMU_ID2_PTFS_64K)) ||
+ (PAGE_SIZE != SZ_4K && PAGE_SIZE != SZ_64K) )
+ {
+ smmu_err(smmu, "CPU page size 0x%lx unsupported\n",
+ PAGE_SIZE);
+ return -ENODEV;
+ }
+ }
+
+ smmu_info(smmu, "\t%lu-bit VA, %lu-bit IPA, %lu-bit PA\n",
+ smmu->input_size, smmu->s1_output_size, smmu->s2_output_size);
+ return 0;
+}
+
+static __init void arm_smmu_device_reset(struct arm_smmu_device *smmu)
+{
+ void __iomem *gr0_base = SMMU_GR0(smmu);
+ void __iomem *cb_base;
+ int i = 0;
+ u32 reg;
+
+ smmu_dbg(smmu, "device reset\n");
+
+ /* Clear Global FSR */
+ reg = readl_relaxed(SMMU_GR0_NS(smmu) + SMMU_GR0_sGFSR);
+ writel(reg, SMMU_GR0_NS(smmu) + SMMU_GR0_sGFSR);
+
+ /* Mark all SMRn as invalid and all S2CRn as bypass */
+ for ( i = 0; i < smmu->num_mapping_groups; ++i )
+ {
+ writel_relaxed(~SMMU_SMR_VALID, gr0_base + SMMU_GR0_SMR(i));
+ writel_relaxed(SMMU_S2CR_TYPE_FAULT, gr0_base + SMMU_GR0_S2CR(i));
+ }
+
+ /* Make sure all context banks are disabled and clear CB_FSR */
+ for ( i = 0; i < smmu->num_context_banks; ++i )
+ {
+ cb_base = SMMU_CB_BASE(smmu) + SMMU_CB(smmu, i);
+ writel_relaxed(0, cb_base + SMMU_CB_SCTLR);
+ writel_relaxed(SMMU_FSR_FAULT, cb_base + SMMU_CB_FSR);
+ }
+
+ /* Invalidate the TLB, just in case */
+ writel_relaxed(0, gr0_base + SMMU_GR0_STLBIALL);
+ writel_relaxed(0, gr0_base + SMMU_GR0_TLBIALLH);
+ writel_relaxed(0, gr0_base + SMMU_GR0_TLBIALLNSNH);
+
+ reg = readl_relaxed(SMMU_GR0_NS(smmu) + SMMU_GR0_sCR0);
+
+ /* Enable fault reporting */
+ reg |= (SMMU_sCR0_GFRE | SMMU_sCR0_GFIE |
+ SMMU_sCR0_GCFGFRE | SMMU_sCR0_GCFGFIE);
+
+ /* Disable TLB broadcasting. */
+ reg |= (SMMU_sCR0_VMIDPNE | SMMU_sCR0_PTM);
+
+ /* Enable client access, generate a fault if no mapping is found */
+ reg &= ~(SMMU_sCR0_CLIENTPD);
+ reg |= SMMU_sCR0_USFCFG;
+
+ /* Disable forced broadcasting */
+ reg &= ~SMMU_sCR0_FB;
+
+ /* Don't upgrade barriers */
+ reg &= ~(SMMU_sCR0_BSU_MASK << SMMU_sCR0_BSU_SHIFT);
+
+ /* Push the button */
+ arm_smmu_tlb_sync(smmu);
+ writel_relaxed(reg, SMMU_GR0_NS(smmu) + SMMU_GR0_sCR0);
+}
+
+int arm_smmu_iommu_domain_init(struct domain *d)
+{
+ struct arm_smmu_domain *smmu_domain;
+
+ smmu_domain = xzalloc(struct arm_smmu_domain);
+ if ( !smmu_domain )
+ return -ENOMEM;
+
+ spin_lock_init(&smmu_domain->lock);
+ INIT_LIST_HEAD(&smmu_domain->contexts);
+
+ domain_hvm_iommu(d)->arch.priv = smmu_domain;
+
+ return 0;
+}
+
+void arm_smmu_iommu_dom0_init(struct domain *d)
+{
+ struct arm_smmu_device *smmu;
+ struct rb_node *node;
+
+ printk(XENLOG_DEBUG "arm-smmu: Initialize dom0\n");
+
+ list_for_each_entry( smmu, &arm_smmu_devices, list )
+ {
+ for ( node = rb_first(&smmu->masters); node; node = rb_next(node) )
+ {
+ struct arm_smmu_master *master;
+
+ master = container_of(node, struct arm_smmu_master, node);
+
+ if ( dt_device_used_by(master->dt_node) == DOMID_XEN ||
+ platform_device_is_blacklisted(master->dt_node) )
+ continue;
+
+ arm_smmu_attach_dev(d, master->dt_node);
+ }
+ }
+}
+
+void arm_smmu_iommu_domain_teardown(struct domain *d)
+{
+ struct arm_smmu_domain *smmu_domain = domain_hvm_iommu(d)->arch.priv;
+ struct arm_smmu_domain_cfg *cfg, *_cfg;
+
+ spin_lock(&smmu_domain->lock);
+
+ list_for_each_entry_safe( cfg, _cfg, &smmu_domain->contexts, list )
+ {
+ struct arm_smmu_master *master, *_master;
+
+ list_for_each_entry_safe( master, _master, &cfg->masters, list )
+ arm_smmu_domain_remove_master(master);
+ arm_smmu_destroy_domain_context(cfg);
+ }
+
+ spin_unlock(&smmu_domain->lock);
+
+ xfree(smmu_domain);
+}
+
+static const struct iommu_ops arm_smmu_iommu_ops = {
+ .init = arm_smmu_iommu_domain_init,
+ .dom0_init = arm_smmu_iommu_dom0_init,
+ .teardown = arm_smmu_iommu_domain_teardown,
+ .iotlb_flush = arm_smmu_iotlb_flush,
+ .iotlb_flush_all = arm_smmu_iotlb_flush_all,
+};
+
+static int __init smmu_init(struct dt_device_node *dev,
+ const void *data)
+{
+ struct arm_smmu_device *smmu;
+ int res;
+ u64 addr, size;
+ unsigned int num_irqs, i;
+ struct dt_phandle_args masterspec;
+ struct rb_node *node;
+
+ /* Even if the device can't be initialized, we don't want to give to
+ * dom0 the smmu device
+ */
+ dt_device_set_used_by(dev, DOMID_XEN);
+
+ smmu = xzalloc(struct arm_smmu_device);
+ if ( !smmu )
+ {
+ printk(XENLOG_ERR "%s: failed to allocate arm_smmu_device\n",
+ dt_node_full_name(dev));
+ return -ENOMEM;
+ }
+
+ smmu->node = dev;
+ check_driver_options(smmu);
+
+ res = dt_device_get_address(smmu->node, 0, &addr, &size);
+ if ( res )
+ {
+ smmu_err(smmu, "unable to retrieve the base address of the SMMU\n");
+ goto out_err;
+ }
+
+ smmu->base = ioremap_nocache(addr, size);
+ if ( !smmu->base )
+ {
+ smmu_err(smmu, "unable to map the SMMU memory\n");
+ goto out_err;
+ }
+
+ smmu->size = size;
+
+ if ( !dt_property_read_u32(smmu->node, "#global-interrupts",
+ &smmu->num_global_irqs) )
+ {
+ smmu_err(smmu, "missing #global-interrupts\n");
+ goto out_unmap;
+ }
+
+ num_irqs = dt_number_of_irq(smmu->node);
+ if ( num_irqs > smmu->num_global_irqs )
+ smmu->num_context_irqs = num_irqs - smmu->num_global_irqs;
+
+ if ( !smmu->num_context_irqs )
+ {
+ smmu_err(smmu, "found %d interrupts but expected at least %d\n",
+ num_irqs, smmu->num_global_irqs + 1);
+ goto out_unmap;
+ }
+
+ smmu->irqs = xzalloc_array(struct dt_irq, num_irqs);
+ if ( !smmu->irqs )
+ {
+ smmu_err(smmu, "failed to allocated %d irqs\n", num_irqs);
+ goto out_unmap;
+ }
+
+ for ( i = 0; i < num_irqs; i++ )
+ {
+ res = dt_device_get_irq(smmu->node, i, &smmu->irqs[i]);
+ if ( res )
+ {
+ smmu_err(smmu, "failed to get irq index %d\n", i);
+ goto out_free_irqs;
+ }
+ }
+
+ smmu->sids = xzalloc_array(unsigned long,
+ BITS_TO_LONGS(SMMU_MAX_STREAMIDS));
+ if ( !smmu->sids )
+ {
+ smmu_err(smmu, "failed to allocated bitmap for stream ID tracking\n");
+ goto out_free_masters;
+ }
+
+
+ i = 0;
+ smmu->masters = RB_ROOT;
+ while ( !dt_parse_phandle_with_args(smmu->node, "mmu-masters",
+ "#stream-id-cells", i, &masterspec) )
+ {
+ res = register_smmu_master(smmu, &masterspec);
+ if ( res )
+ {
+ smmu_err(smmu, "failed to add master %s\n",
+ masterspec.np->name);
+ goto out_free_masters;
+ }
+ i++;
+ }
+
+ smmu_info(smmu, "registered %d master devices\n", i);
+
+ res = arm_smmu_device_cfg_probe(smmu);
+ if ( res )
+ {
+ smmu_err(smmu, "failed to probe the SMMU\n");
+ goto out_free_masters;
+ }
+
+ if ( smmu->version > 1 &&
+ smmu->num_context_banks != smmu->num_context_irqs )
+ {
+ smmu_err(smmu,
+ "found only %d context interrupt(s) but %d required\n",
+ smmu->num_context_irqs, smmu->num_context_banks);
+ goto out_free_masters;
+ }
+
+ smmu_dbg(smmu, "register global IRQs handler\n");
+
+ for ( i = 0; i < smmu->num_global_irqs; ++i )
+ {
+ smmu_dbg(smmu, "\t- global IRQ %u\n", smmu->irqs[i].irq);
+ res = request_dt_irq(&smmu->irqs[i], arm_smmu_global_fault,
+ "arm-smmu global fault", smmu);
+ if ( res )
+ {
+ smmu_err(smmu, "failed to request global IRQ %d (%u)\n",
+ i, smmu->irqs[i].irq);
+ goto out_release_irqs;
+ }
+ }
+
+ INIT_LIST_HEAD(&smmu->list);
+ list_add(&smmu->list, &arm_smmu_devices);
+
+ arm_smmu_device_reset(smmu);
+
+ iommu_set_ops(&arm_smmu_iommu_ops);
+
+ /* sids field can be freed... */
+ xfree(smmu->sids);
+ smmu->sids = NULL;
+
+ return 0;
+
+out_release_irqs:
+ while (i--)
+ release_dt_irq(&smmu->irqs[i], smmu);
+
+out_free_masters:
+ for ( node = rb_first(&smmu->masters); node; node = rb_next(node) )
+ {
+ struct arm_smmu_master *master;
+
+ master = container_of(node, struct arm_smmu_master, node);
+ xfree(master);
+ }
+
+ xfree(smmu->sids);
+
+out_free_irqs:
+ xfree(smmu->irqs);
+
+out_unmap:
+ iounmap(smmu->base);
+
+out_err:
+ xfree(smmu);
+
+ return -ENODEV;
+}
+
+static const char * const smmu_dt_compat[] __initconst =
+{
+ "arm,mmu-400",
+ NULL
+};
+
+DT_DEVICE_START(smmu, "ARM SMMU", DEVICE_IOMMU)
+ .compatible = smmu_dt_compat,
+ .init = smmu_init,
+DT_DEVICE_END
+
+/*
+ * Local variables:
+ * mode: C
+ * c-file-style: "BSD"
+ * c-basic-offset: 4
+ * indent-tabs-mode: nil
+ * End:
+ */
--
1.7.10.4
^ permalink raw reply related [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 00/12] IOMMU support for ARM
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
` (11 preceding siblings ...)
2014-02-07 17:43 ` [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers Julien Grall
@ 2014-02-07 17:51 ` Julien Grall
12 siblings, 0 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-07 17:51 UTC (permalink / raw)
To: Julien Grall; +Cc: xen-devel, tim, ian.campbell, stefano.stabellini, patches
I forgot to add a link to repository:
git://xenbits.xen.org/people/julieng/xen-unstable.git branch smmu
This series has also dependency on:
- early printk series : http://lists.xen.org/archives/html/xen-devel/2014-01/msg00288.html
- interrupt series: http://lists.xen.org/archives/html/xen-devel/2014-01/msg02139.html
- few bug fixes on the previous series
On 02/07/2014 05:42 PM, Julien Grall wrote:
> Hello,
>
> This patch series add support for IOMMU on ARM. It have also added ARM SMMU
> driver which is used for instance on Midway.
>
> The IOMMU architecture for ARM is relying on the page table is shared between
> the processor and each IOMMU.
>
> The patch series is divided following:
> - #1: fixing grant-table with IOMMU. Will be necessary for ARM later
> - #2-#3: Make static some vtd functions
> - #4-#5: Adding new device tree functions
> - #6-#9: Prepare IOMMU code to add support for ARM
> - #10-#11: Add IOMMU architecture for ARM
> - #12: Add SMMU drivers
>
> For now the 1:1 workaround is not removed because a same platform can have
> DMA-capable device which are under an IOMMU and some not. It's a problem for
> the swiotlb which needs to know if the device is procted or not when foreign
> mapping is mapped in dom0.
