vfio: QOMify VFIOContainer
v2 cover vfio: QOMify VFIOContainer
Hello,
The series starts with simple changes (patch 1-4). Two of which were initially sent by Joao in a series adding VFIO migration support with vIOMMU [1].
The changes following prepare VFIOContainer for QOMification, switch the container models to QOM when ready and add some final cleanups.
Applies on top of :
* [v7] Add a host IOMMU device abstraction to check with vIOMMU
https://lore.kernel.org/all/[email protected]
* [v4] VIRTIO-IOMMU/VFIO: Fix host iommu geometry
https://lore.kernel.org/all/[email protected]
Thanks,
C.
[1] https://lore.kernel.org/qemu-devel/[email protected]/
Changes in v2:
- Used OBJECT_DECLARE_SIMPLE_TYPE
- Introduced a instance_finalize() handler
Avihai Horon (1):
vfio/common: Extract vIOMMU code from vfio_sync_dirty_bitmap()
Cédric Le Goater (15):
vfio: Make vfio_devices_dma_logging_start() return bool
vfio: Remove unused declarations from vfio-common.h vfio/container: Introduce vfio_address_space_insert() vfio/container: Simplify vfio_container_init() vfio/container: Modify vfio_get_iommu_type() to use a container fd vfio/container: Introduce vfio_get_iommu_class_name() vfio/container: Introduce vfio_create_container() vfio/container: Discover IOMMU type before creating the container vfio/container: Change VFIOContainerBase to use QOM vfio/container: Switch to QOM
vfio/container: Introduce an instance_init() handler vfio/container: Remove VFIOContainerBase::ops vfio/container: Remove vfio_container_init() vfio/container: Introduce vfio_iommu_legacy_instance_init() vfio/container: Move vfio_container_destroy() to an
instance_finalize() handler
Joao Martins (1):
vfio/common: Move dirty tracking ranges update to helper
include/hw/vfio/vfio-common.h | 10 ++- include/hw/vfio/vfio-container-base.h | 19 +--- hw/vfio/common.c | 124 ++++++++++++++++---------- hw/vfio/container-base.c | 70 +++++++++------ hw/vfio/container.c | 106 ++++++++++++---------- hw/vfio/iommufd.c | 13 ++- hw/vfio/pci.c | 4 +- hw/vfio/spapr.c | 3 + 8 files changed, 196 insertions(+), 153 deletions(-)
Signature · 1 line
2.45.2
PATCH v2 1/17 vfio: Make vfio_devices_dma_logging_start() return bool
Since vfio_devices_dma_logging_start() takes an 'Error **' argument, best practices suggest to return a bool. See the api/error.h Rules section. It will simplify potential changes coming after.
vfio_container_set_dirty_page_tracking() could be modified in the same way but the errno value can be saved in the migration stream when called from vfio_listener_log_global_stop().
--- hw/vfio/common.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-)
hw/vfio/common.c +7 -7
@@ -1020,7 +1020,7 @@ static void vfio_device_feature_dma_logging_start_destroy(g_free (feature); } -static int vfio_devices_dma_logging_start (VFIOContainerBase *bcontainer, +static bool vfio_devices_dma_logging_start (VFIOContainerBase *bcontainer,Error **errp) {struct vfio_device_feature *feature;
@@ -1033,7 +1033,7 @@ static int vfio_devices_dma_logging_start(VFIOContainerBase *bcontainer, &ranges);if (!feature) {error_setg_errno (errp, errno,"Failed to prepare DMA logging" ); -return -errno; +return false; } QLIST_FOREACH(vbasedev, &bcontainer->device_list, container_next) {
@@ -1058,7 +1058,7 @@ out:vfio_device_feature_dma_logging_start_destroy (feature); -return ret; +return ret ==0 ; } static bool vfio_listener_log_global_start(MemoryListener *listener,
@@ -1067,18 +1067,18 @@ static bool vfio_listener_log_global_start(MemoryListener *listener,ERRP_GUARD ();VFIOContainerBase *bcontainer =container_of (listener, VFIOContainerBase, listener); -int ret; +bool ret;if (vfio_devices_all_device_dirty_tracking (bcontainer)) { ret =vfio_devices_dma_logging_start (bcontainer, errp); }else { - ret =vfio_container_set_dirty_page_tracking (bcontainer, true, errp); + ret =vfio_container_set_dirty_page_tracking (bcontainer, true, errp) ==0 ; } -if (ret) { +if (!ret) {error_prepend (errp,"vfio: Could not start dirty page tracking - " ); } -return !ret; +return ret; }static void vfio_listener_log_global_stop (MemoryListener *listener)
Signature · 1 line
2.45.2
Re: [PATCH v2 01/17] vfio: Make vfio_devices_dma_logging_start() return bool
Hi Cédric,
Quoted 3 lines · Cédric Le Goater
Since
vfio_devices_dma_logging_start()takes an 'Error **' argument, best practices suggest to return a bool. See theapi/error.hRules section. It will simplify potential changes coming after.
As I already mentionned the Rules section does not say that, as far as I understand:
It is allowed to either return a bool, a pointer-value, an integer, along with an error handle:
"
* • bool-valued functions return true on success / false on failure, * • pointer-valued functions return non-null / null pointer, and * • integer-valued functions return non-negative / negative. "
Personally I don't like much returning a bool as I think it rather complexifies the code and to me that kind of change is error prone.
Quoted 43 lines
vfio_container_set_dirty_page_tracking()could be modified in the same way but the errno value can be saved in the migration stream when called fromvfio_listener_log_global_stop().Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>--- hw/vfio/common.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-)hw/vfio/common.c +3
-3@@ -1020,7 +1020,7 @@ static void vfio_device_feature_dma_logging_start_destroy(g_free (feature); } -static int vfio_devices_dma_logging_start (VFIOContainerBase *bcontainer, +static bool vfio_devices_dma_logging_start (VFIOContainerBase *bcontainer,Error **errp) {struct vfio_device_feature *feature;@@ -1033,7 +1033,7 @@ static int vfio_devices_dma_logging_start(VFIOContainerBase *bcontainer, &ranges);if (!feature) {error_setg_errno (errp, errno,"Failed to prepare DMA logging" ); -return -errno; +return false; } QLIST_FOREACH(vbasedev, &bcontainer->device_list, container_next) {@@ -1058,7 +1058,7 @@ out:vfio_device_feature_dma_logging_start_destroy (feature); -return ret; +return ret ==0 ; } static bool vfio_listener_log_global_start(MemoryListener *listener,@@ -1067,18 +1067,18 @@ static bool vfio_listener_log_global_start(MemoryListener *listener,ERRP_GUARD ();
hw/vfio/common.c static bool vfio_listener_log_global_start(MemoryListener *listener, · line 1075
Quoted 10 lines
hw/vfio/common.c +2
-2VFIOContainerBase *bcontainer =container_of (listener, VFIOContainerBase, listener); -int ret; +bool ret;if (vfio_devices_all_device_dirty_tracking (bcontainer)) { ret =vfio_devices_dma_logging_start (bcontainer, errp); }else { - ret =vfio_container_set_dirty_page_tracking (bcontainer, true, errp); + ret =vfio_container_set_dirty_page_tracking (bcontainer, true, errp) ==0 ;
why vfio_container_set_dirty_page_tracking(bcontainer, true, errp) doesn't return a bool then?
Eric
Quoted 11 lines
hw/vfio/common.c +2
-2} -if (ret) { +if (!ret) {error_prepend (errp,"vfio: Could not start dirty page tracking - " ); } -return !ret; +return ret; }static void vfio_listener_log_global_stop (MemoryListener *listener)
Re: [PATCH v2 01/17] vfio: Make vfio_devices_dma_logging_start() return bool
Hello Eric,
Quoted 13 lines · Eric Auger
Hi Cédric,
Quoted 3 lines · Cédric Le Goater
Since
vfio_devices_dma_logging_start()takes an 'Error **' argument, best practices suggest to return a bool. See theapi/error.hRules section. It will simplify potential changes coming after.As I already mentionned the Rules section does not say that, as far as I understand:
It is allowed to either return a bool, a pointer-value, an integer, along with an error handle:
Quoted 8 lines
"
* • bool-valued functions return true on success / false on failure, * • pointer-valued functions return non-null / null pointer, and * • integer-valued functions return non-negative / negative.
"Personally I don't like much returning a bool as I think it rather complexifies the code and to me that kind of change is error prone.
Returning an int could be misleading too, since there are multiple ways it could be interpreted. You can find in QEMU routines returning -1 for error which is later used as an errno :/
The error framework in QEMU provides a way to to save and return any kind of errors, using the error_seg_*() routines. I tend to prefer the basic approach: return fail or pass and use the Error parameter for details.
Now, the problem, as always, is doing the conversion in all the code. This is probably why people, with a kernel background, find it confusing.
Quoted 50 lines
Quoted 50 lines
vfio_container_set_dirty_page_tracking()could be modified in the same way but the errno value can be saved in the migration stream when called fromvfio_listener_log_global_stop().Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>--- hw/vfio/common.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-)hw/vfio/common.c +4
-4@@ -1020,7 +1020,7 @@ static void vfio_device_feature_dma_logging_start_destroy(g_free (feature); } -static int vfio_devices_dma_logging_start (VFIOContainerBase *bcontainer, +static bool vfio_devices_dma_logging_start (VFIOContainerBase *bcontainer,Error **errp) {struct vfio_device_feature *feature;@@ -1033,7 +1033,7 @@ static int vfio_devices_dma_logging_start(VFIOContainerBase *bcontainer, &ranges);if (!feature) {error_setg_errno (errp, errno,"Failed to prepare DMA logging" ); -return -errno; +return false; } QLIST_FOREACH(vbasedev, &bcontainer->device_list, container_next) {@@ -1058,7 +1058,7 @@ out:vfio_device_feature_dma_logging_start_destroy (feature); -return ret; +return ret ==0 ; } static bool vfio_listener_log_global_start(MemoryListener *listener,@@ -1067,18 +1067,18 @@ static bool vfio_listener_log_global_start(MemoryListener *listener,ERRP_GUARD ();VFIOContainerBase *bcontainer =container_of (listener, VFIOContainerBase, listener); -int ret; +bool ret;
Quoted 7 lines
hw/vfio/common.c static bool vfio_listener_log_global_start(MemoryListener *listener, · line 1075
Quoted 5 lines
hw/vfio/common.c +1
-1if (vfio_devices_all_device_dirty_tracking (bcontainer)) { ret =vfio_devices_dma_logging_start (bcontainer, errp); }else { - ret =vfio_container_set_dirty_page_tracking (bcontainer, true, errp); + ret =vfio_container_set_dirty_page_tracking (bcontainer, true, errp) ==0 ;why
vfio_container_set_dirty_page_tracking(bcontainer, true, errp)doesn't return a bool then?
The errno value can be saved in the migration stream when called from vfio_listener_log_global_stop(). So this would require changes in the migration subsystem, like we did for vfio_listener_log_global_start().
Thanks,
C.
Quoted 15 lines
Eric
Quoted 11 lines
hw/vfio/common.c +2
-2} -if (ret) { +if (!ret) {error_prepend (errp,"vfio: Could not start dirty page tracking - " ); } -return !ret; +return ret; }static void vfio_listener_log_global_stop (MemoryListener *listener)
Eric Auger
Reviewed-byRe: [PATCH v2 01/17] vfio: Make vfio_devices_dma_logging_start() return bool
Hi Cédric,
Quoted 22 lines · Cédric Le Goater
Hello Eric,
Quoted 19 lines · Eric Auger
Hi Cédric,
Quoted 3 lines · Cédric Le Goater
Since
vfio_devices_dma_logging_start()takes an 'Error **' argument, best practices suggest to return a bool. See theapi/error.hRules section. It will simplify potential changes coming after.As I already mentionned the Rules section does not say that, as far as I understand:
It is allowed to either return a bool, a pointer-value, an integer, along with an error handle:"
* • bool-valued functions return true on success / false on failure, * • pointer-valued functions return non-null / null pointer, and * • integer-valued functions return non-negative / negative. "
Quoted 6 lines
Quoted 2 lines
Personally I don't like much returning a bool as I think it rather complexifies the code and to me that kind of change is error prone.