>
> When I talked with Stefano, 2 solutions came:
> - Having a property in each "protected" device
> - List in the hypervisor node the procted devices
>
> I didn't yet decide which solution I will use.
>
> Any comments, questions are welcomed.
>
> Sincerely yours,
>
> Julien Grall (12):
> xen/common: grant-table: only call IOMMU if paging mode translate is
> disabled
> xen/passthrough: vtd: Don't export iommu_domain_teardown
> xen/passthrough: vtd: Don't export iommu_set_pgd
> xen/dts: Add dt_property_read_bool
> xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle
> xen/passthrough: rework dom0_pvh_reqs to use it also on ARM
> xen/passthrough: iommu: Don't need to map dom0 page when the PT is
> shared
> xen/passthrough: iommu: Split generic IOMMU code
> xen/passthrough: iommu: Introduce arch specific code
> xen/passthrough: Introduce IOMMU ARM architure
> MAINTAINERS: Add drivers/passthrough/arm
> drivers/passthrough: arm: Add support for SMMU drivers
>
> MAINTAINERS | 1 +
> xen/arch/arm/Rules.mk | 1 +
> xen/arch/arm/domain.c | 7 +
> xen/arch/arm/domain_build.c | 2 +
> xen/arch/arm/p2m.c | 4 +
> xen/arch/arm/setup.c | 2 +
> xen/arch/x86/domctl.c | 6 +-
> xen/arch/x86/hvm/io.c | 2 +-
> xen/arch/x86/tboot.c | 3 +-
> xen/common/device_tree.c | 157 ++-
> xen/common/grant_table.c | 7 +-
> xen/drivers/passthrough/Makefile | 7 +-
> xen/drivers/passthrough/amd/iommu_cmd.c | 2 +-
> xen/drivers/passthrough/amd/iommu_guest.c | 8 +-
> xen/drivers/passthrough/amd/iommu_map.c | 56 +-
> xen/drivers/passthrough/amd/pci_amd_iommu.c | 53 +-
> xen/drivers/passthrough/arm/Makefile | 2 +
> xen/drivers/passthrough/arm/iommu.c | 65 +
> xen/drivers/passthrough/arm/smmu.c | 1701 +++++++++++++++++++++++++++
> xen/drivers/passthrough/iommu.c | 525 +--------
> xen/drivers/passthrough/iommu_pci.c | 468 ++++++++
> xen/drivers/passthrough/iommu_x86.c | 106 ++
> xen/drivers/passthrough/vtd/iommu.c | 124 +-
> xen/include/asm-arm/device.h | 3 +-
> xen/include/asm-arm/domain.h | 2 +
> xen/include/asm-arm/hvm/iommu.h | 10 +
> xen/include/asm-arm/iommu.h | 36 +
> xen/include/asm-x86/hvm/iommu.h | 29 +
> xen/include/asm-x86/iommu.h | 50 +
> xen/include/xen/device_tree.h | 75 ++
> xen/include/xen/hvm/iommu.h | 27 +-
> xen/include/xen/iommu.h | 51 +-
> 32 files changed, 2891 insertions(+), 701 deletions(-)
> create mode 100644 xen/drivers/passthrough/arm/Makefile
> create mode 100644 xen/drivers/passthrough/arm/iommu.c
> create mode 100644 xen/drivers/passthrough/arm/smmu.c
> create mode 100644 xen/drivers/passthrough/iommu_pci.c
> create mode 100644 xen/drivers/passthrough/iommu_x86.c
> create mode 100644 xen/include/asm-arm/hvm/iommu.h
> create mode 100644 xen/include/asm-arm/iommu.h
> create mode 100644 xen/include/asm-x86/iommu.h
>
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 01/12] xen/common: grant-table: only call IOMMU if paging mode translate is disabled
2014-02-07 17:43 ` [RFC for-4.5 01/12] xen/common: grant-table: only call IOMMU if paging mode translate is disabled Julien Grall
@ 2014-02-10 8:02 ` Jan Beulich
2014-02-19 12:25 ` Ian Campbell
1 sibling, 0 replies; 51+ messages in thread
From: Jan Beulich @ 2014-02-10 8:02 UTC (permalink / raw)
To: Julien Grall
Cc: Keir Fraser, ian.campbell, patches, tim, stefano.stabellini, xen-devel
>>> On 07.02.14 at 18:43, Julien Grall <julien.grall@linaro.org> wrote:
> From Xen point of view, ARM guests are PV guest with paging auto translate
> enabled.
>
> When IOMMU support will be added for ARM, mapping grant ref will always
> crash
> Xen due to the BUG_ON in __gnttab_map_grant_ref.
>
> On x86:
> - PV guests always have paging mode translate disabled
> - PVH and HVM guests have always paging mode translate enabled
>
> It means that we can safely replace the check that the domain is a PV guests
> by checking if the guest has paging mode translate enabled.
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
> Cc: Keir Fraser <keir@xen.org>
> ---
> xen/common/grant_table.c | 7 ++-----
> 1 file changed, 2 insertions(+), 5 deletions(-)
>
> diff --git a/xen/common/grant_table.c b/xen/common/grant_table.c
> index 107b000..778bdb7 100644
> --- a/xen/common/grant_table.c
> +++ b/xen/common/grant_table.c
> @@ -721,12 +721,10 @@ __gnttab_map_grant_ref(
>
> double_gt_lock(lgt, rgt);
>
> - if ( is_pv_domain(ld) && need_iommu(ld) )
> + if ( !paging_mode_translate(ld) && need_iommu(ld) )
> {
> unsigned int wrc, rdc;
> int err = 0;
> - /* Shouldn't happen, because you can't use iommu in a HVM domain. */
> - BUG_ON(paging_mode_translate(ld));
> /* We're not translated, so we know that gmfns and mfns are
> the same things, so the IOMMU entry is always 1-to-1. */
> mapcount(lgt, rd, frame, &wrc, &rdc);
> @@ -931,11 +929,10 @@ __gnttab_unmap_common(
> act->pin -= GNTPIN_hstw_inc;
> }
>
> - if ( is_pv_domain(ld) && need_iommu(ld) )
> + if ( !paging_mode_translate(ld) && need_iommu(ld) )
> {
> unsigned int wrc, rdc;
> int err = 0;
> - BUG_ON(paging_mode_translate(ld));
> mapcount(lgt, rd, op->frame, &wrc, &rdc);
> if ( (wrc + rdc) == 0 )
> err = iommu_unmap_page(ld, op->frame);
> --
> 1.7.10.4
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 03/12] xen/passthrough: vtd: Don't export iommu_set_pgd
2014-02-07 17:43 ` [RFC for-4.5 03/12] xen/passthrough: vtd: Don't export iommu_set_pgd Julien Grall
@ 2014-02-10 8:04 ` Jan Beulich
2014-02-10 8:07 ` Zhang, Xiantao
2014-02-19 12:27 ` Ian Campbell
1 sibling, 1 reply; 51+ messages in thread
From: Jan Beulich @ 2014-02-10 8:04 UTC (permalink / raw)
To: Julien Grall
Cc: ian.campbell, patches, tim, stefano.stabellini, xen-devel, Xiantoa Zhang
>>> On 07.02.14 at 18:43, Julien Grall <julien.grall@linaro.org> wrote:
> iommu_set_pgd is only used internally in
> xen/drivers/passthrough/vtd/iommu.c
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Xiantoa Zhang <xiantao.zhang@intel.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
> ---
> xen/drivers/passthrough/vtd/iommu.c | 2 +-
> xen/include/xen/iommu.h | 1 -
> 2 files changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/xen/drivers/passthrough/vtd/iommu.c
> b/xen/drivers/passthrough/vtd/iommu.c
> index a8d33fc..d5ce5b7 100644
> --- a/xen/drivers/passthrough/vtd/iommu.c
> +++ b/xen/drivers/passthrough/vtd/iommu.c
> @@ -1826,7 +1826,7 @@ static int vtd_ept_page_compatible(struct iommu *iommu)
> /*
> * set VT-d page table directory to EPT table if allowed
> */
> -void iommu_set_pgd(struct domain *d)
> +static void iommu_set_pgd(struct domain *d)
> {
> struct hvm_iommu *hd = domain_hvm_iommu(d);
> mfn_t pgd_mfn;
> diff --git a/xen/include/xen/iommu.h b/xen/include/xen/iommu.h
> index 8bb0a1d..fcbc432 100644
> --- a/xen/include/xen/iommu.h
> +++ b/xen/include/xen/iommu.h
> @@ -68,7 +68,6 @@ int iommu_map_page(struct domain *d, unsigned long gfn,
> unsigned long mfn,
> unsigned int flags);
> int iommu_unmap_page(struct domain *d, unsigned long gfn);
> void iommu_pte_flush(struct domain *d, u64 gfn, u64 *pte, int order, int
> present);
> -void iommu_set_pgd(struct domain *d);
> void iommu_domain_teardown(struct domain *d);
>
> void pt_pci_init(void);
> --
> 1.7.10.4
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM
2014-02-07 17:43 ` [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM Julien Grall
@ 2014-02-10 8:07 ` Jan Beulich
2014-02-10 11:42 ` Julien Grall
2014-02-10 16:10 ` Julien Grall
1 sibling, 1 reply; 51+ messages in thread
From: Jan Beulich @ 2014-02-10 8:07 UTC (permalink / raw)
To: Julien Grall
Cc: ian.campbell, patches, tim, stefano.stabellini, xen-devel, Xiantao Zhang
>>> On 07.02.14 at 18:43, Julien Grall <julien.grall@linaro.org> wrote:
> DOM0 on ARM will have the same requirements as DOM0 PVH when iommu is
> enabled.
> Both PVH and ARM guest has paging mode translate enabled, so Xen can use it
> to know if it needs to check the requirements.
>
> Rename the function and remove "pvh" word in the commit message.
s/commit/panic/ ?
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
Other than the above,
Acked-by: Jan Beulich <jbeulich@suse.com>
> ---
> xen/drivers/passthrough/iommu.c | 14 +++++++++-----
> 1 file changed, 9 insertions(+), 5 deletions(-)
>
> diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
> index 19b0e23..26a5d91 100644
> --- a/xen/drivers/passthrough/iommu.c
> +++ b/xen/drivers/passthrough/iommu.c
> @@ -130,13 +130,18 @@ int iommu_domain_init(struct domain *d)
> return hd->platform_ops->init(d);
> }
>
> -static __init void check_dom0_pvh_reqs(struct domain *d)
> +static __init void check_dom0_reqs(struct domain *d)
> {
> + if ( !paging_mode_translate(d) )
> + return;
> +
> if ( !iommu_enabled )
> - panic("Presently, iommu must be enabled for pvh dom0\n");
> + panic("Presently, iommu must be enabled to use dom0 with translate
> "
> + "paging mode\n");
>
> if ( iommu_passthrough )
> - panic("For pvh dom0, dom0-passthrough must not be enabled\n");
> + panic("Dom0 uses translate paging mode, dom0-passthrough must not be
> "
> + "enabled\n");
>
> iommu_dom0_strict = 1;
> }
> @@ -145,8 +150,7 @@ void __init iommu_dom0_init(struct domain *d)
> {
> struct hvm_iommu *hd = domain_hvm_iommu(d);
>
> - if ( is_pvh_domain(d) )
> - check_dom0_pvh_reqs(d);
> + check_dom0_reqs(d);
>
> if ( !iommu_enabled )
> return;
> --
> 1.7.10.4
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 03/12] xen/passthrough: vtd: Don't export iommu_set_pgd
2014-02-10 8:04 ` Jan Beulich
@ 2014-02-10 8:07 ` Zhang, Xiantao
0 siblings, 0 replies; 51+ messages in thread
From: Zhang, Xiantao @ 2014-02-10 8:07 UTC (permalink / raw)
To: Jan Beulich, Julien Grall
Cc: xen-devel, tim, stefano.stabellini, ian.campbell, patches
Acked-by: Xiantao Zhang <Xiantao.zhang@intel.com>
Xiantao
-----Original Message-----
From: Jan Beulich [mailto:JBeulich@suse.com]
Sent: Monday, February 10, 2014 4:05 PM
To: Julien Grall
Cc: ian.campbell@citrix.com; stefano.stabellini@citrix.com; Zhang, Xiantao; patches@linaro.org; xen-devel@lists.xenproject.org; tim@xen.org
Subject: Re: [RFC for-4.5 03/12] xen/passthrough: vtd: Don't export iommu_set_pgd
>>> On 07.02.14 at 18:43, Julien Grall <julien.grall@linaro.org> wrote:
> iommu_set_pgd is only used internally in
> xen/drivers/passthrough/vtd/iommu.c
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Xiantoa Zhang <xiantao.zhang@intel.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
> ---
> xen/drivers/passthrough/vtd/iommu.c | 2 +-
> xen/include/xen/iommu.h | 1 -
> 2 files changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/xen/drivers/passthrough/vtd/iommu.c
> b/xen/drivers/passthrough/vtd/iommu.c
> index a8d33fc..d5ce5b7 100644
> --- a/xen/drivers/passthrough/vtd/iommu.c
> +++ b/xen/drivers/passthrough/vtd/iommu.c
> @@ -1826,7 +1826,7 @@ static int vtd_ept_page_compatible(struct iommu
> *iommu)
> /*
> * set VT-d page table directory to EPT table if allowed
> */
> -void iommu_set_pgd(struct domain *d)
> +static void iommu_set_pgd(struct domain *d)
> {
> struct hvm_iommu *hd = domain_hvm_iommu(d);
> mfn_t pgd_mfn;
> diff --git a/xen/include/xen/iommu.h b/xen/include/xen/iommu.h index
> 8bb0a1d..fcbc432 100644
> --- a/xen/include/xen/iommu.h
> +++ b/xen/include/xen/iommu.h
> @@ -68,7 +68,6 @@ int iommu_map_page(struct domain *d, unsigned long
> gfn, unsigned long mfn,
> unsigned int flags); int iommu_unmap_page(struct
> domain *d, unsigned long gfn); void iommu_pte_flush(struct domain *d,
> u64 gfn, u64 *pte, int order, int present); -void iommu_set_pgd(struct
> domain *d); void iommu_domain_teardown(struct domain *d);
>
> void pt_pci_init(void);
> --
> 1.7.10.4
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 07/12] xen/passthrough: iommu: Don't need to map dom0 page when the PT is shared
2014-02-07 17:43 ` [RFC for-4.5 07/12] xen/passthrough: iommu: Don't need to map dom0 page when the PT is shared Julien Grall
@ 2014-02-10 8:11 ` Jan Beulich
2014-02-19 12:37 ` Ian Campbell
1 sibling, 0 replies; 51+ messages in thread
From: Jan Beulich @ 2014-02-10 8:11 UTC (permalink / raw)
To: Julien Grall
Cc: ian.campbell, patches, tim, stefano.stabellini, xen-devel, Xiantao Zhang
>>> On 07.02.14 at 18:43, Julien Grall <julien.grall@linaro.org> wrote:
> Currently iommu_init_dom0 is browsing the page list and call map_page
> callback
> on each page.