Returning an int could be misleading too, since there are multiple ways it could be interpreted. You can find in QEMU routines returning -1 for error which is later used as an errno :/
yes no good!
Quoted 77 lines
The error framework in QEMU provides a way to to save and return any kind of errors, using the error_seg_*() routines. I tend to prefer the basic approach: return fail or pass and use the Error parameter for details.
Now, the problem, as always, is doing the conversion in all the code. This is probably why people, with a kernel background, find it confusing.
Quoted 57 lines
Quoted 57 lines
vfio_container_set_dirty_page_tracking()could be modified in the same way but the errno value can be saved in the migration stream when called fromvfio_listener_log_global_stop().Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>---
hw/vfio/common.c| 14 +++++++-------1 file changed, 7 insertions(+), 7 deletions(-)hw/vfio/common.c +0
-0index
9e4c0cc95ff90209d3e8184035af0806a2bf890b..d48cd9b9361a92d184e423ffc60aabaff40fb487100644hw/vfio/common.c +0
-0@@ -1020,7 +1020,7 @@ static voidvfio_device_feature_dma_logging_start_destroy(
g_free(feature);
}
-static int vfio_devices_dma_logging_start(VFIOContainerBase *bcontainer,
+static bool vfio_devices_dma_logging_start(VFIOContainerBase *bcontainer,
Error **errp) {
structvfio_device_feature*feature;hw/vfio/common.c +0
-0@@ -1033,7 +1033,7 @@ static intvfio_devices_dma_logging_start (VFIOContainerBase *bcontainer, &ranges); if (!feature) {
error_setg_errno(errp, errno, "Failed to prepare DMA logging");
- return -errno;
+ return false;
}
QLIST_FOREACH(vbasedev, &bcontainer->device_list, container_next){hw/vfio/common.c +1
-0@@ -1058,7 +1058,7 @@ out: vfio_device_feature_dma_logging_start_destroy(feature); - return ret; + return ret == 0; }static bool vfio_listener_log_global_start(MemoryListener *listener,
hw/vfio/common.c +1
-1@@ -1067,18 +1067,18 @@ static boolvfio_listener_log_global_start (MemoryListener *listener, ERRP_GUARD(); VFIOContainerBase *bcontainer = container_of(listener, VFIOContainerBase, listener); - int ret; + bool ret; if (vfio_devices_all_device_dirty_tracking(bcontainer)) { ret = vfio_devices_dma_logging_start(bcontainer, errp);Quoted 8 lines
hw/vfio/common.c static bool · line 1078
Quoted 5 lines
hw/vfio/common.c +1
-1} else { - ret = vfio_container_set_dirty_page_tracking(bcontainer, true, errp); + ret = vfio_container_set_dirty_page_tracking(bcontainer, true, errp) ==0 ;why
vfio_container_set_dirty_page_tracking(bcontainer, true, errp)doesn't return a bool then?
Quoted 3 lines
The errno value can be saved in the migration stream when called from
vfio_listener_log_global_stop(). So this would require changes in the migration subsystem, like we did forvfio_listener_log_global_start().
OK
so although I am not a big fan of that kind of change and once highlighted this is not mandated by the error.h doc, please feel free to add my
Thanks
Eric
Quoted 24 lines
Thanks,
C.
Quoted 14 lines
Eric
Quoted 10 lines
hw/vfio/common.c +1
-0} - if (ret) { + if (!ret) {
error_prepend(errp, "vfio: Could not start dirty page tracking - ");
}
- return !ret;
+ return ret;
}
static voidvfio_listener_log_global_stop(MemoryListener *listener)
PATCH v2 2/17 vfio: Remove unused declarations from vfio-common.h
In reply to Cédric Le GoaterThese were forgotten in the recent cleanups.
--- include/hw/vfio/vfio-common.h | 4 ---- 1 file changed, 4 deletions(-)
include/hw/vfio/vfio-common.h +0 -4
@@ -207,10 +207,6 @@ typedef struct VFIODisplay {VFIOAddressSpace *vfio_get_address_space (AddressSpace *as);void vfio_put_address_space (VFIOAddressSpace *space); -/* SPAPR specific */ -int vfio_spapr_container_init (VFIOContainer *container,Error **errp); -void vfio_spapr_container_deinit (VFIOContainer *container); -void vfio_disable_irqindex (VFIODevice *vbasedev,int index);void vfio_unmask_single_irqindex (VFIODevice *vbasedev,int index);void vfio_mask_single_irqindex (VFIODevice *vbasedev,int index);
Signature · 1 line
2.45.2
Re: [PATCH v2 02/17] vfio: Remove unused declarations from vfio-common.h
Quoted 4 lines · Cédric Le Goater
These were forgotten in the recent cleanups.
Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>
Eric
Quoted 19 lines
--- include/hw/vfio/vfio-common.h | 4 ---- 1 file changed, 4 deletions(-)include/hw/vfio/vfio-common.h +0
-4@@ -207,10 +207,6 @@ typedef struct VFIODisplay {VFIOAddressSpace *vfio_get_address_space (AddressSpace *as);void vfio_put_address_space (VFIOAddressSpace *space); -/* SPAPR specific */ -int vfio_spapr_container_init (VFIOContainer *container,Error **errp); -void vfio_spapr_container_deinit (VFIOContainer *container); -void vfio_disable_irqindex (VFIODevice *vbasedev,int index);void vfio_unmask_single_irqindex (VFIODevice *vbasedev,int index);void vfio_mask_single_irqindex (VFIODevice *vbasedev,int index);
PATCH v2 3/17 vfio/common: Move dirty tracking ranges update to helper
In reply to Cédric Le GoaterFrom: Joao Martins <[email protected]>
Separate the changes that updates the ranges from the listener, to make it reusable in preparation to expand its use to vIOMMU support.
[ clg: - Rebased on upstream
- Introduced vfio_dirty_tracking_update_range() ]
--- hw/vfio/common.c | 38 ++++++++++++++++++++++---------------- 1 file changed, 22 insertions(+), 16 deletions(-)
hw/vfio/common.c +22 -16
@@ -839,20 +839,11 @@ static bool vfio_section_is_vfio_pci(MemoryRegionSection *section,return false; } -static void vfio_dirty_tracking_update (MemoryListener *listener, -MemoryRegionSection *section) +static void vfio_dirty_tracking_update_range (VFIODirtyRanges *range, +hwaddr iova,hwaddr end, +bool update_pci) { -VFIODirtyRangesListener *dirty =container_of (listener, - VFIODirtyRangesListener, - listener); -VFIODirtyRanges *range = &dirty->ranges; -hwaddr iova, end, *min, *max; - -if (!vfio_listener_valid_section (section,"tracking_update" ) || - !vfio_get_section_iova_range (dirty->bcontainer, section, - &iova, &end,NULL )) { -return ; - } +hwaddr *min, *max;/* * The address space passed to the dirty tracker is reduced to three ranges:
@@ -873,8 +864,7 @@ static void vfio_dirty_tracking_update(MemoryListener *listener,* The alternative would be an IOVATree but that has a much bigger runtime * overhead and unnecessary complexity. */ -if (vfio_section_is_vfio_pci (section, dirty->bcontainer) && - iova >=UINT32_MAX ) { +if (update_pci && iova >=UINT32_MAX ) { min = &range->minpci64; max = &range->maxpci64; }else {
@@ -889,7 +879,23 @@ static void vfio_dirty_tracking_update(MemoryListener *listener, }trace_vfio_device_dirty_tracking_update (iova, end, *min, *max); -return ; +} + +static void vfio_dirty_tracking_update (MemoryListener *listener, +MemoryRegionSection *section) +{ +VFIODirtyRangesListener *dirty = +container_of (listener, VFIODirtyRangesListener, listener); +hwaddr iova, end; + +if (!vfio_listener_valid_section (section,"tracking_update" ) || + !vfio_get_section_iova_range (dirty->bcontainer, section, + &iova, &end,NULL )) { +return ; + } + +vfio_dirty_tracking_update_range (&dirty->ranges, iova, end, +vfio_section_is_vfio_pci (section, dirty->bcontainer)); }static const MemoryListener vfio_dirty_tracking_listener = {
Signature · 1 line
2.45.2
Re: [PATCH v2 03/17] vfio/common: Move dirty tracking ranges update to helper
Hi Cédric,
Quoted 3 lines · Cédric Le Goater
From: Joao Martins <[email protected]>
Separate the changes that updates the ranges from the listener, to
s/updates/update
Quoted 52 lines
make it reusable in preparation to expand its use to vIOMMU support.
[ clg: - Rebased on upstream
- Introducedvfio_dirty_tracking_update_range()]Signed-off-by Joao Martins <[email protected]>Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>--- hw/vfio/common.c | 38 ++++++++++++++++++++++---------------- 1 file changed, 22 insertions(+), 16 deletions(-)hw/vfio/common.c +5
-15@@ -839,20 +839,11 @@ static bool vfio_section_is_vfio_pci(MemoryRegionSection *section,return false; } -static void vfio_dirty_tracking_update (MemoryListener *listener, -MemoryRegionSection *section) +static void vfio_dirty_tracking_update_range (VFIODirtyRanges *range, +hwaddr iova,hwaddr end, +bool update_pci) { -VFIODirtyRangesListener *dirty =container_of (listener, - VFIODirtyRangesListener, - listener); -VFIODirtyRanges *range = &dirty->ranges; -hwaddr iova, end, *min, *max; - -if (!vfio_listener_valid_section (section,"tracking_update" ) || - !vfio_get_section_iova_range (dirty->bcontainer, section, - &iova, &end,NULL )) { -return ; - } +hwaddr *min, *max;/* * The address space passed to the dirty tracker is reduced to three ranges: @@ -873,8 +864,7 @@ static void vfio_dirty_tracking_update(MemoryListener *listener,* The alternative would be an IOVATree but that has a much bigger runtime * overhead and unnecessary complexity. */ -if (vfio_section_is_vfio_pci (section, dirty->bcontainer) && - iova >=UINT32_MAX ) { +if (update_pci && iova >=UINT32_MAX ) { min = &range->minpci64; max = &range->maxpci64; }else {
hw/vfio/common.c static void vfio_dirty_tracking_update(MemoryListener *listener, · line 881
Quoted 3 lines
hw/vfio/common.c +0
-0@@ -889,7 +879,23 @@ static void vfio_dirty_tracking_update(MemoryListener *listener, }trace_vfio_device_dirty_tracking_update (iova, end, *min, *max);
don't you want to update the name of the trace point?
Quoted 11 lines
hw/vfio/common.c +10
-1-return ; +} + +static void vfio_dirty_tracking_update (MemoryListener *listener, +MemoryRegionSection *section) +{ +VFIODirtyRangesListener *dirty = +container_of (listener, VFIODirtyRangesListener, listener); +hwaddr iova, end; + + if (!vfio_listener_valid_section(section, "tracking_update") ||
hw/vfio/common.c static void vfio_dirty_tracking_update(MemoryListener *listener, · line 901
Quoted 10 lines
hw/vfio/common.c +7
-0+ !vfio_get_section_iova_range(dirty->bcontainer, section, + &iova, &end, NULL)) { +return ; + } + +vfio_dirty_tracking_update_range (&dirty->ranges, iova, end, +vfio_section_is_vfio_pci (section, dirty->bcontainer)); }static const MemoryListener vfio_dirty_tracking_listener = {
Besides
Eric
Re: [PATCH v2 03/17] vfio/common: Move dirty tracking ranges update to helper
Quoted 6 lines · Eric Auger
Hi Cédric,
Quoted 3 lines · Cédric Le Goater
From: Joao Martins <[email protected]>
Separate the changes that updates the ranges from the listener, to
s/updates/update
fixed.