>
> On both AMD and VTD drivers, the function will directly return if the page
> table is shared with the processor. So Xen can safely avoid to run through
> the page list.
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
> ---
> xen/drivers/passthrough/iommu.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
> index 26a5d91..0a26956 100644
> --- a/xen/drivers/passthrough/iommu.c
> +++ b/xen/drivers/passthrough/iommu.c
> @@ -157,7 +157,7 @@ void __init iommu_dom0_init(struct domain *d)
>
> register_keyhandler('o', &iommu_p2m_table);
> d->need_iommu = !!iommu_dom0_strict;
> - if ( need_iommu(d) )
> + if ( need_iommu(d) && !iommu_use_hap_pt(d) )
> {
> struct page_info *page;
> unsigned int i = 0;
> --
> 1.7.10.4
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 08/12] xen/passthrough: iommu: Split generic IOMMU code
2014-02-07 17:43 ` [RFC for-4.5 08/12] xen/passthrough: iommu: Split generic IOMMU code Julien Grall
@ 2014-02-10 8:22 ` Jan Beulich
2014-02-10 11:52 ` Julien Grall
2014-02-19 12:39 ` Ian Campbell
1 sibling, 1 reply; 51+ messages in thread
From: Jan Beulich @ 2014-02-10 8:22 UTC (permalink / raw)
To: Julien Grall
Cc: ian.campbell, patches, tim, stefano.stabellini, xen-devel, Xiantao Zhang
>>> On 07.02.14 at 18:43, Julien Grall <julien.grall@linaro.org> wrote:
> The generic IOMMU framework code (xen/drivers/passthrough/iommu.c) contains
> functions specific to x86 and PCI.
>
> Split the framework in 3 distincts files:
> - iommu.c: contains generic functions shared between x86 and ARM
> (when it will be supported)
> - iommu_pci.c: contains specific functions for PCI passthrough
> - iommu_x86.c: contains specific functions for x86
>
> iommu_pci.c will be only compiled when PCI is supported by the architecture
> (eg. HAS_PCI is defined).
>
> This patch is mostly code movement in new files.
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
> Cc: Jan Beulich <jbeulich@suse.com>
> ---
> xen/drivers/passthrough/Makefile | 6 +-
> xen/drivers/passthrough/iommu.c | 473 +----------------------------------
> xen/drivers/passthrough/iommu_pci.c | 468 ++++++++++++++++++++++++++++++++++
There's xen/drivers/passthrough/pci.c already - any reason not to
move the code there?
> xen/drivers/passthrough/iommu_x86.c | 65 +++++
Same here for xen/drivers/passthrough/x86/.
> @@ -696,125 +344,6 @@ void iommu_crash_shutdown(void)
> iommu_enabled = iommu_intremap = 0;
> }
>
> -int iommu_do_domctl(
> - struct xen_domctl *domctl, struct domain *d,
> - XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)
The function itself should probably not be moved out. Either the
PCI-specific pieces of it should be made conditional, or a
descendant function be created. Since (afaict) you'll need all of
the domctl-s (with different arguments) too for pass-through on
ARM - what's your plan for them?
> --- a/xen/drivers/passthrough/vtd/iommu.c
> +++ b/xen/drivers/passthrough/vtd/iommu.c
> @@ -1784,31 +1784,31 @@ static int intel_iommu_unmap_page(struct domain *d,
> unsigned long gfn)
>
> void iommu_pte_flush(struct domain *d, u64 gfn, u64 *pte,
> int order, int present)
> -{
> - struct acpi_drhd_unit *drhd;
> - struct iommu *iommu = NULL;
> - struct hvm_iommu *hd = domain_hvm_iommu(d);
> - int flush_dev_iotlb;
> - int iommu_domid;
> + {
> + struct acpi_drhd_unit *drhd;
> + struct iommu *iommu = NULL;
> + struct hvm_iommu *hd = domain_hvm_iommu(d);
> + int flush_dev_iotlb;
> + int iommu_domid;
>
> - iommu_flush_cache_entry(pte, sizeof(struct dma_pte));
> + iommu_flush_cache_entry(pte, sizeof(struct dma_pte));
>
> - for_each_drhd_unit ( drhd )
> - {
> - iommu = drhd->iommu;
> - if ( !test_bit(iommu->index, &hd->iommu_bitmap) )
> - continue;
> + for_each_drhd_unit ( drhd )
> + {
> + iommu = drhd->iommu;
> + if ( !test_bit(iommu->index, &hd->iommu_bitmap) )
> + continue;
>
> - flush_dev_iotlb = find_ats_dev_drhd(iommu) ? 1 : 0;
> - iommu_domid= domain_iommu_domid(d, iommu);
> - if ( iommu_domid == -1 )
> - continue;
> - if ( iommu_flush_iotlb_psi(iommu, iommu_domid,
> - (paddr_t)gfn << PAGE_SHIFT_4K,
> - order, !present, flush_dev_iotlb) )
> - iommu_flush_write_buffer(iommu);
> + flush_dev_iotlb = find_ats_dev_drhd(iommu) ? 1 : 0;
> + iommu_domid= domain_iommu_domid(d, iommu);
> + if ( iommu_domid == -1 )
> + continue;
> + if ( iommu_flush_iotlb_psi(iommu, iommu_domid,
> + (paddr_t)gfn << PAGE_SHIFT_4K,
> + order, !present, flush_dev_iotlb) )
> + iommu_flush_write_buffer(iommu);
> + }
> }
> -}
What are these changes to indentation about? Are you
deliberately breaking common rules here, or is this some sort of
unintentional leftover?
Jan
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM
2014-02-10 8:07 ` Jan Beulich
@ 2014-02-10 11:42 ` Julien Grall
0 siblings, 0 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-10 11:42 UTC (permalink / raw)
To: Jan Beulich
Cc: ian.campbell, patches, tim, stefano.stabellini, xen-devel, Xiantao Zhang
Hello Jan,
On 10/02/14 08:07, Jan Beulich wrote:
>>>> On 07.02.14 at 18:43, Julien Grall <julien.grall@linaro.org> wrote:
>> DOM0 on ARM will have the same requirements as DOM0 PVH when iommu is
>> enabled.
>> Both PVH and ARM guest has paging mode translate enabled, so Xen can use it
>> to know if it needs to check the requirements.
>>
>> Rename the function and remove "pvh" word in the commit message.
>
> s/commit/panic/ ?
It's an error when I wrote the commit message. I will replace it on the
next version.
>>
>> Signed-off-by: Julien Grall <julien.grall@linaro.org>
>> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
>
> Other than the above,
> Acked-by: Jan Beulich <jbeulich@suse.com>
Thanks.
Regards,
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 08/12] xen/passthrough: iommu: Split generic IOMMU code
2014-02-10 8:22 ` Jan Beulich
@ 2014-02-10 11:52 ` Julien Grall
0 siblings, 0 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-10 11:52 UTC (permalink / raw)
To: Jan Beulich
Cc: ian.campbell, patches, tim, stefano.stabellini, xen-devel, Xiantao Zhang
Hello Jan,
Thanks for the review.
On 10/02/14 08:22, Jan Beulich wrote:
>>>> On 07.02.14 at 18:43, Julien Grall <julien.grall@linaro.org> wrote:
>> The generic IOMMU framework code (xen/drivers/passthrough/iommu.c) contains
>> functions specific to x86 and PCI.
>>
>> Split the framework in 3 distincts files:
>> - iommu.c: contains generic functions shared between x86 and ARM
>> (when it will be supported)
>> - iommu_pci.c: contains specific functions for PCI passthrough
>> - iommu_x86.c: contains specific functions for x86
>>
>> iommu_pci.c will be only compiled when PCI is supported by the architecture
>> (eg. HAS_PCI is defined).
>>
>> This patch is mostly code movement in new files.
>>
>> Signed-off-by: Julien Grall <julien.grall@linaro.org>
>> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
>> Cc: Jan Beulich <jbeulich@suse.com>
>> ---
>> xen/drivers/passthrough/Makefile | 6 +-
>> xen/drivers/passthrough/iommu.c | 473 +----------------------------------
>> xen/drivers/passthrough/iommu_pci.c | 468 ++++++++++++++++++++++++++++++++++
>
> There's xen/drivers/passthrough/pci.c already - any reason not to
> move the code there?
I didn't specific about moving the code directly in passthrough/pci.c. I
will do on the next version.
>
>> xen/drivers/passthrough/iommu_x86.c | 65 +++++
>
> Same here for xen/drivers/passthrough/x86/.
>
>> @@ -696,125 +344,6 @@ void iommu_crash_shutdown(void)
>> iommu_enabled = iommu_intremap = 0;
>> }
>>
>> -int iommu_do_domctl(
>> - struct xen_domctl *domctl, struct domain *d,
>> - XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)
>
> The function itself should probably not be moved out. Either the
> PCI-specific pieces of it should be made conditional, or a
> descendant function be created. Since (afaict) you'll need all of
> the domctl-s (with different arguments) too for pass-through on
> ARM - what's your plan for them?
PCI passthrough will be supported on ARM in the future. But we will also
have to support passthrough (via the device tree). I think, we will add
new DOMCTL for that. Didn't really think about that know.
I will add a descendant function to handle PCI DOMCTL.
>
>> --- a/xen/drivers/passthrough/vtd/iommu.c
>> +++ b/xen/drivers/passthrough/vtd/iommu.c
>> @@ -1784,31 +1784,31 @@ static int intel_iommu_unmap_page(struct domain *d,
>> unsigned long gfn)
>>
>> void iommu_pte_flush(struct domain *d, u64 gfn, u64 *pte,
>> int order, int present)
>> -{
>> - struct acpi_drhd_unit *drhd;
>> - struct iommu *iommu = NULL;
>> - struct hvm_iommu *hd = domain_hvm_iommu(d);
>> - int flush_dev_iotlb;
>> - int iommu_domid;
>> + {
>> + struct acpi_drhd_unit *drhd;
>> + struct iommu *iommu = NULL;
>> + struct hvm_iommu *hd = domain_hvm_iommu(d);
>> + int flush_dev_iotlb;
>> + int iommu_domid;
>>
>> - iommu_flush_cache_entry(pte, sizeof(struct dma_pte));
>> + iommu_flush_cache_entry(pte, sizeof(struct dma_pte));
>>
>> - for_each_drhd_unit ( drhd )
>> - {
>> - iommu = drhd->iommu;
>> - if ( !test_bit(iommu->index, &hd->iommu_bitmap) )
>> - continue;
>> + for_each_drhd_unit ( drhd )
>> + {
>> + iommu = drhd->iommu;
>> + if ( !test_bit(iommu->index, &hd->iommu_bitmap) )
>> + continue;
>>
>> - flush_dev_iotlb = find_ats_dev_drhd(iommu) ? 1 : 0;
>> - iommu_domid= domain_iommu_domid(d, iommu);
>> - if ( iommu_domid == -1 )
>> - continue;
>> - if ( iommu_flush_iotlb_psi(iommu, iommu_domid,
>> - (paddr_t)gfn << PAGE_SHIFT_4K,
>> - order, !present, flush_dev_iotlb) )
>> - iommu_flush_write_buffer(iommu);
>> + flush_dev_iotlb = find_ats_dev_drhd(iommu) ? 1 : 0;
>> + iommu_domid= domain_iommu_domid(d, iommu);
>> + if ( iommu_domid == -1 )
>> + continue;
>> + if ( iommu_flush_iotlb_psi(iommu, iommu_domid,
>> + (paddr_t)gfn << PAGE_SHIFT_4K,
>> + order, !present, flush_dev_iotlb) )
>> + iommu_flush_write_buffer(iommu);
>> + }
>> }
>> -}
>
> What are these changes to indentation about? Are you
> deliberately breaking common rules here, or is this some sort of
> unintentional leftover?
It's an error when I have created this patch. I will remove it.
Regards,
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM
2014-02-07 17:43 ` [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM Julien Grall
2014-02-10 8:07 ` Jan Beulich
@ 2014-02-10 16:10 ` Julien Grall
2014-02-10 16:35 ` Jan Beulich
1 sibling, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-10 16:10 UTC (permalink / raw)
To: Julien Grall
Cc: ian.campbell, patches, tim, stefano.stabellini, Jan Beulich,
xen-devel, Xiantao Zhang
On 02/07/2014 05:43 PM, Julien Grall wrote:
> DOM0 on ARM will have the same requirements as DOM0 PVH when iommu is enabled.
> Both PVH and ARM guest has paging mode translate enabled, so Xen can use it
> to know if it needs to check the requirements.
>
> Rename the function and remove "pvh" word in the commit message.
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
> Cc: Jan Beulich <jbeulich@suse.com>
> ---
> xen/drivers/passthrough/iommu.c | 14 +++++++++-----
> 1 file changed, 9 insertions(+), 5 deletions(-)
>
> diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
> index 19b0e23..26a5d91 100644
> --- a/xen/drivers/passthrough/iommu.c
> +++ b/xen/drivers/passthrough/iommu.c
> @@ -130,13 +130,18 @@ int iommu_domain_init(struct domain *d)
> return hd->platform_ops->init(d);
> }
>
> -static __init void check_dom0_pvh_reqs(struct domain *d)
> +static __init void check_dom0_reqs(struct domain *d)
> {
> + if ( !paging_mode_translate(d) )
> + return;
> +
> if ( !iommu_enabled )
> - panic("Presently, iommu must be enabled for pvh dom0\n");
> + panic("Presently, iommu must be enabled to use dom0 with translate "
> + "paging mode\n");
Hmmm... this change is wrong. I forgot that iommu doesn't exist on some
ARM platform (for instance Arndale).
Do we really this check for PVH? If yes, I will add replace the check
by: is_pvh_domain(d) && !iommu_enabled.
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM
2014-02-10 16:10 ` Julien Grall
@ 2014-02-10 16:35 ` Jan Beulich
2014-02-10 17:42 ` Julien Grall
0 siblings, 1 reply; 51+ messages in thread
From: Jan Beulich @ 2014-02-10 16:35 UTC (permalink / raw)
To: Julien Grall
Cc: ian.campbell, patches, tim, stefano.stabellini, xen-devel, Xiantao Zhang
>>> On 10.02.14 at 17:10, Julien Grall <julien.grall@linaro.org> wrote:
> On 02/07/2014 05:43 PM, Julien Grall wrote:
>> DOM0 on ARM will have the same requirements as DOM0 PVH when iommu is
> enabled.
>> Both PVH and ARM guest has paging mode translate enabled, so Xen can use it
>> to know if it needs to check the requirements.
>>
>> Rename the function and remove "pvh" word in the commit message.
>>
>> Signed-off-by: Julien Grall <julien.grall@linaro.org>
>> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
>> Cc: Jan Beulich <jbeulich@suse.com>
>> ---
>> xen/drivers/passthrough/iommu.c | 14 +++++++++-----
>> 1 file changed, 9 insertions(+), 5 deletions(-)
>>
>> diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
>> index 19b0e23..26a5d91 100644
>> --- a/xen/drivers/passthrough/iommu.c
>> +++ b/xen/drivers/passthrough/iommu.c
>> @@ -130,13 +130,18 @@ int iommu_domain_init(struct domain *d)
>> return hd->platform_ops->init(d);
>> }
>>
>> -static __init void check_dom0_pvh_reqs(struct domain *d)
>> +static __init void check_dom0_reqs(struct domain *d)
>> {
>> + if ( !paging_mode_translate(d) )
>> + return;
>> +
>> if ( !iommu_enabled )
>> - panic("Presently, iommu must be enabled for pvh dom0\n");
>> + panic("Presently, iommu must be enabled to use dom0 with translate
> "
>> + "paging mode\n");
>
> Hmmm... this change is wrong. I forgot that iommu doesn't exist on some
> ARM platform (for instance Arndale).
>
> Do we really this check for PVH? If yes, I will add replace the check
> by: is_pvh_domain(d) && !iommu_enabled.
Of course we need it: How would PVH Dom0 be able to do any kind
of DMA without an IOMMU?
Jan
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM
2014-02-10 16:35 ` Jan Beulich
@ 2014-02-10 17:42 ` Julien Grall
2014-02-11 7:47 ` Jan Beulich
0 siblings, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-10 17:42 UTC (permalink / raw)
To: Jan Beulich
Cc: ian.campbell, patches, tim, stefano.stabellini, xen-devel, Xiantao Zhang
On 02/10/2014 04:35 PM, Jan Beulich wrote:
>>>> On 10.02.14 at 17:10, Julien Grall <julien.grall@linaro.org> wrote:
>> On 02/07/2014 05:43 PM, Julien Grall wrote:
>>> DOM0 on ARM will have the same requirements as DOM0 PVH when iommu is
>> enabled.
>>> Both PVH and ARM guest has paging mode translate enabled, so Xen can use it
>>> to know if it needs to check the requirements.
>>>
>>> Rename the function and remove "pvh" word in the commit message.
>>>
>>> Signed-off-by: Julien Grall <julien.grall@linaro.org>
>>> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
>>> Cc: Jan Beulich <jbeulich@suse.com>
>>> ---
>>> xen/drivers/passthrough/iommu.c | 14 +++++++++-----
>>> 1 file changed, 9 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
>>> index 19b0e23..26a5d91 100644
>>> --- a/xen/drivers/passthrough/iommu.c
>>> +++ b/xen/drivers/passthrough/iommu.c
>>> @@ -130,13 +130,18 @@ int iommu_domain_init(struct domain *d)
>>> return hd->platform_ops->init(d);
>>> }
>>>
>>> -static __init void check_dom0_pvh_reqs(struct domain *d)
>>> +static __init void check_dom0_reqs(struct domain *d)
>>> {
>>> + if ( !paging_mode_translate(d) )
>>> + return;
>>> +
>>> if ( !iommu_enabled )
>>> - panic("Presently, iommu must be enabled for pvh dom0\n");
>>> + panic("Presently, iommu must be enabled to use dom0 with translate
>> "
>>> + "paging mode\n");
>>
>> Hmmm... this change is wrong. I forgot that iommu doesn't exist on some
>> ARM platform (for instance Arndale).
>>
>> Do we really this check for PVH? If yes, I will add replace the check
>> by: is_pvh_domain(d) && !iommu_enabled.
>
> Of course we need it: How would PVH Dom0 be able to do any kind
> of DMA without an IOMMU?
Right, on ARM we have the 1:1 memory mapping to avoid this issue.
I will fix it. Can I keep your ack on this patch?
> Jan
>
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM
2014-02-10 17:42 ` Julien Grall
@ 2014-02-11 7:47 ` Jan Beulich
0 siblings, 0 replies; 51+ messages in thread
From: Jan Beulich @ 2014-02-11 7:47 UTC (permalink / raw)
To: Julien Grall
Cc: ian.campbell, patches, tim, stefano.stabellini, xen-devel, Xiantao Zhang
>>> On 10.02.14 at 18:42, Julien Grall <julien.grall@linaro.org> wrote:
> On 02/10/2014 04:35 PM, Jan Beulich wrote:
>>>>> On 10.02.14 at 17:10, Julien Grall <julien.grall@linaro.org> wrote:
>>> On 02/07/2014 05:43 PM, Julien Grall wrote:
>>>> DOM0 on ARM will have the same requirements as DOM0 PVH when iommu is
>>> enabled.
>>>> Both PVH and ARM guest has paging mode translate enabled, so Xen can use it
>>>> to know if it needs to check the requirements.
>>>>
>>>> Rename the function and remove "pvh" word in the commit message.
>>>>
>>>> Signed-off-by: Julien Grall <julien.grall@linaro.org>
>>>> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
>>>> Cc: Jan Beulich <jbeulich@suse.com>
>>>> ---
>>>> xen/drivers/passthrough/iommu.c | 14 +++++++++-----
>>>> 1 file changed, 9 insertions(+), 5 deletions(-)
>>>>
>>>> diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
>>>> index 19b0e23..26a5d91 100644
>>>> --- a/xen/drivers/passthrough/iommu.c
>>>> +++ b/xen/drivers/passthrough/iommu.c
>>>> @@ -130,13 +130,18 @@ int iommu_domain_init(struct domain *d)
>>>> return hd->platform_ops->init(d);
>>>> }
>>>>
>>>> -static __init void check_dom0_pvh_reqs(struct domain *d)
>>>> +static __init void check_dom0_reqs(struct domain *d)
>>>> {
>>>> + if ( !paging_mode_translate(d) )
>>>> + return;
>>>> +
>>>> if ( !iommu_enabled )
>>>> - panic("Presently, iommu must be enabled for pvh dom0\n");
>>>> + panic("Presently, iommu must be enabled to use dom0 with translate
>>> "
>>>> + "paging mode\n");
>>>
>>> Hmmm... this change is wrong. I forgot that iommu doesn't exist on some
>>> ARM platform (for instance Arndale).
>>>
>>> Do we really this check for PVH? If yes, I will add replace the check
>>> by: is_pvh_domain(d) && !iommu_enabled.
>>
>> Of course we need it: How would PVH Dom0 be able to do any kind
>> of DMA without an IOMMU?
>
> Right, on ARM we have the 1:1 memory mapping to avoid this issue.
>
> I will fix it. Can I keep your ack on this patch?
For this simple an adjustment - sure.
Jan
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 01/12] xen/common: grant-table: only call IOMMU if paging mode translate is disabled
2014-02-07 17:43 ` [RFC for-4.5 01/12] xen/common: grant-table: only call IOMMU if paging mode translate is disabled Julien Grall
2014-02-10 8:02 ` Jan Beulich
@ 2014-02-19 12:25 ` Ian Campbell
1 sibling, 0 replies; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 12:25 UTC (permalink / raw)
To: Julien Grall; +Cc: xen-devel, Keir Fraser, tim, stefano.stabellini, patches
On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
> From Xen point of view, ARM guests are PV guest with paging auto translate
> enabled.
>
> When IOMMU support will be added for ARM, mapping grant ref will always crash
> Xen due to the BUG_ON in __gnttab_map_grant_ref.
>
> On x86:
> - PV guests always have paging mode translate disabled
> - PVH and HVM guests have always paging mode translate enabled
>
> It means that we can safely replace the check that the domain is a PV guests
> by checking if the guest has paging mode translate enabled.
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Keir Fraser <keir@xen.org>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 02/12] xen/passthrough: vtd: Don't export iommu_domain_teardown
2014-02-07 17:43 ` [RFC for-4.5 02/12] xen/passthrough: vtd: Don't export iommu_domain_teardown Julien Grall
@ 2014-02-19 12:26 ` Ian Campbell
0 siblings, 0 replies; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 12:26 UTC (permalink / raw)
To: Julien Grall; +Cc: xen-devel, tim, Xiantao Zhang, stefano.stabellini, patches
On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
> iommu_domain_teardown is only used internally in
> xen/drivers/passthrough/vtd/iommu.c
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
Not really my remit but:
Acked-by: Ian Campbell <ian.campbell@citrix.com>
> ---
> xen/drivers/passthrough/vtd/iommu.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/xen/drivers/passthrough/vtd/iommu.c b/xen/drivers/passthrough/vtd/iommu.c
> index 5f10034..a8d33fc 100644
> --- a/xen/drivers/passthrough/vtd/iommu.c
> +++ b/xen/drivers/passthrough/vtd/iommu.c
> @@ -1701,7 +1701,7 @@ static int reassign_device_ownership(
> return ret;
> }
>
> -void iommu_domain_teardown(struct domain *d)
> +static void iommu_domain_teardown(struct domain *d)
> {
> struct hvm_iommu *hd = domain_hvm_iommu(d);
>
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 03/12] xen/passthrough: vtd: Don't export iommu_set_pgd
2014-02-07 17:43 ` [RFC for-4.5 03/12] xen/passthrough: vtd: Don't export iommu_set_pgd Julien Grall
2014-02-10 8:04 ` Jan Beulich
@ 2014-02-19 12:27 ` Ian Campbell
1 sibling, 0 replies; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 12:27 UTC (permalink / raw)
To: Julien Grall
Cc: patches, tim, stefano.stabellini, Jan Beulich, xen-devel, Xiantoa Zhang
On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
> iommu_set_pgd is only used internally in
> xen/drivers/passthrough/vtd/iommu.c
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Xiantoa Zhang <xiantao.zhang@intel.com>
> Cc: Jan Beulich <jbeulich@suse.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
(could collapse with the previous patch into a "don't export unnecessary
stuff" patch)
> ---
> xen/drivers/passthrough/vtd/iommu.c | 2 +-
> xen/include/xen/iommu.h | 1 -
> 2 files changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/xen/drivers/passthrough/vtd/iommu.c b/xen/drivers/passthrough/vtd/iommu.c
> index a8d33fc..d5ce5b7 100644
> --- a/xen/drivers/passthrough/vtd/iommu.c
> +++ b/xen/drivers/passthrough/vtd/iommu.c
> @@ -1826,7 +1826,7 @@ static int vtd_ept_page_compatible(struct iommu *iommu)
> /*
> * set VT-d page table directory to EPT table if allowed
> */
> -void iommu_set_pgd(struct domain *d)
> +static void iommu_set_pgd(struct domain *d)
> {
> struct hvm_iommu *hd = domain_hvm_iommu(d);
> mfn_t pgd_mfn;
> diff --git a/xen/include/xen/iommu.h b/xen/include/xen/iommu.h
> index 8bb0a1d..fcbc432 100644
> --- a/xen/include/xen/iommu.h
> +++ b/xen/include/xen/iommu.h
> @@ -68,7 +68,6 @@ int iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,
> unsigned int flags);
> int iommu_unmap_page(struct domain *d, unsigned long gfn);
> void iommu_pte_flush(struct domain *d, u64 gfn, u64 *pte, int order, int present);
> -void iommu_set_pgd(struct domain *d);
> void iommu_domain_teardown(struct domain *d);
>
> void pt_pci_init(void);
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 04/12] xen/dts: Add dt_property_read_bool
2014-02-07 17:43 ` [RFC for-4.5 04/12] xen/dts: Add dt_property_read_bool Julien Grall
@ 2014-02-19 12:28 ` Ian Campbell
2014-02-19 14:57 ` Julien Grall
0 siblings, 1 reply; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 12:28 UTC (permalink / raw)
To: Julien Grall; +Cc: xen-devel, tim, stefano.stabellini, patches
On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
> The function check if property exists in a specific node.