Quoted 54 lines
Quoted 54 lines
make it reusable in preparation to expand its use to vIOMMU support.
[ clg: - Rebased on upstream
- Introducedvfio_dirty_tracking_update_range()]Signed-off-by Joao Martins <[email protected]>Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>--- hw/vfio/common.c | 38 ++++++++++++++++++++++---------------- 1 file changed, 22 insertions(+), 16 deletions(-)hw/vfio/common.c +5
-15@@ -839,20 +839,11 @@ static bool vfio_section_is_vfio_pci(MemoryRegionSection *section,return false; } -static void vfio_dirty_tracking_update (MemoryListener *listener, -MemoryRegionSection *section) +static void vfio_dirty_tracking_update_range (VFIODirtyRanges *range, +hwaddr iova,hwaddr end, +bool update_pci) { -VFIODirtyRangesListener *dirty =container_of (listener, - VFIODirtyRangesListener, - listener); -VFIODirtyRanges *range = &dirty->ranges; -hwaddr iova, end, *min, *max; - -if (!vfio_listener_valid_section (section,"tracking_update" ) || - !vfio_get_section_iova_range (dirty->bcontainer, section, - &iova, &end,NULL )) { -return ; - } +hwaddr *min, *max;/* * The address space passed to the dirty tracker is reduced to three ranges: @@ -873,8 +864,7 @@ static void vfio_dirty_tracking_update(MemoryListener *listener,* The alternative would be an IOVATree but that has a much bigger runtime * overhead and unnecessary complexity. */ -if (vfio_section_is_vfio_pci (section, dirty->bcontainer) && - iova >=UINT32_MAX ) { +if (update_pci && iova >=UINT32_MAX ) { min = &range->minpci64; max = &range->maxpci64; }else {@@ -889,7 +879,23 @@ static void vfio_dirty_tracking_update(MemoryListener *listener, }
Quoted 2 lines
hw/vfio/common.c static void vfio_dirty_tracking_update(MemoryListener *listener, · line 881
Quoted 1 line
hw/vfio/common.c +0
-0trace_vfio_device_dirty_tracking_update (iova, end, *min, *max);don't you want to update the name of the trace point?
There is only one caller. Let's reconsider when there are more users of this routine.
Thanks,
C.
Quoted 26 lines
Quoted 11 lines
hw/vfio/common.c +10
-1-return ; +} + +static void vfio_dirty_tracking_update (MemoryListener *listener, +MemoryRegionSection *section) +{ +VFIODirtyRangesListener *dirty = +container_of (listener, VFIODirtyRangesListener, listener); +hwaddr iova, end; + + if (!vfio_listener_valid_section(section, "tracking_update") ||hw/vfio/common.c static void vfio_dirty_tracking_update(MemoryListener *listener, · line 901
Quoted 10 lines
hw/vfio/common.c +7
-0+ !vfio_get_section_iova_range(dirty->bcontainer, section, + &iova, &end, NULL)) { +return ; + } + +vfio_dirty_tracking_update_range (&dirty->ranges, iova, end, +vfio_section_is_vfio_pci (section, dirty->bcontainer)); }static const MemoryListener vfio_dirty_tracking_listener = {Besides
Reviewed-by Eric Auger <[email protected]>Eric
PATCH v2 4/17 vfio/common: Extract vIOMMU code from vfio_sync_dirty_bitmap()
In reply to Cédric Le GoaterFrom: Avihai Horon <[email protected]>
Extract vIOMMU code from vfio_sync_dirty_bitmap() to a new function and restructure the code.
This is done in preparation for optimizing vIOMMU deviice dirty page tracking. No functional changes intended.
--- hw/vfio/common.c | 63 +++++++++++++++++++++++++++++------------------- 1 file changed, 38 insertions(+), 25 deletions(-)
hw/vfio/common.c +38 -25
@@ -1302,37 +1302,50 @@ vfio_sync_ram_discard_listener_dirty_bitmap(VFIOContainerBase *bcontainer, &vrdl); } +static int vfio_sync_iommu_dirty_bitmap (VFIOContainerBase *bcontainer, +MemoryRegionSection *section) +{ +VFIOGuestIOMMU *giommu; +bool found = false; +Int128 llend; +vfio_giommu_dirty_notifier gdn; +int idx; + + QLIST_FOREACH(giommu, &bcontainer->giommu_list, giommu_next) { +if (MEMORY_REGION (giommu->iommu_mr) == section->mr && + giommu->n.start == section->offset_within_region) { + found = true; +break ; + } + } + +if (!found) { +return 0 ; + } + + gdn.giommu = giommu; + idx =memory_region_iommu_attrs_to_index (giommu->iommu_mr, +MEMTXATTRS_UNSPECIFIED ); + + llend =int128_add (int128_make64 (section->offset_within_region), + section->size); + llend =int128_sub (llend,int128_one ()); + +iommu_notifier_init (&gdn.n, vfio_iommu_map_dirty_notify,IOMMU_NOTIFIER_MAP , + section->offset_within_region,int128_get64 (llend), + idx); +memory_region_iommu_replay (giommu->iommu_mr, &gdn.n); + +return 0 ; +} +static int vfio_sync_dirty_bitmap (VFIOContainerBase *bcontainer,MemoryRegionSection *section,Error **errp) {ram_addr_t ram_addr;if (memory_region_is_iommu (section->mr)) { -VFIOGuestIOMMU *giommu; - - QLIST_FOREACH(giommu, &bcontainer->giommu_list, giommu_next) { -if (MEMORY_REGION (giommu->iommu_mr) == section->mr && - giommu->n.start == section->offset_within_region) { -Int128 llend; -vfio_giommu_dirty_notifier gdn = { .giommu = giommu }; -int idx =memory_region_iommu_attrs_to_index (giommu->iommu_mr, -MEMTXATTRS_UNSPECIFIED ); - - llend =int128_add (int128_make64 (section->offset_within_region), - section->size); - llend =int128_sub (llend,int128_one ()); - -iommu_notifier_init (&gdn.n, - vfio_iommu_map_dirty_notify, -IOMMU_NOTIFIER_MAP , - section->offset_within_region, -int128_get64 (llend), - idx); -memory_region_iommu_replay (giommu->iommu_mr, &gdn.n); -break ; - } - } -return 0 ; +return vfio_sync_iommu_dirty_bitmap (bcontainer, section); }else if (memory_region_has_ram_discard_manager (section->mr)) {int ret;
Signature · 1 line
2.45.2
Re: [PATCH v2 04/17] vfio/common: Extract vIOMMU code from vfio_sync_dirty_bitmap()
Hi Cédric,
Quoted 6 lines · Cédric Le Goater
From: Avihai Horon <[email protected]>
Extract vIOMMU code from
vfio_sync_dirty_bitmap()to a new function and restructure the code.This is done in preparation for optimizing vIOMMU deviice dirty page
device
Quoted 44 lines
tracking. No functional changes intended.
Signed-off-by Avihai Horon <[email protected]>Signed-off-by Joao Martins <[email protected]>[ clg: - Rebased on upstream ]Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>--- hw/vfio/common.c | 63 +++++++++++++++++++++++++++++------------------- 1 file changed, 38 insertions(+), 25 deletions(-)hw/vfio/common.c +25
-0@@ -1302,37 +1302,50 @@ vfio_sync_ram_discard_listener_dirty_bitmap(VFIOContainerBase *bcontainer, &vrdl); } +static int vfio_sync_iommu_dirty_bitmap (VFIOContainerBase *bcontainer, +MemoryRegionSection *section) +{ +VFIOGuestIOMMU *giommu; +bool found = false; +Int128 llend; +vfio_giommu_dirty_notifier gdn; +int idx; + + QLIST_FOREACH(giommu, &bcontainer->giommu_list, giommu_next) { +if (MEMORY_REGION (giommu->iommu_mr) == section->mr && + giommu->n.start == section->offset_within_region) { + found = true; +break ; + } + } + +if (!found) { +return 0 ; + } + + gdn.giommu = giommu; + idx =memory_region_iommu_attrs_to_index (giommu->iommu_mr, +MEMTXATTRS_UNSPECIFIED ); +
hw/vfio/common.c vfio_sync_ram_discard_listener_dirty_bitmap(VFIOContainerBase *bcontainer, · line 1339
Quoted 10 lines
hw/vfio/common.c +10
-0+ llend =int128_add (int128_make64 (section->offset_within_region), + section->size); + llend =int128_sub (llend,int128_one ()); + +iommu_notifier_init (&gdn.n, vfio_iommu_map_dirty_notify,IOMMU_NOTIFIER_MAP , + section->offset_within_region,int128_get64 (llend), + idx); +memory_region_iommu_replay (giommu->iommu_mr, &gdn.n); + +return 0 ;
if this does not return anything else than 0, shouldn't it be void?
Thanks
Eric
Quoted 37 lines
hw/vfio/common.c +3
-25+} +static int vfio_sync_dirty_bitmap (VFIOContainerBase *bcontainer,MemoryRegionSection *section,Error **errp) {ram_addr_t ram_addr;if (memory_region_is_iommu (section->mr)) { -VFIOGuestIOMMU *giommu; - - QLIST_FOREACH(giommu, &bcontainer->giommu_list, giommu_next) { -if (MEMORY_REGION (giommu->iommu_mr) == section->mr && - giommu->n.start == section->offset_within_region) { -Int128 llend; -vfio_giommu_dirty_notifier gdn = { .giommu = giommu }; -int idx =memory_region_iommu_attrs_to_index (giommu->iommu_mr, -MEMTXATTRS_UNSPECIFIED ); - - llend =int128_add (int128_make64 (section->offset_within_region), - section->size); - llend =int128_sub (llend,int128_one ()); - -iommu_notifier_init (&gdn.n, - vfio_iommu_map_dirty_notify, -IOMMU_NOTIFIER_MAP , - section->offset_within_region, -int128_get64 (llend), - idx); -memory_region_iommu_replay (giommu->iommu_mr, &gdn.n); -break ; - } - } -return 0 ; +return vfio_sync_iommu_dirty_bitmap (bcontainer, section); }else if (memory_region_has_ram_discard_manager (section->mr)) {int ret;
Re: [PATCH v2 04/17] vfio/common: Extract vIOMMU code from vfio_sync_dirty_bitmap()
Quoted 10 lines · Eric Auger
Hi Cédric,
Quoted 6 lines · Cédric Le Goater
From: Avihai Horon <[email protected]>
Extract vIOMMU code from
vfio_sync_dirty_bitmap()to a new function and restructure the code.This is done in preparation for optimizing vIOMMU deviice dirty page
device
fixed.
Quoted 47 lines
Quoted 47 lines
tracking. No functional changes intended.