^a
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 05/12] xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle
2014-02-07 17:43 ` [RFC for-4.5 05/12] xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle Julien Grall
@ 2014-02-19 12:34 ` Ian Campbell
2014-02-19 16:17 ` Julien Grall
0 siblings, 1 reply; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 12:34 UTC (permalink / raw)
To: Julien Grall; +Cc: xen-devel, tim, stefano.stabellini, patches
On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
> Code adapted from linux drivers/of/base.c (commit ef42c58).
On that basis I only took a cursory glance through that monster
function.
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> diff --git a/xen/include/xen/device_tree.h b/xen/include/xen/device_tree.h
> index 7c075d9..d429e60 100644
> --- a/xen/include/xen/device_tree.h
> +++ b/xen/include/xen/device_tree.h
> @@ -112,6 +112,13 @@ struct dt_device_node {
>
> };
>
> +#define MAX_PHANDLE_ARGS 16
> +struct dt_phandle_args {
> + struct dt_device_node *np;
> + int args_count;
> + uint32_t args[MAX_PHANDLE_ARGS];
> +};
> +
> /**
> * IRQ line type.
> *
> @@ -621,6 +628,53 @@ void dt_set_range(__be32 **cellp, const struct dt_device_node *np,
> void dt_get_range(const __be32 **cellp, const struct dt_device_node *np,
> u64 *address, u64 *size);
>
> +/**
> + * dt_parse_phandle - Resolve a phandle property to a device_node pointer
> + * @np: Pointer to device node holding phandle property
> + * @phandle_name: Name of property holding a phandle value
> + * @index: For properties holding a table of phandles, this is the index into
> + * the table
Otherwise it is -1 or something else?
> + *
> + * Returns the device_node pointer.
> + */
> +struct dt_device_node *dt_parse_phandle(const struct dt_device_node *np,
> + const char *phandle_name,
Stray hard tabs?
> + int index);
> +
> +/**
> + * dt_parse_phandle_with_args() - Find a node pointed by phandle in a list
> + * @np: pointer to a device tree node containing a list
> + * @list_name: property name that contains a list
> + * @cells_name: property name that specifies phandles' arguments count
> + * @index: index of a phandle to parse out
> + * @out_args: optional pointer to output arguments structure (will be filled)
> + *
> + * This function is useful to parse lists of phandles and their arguments.
> + * Returns 0 on success and fills out_args, on error returns appropriate
> + * errno value.
> + *
> + * Example:
> + *
> + * phandle1: node1 {
> + * #list-cells = <2>;
> + * }
> + *
> + * phandle2: node2 {
> + * #list-cells = <1>;
> + * }
> + *
> + * node3 {
> + * list = <&phandle1 1 2 &phandle2 3>;
> + * }
> + *
> + * To get a device_node of the `node2' node you may call this:
> + * dt_parse_phandle_with_args(node3, "list", "#list-cells", 1, &args);
Wow, what an exciting function!
How do I decide the correct size for out_args?
> + */
> +int dt_parse_phandle_with_args(const struct dt_device_node *np,
> + const char *list_name,
> + const char *cells_name, int index,
> + struct dt_phandle_args *out_args);
> +
> #endif /* __XEN_DEVICE_TREE_H */
>
> /*
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 07/12] xen/passthrough: iommu: Don't need to map dom0 page when the PT is shared
2014-02-07 17:43 ` [RFC for-4.5 07/12] xen/passthrough: iommu: Don't need to map dom0 page when the PT is shared Julien Grall
2014-02-10 8:11 ` Jan Beulich
@ 2014-02-19 12:37 ` Ian Campbell
1 sibling, 0 replies; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 12:37 UTC (permalink / raw)
To: Julien Grall
Cc: patches, tim, stefano.stabellini, Jan Beulich, xen-devel, Xiantao Zhang
On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
> Currently iommu_init_dom0 is browsing the page list and call map_page callback
> on each page.
>
> On both AMD and VTD drivers, the function will directly return if the page
> table is shared with the processor. So Xen can safely avoid to run through
> the page list.
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
> Cc: Jan Beulich <jbeulich@suse.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
> ---
> xen/drivers/passthrough/iommu.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
> index 26a5d91..0a26956 100644
> --- a/xen/drivers/passthrough/iommu.c
> +++ b/xen/drivers/passthrough/iommu.c
> @@ -157,7 +157,7 @@ void __init iommu_dom0_init(struct domain *d)
>
> register_keyhandler('o', &iommu_p2m_table);
> d->need_iommu = !!iommu_dom0_strict;
> - if ( need_iommu(d) )
> + if ( need_iommu(d) && !iommu_use_hap_pt(d) )
> {
> struct page_info *page;
> unsigned int i = 0;
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 08/12] xen/passthrough: iommu: Split generic IOMMU code
2014-02-07 17:43 ` [RFC for-4.5 08/12] xen/passthrough: iommu: Split generic IOMMU code Julien Grall
2014-02-10 8:22 ` Jan Beulich
@ 2014-02-19 12:39 ` Ian Campbell
2014-02-19 16:23 ` Julien Grall
1 sibling, 1 reply; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 12:39 UTC (permalink / raw)
To: Julien Grall
Cc: patches, tim, stefano.stabellini, Jan Beulich, xen-devel, Xiantao Zhang
On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
> The generic IOMMU framework code (xen/drivers/passthrough/iommu.c) contains
> functions specific to x86 and PCI.
>
> Split the framework in 3 distincts files:
> - iommu.c: contains generic functions shared between x86 and ARM
> (when it will be supported)
> - iommu_pci.c: contains specific functions for PCI passthrough
> - iommu_x86.c: contains specific functions for x86
>
> iommu_pci.c will be only compiled when PCI is supported by the architecture
> (eg. HAS_PCI is defined).
>
> This patch is mostly code movement in new files.
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
> Cc: Jan Beulich <jbeulich@suse.com>
> ---
> xen/drivers/passthrough/Makefile | 6 +-
> xen/drivers/passthrough/iommu.c | 473 +----------------------------------
> xen/drivers/passthrough/iommu_pci.c | 468 ++++++++++++++++++++++++++++++++++
> xen/drivers/passthrough/iommu_x86.c | 65 +++++
> xen/drivers/passthrough/vtd/iommu.c | 42 ++--
> xen/include/asm-x86/iommu.h | 46 ++++
> xen/include/xen/hvm/iommu.h | 1 +
> xen/include/xen/iommu.h | 42 ++--
> 8 files changed, 625 insertions(+), 518 deletions(-)
> create mode 100644 xen/drivers/passthrough/iommu_pci.c
> create mode 100644 xen/drivers/passthrough/iommu_x86.c
> create mode 100644 xen/include/asm-x86/iommu.h
>
> diff --git a/xen/drivers/passthrough/Makefile b/xen/drivers/passthrough/Makefile
> index 7c40fa5..51e0a0d 100644
> --- a/xen/drivers/passthrough/Makefile
> +++ b/xen/drivers/passthrough/Makefile
> @@ -3,5 +3,7 @@ subdir-$(x86) += amd
> subdir-$(x86_64) += x86
>
> obj-y += iommu.o
> -obj-y += io.o
> -obj-y += pci.o
> +obj-$(x86) += iommu_x86.o
> +obj-$(HAS_PCI) += iommu_pci.o
> +obj-$(x86) += io.o
io.[co] not mentioned in the description?
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 09/12] xen/passthrough: iommu: Introduce arch specific code
2014-02-07 17:43 ` [RFC for-4.5 09/12] xen/passthrough: iommu: Introduce arch specific code Julien Grall
@ 2014-02-19 12:40 ` Ian Campbell
2014-02-19 16:25 ` Julien Grall
0 siblings, 1 reply; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 12:40 UTC (permalink / raw)
To: Julien Grall
Cc: Keir Fraser, Suravee Suthikulpanit, patches, Shane Wang,
Joseph Cihula, tim, stefano.stabellini, Jan Beulich, xen-devel,
Gang Wei, Xiantao Zhang
On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
> Currently the structure hvm_iommu (xen/include/xen/hvm/iommu.h) contains
> x86 specific fields.
>
> This patch creates:
> - arch_hvm_iommu structure which will contain architecture depend
> fields
> - arch_iommu_domain_{init,destroy} function to execute arch
> specific during domain creation/destruction
>
> Also move iommu_use_hap_pt and domain_hvm_iommu in asm-x86/iommu.h.
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Keir Fraser <keir@xen.org>
> Cc: Jan Beulich <jbeulich@suse.com>
> Cc: Joseph Cihula <joseph.cihula@intel.com>
> Cc: Gang Wei <gang.wei@intel.com>
> Cc: Shane Wang <shane.wang@intel.com>
> Cc: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
Mostly mechanical I guess?
Acked-by: Ian Campbell <ian.campbell@citrix.com>
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 10/12] xen/passthrough: Introduce IOMMU ARM architure
2014-02-07 17:43 ` [RFC for-4.5 10/12] xen/passthrough: Introduce IOMMU ARM architure Julien Grall
@ 2014-02-19 12:49 ` Ian Campbell
2014-02-19 16:50 ` Julien Grall
0 siblings, 1 reply; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 12:49 UTC (permalink / raw)
To: Julien Grall
Cc: patches, tim, stefano.stabellini, Jan Beulich, xen-devel, Xiantao Zhang
On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
> This patch contains the architecture to use IOMMU on ARM. There is no
s/IOMMU/IOMMUs/
> IOMMU drivers on this patch.
>
> The code will run through the device tree and will initialize every IOMMUs.
s/IOMMUs/IOMMU/
> It's possible to have multiple IOMMUs on the same platform, but they should
s/should/must/ I think for now?
> be handled with the same drivers.
s/drivers/driver/
> For now, there is no support for using multiple iommu drivers at runtime.
But this is in principal possible in the future, right? (I shudder to
think what the designers of an SoC with multiple different SMMUs on it
would have to be smoking...)
You should mention somewhere that on ARM these are called SMMUs not
IOMMUs.
> Each new IOMMU drivers should contain:
>
> static const char * const myiommu_dt_compat[] __initcontst =
"__initconst".
> {
> /* list of device compatible with the drivers. Will be matched with
> * the "compatible" property on the device tree
> */
> NULL,
> };
>
> DT_DEVICE_START(myiommu, "MY IOMMU", DEVICE_IOMMU)
> .compatible = myiommu_compatible,
> .init = myiommu_init,
> DT_DEVICE_END
This is the same as for any other driver, right?
> @@ -568,6 +571,10 @@ fail:
>
> void arch_domain_destroy(struct domain *d)
> {
> + /* IOMMU page table is shared with P2M, always call
> + * iommu_domain_destroy() before p2m_teardown().
It would be worth adding some commentary on the design we are using
(decisions such as whether to share PTs with the stage 2 MMU) in the
commit message as well.
I suppose this requirement puts constraints on the SMMU hardware we can
support. I think it is fine to live with that until someone shows up
with an SMMU with pagetables that are incompatible with the stage 2
paging ones.
> diff --git a/xen/arch/arm/setup.c b/xen/arch/arm/setup.c
> index 1f6d713..5a687d1 100644
> --- a/xen/arch/arm/setup.c
> +++ b/xen/arch/arm/setup.c
> @@ -725,6 +725,8 @@ void __init start_xen(unsigned long boot_phys_offset,
> local_irq_enable();
> local_abort_enable();
>
> + iommu_setup(); /* setup iommu if available */
Comment is a bit redundant ;-)
> diff --git a/xen/drivers/passthrough/arm/iommu.c b/xen/drivers/passthrough/arm/iommu.c
> new file mode 100644
> index 0000000..7cf36cd
> --- /dev/null
> +++ b/xen/drivers/passthrough/arm/iommu.c
> @@ -0,0 +1,65 @@
> +/*
> + * xen/drivers/passthrough/arm/iommu.c
> + *
> + * Generic IOMMU framework via the device tree
> + *
> + * Julien Grall <julien.grall@linaro.org>
> + * Copyright (c) 2013 Linaro Limited.
It's not 2013 any more...
> diff --git a/xen/include/asm-arm/hvm/iommu.h b/xen/include/asm-arm/hvm/iommu.h
> new file mode 100644
> index 0000000..461c8cf
> --- /dev/null
> +++ b/xen/include/asm-arm/hvm/iommu.h
> @@ -0,0 +1,10 @@
> +#ifndef __ASM_ARM_HVM_IOMMU_H_
> +#define __ASM_ARM_HVM_IOMMU_H_
> +
> +struct arch_hvm_iommu
> +{
> + /* Private information for the IOMMU drivers */
> + void *priv;
> +};
> +
> +#endif /* __ASM_ARM_HVM_IOMMU_H_ */
Emacs magic block please.
That was a surprisingly small and uncontroversial patch. Well done.
Ian.
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 11/12] MAINTAINERS: Add drivers/passthrough/arm
2014-02-07 17:43 ` [RFC for-4.5 11/12] MAINTAINERS: Add drivers/passthrough/arm Julien Grall
@ 2014-02-19 12:49 ` Ian Campbell
0 siblings, 0 replies; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 12:49 UTC (permalink / raw)
To: Julien Grall; +Cc: xen-devel, Keir Fraser, tim, stefano.stabellini, patches
On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
> Add the ARM IOMMU directory to "ARM ARCHITECTURE" part
>
> Signed-off-by: Julien Grall <julien.grall@linaro.org>
> Cc: Keir Fraser <keir@xen.org>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
> ---
> MAINTAINERS | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/MAINTAINERS b/MAINTAINERS
> index 7757cdd..ad6c8a9 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -130,6 +130,7 @@ S: Supported
> L: xen-devel@lists.xen.org
> F: xen/arch/arm/
> F: xen/include/asm-arm/
> +F: xen/drivers/passthrough/arm
>
> CPU POOLS
> M: Juergen Gross <juergen.gross@ts.fujitsu.com>
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers
2014-02-07 17:43 ` [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers Julien Grall
@ 2014-02-19 14:15 ` Ian Campbell
2014-02-19 17:21 ` Julien Grall
2014-02-19 17:23 ` Julien Grall
0 siblings, 2 replies; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 14:15 UTC (permalink / raw)
To: Julien Grall
Cc: patches, tim, stefano.stabellini, Jan Beulich, xen-devel, Xiantao Zhang
On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
> This patch add support for ARM architected SMMU driver. It's based on the
> linux drivers (drivers/iommu/arm-smmu) commit 89ac23cd.
Aha, here comes all the hard stuff ;-)
Could you try and briefly enumerate the areas which you had to change
please.
Some comments on e.g. which translation context type we are using and
how we are configuring things etc might also be helpful here in
understanding what is going on.
Also could you give details of the test setup you used, was it just
booting dom0 on Midway with these patches? Was the DTB complete etc?
> + * This driver currently supports:
> + * - SMMUv1 and v2 implementations (didn't try v2 SMMU)
I guess this is Will's original comment, I thought SMMU-400, which
you've tried, was v2?
> + * - Stream-matching and stream-indexing
> + * - v7/v8 long-descriptor format
> + * - Non-secure access to the SMMU
> + * - 4k pages, p2m shared with the processor
> + * - Up to 40-bit addressing
> + * - Context fault reporting
> + */
> +
> +#include <xen/config.h>
> +#include <xen/delay.h>
> +#include <xen/errno.h>
> +#include <xen/irq.h>
> +#include <xen/lib.h>
> +#include <xen/list.h>
> +#include <xen/mm.h>
> +#include <xen/rbtree.h>
> +#include <xen/sched.h>
> +#include <asm/atomic.h>
> +#include <asm/device.h>
> +#include <asm/io.h>
> +#include <asm/platform.h>
> +
> +#define SZ_4K (1 << 12)
> +#define SZ_64K (1 << 16)
> +
> +/* Driver options */
> +#define SMMU_OPT_SECURE_CONFIG_ACCESS (1 << 0)
Is this just retained to reduce the deviation from the Linux driver?
It's no use to us I think. (I suppose that goes for a bunch of other
stuff, eg.. the PGSZ_4K stuff, which I will avoid commenting on
further).
> +
> +void arm_smmu_iommu_dom0_init(struct domain *d)
> +{
> + struct arm_smmu_device *smmu;
> + struct rb_node *node;
> +
> + printk(XENLOG_DEBUG "arm-smmu: Initialize dom0\n");
> +
> + list_for_each_entry( smmu, &arm_smmu_devices, list )
> + {
> + for ( node = rb_first(&smmu->masters); node; node = rb_next(node) )
> + {
> + struct arm_smmu_master *master;
> +
> + master = container_of(node, struct arm_smmu_master, node);
> +
> + if ( dt_device_used_by(master->dt_node) == DOMID_XEN ||
> + platform_device_is_blacklisted(master->dt_node) )
> + continue;
> +
> + arm_smmu_attach_dev(d, master->dt_node);
Should this not be driven from the same loop as the MMIO mapping setup
in dom0 build? Otherwise isn't there a danger that they won't
correspond?
A "master" here is a "bus master" i.e. a device, right?
> + /* Even if the device can't be initialized, we don't want to give to
> + * dom0 the smmu device
"give the smmu device to dom0"
I obviously haven't reviewed all this code in detail, but I have skimmed
it and nothing leapt out.
Ian.
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 04/12] xen/dts: Add dt_property_read_bool
2014-02-19 12:28 ` Ian Campbell
@ 2014-02-19 14:57 ` Julien Grall
0 siblings, 0 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-19 14:57 UTC (permalink / raw)
To: Ian Campbell; +Cc: xen-devel, tim, stefano.stabellini, patches
Hi Ian,
On 02/19/2014 12:28 PM, Ian Campbell wrote:
> On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
>> The function check if property exists in a specific node.
>
> ^a
I will fix it.
>>
>> Signed-off-by: Julien Grall <julien.grall@linaro.org>
>
> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Thanks,
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 05/12] xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle
2014-02-19 12:34 ` Ian Campbell
@ 2014-02-19 16:17 ` Julien Grall
2014-02-19 16:26 ` Ian Campbell
0 siblings, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-19 16:17 UTC (permalink / raw)
To: Ian Campbell; +Cc: xen-devel, tim, stefano.stabellini, patches
Hello Ian,
On 02/19/2014 12:34 PM, Ian Campbell wrote:
> On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
>> Code adapted from linux drivers/of/base.c (commit ef42c58).
>
> On that basis I only took a cursory glance through that monster
> function.
>
>>
>> Signed-off-by: Julien Grall <julien.grall@linaro.org>
>> diff --git a/xen/include/xen/device_tree.h b/xen/include/xen/device_tree.h
>> index 7c075d9..d429e60 100644
>> --- a/xen/include/xen/device_tree.h
>> +++ b/xen/include/xen/device_tree.h
>> @@ -112,6 +112,13 @@ struct dt_device_node {
>>
>> };
>>
>> +#define MAX_PHANDLE_ARGS 16
>> +struct dt_phandle_args {
>> + struct dt_device_node *np;
>> + int args_count;
>> + uint32_t args[MAX_PHANDLE_ARGS];
>> +};
>> +
>> /**
>> * IRQ line type.
>> *
>> @@ -621,6 +628,53 @@ void dt_set_range(__be32 **cellp, const struct dt_device_node *np,
>> void dt_get_range(const __be32 **cellp, const struct dt_device_node *np,
>> u64 *address, u64 *size);
>>
>> +/**
>> + * dt_parse_phandle - Resolve a phandle property to a device_node pointer
>> + * @np: Pointer to device node holding phandle property
>> + * @phandle_name: Name of property holding a phandle value
>> + * @index: For properties holding a table of phandles, this is the index into
>> + * the table
>
> Otherwise it is -1 or something else?
Even with one element, the property contains a table of phandles. So,
the first index is always 0.
>
>> + *
>> + * Returns the device_node pointer.
>> + */
>> +struct dt_device_node *dt_parse_phandle(const struct dt_device_node *np,
>> + const char *phandle_name,
>
> Stray hard tabs?
Oops, I forgot to check the hard tabs on this function. I will fix it.
>> + int index);
>> +
>> +/**
>> + * dt_parse_phandle_with_args() - Find a node pointed by phandle in a list
>> + * @np: pointer to a device tree node containing a list
>> + * @list_name: property name that contains a list
>> + * @cells_name: property name that specifies phandles' arguments count
>> + * @index: index of a phandle to parse out
>> + * @out_args: optional pointer to output arguments structure (will be filled)
>> + *
>> + * This function is useful to parse lists of phandles and their arguments.
>> + * Returns 0 on success and fills out_args, on error returns appropriate
>> + * errno value.
>> + *
>> + * Example:
>> + *
>> + * phandle1: node1 {
>> + * #list-cells = <2>;
>> + * }
>> + *
>> + * phandle2: node2 {
>> + * #list-cells = <1>;
>> + * }
>> + *
>> + * node3 {
>> + * list = <&phandle1 1 2 &phandle2 3>;
>> + * }
>> + *
>> + * To get a device_node of the `node2' node you may call this:
>> + * dt_parse_phandle_with_args(node3, "list", "#list-cells", 1, &args);
>
> Wow, what an exciting function!
>
> How do I decide the correct size for out_args?
out_args is a structure dt_phandle_args which contains the node and an
array of arguments.
The function is designed to parse one phandle by one phandle. So
out_args is either NULL (we don't want to get the info) or contains a
pointer to the structure to fill.
The description of out_args seems clear to me. Does it miss something?
Cheers,
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 08/12] xen/passthrough: iommu: Split generic IOMMU code
2014-02-19 12:39 ` Ian Campbell
@ 2014-02-19 16:23 ` Julien Grall
0 siblings, 0 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-19 16:23 UTC (permalink / raw)
To: Ian Campbell
Cc: patches, tim, stefano.stabellini, Jan Beulich, xen-devel, Xiantao Zhang
Hi Ian,
On 02/19/2014 12:39 PM, Ian Campbell wrote:
> On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
>> The generic IOMMU framework code (xen/drivers/passthrough/iommu.c) contains
>> functions specific to x86 and PCI.
>>
>> Split the framework in 3 distincts files:
>> - iommu.c: contains generic functions shared between x86 and ARM
>> (when it will be supported)
>> - iommu_pci.c: contains specific functions for PCI passthrough
>> - iommu_x86.c: contains specific functions for x86
>>
>> iommu_pci.c will be only compiled when PCI is supported by the architecture
>> (eg. HAS_PCI is defined).
>>
>> This patch is mostly code movement in new files.
>>
>> Signed-off-by: Julien Grall <julien.grall@linaro.org>
>> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
>> Cc: Jan Beulich <jbeulich@suse.com>
>> ---
>> xen/drivers/passthrough/Makefile | 6 +-
>> xen/drivers/passthrough/iommu.c | 473 +----------------------------------
>> xen/drivers/passthrough/iommu_pci.c | 468 ++++++++++++++++++++++++++++++++++
>> xen/drivers/passthrough/iommu_x86.c | 65 +++++
>> xen/drivers/passthrough/vtd/iommu.c | 42 ++--
>> xen/include/asm-x86/iommu.h | 46 ++++
>> xen/include/xen/hvm/iommu.h | 1 +
>> xen/include/xen/iommu.h | 42 ++--
>> 8 files changed, 625 insertions(+), 518 deletions(-)
>> create mode 100644 xen/drivers/passthrough/iommu_pci.c
>> create mode 100644 xen/drivers/passthrough/iommu_x86.c
>> create mode 100644 xen/include/asm-x86/iommu.h
>>
>> diff --git a/xen/drivers/passthrough/Makefile b/xen/drivers/passthrough/Makefile
>> index 7c40fa5..51e0a0d 100644
>> --- a/xen/drivers/passthrough/Makefile
>> +++ b/xen/drivers/passthrough/Makefile
>> @@ -3,5 +3,7 @@ subdir-$(x86) += amd
>> subdir-$(x86_64) += x86
>>
>> obj-y += iommu.o
>> -obj-y += io.o
>> -obj-y += pci.o
>> +obj-$(x86) += iommu_x86.o
>> +obj-$(HAS_PCI) += iommu_pci.o
>> +obj-$(x86) += io.o
>
> io.[co] not mentioned in the description?
Right, I will udpate the commit message.
Cheers,
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 09/12] xen/passthrough: iommu: Introduce arch specific code
2014-02-19 12:40 ` Ian Campbell
@ 2014-02-19 16:25 ` Julien Grall
0 siblings, 0 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-19 16:25 UTC (permalink / raw)
To: Ian Campbell
Cc: Keir Fraser, Suravee Suthikulpanit, patches, Shane Wang,
Joseph Cihula, tim, stefano.stabellini, Jan Beulich, xen-devel,
Gang Wei, Xiantao Zhang
Hi Ian,
On 02/19/2014 12:40 PM, Ian Campbell wrote:
> On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
>> Currently the structure hvm_iommu (xen/include/xen/hvm/iommu.h) contains
>> x86 specific fields.