Signed-off-by Avihai Horon <[email protected]>Signed-off-by Joao Martins <[email protected]>[ clg: - Rebased on upstream ]Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>--- hw/vfio/common.c | 63 +++++++++++++++++++++++++++++------------------- 1 file changed, 38 insertions(+), 25 deletions(-)hw/vfio/common.c +28
-0@@ -1302,37 +1302,50 @@ vfio_sync_ram_discard_listener_dirty_bitmap(VFIOContainerBase *bcontainer, &vrdl); } +static int vfio_sync_iommu_dirty_bitmap (VFIOContainerBase *bcontainer, +MemoryRegionSection *section) +{ +VFIOGuestIOMMU *giommu; +bool found = false; +Int128 llend; +vfio_giommu_dirty_notifier gdn; +int idx; + + QLIST_FOREACH(giommu, &bcontainer->giommu_list, giommu_next) { +if (MEMORY_REGION (giommu->iommu_mr) == section->mr && + giommu->n.start == section->offset_within_region) { + found = true; +break ; + } + } + +if (!found) { +return 0 ; + } + + gdn.giommu = giommu; + idx =memory_region_iommu_attrs_to_index (giommu->iommu_mr, +MEMTXATTRS_UNSPECIFIED ); + + llend =int128_add (int128_make64 (section->offset_within_region), + section->size); + llend =int128_sub (llend,int128_one ());
Quoted 8 lines
hw/vfio/common.c vfio_sync_ram_discard_listener_dirty_bitmap(VFIOContainerBase *bcontainer, · line 1339
Quoted 7 lines
hw/vfio/common.c +7
-0+ +iommu_notifier_init (&gdn.n, vfio_iommu_map_dirty_notify,IOMMU_NOTIFIER_MAP , + section->offset_within_region,int128_get64 (llend), + idx); +memory_region_iommu_replay (giommu->iommu_mr, &gdn.n); + +return 0 ;if this does not return anything else than 0, shouldn't it be void?
Yes it could. The return value is practical for the caller below though. Let's keep it that way for now, it's harmless.
Thanks,
C.
Quoted 42 lines
Thanks
Eric
Quoted 37 lines
hw/vfio/common.c +3
-25+} +static int vfio_sync_dirty_bitmap (VFIOContainerBase *bcontainer,MemoryRegionSection *section,Error **errp) {ram_addr_t ram_addr;if (memory_region_is_iommu (section->mr)) { -VFIOGuestIOMMU *giommu; - - QLIST_FOREACH(giommu, &bcontainer->giommu_list, giommu_next) { -if (MEMORY_REGION (giommu->iommu_mr) == section->mr && - giommu->n.start == section->offset_within_region) { -Int128 llend; -vfio_giommu_dirty_notifier gdn = { .giommu = giommu }; -int idx =memory_region_iommu_attrs_to_index (giommu->iommu_mr, -MEMTXATTRS_UNSPECIFIED ); - - llend =int128_add (int128_make64 (section->offset_within_region), - section->size); - llend =int128_sub (llend,int128_one ()); - -iommu_notifier_init (&gdn.n, - vfio_iommu_map_dirty_notify, -IOMMU_NOTIFIER_MAP , - section->offset_within_region, -int128_get64 (llend), - idx); -memory_region_iommu_replay (giommu->iommu_mr, &gdn.n); -break ; - } - } -return 0 ; +return vfio_sync_iommu_dirty_bitmap (bcontainer, section); }else if (memory_region_has_ram_discard_manager (section->mr)) {int ret;
PATCH v2 5/17 vfio/container: Introduce vfio_address_space_insert()
In reply to Cédric Le GoaterIt will ease future changes.
--- include/hw/vfio/vfio-common.h | 2 ++ hw/vfio/common.c | 6 ++++++ hw/vfio/container.c | 2 +- hw/vfio/iommufd.c | 2 +- 4 files changed, 10 insertions(+), 2 deletions(-)
include/hw/vfio/vfio-common.h +2 -0
@@ -206,6 +206,8 @@ typedef struct VFIODisplay {VFIOAddressSpace *vfio_get_address_space (AddressSpace *as);void vfio_put_address_space (VFIOAddressSpace *space); +void vfio_address_space_insert (VFIOAddressSpace *space, +VFIOContainerBase *bcontainer);void vfio_disable_irqindex (VFIODevice *vbasedev,int index);void vfio_unmask_single_irqindex (VFIODevice *vbasedev,int index);
hw/vfio/common.c +6 -0
@@ -1508,6 +1508,12 @@ void vfio_put_address_space(VFIOAddressSpace *space) } } +void vfio_address_space_insert (VFIOAddressSpace *space, +VFIOContainerBase *bcontainer) +{ +QLIST_INSERT_HEAD (&space->containers, bcontainer, next); +} +struct vfio_device_info *vfio_get_device_info (int fd) {struct vfio_device_info *info;
hw/vfio/container.c +1 -1
@@ -637,7 +637,7 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as,vfio_kvm_device_add_group (group);QLIST_INIT (&container->group_list); -QLIST_INSERT_HEAD (&space->containers, bcontainer, next); +vfio_address_space_insert (space, bcontainer); group->container = container;QLIST_INSERT_HEAD (&container->group_list, group, container_next);
hw/vfio/iommufd.c +1 -1
@@ -358,7 +358,7 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev, bcontainer = &container->bcontainer;vfio_container_init (bcontainer, space, iommufd_vioc); -QLIST_INSERT_HEAD (&space->containers, bcontainer, next); +vfio_address_space_insert (space, bcontainer);if (!iommufd_cdev_attach_container (vbasedev, container, errp)) { goto err_attach_container;
Signature · 1 line
2.45.2
Re: [PATCH v2 05/17] vfio/container: Introduce vfio_address_space_insert()
Quoted 1 line · Cédric Le Goater
It will ease future changes.
Does it, really?
Besides
Eric
Quoted 66 lines
Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>--- include/hw/vfio/vfio-common.h | 2 ++ hw/vfio/common.c | 6 ++++++ hw/vfio/container.c | 2 +- hw/vfio/iommufd.c | 2 +- 4 files changed, 10 insertions(+), 2 deletions(-)include/hw/vfio/vfio-common.h +2
-0@@ -206,6 +206,8 @@ typedef struct VFIODisplay {VFIOAddressSpace *vfio_get_address_space (AddressSpace *as);void vfio_put_address_space (VFIOAddressSpace *space); +void vfio_address_space_insert (VFIOAddressSpace *space, +VFIOContainerBase *bcontainer);void vfio_disable_irqindex (VFIODevice *vbasedev,int index);void vfio_unmask_single_irqindex (VFIODevice *vbasedev,int index);hw/vfio/common.c +6
-0@@ -1508,6 +1508,12 @@ void vfio_put_address_space(VFIOAddressSpace *space) } } +void vfio_address_space_insert (VFIOAddressSpace *space, +VFIOContainerBase *bcontainer) +{ +QLIST_INSERT_HEAD (&space->containers, bcontainer, next); +} +struct vfio_device_info *vfio_get_device_info (int fd) {struct vfio_device_info *info;hw/vfio/container.c +1
-1@@ -637,7 +637,7 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as,vfio_kvm_device_add_group (group);QLIST_INIT (&container->group_list); -QLIST_INSERT_HEAD (&space->containers, bcontainer, next); +vfio_address_space_insert (space, bcontainer); group->container = container;QLIST_INSERT_HEAD (&container->group_list, group, container_next);hw/vfio/iommufd.c +1
-1@@ -358,7 +358,7 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev, bcontainer = &container->bcontainer;vfio_container_init (bcontainer, space, iommufd_vioc); -QLIST_INSERT_HEAD (&space->containers, bcontainer, next); +vfio_address_space_insert (space, bcontainer);if (!iommufd_cdev_attach_container (vbasedev, container, errp)) { goto err_attach_container;
Re: [PATCH v2 05/17] vfio/container: Introduce vfio_address_space_insert()
Quoted 5 lines · Eric Auger
Quoted 1 line · Cédric Le Goater
It will ease future changes.
Does it, really?
Changed to :
It prepares gound for a future change initializing the 'space' pointer of VFIOContainerBase. The goal is to replace vfio_container_init() by an .instance_init() handler when VFIOContainerBase is QOMified.
Thanks,
C.
Quoted 72 lines
Besides
Reviewed-by Eric Auger <[email protected]>Eric
Quoted 66 lines
Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>--- include/hw/vfio/vfio-common.h | 2 ++ hw/vfio/common.c | 6 ++++++ hw/vfio/container.c | 2 +- hw/vfio/iommufd.c | 2 +- 4 files changed, 10 insertions(+), 2 deletions(-)include/hw/vfio/vfio-common.h +2
-0@@ -206,6 +206,8 @@ typedef struct VFIODisplay {VFIOAddressSpace *vfio_get_address_space (AddressSpace *as);void vfio_put_address_space (VFIOAddressSpace *space); +void vfio_address_space_insert (VFIOAddressSpace *space, +VFIOContainerBase *bcontainer);void vfio_disable_irqindex (VFIODevice *vbasedev,int index);void vfio_unmask_single_irqindex (VFIODevice *vbasedev,int index);hw/vfio/common.c +6
-0@@ -1508,6 +1508,12 @@ void vfio_put_address_space(VFIOAddressSpace *space) } } +void vfio_address_space_insert (VFIOAddressSpace *space, +VFIOContainerBase *bcontainer) +{ +QLIST_INSERT_HEAD (&space->containers, bcontainer, next); +} +struct vfio_device_info *vfio_get_device_info (int fd) {struct vfio_device_info *info;hw/vfio/container.c +1
-1@@ -637,7 +637,7 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as,vfio_kvm_device_add_group (group);QLIST_INIT (&container->group_list); -QLIST_INSERT_HEAD (&space->containers, bcontainer, next); +vfio_address_space_insert (space, bcontainer); group->container = container;QLIST_INSERT_HEAD (&container->group_list, group, container_next);hw/vfio/iommufd.c +1
-1@@ -358,7 +358,7 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev, bcontainer = &container->bcontainer;vfio_container_init (bcontainer, space, iommufd_vioc); -QLIST_INSERT_HEAD (&space->containers, bcontainer, next); +vfio_address_space_insert (space, bcontainer);if (!iommufd_cdev_attach_container (vbasedev, container, errp)) { goto err_attach_container;
PATCH v2 6/17 vfio/container: Simplify vfio_container_init()
In reply to Cédric Le GoaterAssign the base container VFIOAddressSpace 'space' pointer in vfio_address_space_insert().
To be noted that vfio_connect_container() will assign the 'space' pointer later in the execution flow. This should not have any consequence.
--- include/hw/vfio/vfio-container-base.h | 1 - hw/vfio/common.c | 1 + hw/vfio/container-base.c | 3 +-- hw/vfio/container.c | 6 +++--- hw/vfio/iommufd.c | 2 +- 5 files changed, 6 insertions(+), 7 deletions(-)
include/hw/vfio/vfio-container-base.h +0 -1
@@ -87,7 +87,6 @@ int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer, VFIOBitmap *vbmap, hwaddr iova, hwaddr size, Error **errp);void vfio_container_init (VFIOContainerBase *bcontainer, -VFIOAddressSpace *space,const VFIOIOMMUClass *ops);void vfio_container_destroy (VFIOContainerBase *bcontainer);
hw/vfio/common.c +1 -0
@@ -1512,6 +1512,7 @@ void vfio_address_space_insert(VFIOAddressSpace *space, VFIOContainerBase *bcontainer) {QLIST_INSERT_HEAD (&space->containers, bcontainer, next); + bcontainer->space = space; }struct vfio_device_info *vfio_get_device_info (int fd)
hw/vfio/container-base.c +1 -2
@@ -71,11 +71,10 @@ int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer, errp); } -void vfio_container_init (VFIOContainerBase *bcontainer,VFIOAddressSpace *space, +void vfio_container_init (VFIOContainerBase *bcontainer,const VFIOIOMMUClass *ops) { bcontainer->ops = ops; - bcontainer->space = space; bcontainer->error =NULL ; bcontainer->dirty_pages_supported = false; bcontainer->dma_max_mappings =0 ;
hw/vfio/container.c +3 -3
@@ -394,7 +394,7 @@ static const VFIOIOMMUClass *vfio_get_iommu_class(int iommu_type, Error **errp) }static bool vfio_set_iommu (VFIOContainer *container,int group_fd, -VFIOAddressSpace *space,Error **errp) +Error **errp) {int iommu_type;const VFIOIOMMUClass *vioc;
@@ -432,7 +432,7 @@ static bool vfio_set_iommu(VFIOContainer *container, int group_fd,return false; } -vfio_container_init (&container->bcontainer, space, vioc); +vfio_container_init (&container->bcontainer, vioc);return true; }
@@ -614,7 +614,7 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as, container->fd = fd; bcontainer = &container->bcontainer; -if (!vfio_set_iommu (container, group->fd, space, errp)) { +if (!vfio_set_iommu (container, group->fd, errp)) { goto free_container_exit; }
hw/vfio/iommufd.c +1 -1
@@ -357,7 +357,7 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev, container->ioas_id = ioas_id; bcontainer = &container->bcontainer; -vfio_container_init (bcontainer, space, iommufd_vioc); +vfio_container_init (bcontainer, iommufd_vioc);vfio_address_space_insert (space, bcontainer);if (!iommufd_cdev_attach_container (vbasedev, container, errp)) {
Signature · 1 line
2.45.2
Re: [PATCH v2 06/17] vfio/container: Simplify vfio_container_init()
Hi Cédric,
Quoted 2 lines · Cédric Le Goater
Assign the base container VFIOAddressSpace 'space' pointer in
vfio_address_space_insert().