>>
>> This patch creates:
>> - arch_hvm_iommu structure which will contain architecture depend
>> fields
>> - arch_iommu_domain_{init,destroy} function to execute arch
>> specific during domain creation/destruction
>>
>> Also move iommu_use_hap_pt and domain_hvm_iommu in asm-x86/iommu.h.
>>
>> Signed-off-by: Julien Grall <julien.grall@linaro.org>
>> Cc: Keir Fraser <keir@xen.org>
>> Cc: Jan Beulich <jbeulich@suse.com>
>> Cc: Joseph Cihula <joseph.cihula@intel.com>
>> Cc: Gang Wei <gang.wei@intel.com>
>> Cc: Shane Wang <shane.wang@intel.com>
>> Cc: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
>> Cc: Xiantao Zhang <xiantao.zhang@intel.com>
>
> Mostly mechanical I guess?
Yes. Mostly replace hd->field by hd->arch.field.
> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Thanks,
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 05/12] xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle
2014-02-19 16:17 ` Julien Grall
@ 2014-02-19 16:26 ` Ian Campbell
2014-02-19 16:52 ` Julien Grall
0 siblings, 1 reply; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 16:26 UTC (permalink / raw)
To: Julien Grall; +Cc: xen-devel, tim, stefano.stabellini, patches
On Wed, 2014-02-19 at 16:17 +0000, Julien Grall wrote:
> >> +/**
> >> + * dt_parse_phandle - Resolve a phandle property to a device_node pointer
> >> + * @np: Pointer to device node holding phandle property
> >> + * @phandle_name: Name of property holding a phandle value
> >> + * @index: For properties holding a table of phandles, this is the index into
> >> + * the table
> >
> > Otherwise it is -1 or something else?
>
> Even with one element, the property contains a table of phandles. So,
> the first index is always 0.
OK, so the "For a properties holding a table..." bit is misleading
(that's the "otherwise" I was referring too, e..g properties which do
not hold a table0, since this is always the index, even if the table has
a single element.
> > How do I decide the correct size for out_args?
>
> out_args is a structure dt_phandle_args which contains the node and an
> array of arguments.
>
> The function is designed to parse one phandle by one phandle. So
> out_args is either NULL (we don't want to get the info) or contains a
> pointer to the structure to fill.
>
> The description of out_args seems clear to me. Does it miss something?
No, I didn't realise that the array was within dt_phandle_args rather
than out_args itself being a pointer to an array (the plural name is a
bit confusing in that regard).
I see the name comes from Linux, so no need to change.
Ian.
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 10/12] xen/passthrough: Introduce IOMMU ARM architure
2014-02-19 12:49 ` Ian Campbell
@ 2014-02-19 16:50 ` Julien Grall
0 siblings, 0 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-19 16:50 UTC (permalink / raw)
To: Ian Campbell
Cc: patches, tim, stefano.stabellini, Jan Beulich, xen-devel, Xiantao Zhang
Hi Ian,
On 02/19/2014 12:49 PM, Ian Campbell wrote:
> On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
>> This patch contains the architecture to use IOMMU on ARM. There is no
>
> s/IOMMU/IOMMUs/
>
>> IOMMU drivers on this patch.
>>
>> The code will run through the device tree and will initialize every IOMMUs.
>
> s/IOMMUs/IOMMU/
>
>> It's possible to have multiple IOMMUs on the same platform, but they should
>
> s/should/must/ I think for now?
Right.
>> be handled with the same drivers.
>
> s/drivers/driver/
>
>> For now, there is no support for using multiple iommu drivers at runtime.
>
> But this is in principal possible in the future, right? (I shudder to
> think what the designers of an SoC with multiple different SMMUs on it
> would have to be smoking...)
In theory yes. I will wait that someone want to support a such platfomr
on Xen before implement it :).
>
> You should mention somewhere that on ARM these are called SMMUs not
> IOMMUs.
I think SMMU is the name used for ARM IOMMU. Samsung and TI doesn't use
the term SMMU.
>
>> Each new IOMMU drivers should contain:
>>
>> static const char * const myiommu_dt_compat[] __initcontst =
>
> "__initconst".
Oops. Will fix it.
>> {
>> /* list of device compatible with the drivers. Will be matched with
>> * the "compatible" property on the device tree
>> */
>> NULL,
>> };
>>
>> DT_DEVICE_START(myiommu, "MY IOMMU", DEVICE_IOMMU)
>> .compatible = myiommu_compatible,
>> .init = myiommu_init,
>> DT_DEVICE_END
>
> This is the same as for any other driver, right?
Yes, the only change is we need to specify DEVICE_IOMMU in DT_DEVICE_START.
>
>> @@ -568,6 +571,10 @@ fail:
>>
>> void arch_domain_destroy(struct domain *d)
>> {
>> + /* IOMMU page table is shared with P2M, always call
>> + * iommu_domain_destroy() before p2m_teardown().
>
> It would be worth adding some commentary on the design we are using
> (decisions such as whether to share PTs with the stage 2 MMU) in the
> commit message as well.
I will do.
> I suppose this requirement puts constraints on the SMMU hardware we can
> support. I think it is fine to live with that until someone shows up
> with an SMMU with pagetables that are incompatible with the stage 2
> paging ones.
Adding support for a such hardware should not need too much code.
>> diff --git a/xen/arch/arm/setup.c b/xen/arch/arm/setup.c
>> index 1f6d713..5a687d1 100644
>> --- a/xen/arch/arm/setup.c
>> +++ b/xen/arch/arm/setup.c
>> @@ -725,6 +725,8 @@ void __init start_xen(unsigned long boot_phys_offset,
>> local_irq_enable();
>> local_abort_enable();
>>
>> + iommu_setup(); /* setup iommu if available */
>
> Comment is a bit redundant ;-)
Copied for x86/setup.c. I will remove it.
>
>> diff --git a/xen/drivers/passthrough/arm/iommu.c b/xen/drivers/passthrough/arm/iommu.c
>> new file mode 100644
>> index 0000000..7cf36cd
>> --- /dev/null
>> +++ b/xen/drivers/passthrough/arm/iommu.c
>> @@ -0,0 +1,65 @@
>> +/*
>> + * xen/drivers/passthrough/arm/iommu.c
>> + *
>> + * Generic IOMMU framework via the device tree
>> + *
>> + * Julien Grall <julien.grall@linaro.org>
>> + * Copyright (c) 2013 Linaro Limited.
>
> It's not 2013 any more...
I have started the implementation in 2013 and forgot to update it.
>
>> diff --git a/xen/include/asm-arm/hvm/iommu.h b/xen/include/asm-arm/hvm/iommu.h
>> new file mode 100644
>> index 0000000..461c8cf
>> --- /dev/null
>> +++ b/xen/include/asm-arm/hvm/iommu.h
>> @@ -0,0 +1,10 @@
>> +#ifndef __ASM_ARM_HVM_IOMMU_H_
>> +#define __ASM_ARM_HVM_IOMMU_H_
>> +
>> +struct arch_hvm_iommu
>> +{
>> + /* Private information for the IOMMU drivers */
>> + void *priv;
>> +};
>> +
>> +#endif /* __ASM_ARM_HVM_IOMMU_H_ */
>
> Emacs magic block please.
I will do.
> That was a surprisingly small and uncontroversial patch. Well done.
The IOMMU framework in Xen was well designed.
cheers,
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 05/12] xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle
2014-02-19 16:26 ` Ian Campbell
@ 2014-02-19 16:52 ` Julien Grall
2014-02-19 16:57 ` Ian Campbell
0 siblings, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-19 16:52 UTC (permalink / raw)
To: Ian Campbell; +Cc: xen-devel, tim, stefano.stabellini, patches
On 02/19/2014 04:26 PM, Ian Campbell wrote:
> On Wed, 2014-02-19 at 16:17 +0000, Julien Grall wrote:
>>>> +/**
>>>> + * dt_parse_phandle - Resolve a phandle property to a device_node pointer
>>>> + * @np: Pointer to device node holding phandle property
>>>> + * @phandle_name: Name of property holding a phandle value
>>>> + * @index: For properties holding a table of phandles, this is the index into
>>>> + * the table
>>>
>>> Otherwise it is -1 or something else?
>>
>> Even with one element, the property contains a table of phandles. So,
>> the first index is always 0.
>
> OK, so the "For a properties holding a table..." bit is misleading
> (that's the "otherwise" I was referring too, e..g properties which do
> not hold a table0, since this is always the index, even if the table has
> a single element.
It was copied from Linux. Shall I update the comment?
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 05/12] xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle
2014-02-19 16:52 ` Julien Grall
@ 2014-02-19 16:57 ` Ian Campbell
0 siblings, 0 replies; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 16:57 UTC (permalink / raw)
To: Julien Grall; +Cc: xen-devel, tim, stefano.stabellini, patches
On Wed, 2014-02-19 at 16:52 +0000, Julien Grall wrote:
> On 02/19/2014 04:26 PM, Ian Campbell wrote:
> > On Wed, 2014-02-19 at 16:17 +0000, Julien Grall wrote:
> >>>> +/**
> >>>> + * dt_parse_phandle - Resolve a phandle property to a device_node pointer
> >>>> + * @np: Pointer to device node holding phandle property
> >>>> + * @phandle_name: Name of property holding a phandle value
> >>>> + * @index: For properties holding a table of phandles, this is the index into
> >>>> + * the table
> >>>
> >>> Otherwise it is -1 or something else?
> >>
> >> Even with one element, the property contains a table of phandles. So,
> >> the first index is always 0.
> >
> > OK, so the "For a properties holding a table..." bit is misleading
> > (that's the "otherwise" I was referring too, e..g properties which do
> > not hold a table0, since this is always the index, even if the table has
> > a single element.
>
> It was copied from Linux. Shall I update the comment?
I suppose it is better to not deviate more than necessary, so feel free
to leave it.
Ian.
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers
2014-02-19 14:15 ` Ian Campbell
@ 2014-02-19 17:21 ` Julien Grall
2014-02-19 17:27 ` Ian Campbell
2014-02-19 17:23 ` Julien Grall
1 sibling, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-19 17:21 UTC (permalink / raw)
To: Ian Campbell
Cc: patches, tim, stefano.stabellini, Jan Beulich, xen-devel, Xiantao Zhang
On 02/19/2014 02:15 PM, Ian Campbell wrote:
> On Fri, 2014-02-07 at 17:43 +0000, Julien Grall wrote:
>> This patch add support for ARM architected SMMU driver. It's based on the
>> linux drivers (drivers/iommu/arm-smmu) commit 89ac23cd.
>
> Aha, here comes all the hard stuff ;-)
>
> Could you try and briefly enumerate the areas which you had to change
> please.
The main changes are:
- Fault by default if the SMMU is enabled to translate an
address (Linux is bypassing the SMMU)
- Using P2M page table instead of creating new one
- Dropped stage-1 support
- Dropped chained SMMUs support for now
- Reworking device assignment and the structure
> Some comments on e.g. which translation context type we are using and
> how we are configuring things etc might also be helpful here in
> understanding what is going on.
Honestly, the configuration part was mostly copied from Linux. Some bits
are still fuzzy for me. I will try to improve the commit message.
> Also could you give details of the test setup you used, was it just
> booting dom0 on Midway with these patches? Was the DTB complete etc?
I had to modify the device tree by applying this following patch:
http://www.spinics.net/lists/arm-kernel/msg301163.html
I've tried to boot dom0 and a guest with LVM (a patch is coming to
remove swiotlb when the device is protected by an IOMMU).
>
>> + * This driver currently supports:
>> + * - SMMUv1 and v2 implementations (didn't try v2 SMMU)
>
> I guess this is Will's original comment, I thought SMMU-400, which
> you've tried, was v2?
No it's SMMU v2. As I understand:
- SMMU-400 => SMMU v1
- SMMU-500 => SMMU v2
The words in parenthesis was added by me because I don't have a test box
with SMMU v2.
>
>> + * - Stream-matching and stream-indexing
>> + * - v7/v8 long-descriptor format
>> + * - Non-secure access to the SMMU
>> + * - 4k pages, p2m shared with the processor
>> + * - Up to 40-bit addressing
>> + * - Context fault reporting
>> + */
>> +
>> +#include <xen/config.h>
>> +#include <xen/delay.h>
>> +#include <xen/errno.h>
>> +#include <xen/irq.h>
>> +#include <xen/lib.h>
>> +#include <xen/list.h>
>> +#include <xen/mm.h>
>> +#include <xen/rbtree.h>
>> +#include <xen/sched.h>
>> +#include <asm/atomic.h>
>> +#include <asm/device.h>
>> +#include <asm/io.h>
>> +#include <asm/platform.h>
>> +
>> +#define SZ_4K (1 << 12)
>> +#define SZ_64K (1 << 16)
>> +
>> +/* Driver options */
>> +#define SMMU_OPT_SECURE_CONFIG_ACCESS (1 << 0)
>
> Is this just retained to reduce the deviation from the Linux driver?