OK I get it now. Maybe in the previous patch, say that the
vfio_address_space_insert() will be enhanced with additional initializations.
Quoted 4 lines
To be noted that
vfio_connect_container()will assign the 'space' pointer later in the execution flow. This should not have any consequence.
You do not explain the motivation of that change.
Quoted 88 lines
Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>--- include/hw/vfio/vfio-container-base.h | 1 - hw/vfio/common.c | 1 + hw/vfio/container-base.c | 3 +-- hw/vfio/container.c | 6 +++--- hw/vfio/iommufd.c | 2 +- 5 files changed, 6 insertions(+), 7 deletions(-)include/hw/vfio/vfio-container-base.h +0
-1@@ -87,7 +87,6 @@ int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer, VFIOBitmap *vbmap, hwaddr iova, hwaddr size, Error **errp);void vfio_container_init (VFIOContainerBase *bcontainer, -VFIOAddressSpace *space,const VFIOIOMMUClass *ops);void vfio_container_destroy (VFIOContainerBase *bcontainer);hw/vfio/common.c +1
-0@@ -1512,6 +1512,7 @@ void vfio_address_space_insert(VFIOAddressSpace *space, VFIOContainerBase *bcontainer) {QLIST_INSERT_HEAD (&space->containers, bcontainer, next); + bcontainer->space = space; }struct vfio_device_info *vfio_get_device_info (int fd)hw/vfio/container-base.c +1
-2@@ -71,11 +71,10 @@ int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer, errp); } -void vfio_container_init (VFIOContainerBase *bcontainer,VFIOAddressSpace *space, +void vfio_container_init (VFIOContainerBase *bcontainer,const VFIOIOMMUClass *ops) { bcontainer->ops = ops; - bcontainer->space = space; bcontainer->error =NULL ; bcontainer->dirty_pages_supported = false; bcontainer->dma_max_mappings =0 ;hw/vfio/container.c +3
-3@@ -394,7 +394,7 @@ static const VFIOIOMMUClass *vfio_get_iommu_class(int iommu_type, Error **errp) }static bool vfio_set_iommu (VFIOContainer *container,int group_fd, -VFIOAddressSpace *space,Error **errp) +Error **errp) {int iommu_type;const VFIOIOMMUClass *vioc;@@ -432,7 +432,7 @@ static bool vfio_set_iommu(VFIOContainer *container, int group_fd,return false; } -vfio_container_init (&container->bcontainer, space, vioc); +vfio_container_init (&container->bcontainer, vioc);return true; }@@ -614,7 +614,7 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as, container->fd = fd; bcontainer = &container->bcontainer; -if (!vfio_set_iommu (container, group->fd, space, errp)) { +if (!vfio_set_iommu (container, group->fd, errp)) { goto free_container_exit; }hw/vfio/iommufd.c +0
-0
hw/vfio/iommufd.c static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev, · line 363
Quoted 8 lines
hw/vfio/iommufd.c +1
-1@@ -357,7 +357,7 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev, container->ioas_id = ioas_id; bcontainer = &container->bcontainer; -vfio_container_init (bcontainer, space, iommufd_vioc); +vfio_container_init (bcontainer, iommufd_vioc);vfio_address_space_insert (space, bcontainer);if (!iommufd_cdev_attach_container (vbasedev, container, errp)) {
Thanks
Eric
Re: [PATCH v2 06/17] vfio/container: Simplify vfio_container_init()
Quoted 14 lines · Eric Auger
Hi Cédric,
Quoted 2 lines · Cédric Le Goater
Assign the base container VFIOAddressSpace 'space' pointer in
vfio_address_space_insert().OK I get it now. Maybe in the previous patch, say that the
vfio_address_space_insert()will be enhanced with additional initializations.Quoted 1 line
Quoted 5 lines
Quoted 3 lines
To be noted that
vfio_connect_container()will assign the 'space' pointer later in the execution flow. This should not have any consequence.You do not explain the motivation of that change.
Changed to:
Assign the base container VFIOAddressSpace 'space' pointer in vfio_address_space_insert(). The ultimate goal is to remove vfio_container_init() and instead rely on an .instance_init() handler to perfom the initialization of VFIOContainerBase.
To be noted that vfio_connect_container() will assign the 'space' pointer later in the execution flow. This should not have any consequence.
Thanks,
C.
Quoted 101 lines
Quoted 89 lines
Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>--- include/hw/vfio/vfio-container-base.h | 1 - hw/vfio/common.c | 1 + hw/vfio/container-base.c | 3 +-- hw/vfio/container.c | 6 +++--- hw/vfio/iommufd.c | 2 +- 5 files changed, 6 insertions(+), 7 deletions(-)include/hw/vfio/vfio-container-base.h +0
-1@@ -87,7 +87,6 @@ int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer, VFIOBitmap *vbmap, hwaddr iova, hwaddr size, Error **errp);void vfio_container_init (VFIOContainerBase *bcontainer, -VFIOAddressSpace *space,const VFIOIOMMUClass *ops);void vfio_container_destroy (VFIOContainerBase *bcontainer);hw/vfio/common.c +1
-0@@ -1512,6 +1512,7 @@ void vfio_address_space_insert(VFIOAddressSpace *space, VFIOContainerBase *bcontainer) {QLIST_INSERT_HEAD (&space->containers, bcontainer, next); + bcontainer->space = space; }struct vfio_device_info *vfio_get_device_info (int fd)hw/vfio/container-base.c +1
-2@@ -71,11 +71,10 @@ int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer, errp); } -void vfio_container_init (VFIOContainerBase *bcontainer,VFIOAddressSpace *space, +void vfio_container_init (VFIOContainerBase *bcontainer,const VFIOIOMMUClass *ops) { bcontainer->ops = ops; - bcontainer->space = space; bcontainer->error =NULL ; bcontainer->dirty_pages_supported = false; bcontainer->dma_max_mappings =0 ;hw/vfio/container.c +3
-3@@ -394,7 +394,7 @@ static const VFIOIOMMUClass *vfio_get_iommu_class(int iommu_type, Error **errp) }static bool vfio_set_iommu (VFIOContainer *container,int group_fd, -VFIOAddressSpace *space,Error **errp) +Error **errp) {int iommu_type;const VFIOIOMMUClass *vioc;@@ -432,7 +432,7 @@ static bool vfio_set_iommu(VFIOContainer *container, int group_fd,return false; } -vfio_container_init (&container->bcontainer, space, vioc); +vfio_container_init (&container->bcontainer, vioc);return true; }@@ -614,7 +614,7 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as, container->fd = fd; bcontainer = &container->bcontainer; -if (!vfio_set_iommu (container, group->fd, space, errp)) { +if (!vfio_set_iommu (container, group->fd, errp)) { goto free_container_exit; }hw/vfio/iommufd.c +0
-0hw/vfio/iommufd.c static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev, · line 363
Quoted 8 lines
hw/vfio/iommufd.c +1
-1@@ -357,7 +357,7 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev, container->ioas_id = ioas_id; bcontainer = &container->bcontainer; -vfio_container_init (bcontainer, space, iommufd_vioc); +vfio_container_init (bcontainer, iommufd_vioc);vfio_address_space_insert (space, bcontainer);if (!iommufd_cdev_attach_container (vbasedev, container, errp)) {Thanks
Eric
PATCH v2 7/17 vfio/container: Modify vfio_get_iommu_type() to use a container fd
In reply to Cédric Le GoaterThe 'container' pointer has no other use than its 'fd' attribute. Simplify the prototype to ease future changes.
--- hw/vfio/container.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
hw/vfio/container.c +3 -3
@@ -354,7 +354,7 @@ static void vfio_kvm_device_del_group(VFIOGroup *group)/* * vfio_get_iommu_type - selects the richest iommu_type (v2 first) */ -static int vfio_get_iommu_type (VFIOContainer *container, +static int vfio_get_iommu_type (int container_fd,Error **errp) {int iommu_types[] = { VFIO_TYPE1v2_IOMMU,VFIO_TYPE1_IOMMU ,
@@ -362,7 +362,7 @@ static int vfio_get_iommu_type(VFIOContainer *container,int i;for (i =0 ; i <ARRAY_SIZE (iommu_types); i++) { -if (ioctl (container->fd,VFIO_CHECK_EXTENSION , iommu_types[i])) { +if (ioctl (container_fd,VFIO_CHECK_EXTENSION , iommu_types[i])) {return iommu_types[i]; } }
@@ -399,7 +399,7 @@ static bool vfio_set_iommu(VFIOContainer *container, int group_fd,int iommu_type;const VFIOIOMMUClass *vioc; - iommu_type =vfio_get_iommu_type (container, errp); + iommu_type =vfio_get_iommu_type (container->fd, errp);if (iommu_type <0 ) {return false; }
Signature · 1 line
2.45.2
Re: [PATCH v2 07/17] vfio/container: Modify vfio_get_iommu_type() to use a container fd
Quoted 5 lines · Cédric Le Goater
The 'container' pointer has no other use than its 'fd' attribute. Simplify the prototype to ease future changes.
Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>
Eric
Quoted 35 lines
--- hw/vfio/container.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)hw/vfio/container.c +3
-3@@ -354,7 +354,7 @@ static void vfio_kvm_device_del_group(VFIOGroup *group)/* * vfio_get_iommu_type - selects the richest iommu_type (v2 first) */ -static int vfio_get_iommu_type (VFIOContainer *container, +static int vfio_get_iommu_type (int container_fd,Error **errp) {int iommu_types[] = { VFIO_TYPE1v2_IOMMU,VFIO_TYPE1_IOMMU ,@@ -362,7 +362,7 @@ static int vfio_get_iommu_type(VFIOContainer *container,int i;for (i =0 ; i <ARRAY_SIZE (iommu_types); i++) { -if (ioctl (container->fd,VFIO_CHECK_EXTENSION , iommu_types[i])) { +if (ioctl (container_fd,VFIO_CHECK_EXTENSION , iommu_types[i])) {return iommu_types[i]; } }@@ -399,7 +399,7 @@ static bool vfio_set_iommu(VFIOContainer *container, int group_fd,int iommu_type;const VFIOIOMMUClass *vioc; - iommu_type =vfio_get_iommu_type (container, errp); + iommu_type =vfio_get_iommu_type (container->fd, errp);if (iommu_type <0 ) {return false; }
PATCH v2 8/17 vfio/container: Introduce vfio_get_iommu_class_name()
In reply to Cédric Le GoaterRework vfio_get_iommu_class() to return a literal class name instead of a class object. We will need this name to instantiate the object later on. Since the default case asserts, remove the error report as QEMU will simply abort before.
--- hw/vfio/container.c | 18 ++++++------------ 1 file changed, 6 insertions(+), 12 deletions(-)
hw/vfio/container.c +6 -12
@@ -373,24 +373,20 @@ static int vfio_get_iommu_type(int container_fd,/* * vfio_get_iommu_ops - get a VFIOIOMMUClass associated with a type */ -static const VFIOIOMMUClass *vfio_get_iommu_class (int iommu_type,Error **errp) +static const char *vfio_get_iommu_class_name (int iommu_type) { -ObjectClass *klass =NULL ; -switch (iommu_type) {case VFIO_TYPE1v2_IOMMU:case VFIO_TYPE1_IOMMU : - klass =object_class_by_name (TYPE_VFIO_IOMMU_LEGACY ); +return TYPE_VFIO_IOMMU_LEGACY ;break ;case VFIO_SPAPR_TCE_v2_IOMMU:case VFIO_SPAPR_TCE_IOMMU : - klass =object_class_by_name (TYPE_VFIO_IOMMU_SPAPR ); +return TYPE_VFIO_IOMMU_SPAPR ;break ;default :g_assert_not_reached (); }; - -return VFIO_IOMMU_CLASS (klass); } static bool vfio_set_iommu(VFIOContainer *container, int group_fd,
@@ -398,6 +394,7 @@ static bool vfio_set_iommu(VFIOContainer *container, int group_fd, {int iommu_type;const VFIOIOMMUClass *vioc; +const char *vioc_name; iommu_type =vfio_get_iommu_type (container->fd, errp);if (iommu_type <0 ) {
@@ -426,11 +423,8 @@ static bool vfio_set_iommu(VFIOContainer *container, int group_fd, container->iommu_type = iommu_type; - vioc =vfio_get_iommu_class (iommu_type, errp); -if (!vioc) { -error_setg (errp,"No available IOMMU models" ); -return false; - } + vioc_name =vfio_get_iommu_class_name (iommu_type); + vioc =VFIO_IOMMU_CLASS (object_class_by_name (vioc_name));vfio_container_init (&container->bcontainer, vioc);return true;
Signature · 1 line
2.45.2
Re: [PATCH v2 08/17] vfio/container: Introduce vfio_get_iommu_class_name()
Quoted 7 lines · Cédric Le Goater
Rework
vfio_get_iommu_class()to return a literal class name instead of a class object. We will need this name to instantiate the object later on. Since the default case asserts, remove the error report as QEMU will simply abort before.Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>
Eric
Quoted 58 lines
--- hw/vfio/container.c | 18 ++++++------------ 1 file changed, 6 insertions(+), 12 deletions(-)hw/vfio/container.c +6
-12@@ -373,24 +373,20 @@ static int vfio_get_iommu_type(int container_fd,/* * vfio_get_iommu_ops - get a VFIOIOMMUClass associated with a type */ -static const VFIOIOMMUClass *vfio_get_iommu_class (int iommu_type,Error **errp) +static const char *vfio_get_iommu_class_name (int iommu_type) { -ObjectClass *klass =NULL ; -switch (iommu_type) {case VFIO_TYPE1v2_IOMMU:case VFIO_TYPE1_IOMMU : - klass =object_class_by_name (TYPE_VFIO_IOMMU_LEGACY ); +return TYPE_VFIO_IOMMU_LEGACY ;break ;case VFIO_SPAPR_TCE_v2_IOMMU:case VFIO_SPAPR_TCE_IOMMU : - klass =object_class_by_name (TYPE_VFIO_IOMMU_SPAPR ); +return TYPE_VFIO_IOMMU_SPAPR ;break ;default :g_assert_not_reached (); }; - -return VFIO_IOMMU_CLASS (klass); } static bool vfio_set_iommu(VFIOContainer *container, int group_fd,@@ -398,6 +394,7 @@ static bool vfio_set_iommu(VFIOContainer *container, int group_fd, {int iommu_type;const VFIOIOMMUClass *vioc; +const char *vioc_name; iommu_type =vfio_get_iommu_type (container->fd, errp);if (iommu_type <0 ) {@@ -426,11 +423,8 @@ static bool vfio_set_iommu(VFIOContainer *container, int group_fd, container->iommu_type = iommu_type; - vioc =vfio_get_iommu_class (iommu_type, errp); -if (!vioc) { -error_setg (errp,"No available IOMMU models" ); -return false; - } + vioc_name =vfio_get_iommu_class_name (iommu_type); + vioc =VFIO_IOMMU_CLASS (object_class_by_name (vioc_name));vfio_container_init (&container->bcontainer, vioc);return true;
PATCH v2 9/17 vfio/container: Introduce vfio_create_container()
In reply to Cédric Le GoaterThis routine allocates the QEMU struct type representing the VFIO container. It is minimal currently and future changes will do more initialization.
--- hw/vfio/container.c | 19 +++++++++++++++---- 1 file changed, 15 insertions(+), 4 deletions(-)
hw/vfio/container.c +15 -4
@@ -430,6 +430,16 @@ static bool vfio_set_iommu(VFIOContainer *container, int group_fd,return true; } +static VFIOContainer *vfio_create_container (int fd,VFIOGroup *group, +Error **errp) +{ +VFIOContainer *container; + + container =g_malloc0 (sizeof (*container)); + container->fd = fd; +return container; +} +static int vfio_get_iommu_info (VFIOContainer *container,struct vfio_iommu_type1_info **info) {
@@ -604,13 +614,14 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as, goto close_fd_exit; } - container =g_malloc0 (sizeof (*container)); - container->fd = fd; - bcontainer = &container->bcontainer; - + container =vfio_create_container (fd, group, errp); +if (!container) { + goto close_fd_exit; + }if (!vfio_set_iommu (container, group->fd, errp)) { goto free_container_exit; } + bcontainer = &container->bcontainer;if (!vfio_cpr_register_container (bcontainer, errp)) { goto free_container_exit;
Signature · 1 line
2.45.2
Re: [PATCH v2 09/17] vfio/container: Introduce vfio_create_container()
Quoted 6 lines · Cédric Le Goater
This routine allocates the QEMU struct type representing the VFIO container. It is minimal currently and future changes will do more initialization.
Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>
Eric
Quoted 44 lines
--- hw/vfio/container.c | 19 +++++++++++++++---- 1 file changed, 15 insertions(+), 4 deletions(-)hw/vfio/container.c +15
-4@@ -430,6 +430,16 @@ static bool vfio_set_iommu(VFIOContainer *container, int group_fd,return true; } +static VFIOContainer *vfio_create_container (int fd,VFIOGroup *group, +Error **errp) +{ +VFIOContainer *container; + + container =g_malloc0 (sizeof (*container)); + container->fd = fd; +return container; +} +static int vfio_get_iommu_info (VFIOContainer *container,struct vfio_iommu_type1_info **info) {@@ -604,13 +614,14 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as, goto close_fd_exit; } - container =g_malloc0 (sizeof (*container)); - container->fd = fd; - bcontainer = &container->bcontainer; - + container =vfio_create_container (fd, group, errp); +if (!container) { + goto close_fd_exit; + }if (!vfio_set_iommu (container, group->fd, errp)) { goto free_container_exit; } + bcontainer = &container->bcontainer;if (!vfio_cpr_register_container (bcontainer, errp)) { goto free_container_exit;
PATCH v2 10/17 vfio/container: Discover IOMMU type before creating the container
In reply to Cédric Le GoaterSince the QEMU struct type representing the VFIO container is deduced from the IOMMU type exposed by the host, this type should be well defined *before* creating the container struct. This will be necessary to instantiate a QOM object of the correct type in future changes.
Rework vfio_set_iommu() to extract the part doing the container initialization and move it under vfio_create_container().
--- hw/vfio/container.c | 47 ++++++++++++++++++++++----------------------- 1 file changed, 23 insertions(+), 24 deletions(-)
hw/vfio/container.c +23 -24
@@ -389,54 +389,56 @@ static const char *vfio_get_iommu_class_name(int iommu_type) }; } -static bool vfio_set_iommu (VFIOContainer *container,int group_fd, -Error **errp) +static bool vfio_set_iommu (int container_fd,int group_fd, +int *iommu_type,Error **errp) { -int iommu_type; -const VFIOIOMMUClass *vioc; -const char *vioc_name; - - iommu_type =vfio_get_iommu_type (container->fd, errp); -if (iommu_type <0 ) { -return false; - } - -if (ioctl (group_fd,VFIO_GROUP_SET_CONTAINER , &container->fd)) { +if (ioctl (group_fd,VFIO_GROUP_SET_CONTAINER , &container_fd)) {error_setg_errno (errp, errno,"Failed to set group container" );return false; } -while (ioctl (container->fd,VFIO_SET_IOMMU , iommu_type)) { -if (iommu_type == VFIO_SPAPR_TCE_v2_IOMMU) { +while (ioctl (container_fd,VFIO_SET_IOMMU , *iommu_type)) { +if (*iommu_type == VFIO_SPAPR_TCE_v2_IOMMU) {/* * On sPAPR, despite the IOMMU subdriver always advertises v1 and * v2, the running platform may not support v2 and there is no * way to guess it until an IOMMU group gets added to the container. * So in case it fails with v2, try v1 as a fallback. */ - iommu_type =VFIO_SPAPR_TCE_IOMMU ; + *iommu_type =VFIO_SPAPR_TCE_IOMMU ;continue ; }error_setg_errno (errp, errno,"Failed to set iommu for container" );return false; } - container->iommu_type = iommu_type; - - vioc_name =vfio_get_iommu_class_name (iommu_type); - vioc =VFIO_IOMMU_CLASS (object_class_by_name (vioc_name)); - -vfio_container_init (&container->bcontainer, vioc);return true; }static VFIOContainer *vfio_create_container (int fd,VFIOGroup *group,Error **errp) { +int iommu_type; +const VFIOIOMMUClass *vioc; +const char *vioc_name;VFIOContainer *container; + iommu_type =vfio_get_iommu_type (fd, errp); +if (iommu_type <0 ) { +return NULL ; + } + +if (!vfio_set_iommu (fd, group->fd, &iommu_type, errp)) { +return NULL ; + } + + vioc_name =vfio_get_iommu_class_name (iommu_type); + vioc =VFIO_IOMMU_CLASS (object_class_by_name (vioc_name)); + container =g_malloc0 (sizeof (*container)); container->fd = fd; + container->iommu_type = iommu_type; +vfio_container_init (&container->bcontainer, vioc);return container; }
@@ -618,9 +620,6 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as,if (!container) { goto close_fd_exit; } -if (!vfio_set_iommu (container, group->fd, errp)) { - goto free_container_exit; - } bcontainer = &container->bcontainer;if (!vfio_cpr_register_container (bcontainer, errp)) {
Signature · 1 line
2.45.2
Re: [PATCH v2 10/17] vfio/container: Discover IOMMU type before creating the container
Quoted 10 lines · Cédric Le Goater
Since the QEMU struct type representing the VFIO container is deduced from the IOMMU type exposed by the host, this type should be well defined *before* creating the container struct. This will be necessary to instantiate a QOM object of the correct type in future changes.