> It's no use to us I think. (I suppose that goes for a bunch of other
> stuff, eg.. the PGSZ_4K stuff, which I will avoid commenting on
> further).
SZ_4K and SZ_64K is used later in the code. The
SMMU_OPT_SECURE_CONFIG_ACCESS is used for midway as the SMMU is broken.
>
>> +
>> +void arm_smmu_iommu_dom0_init(struct domain *d)
>> +{
>> + struct arm_smmu_device *smmu;
>> + struct rb_node *node;
>> +
>> + printk(XENLOG_DEBUG "arm-smmu: Initialize dom0\n");
>> +
>> + list_for_each_entry( smmu, &arm_smmu_devices, list )
>> + {
>> + for ( node = rb_first(&smmu->masters); node; node = rb_next(node) )
>> + {
>> + struct arm_smmu_master *master;
>> +
>> + master = container_of(node, struct arm_smmu_master, node);
>> +
>> + if ( dt_device_used_by(master->dt_node) == DOMID_XEN ||
>> + platform_device_is_blacklisted(master->dt_node) )
>> + continue;
>> +
>> + arm_smmu_attach_dev(d, master->dt_node);
>
> Should this not be driven from the same loop as the MMIO mapping setup
> in dom0 build? Otherwise isn't there a danger that they won't
> correspond?
On my second version of the patch series I moved out this code in
map_device.
>
> A "master" here is a "bus master" i.e. a device, right?
>> + /* Even if the device can't be initialized, we don't want to give to
>> + * dom0 the smmu device
>
> "give the smmu device to dom0"
>
>
> I obviously haven't reviewed all this code in detail, but I have skimmed
> it and nothing leapt out.
Thanks for the review!
Cheers,
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers
2014-02-19 14:15 ` Ian Campbell
2014-02-19 17:21 ` Julien Grall
@ 2014-02-19 17:23 ` Julien Grall
1 sibling, 0 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-19 17:23 UTC (permalink / raw)
To: Ian Campbell
Cc: patches, tim, stefano.stabellini, Jan Beulich, xen-devel, Xiantao Zhang
On 02/19/2014 02:15 PM, Ian Campbell wrote:
> A "master" here is a "bus master" i.e. a device, right?
I forgot to answer to this part: yes.
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers
2014-02-19 17:21 ` Julien Grall
@ 2014-02-19 17:27 ` Ian Campbell
2014-02-19 17:31 ` Julien Grall
0 siblings, 1 reply; 51+ messages in thread
From: Ian Campbell @ 2014-02-19 17:27 UTC (permalink / raw)
To: Julien Grall
Cc: patches, tim, stefano.stabellini, Jan Beulich, xen-devel, Xiantao Zhang
On Wed, 2014-02-19 at 17:21 +0000, Julien Grall wrote:
> >
> >> + * This driver currently supports:
> >> + * - SMMUv1 and v2 implementations (didn't try v2 SMMU)
> >
> > I guess this is Will's original comment, I thought SMMU-400, which
> > you've tried, was v2?
>
> No it's SMMU v2. As I understand:
> - SMMU-400 => SMMU v1
> - SMMU-500 => SMMU v2
Ah yes, that sounds more familiar.
> The words in parenthesis was added by me because I don't have a test box
> with SMMU v2.
OK.
> >> +#define SZ_4K (1 << 12)
> >> +#define SZ_64K (1 << 16)
> >> +
> >> +/* Driver options */
> >> +#define SMMU_OPT_SECURE_CONFIG_ACCESS (1 << 0)
> >
> > Is this just retained to reduce the deviation from the Linux driver?
> > It's no use to us I think. (I suppose that goes for a bunch of other
> > stuff, eg.. the PGSZ_4K stuff, which I will avoid commenting on
> > further).
>
> SZ_4K and SZ_64K is used later in the code.
But they are actually useless to us aren't they?
> The
> SMMU_OPT_SECURE_CONFIG_ACCESS is used for midway as the SMMU is broken.
Ah, so this is effectively a quirk then.
Ian/
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers
2014-02-19 17:27 ` Ian Campbell
@ 2014-02-19 17:31 ` Julien Grall
2014-02-20 8:11 ` Jan Beulich
0 siblings, 1 reply; 51+ messages in thread
From: Julien Grall @ 2014-02-19 17:31 UTC (permalink / raw)
To: Ian Campbell
Cc: patches, tim, stefano.stabellini, Jan Beulich, xen-devel, Xiantao Zhang
On 02/19/2014 05:27 PM, Ian Campbell wrote:
> On Wed, 2014-02-19 at 17:21 +0000, Julien Grall wrote:
>>>
>>>> + * This driver currently supports:
>>>> + * - SMMUv1 and v2 implementations (didn't try v2 SMMU)
>>>
>>> I guess this is Will's original comment, I thought SMMU-400, which
>>> you've tried, was v2?
>>
>> No it's SMMU v2. As I understand:
>> - SMMU-400 => SMMU v1
>> - SMMU-500 => SMMU v2
>
> Ah yes, that sounds more familiar.
>
>> The words in parenthesis was added by me because I don't have a test box
>> with SMMU v2.
>
> OK.
>
>>>> +#define SZ_4K (1 << 12)
>>>> +#define SZ_64K (1 << 16)
>>>> +
>>>> +/* Driver options */
>>>> +#define SMMU_OPT_SECURE_CONFIG_ACCESS (1 << 0)
>>>
>>> Is this just retained to reduce the deviation from the Linux driver?
>>> It's no use to us I think. (I suppose that goes for a bunch of other
>>> stuff, eg.. the PGSZ_4K stuff, which I will avoid commenting on
>>> further).
>>
>> SZ_4K and SZ_64K is used later in the code.
>
> But they are actually useless to us aren't they?
As we use only 4K page in Xen yes. I kept it because there is few place
where the SMMU configuration is not the same.
If we want to support 64K page in the future it will harder to had
support if this stuff is removed.
The constant was added by himself as it doesn't exists on Xen. I can
move it into the generic code.
Cheers,
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers
2014-02-19 17:31 ` Julien Grall
@ 2014-02-20 8:11 ` Jan Beulich
2014-02-20 11:05 ` Julien Grall
0 siblings, 1 reply; 51+ messages in thread
From: Jan Beulich @ 2014-02-20 8:11 UTC (permalink / raw)
To: Julien Grall
Cc: Ian Campbell, patches, tim, stefano.stabellini, xen-devel, Xiantao Zhang
>>> On 19.02.14 at 18:31, Julien Grall <julien.grall@linaro.org> wrote:
> On 02/19/2014 05:27 PM, Ian Campbell wrote:
>> On Wed, 2014-02-19 at 17:21 +0000, Julien Grall wrote:
>>>>> +#define SZ_4K (1 << 12)
>>>>> +#define SZ_64K (1 << 16)
>>>>> +
>>>>> +/* Driver options */
>>>>> +#define SMMU_OPT_SECURE_CONFIG_ACCESS (1 << 0)
>>>>
>>>> Is this just retained to reduce the deviation from the Linux driver?
>>>> It's no use to us I think. (I suppose that goes for a bunch of other
>>>> stuff, eg.. the PGSZ_4K stuff, which I will avoid commenting on
>>>> further).
>>>
>>> SZ_4K and SZ_64K is used later in the code.
>>
>> But they are actually useless to us aren't they?
>
> As we use only 4K page in Xen yes. I kept it because there is few place
> where the SMMU configuration is not the same.
>
> If we want to support 64K page in the future it will harder to had
> support if this stuff is removed.
>
> The constant was added by himself as it doesn't exists on Xen. I can
> move it into the generic code.
But if possible I'd like to encourage you using PAGE_SIZE_4K to
match up with what we already got in the IOMMU code. Unless
deviation from the sources you're cloning is deemed worse than
consistency within our code.
Jan
^ permalink raw reply [flat|nested] 51+ messages in thread
* Re: [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers
2014-02-20 8:11 ` Jan Beulich
@ 2014-02-20 11:05 ` Julien Grall
0 siblings, 0 replies; 51+ messages in thread
From: Julien Grall @ 2014-02-20 11:05 UTC (permalink / raw)
To: Jan Beulich
Cc: Ian Campbell, patches, tim, stefano.stabellini, xen-devel, Xiantao Zhang
Hello Jan,
On 20/02/14 08:11, Jan Beulich wrote:
>>>> On 19.02.14 at 18:31, Julien Grall <julien.grall@linaro.org> wrote:
>> On 02/19/2014 05:27 PM, Ian Campbell wrote:
>>> On Wed, 2014-02-19 at 17:21 +0000, Julien Grall wrote:
>>>>>> +#define SZ_4K (1 << 12)
>>>>>> +#define SZ_64K (1 << 16)
>>>>>> +
>>>>>> +/* Driver options */
>>>>>> +#define SMMU_OPT_SECURE_CONFIG_ACCESS (1 << 0)
>>>>>
>>>>> Is this just retained to reduce the deviation from the Linux driver?
>>>>> It's no use to us I think. (I suppose that goes for a bunch of other
>>>>> stuff, eg.. the PGSZ_4K stuff, which I will avoid commenting on
>>>>> further).
>>>>
>>>> SZ_4K and SZ_64K is used later in the code.
>>>
>>> But they are actually useless to us aren't they?
>>
>> As we use only 4K page in Xen yes. I kept it because there is few place
>> where the SMMU configuration is not the same.
>>
>> If we want to support 64K page in the future it will harder to had
>> support if this stuff is removed.
>>
>> The constant was added by himself as it doesn't exists on Xen. I can
>> move it into the generic code.
>
> But if possible I'd like to encourage you using PAGE_SIZE_4K to
> match up with what we already got in the IOMMU code. Unless
> deviation from the sources you're cloning is deemed worse than
> consistency within our code.
I think this change is fine. It doesn't fundamentally change the code.
I will update the patch for the next version.
Regards,
--
Julien Grall
^ permalink raw reply [flat|nested] 51+ messages in thread
end of thread, other threads:[~2014-02-20 11:05 UTC | newest]
Thread overview: 51+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-02-07 17:42 [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
2014-02-07 17:43 ` [RFC for-4.5 01/12] xen/common: grant-table: only call IOMMU if paging mode translate is disabled Julien Grall
2014-02-10 8:02 ` Jan Beulich
2014-02-19 12:25 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 02/12] xen/passthrough: vtd: Don't export iommu_domain_teardown Julien Grall
2014-02-19 12:26 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 03/12] xen/passthrough: vtd: Don't export iommu_set_pgd Julien Grall
2014-02-10 8:04 ` Jan Beulich
2014-02-10 8:07 ` Zhang, Xiantao
2014-02-19 12:27 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 04/12] xen/dts: Add dt_property_read_bool Julien Grall
2014-02-19 12:28 ` Ian Campbell
2014-02-19 14:57 ` Julien Grall
2014-02-07 17:43 ` [RFC for-4.5 05/12] xen/dts: Add dt_parse_phandle_with_args and dt_parse_phandle Julien Grall
2014-02-19 12:34 ` Ian Campbell
2014-02-19 16:17 ` Julien Grall
2014-02-19 16:26 ` Ian Campbell
2014-02-19 16:52 ` Julien Grall
2014-02-19 16:57 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 06/12] xen/passthrough: rework dom0_pvh_reqs to use it also on ARM Julien Grall
2014-02-10 8:07 ` Jan Beulich
2014-02-10 11:42 ` Julien Grall
2014-02-10 16:10 ` Julien Grall
2014-02-10 16:35 ` Jan Beulich
2014-02-10 17:42 ` Julien Grall
2014-02-11 7:47 ` Jan Beulich
2014-02-07 17:43 ` [RFC for-4.5 07/12] xen/passthrough: iommu: Don't need to map dom0 page when the PT is shared Julien Grall
2014-02-10 8:11 ` Jan Beulich
2014-02-19 12:37 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 08/12] xen/passthrough: iommu: Split generic IOMMU code Julien Grall
2014-02-10 8:22 ` Jan Beulich
2014-02-10 11:52 ` Julien Grall
2014-02-19 12:39 ` Ian Campbell
2014-02-19 16:23 ` Julien Grall
2014-02-07 17:43 ` [RFC for-4.5 09/12] xen/passthrough: iommu: Introduce arch specific code Julien Grall
2014-02-19 12:40 ` Ian Campbell
2014-02-19 16:25 ` Julien Grall
2014-02-07 17:43 ` [RFC for-4.5 10/12] xen/passthrough: Introduce IOMMU ARM architure Julien Grall
2014-02-19 12:49 ` Ian Campbell
2014-02-19 16:50 ` Julien Grall
2014-02-07 17:43 ` [RFC for-4.5 11/12] MAINTAINERS: Add drivers/passthrough/arm Julien Grall
2014-02-19 12:49 ` Ian Campbell
2014-02-07 17:43 ` [RFC for-4.5 12/12] drivers/passthrough: arm: Add support for SMMU drivers Julien Grall
2014-02-19 14:15 ` Ian Campbell
2014-02-19 17:21 ` Julien Grall
2014-02-19 17:27 ` Ian Campbell
2014-02-19 17:31 ` Julien Grall
2014-02-20 8:11 ` Jan Beulich
2014-02-20 11:05 ` Julien Grall
2014-02-19 17:23 ` Julien Grall
2014-02-07 17:51 ` [RFC for-4.5 00/12] IOMMU support for ARM Julien Grall
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.