Rework
vfio_set_iommu()to extract the part doing the container initialization and move it undervfio_create_container().Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>
Eric
Quoted 96 lines
--- hw/vfio/container.c | 47 ++++++++++++++++++++++----------------------- 1 file changed, 23 insertions(+), 24 deletions(-)hw/vfio/container.c +23
-24@@ -389,54 +389,56 @@ static const char *vfio_get_iommu_class_name(int iommu_type) }; } -static bool vfio_set_iommu (VFIOContainer *container,int group_fd, -Error **errp) +static bool vfio_set_iommu (int container_fd,int group_fd, +int *iommu_type,Error **errp) { -int iommu_type; -const VFIOIOMMUClass *vioc; -const char *vioc_name; - - iommu_type =vfio_get_iommu_type (container->fd, errp); -if (iommu_type <0 ) { -return false; - } - -if (ioctl (group_fd,VFIO_GROUP_SET_CONTAINER , &container->fd)) { +if (ioctl (group_fd,VFIO_GROUP_SET_CONTAINER , &container_fd)) {error_setg_errno (errp, errno,"Failed to set group container" );return false; } -while (ioctl (container->fd,VFIO_SET_IOMMU , iommu_type)) { -if (iommu_type == VFIO_SPAPR_TCE_v2_IOMMU) { +while (ioctl (container_fd,VFIO_SET_IOMMU , *iommu_type)) { +if (*iommu_type == VFIO_SPAPR_TCE_v2_IOMMU) {/* * On sPAPR, despite the IOMMU subdriver always advertises v1 and * v2, the running platform may not support v2 and there is no * way to guess it until an IOMMU group gets added to the container. * So in case it fails with v2, try v1 as a fallback. */ - iommu_type =VFIO_SPAPR_TCE_IOMMU ; + *iommu_type =VFIO_SPAPR_TCE_IOMMU ;continue ; }error_setg_errno (errp, errno,"Failed to set iommu for container" );return false; } - container->iommu_type = iommu_type; - - vioc_name =vfio_get_iommu_class_name (iommu_type); - vioc =VFIO_IOMMU_CLASS (object_class_by_name (vioc_name)); - -vfio_container_init (&container->bcontainer, vioc);return true; }static VFIOContainer *vfio_create_container (int fd,VFIOGroup *group,Error **errp) { +int iommu_type; +const VFIOIOMMUClass *vioc; +const char *vioc_name;VFIOContainer *container; + iommu_type =vfio_get_iommu_type (fd, errp); +if (iommu_type <0 ) { +return NULL ; + } + +if (!vfio_set_iommu (fd, group->fd, &iommu_type, errp)) { +return NULL ; + } + + vioc_name =vfio_get_iommu_class_name (iommu_type); + vioc =VFIO_IOMMU_CLASS (object_class_by_name (vioc_name)); + container =g_malloc0 (sizeof (*container)); container->fd = fd; + container->iommu_type = iommu_type; +vfio_container_init (&container->bcontainer, vioc);return container; }@@ -618,9 +620,6 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as,if (!container) { goto close_fd_exit; } -if (!vfio_set_iommu (container, group->fd, errp)) { - goto free_container_exit; - } bcontainer = &container->bcontainer;if (!vfio_cpr_register_container (bcontainer, errp)) {
PATCH v2 11/17 vfio/container: Change VFIOContainerBase to use QOM
In reply to Cédric Le GoaterVFIOContainerBase was made a QOM interface because we believed that a QOM object would expose all the IOMMU backends to the QEMU machine and human interface. This only applies to user creatable devices or objects.
Change the VFIOContainerBase nature from interface to object and make the necessary adjustments in the VFIO_IOMMU hierarchy.
--- include/hw/vfio/vfio-common.h | 4 ++++ include/hw/vfio/vfio-container-base.h | 12 +++--------- hw/vfio/container-base.c | 4 +++- hw/vfio/container.c | 1 + hw/vfio/iommufd.c | 1 + hw/vfio/spapr.c | 3 +++ 6 files changed, 15 insertions(+), 10 deletions(-)
include/hw/vfio/vfio-common.h +4 -0
@@ -84,6 +84,8 @@ typedef struct VFIOContainer { QLIST_HEAD(, VFIOGroup) group_list; } VFIOContainer; +OBJECT_DECLARE_SIMPLE_TYPE (VFIOContainer,VFIO_IOMMU_LEGACY ); + typedef struct VFIOHostDMAWindow { hwaddr min_iova; hwaddr max_iova;
@@ -99,6 +101,8 @@ typedef struct VFIOIOMMUFDContainer {uint32_t ioas_id; } VFIOIOMMUFDContainer; +OBJECT_DECLARE_SIMPLE_TYPE (VFIOIOMMUFDContainer,VFIO_IOMMU_IOMMUFD ); +typedef struct VFIODeviceOps VFIODeviceOps ; typedef struct VFIODevice {
include/hw/vfio/vfio-container-base.h +3 -9
@@ -34,6 +34,7 @@ typedef struct VFIOAddressSpace {* This is the base object for vfio container backends */ typedef struct VFIOContainerBase { + Object parent; const VFIOIOMMUClass *ops; VFIOAddressSpace *space; MemoryListener listener;
@@ -96,17 +97,10 @@ void vfio_container_destroy(VFIOContainerBase *bcontainer);#define TYPE_VFIO_IOMMU_SPAPR TYPE_VFIO_IOMMU "-spapr"#define TYPE_VFIO_IOMMU_IOMMUFD TYPE_VFIO_IOMMU "-iommufd" -/* -* VFIOContainerBase is not an abstract QOM object because it felt -* unnecessary to expose all the IOMMU backends to the QEMU machine -* and human interface. However, we can still abstract the IOMMU -* backend handlers using a QOM interface class. This provides more -* flexibility when referencing the various implementations. -*/ -DECLARE_CLASS_CHECKERS(VFIOIOMMUClass, VFIO_IOMMU, TYPE_VFIO_IOMMU) +OBJECT_DECLARE_TYPE(VFIOContainerBase, VFIOIOMMUClass, VFIO_IOMMU)struct VFIOIOMMUClass { -InterfaceClass parent_class; +ObjectClass parent_class;/* Properties */ const char *hiod_typename;
hw/vfio/container-base.c +3 -1
@@ -102,8 +102,10 @@ void vfio_container_destroy(VFIOContainerBase *bcontainer)static const TypeInfo types[] = { { .name =TYPE_VFIO_IOMMU , - .parent =TYPE_INTERFACE , + .parent =TYPE_OBJECT , + .instance_size =sizeof (VFIOContainerBase), .class_size =sizeof (VFIOIOMMUClass), + .abstract = true, }, };
hw/vfio/container.c +1 -0
@@ -1196,6 +1196,7 @@ static const TypeInfo types[] = { { .name = TYPE_VFIO_IOMMU_LEGACY, .parent = TYPE_VFIO_IOMMU, + .instance_size = sizeof(VFIOContainer), .class_init = vfio_iommu_legacy_class_init, }, { .name = TYPE_HOST_IOMMU_DEVICE_LEGACY_VFIO,
hw/vfio/iommufd.c +1 -0
@@ -672,6 +672,7 @@ static const TypeInfo types[] = { { .name = TYPE_VFIO_IOMMU_IOMMUFD, .parent = TYPE_VFIO_IOMMU, + .instance_size = sizeof(VFIOIOMMUFDContainer), .class_init = vfio_iommu_iommufd_class_init, }, { .name = TYPE_HOST_IOMMU_DEVICE_IOMMUFD_VFIO,
hw/vfio/spapr.c +3 -0
@@ -30,6 +30,8 @@ typedef struct VFIOSpaprContainer { QLIST_HEAD(, VFIOHostDMAWindow) hostwin_list; } VFIOSpaprContainer; +OBJECT_DECLARE_SIMPLE_TYPE (VFIOSpaprContainer,VFIO_IOMMU_SPAPR ); +static bool vfio_prereg_listener_skipped_section (MemoryRegionSection *section) {if (memory_region_is_iommu (section->mr)) {
@@ -548,6 +550,7 @@ static const TypeInfo types[] = { { .name = TYPE_VFIO_IOMMU_SPAPR, .parent = TYPE_VFIO_IOMMU_LEGACY, + .instance_size = sizeof(VFIOSpaprContainer), .class_init = vfio_iommu_spapr_class_init, }, };
Signature · 1 line
2.45.2
Re: [PATCH v2 11/17] vfio/container: Change VFIOContainerBase to use QOM
Quoted 9 lines · Cédric Le Goater
VFIOContainerBase was made a QOM interface because we believed that a QOM object would expose all the IOMMU backends to the QEMU machine and human interface. This only applies to user creatable devices or objects.
Change the VFIOContainerBase nature from interface to object and make the necessary adjustments in the
VFIO_IOMMUhierarchy.Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>
Eric
Quoted 124 lines
--- include/hw/vfio/vfio-common.h | 4 ++++ include/hw/vfio/vfio-container-base.h | 12 +++--------- hw/vfio/container-base.c | 4 +++- hw/vfio/container.c | 1 + hw/vfio/iommufd.c | 1 + hw/vfio/spapr.c | 3 +++ 6 files changed, 15 insertions(+), 10 deletions(-)include/hw/vfio/vfio-common.h +4
-0@@ -84,6 +84,8 @@ typedef struct VFIOContainer { QLIST_HEAD(, VFIOGroup) group_list; } VFIOContainer; +OBJECT_DECLARE_SIMPLE_TYPE (VFIOContainer,VFIO_IOMMU_LEGACY ); + typedef struct VFIOHostDMAWindow { hwaddr min_iova; hwaddr max_iova;@@ -99,6 +101,8 @@ typedef struct VFIOIOMMUFDContainer {uint32_t ioas_id; } VFIOIOMMUFDContainer; +OBJECT_DECLARE_SIMPLE_TYPE (VFIOIOMMUFDContainer,VFIO_IOMMU_IOMMUFD ); +typedef struct VFIODeviceOps VFIODeviceOps ; typedef struct VFIODevice {include/hw/vfio/vfio-container-base.h +3
-9@@ -34,6 +34,7 @@ typedef struct VFIOAddressSpace {* This is the base object for vfio container backends */ typedef struct VFIOContainerBase { + Object parent; const VFIOIOMMUClass *ops; VFIOAddressSpace *space; MemoryListener listener;@@ -96,17 +97,10 @@ void vfio_container_destroy(VFIOContainerBase *bcontainer);#define TYPE_VFIO_IOMMU_SPAPR TYPE_VFIO_IOMMU "-spapr"#define TYPE_VFIO_IOMMU_IOMMUFD TYPE_VFIO_IOMMU "-iommufd" -/* -* VFIOContainerBase is not an abstract QOM object because it felt -* unnecessary to expose all the IOMMU backends to the QEMU machine -* and human interface. However, we can still abstract the IOMMU -* backend handlers using a QOM interface class. This provides more -* flexibility when referencing the various implementations. -*/ -DECLARE_CLASS_CHECKERS(VFIOIOMMUClass, VFIO_IOMMU, TYPE_VFIO_IOMMU) +OBJECT_DECLARE_TYPE(VFIOContainerBase, VFIOIOMMUClass, VFIO_IOMMU)struct VFIOIOMMUClass { -InterfaceClass parent_class; +ObjectClass parent_class;/* Properties */ const char *hiod_typename;hw/vfio/container-base.c +3
-1@@ -102,8 +102,10 @@ void vfio_container_destroy(VFIOContainerBase *bcontainer)static const TypeInfo types[] = { { .name =TYPE_VFIO_IOMMU , - .parent =TYPE_INTERFACE , + .parent =TYPE_OBJECT , + .instance_size =sizeof (VFIOContainerBase), .class_size =sizeof (VFIOIOMMUClass), + .abstract = true, }, };hw/vfio/container.c +1
-0@@ -1196,6 +1196,7 @@ static const TypeInfo types[] = { { .name = TYPE_VFIO_IOMMU_LEGACY, .parent = TYPE_VFIO_IOMMU, + .instance_size = sizeof(VFIOContainer), .class_init = vfio_iommu_legacy_class_init, }, { .name = TYPE_HOST_IOMMU_DEVICE_LEGACY_VFIO,hw/vfio/iommufd.c +1
-0@@ -672,6 +672,7 @@ static const TypeInfo types[] = { { .name = TYPE_VFIO_IOMMU_IOMMUFD, .parent = TYPE_VFIO_IOMMU, + .instance_size = sizeof(VFIOIOMMUFDContainer), .class_init = vfio_iommu_iommufd_class_init, }, { .name = TYPE_HOST_IOMMU_DEVICE_IOMMUFD_VFIO,hw/vfio/spapr.c +3
-0@@ -30,6 +30,8 @@ typedef struct VFIOSpaprContainer { QLIST_HEAD(, VFIOHostDMAWindow) hostwin_list; } VFIOSpaprContainer; +OBJECT_DECLARE_SIMPLE_TYPE (VFIOSpaprContainer,VFIO_IOMMU_SPAPR ); +static bool vfio_prereg_listener_skipped_section (MemoryRegionSection *section) {if (memory_region_is_iommu (section->mr)) {@@ -548,6 +550,7 @@ static const TypeInfo types[] = { { .name = TYPE_VFIO_IOMMU_SPAPR, .parent = TYPE_VFIO_IOMMU_LEGACY, + .instance_size = sizeof(VFIOSpaprContainer), .class_init = vfio_iommu_spapr_class_init, }, };
PATCH v2 12/17 vfio/container: Switch to QOM
In reply to Cédric Le GoaterInstead of allocating the container struct, create a QOM object of the appropriate type.
--- hw/vfio/container.c | 6 +++--- hw/vfio/iommufd.c | 4 ++-- 2 files changed, 5 insertions(+), 5 deletions(-)
hw/vfio/container.c +3 -3
@@ -435,7 +435,7 @@ static VFIOContainer *vfio_create_container(int fd, VFIOGroup *group, vioc_name =vfio_get_iommu_class_name (iommu_type); vioc =VFIO_IOMMU_CLASS (object_class_by_name (vioc_name)); - container =g_malloc0 (sizeof (*container)); + container =VFIO_IOMMU_LEGACY (object_new (vioc_name)); container->fd = fd; container->iommu_type = iommu_type;vfio_container_init (&container->bcontainer, vioc);
@@ -674,7 +674,7 @@ unregister_container_exit:vfio_cpr_unregister_container (bcontainer); free_container_exit: -g_free (container); +object_unref (container); close_fd_exit:close (fd);
@@ -718,7 +718,7 @@ static void vfio_disconnect_container(VFIOGroup *group)trace_vfio_disconnect_container (container->fd);vfio_cpr_unregister_container (bcontainer);close (container->fd); -g_free (container); +object_unref (container);vfio_put_address_space (space); }
hw/vfio/iommufd.c +2 -2
@@ -239,7 +239,7 @@ static void iommufd_cdev_container_destroy(VFIOIOMMUFDContainer *container)memory_listener_unregister (&bcontainer->listener);vfio_container_destroy (bcontainer);iommufd_backend_free_id (container->be, container->ioas_id); -g_free (container); +object_unref (container); }static int iommufd_cdev_ram_block_discard_disable (bool state)
@@ -352,7 +352,7 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev,trace_iommufd_cdev_alloc_ioas (vbasedev->iommufd->fd, ioas_id); - container =g_malloc0 (sizeof (*container)); + container =VFIO_IOMMU_IOMMUFD (object_new (TYPE_VFIO_IOMMU_IOMMUFD )); container->be = vbasedev->iommufd; container->ioas_id = ioas_id;
Signature · 1 line
2.45.2
Re: [PATCH v2 12/17] vfio/container: Switch to QOM
Quoted 5 lines · Cédric Le Goater
Instead of allocating the container struct, create a QOM object of the appropriate type.
Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>
Eric
Quoted 58 lines
--- hw/vfio/container.c | 6 +++--- hw/vfio/iommufd.c | 4 ++-- 2 files changed, 5 insertions(+), 5 deletions(-)hw/vfio/container.c +3
-3@@ -435,7 +435,7 @@ static VFIOContainer *vfio_create_container(int fd, VFIOGroup *group, vioc_name =vfio_get_iommu_class_name (iommu_type); vioc =VFIO_IOMMU_CLASS (object_class_by_name (vioc_name)); - container =g_malloc0 (sizeof (*container)); + container =VFIO_IOMMU_LEGACY (object_new (vioc_name)); container->fd = fd; container->iommu_type = iommu_type;vfio_container_init (&container->bcontainer, vioc);@@ -674,7 +674,7 @@ unregister_container_exit:vfio_cpr_unregister_container (bcontainer); free_container_exit: -g_free (container); +object_unref (container); close_fd_exit:close (fd);@@ -718,7 +718,7 @@ static void vfio_disconnect_container(VFIOGroup *group)trace_vfio_disconnect_container (container->fd);vfio_cpr_unregister_container (bcontainer);close (container->fd); -g_free (container); +object_unref (container);vfio_put_address_space (space); }hw/vfio/iommufd.c +2
-2@@ -239,7 +239,7 @@ static void iommufd_cdev_container_destroy(VFIOIOMMUFDContainer *container)memory_listener_unregister (&bcontainer->listener);vfio_container_destroy (bcontainer);iommufd_backend_free_id (container->be, container->ioas_id); -g_free (container); +object_unref (container); }static int iommufd_cdev_ram_block_discard_disable (bool state)@@ -352,7 +352,7 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev,trace_iommufd_cdev_alloc_ioas (vbasedev->iommufd->fd, ioas_id); - container =g_malloc0 (sizeof (*container)); + container =VFIO_IOMMU_IOMMUFD (object_new (TYPE_VFIO_IOMMU_IOMMUFD )); container->be = vbasedev->iommufd; container->ioas_id = ioas_id;
PATCH v2 13/17 vfio/container: Introduce an instance_init() handler
In reply to Cédric Le GoaterThis allows us to move the initialization code from vfio_container_init(), which we will soon remove.
--- hw/vfio/container-base.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)
hw/vfio/container-base.c +13 -6
@@ -75,12 +75,6 @@ void vfio_container_init(VFIOContainerBase *bcontainer, const VFIOIOMMUClass *ops) { bcontainer->ops = ops; - bcontainer->error =NULL ; - bcontainer->dirty_pages_supported = false; - bcontainer->dma_max_mappings =0 ; - bcontainer->iova_ranges =NULL ; -QLIST_INIT (&bcontainer->giommu_list); -QLIST_INIT (&bcontainer->vrdl_list); }void vfio_container_destroy (VFIOContainerBase *bcontainer)
@@ -99,10 +93,23 @@ void vfio_container_destroy(VFIOContainerBase *bcontainer)g_list_free_full (bcontainer->iova_ranges, g_free); } +static void vfio_container_instance_init (Object *obj) +{ +VFIOContainerBase *bcontainer =VFIO_IOMMU (obj); + + bcontainer->error =NULL ; + bcontainer->dirty_pages_supported = false; + bcontainer->dma_max_mappings =0 ; + bcontainer->iova_ranges =NULL ; +QLIST_INIT (&bcontainer->giommu_list); +QLIST_INIT (&bcontainer->vrdl_list); +} +static const TypeInfo types[] = { { .name =TYPE_VFIO_IOMMU , .parent =TYPE_OBJECT , + .instance_init = vfio_container_instance_init, .instance_size =sizeof (VFIOContainerBase), .class_size =sizeof (VFIOIOMMUClass), .abstract = true,
Signature · 1 line
2.45.2
Re: [PATCH v2 13/17] vfio/container: Introduce an instance_init() handler
Quoted 5 lines · Cédric Le Goater
This allows us to move the initialization code from
vfio_container_init(), which we will soon remove.Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>
Eric
Quoted 45 lines
--- hw/vfio/container-base.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)hw/vfio/container-base.c +13
-6@@ -75,12 +75,6 @@ void vfio_container_init(VFIOContainerBase *bcontainer, const VFIOIOMMUClass *ops) { bcontainer->ops = ops; - bcontainer->error =NULL ; - bcontainer->dirty_pages_supported = false; - bcontainer->dma_max_mappings =0 ; - bcontainer->iova_ranges =NULL ; -QLIST_INIT (&bcontainer->giommu_list); -QLIST_INIT (&bcontainer->vrdl_list); }void vfio_container_destroy (VFIOContainerBase *bcontainer)@@ -99,10 +93,23 @@ void vfio_container_destroy(VFIOContainerBase *bcontainer)g_list_free_full (bcontainer->iova_ranges, g_free); } +static void vfio_container_instance_init (Object *obj) +{ +VFIOContainerBase *bcontainer =VFIO_IOMMU (obj); + + bcontainer->error =NULL ; + bcontainer->dirty_pages_supported = false; + bcontainer->dma_max_mappings =0 ; + bcontainer->iova_ranges =NULL ; +QLIST_INIT (&bcontainer->giommu_list); +QLIST_INIT (&bcontainer->vrdl_list); +} +static const TypeInfo types[] = { { .name =TYPE_VFIO_IOMMU , .parent =TYPE_OBJECT , + .instance_init = vfio_container_instance_init, .instance_size =sizeof (VFIOContainerBase), .class_size =sizeof (VFIOIOMMUClass), .abstract = true,
PATCH v2 14/17 vfio/container: Remove VFIOContainerBase::ops
In reply to Cédric Le GoaterInstead, use VFIO_IOMMU_GET_CLASS() to get the class pointer.
--- include/hw/vfio/vfio-container-base.h | 1 - hw/vfio/common.c | 2 +- hw/vfio/container-base.c | 37 +++++++++++++++++---------- hw/vfio/container.c | 15 ++++++----- hw/vfio/iommufd.c | 4 +-- hw/vfio/pci.c | 4 +-- 6 files changed, 38 insertions(+), 25 deletions(-)
include/hw/vfio/vfio-container-base.h +0 -1
@@ -35,7 +35,6 @@ typedef struct VFIOAddressSpace {*/ typedef struct VFIOContainerBase { Object parent; - const VFIOIOMMUClass *ops; VFIOAddressSpace *space; MemoryListener listener; Error *error;
hw/vfio/common.c +1 -1
@@ -1573,5 +1573,5 @@ void vfio_detach_device(VFIODevice *vbasedev)return ; }object_unref (vbasedev->hiod); - vbasedev->bcontainer->ops->detach_device (vbasedev); +VFIO_IOMMU_GET_CLASS (vbasedev->bcontainer)->detach_device (vbasedev); }
hw/vfio/container-base.c +24 -13
@@ -19,62 +19,73 @@ int vfio_container_dma_map(VFIOContainerBase *bcontainer, hwaddr iova, ram_addr_t size, void *vaddr, bool readonly) { - g_assert(bcontainer->ops->dma_map); -return bcontainer->ops->dma_map (bcontainer, iova, size, vaddr, readonly); + VFIOIOMMUClass *vioc = VFIO_IOMMU_GET_CLASS(bcontainer); + +g_assert (vioc->dma_map); +return vioc->dma_map (bcontainer, iova, size, vaddr, readonly); }int vfio_container_dma_unmap (VFIOContainerBase *bcontainer,hwaddr iova,ram_addr_t size,IOMMUTLBEntry *iotlb) { -g_assert (bcontainer->ops->dma_unmap); -return bcontainer->ops->dma_unmap (bcontainer, iova, size, iotlb); +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer); + +g_assert (vioc->dma_unmap); +return vioc->dma_unmap (bcontainer, iova, size, iotlb); }bool vfio_container_add_section_window (VFIOContainerBase *bcontainer,MemoryRegionSection *section,Error **errp) { -if (!bcontainer->ops->add_window) { +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer); + +if (!vioc->add_window) {return true; } -return bcontainer->ops->add_window (bcontainer, section, errp); +return vioc->add_window (bcontainer, section, errp); }void vfio_container_del_section_window (VFIOContainerBase *bcontainer,MemoryRegionSection *section) { -if (!bcontainer->ops->del_window) { +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer); + +if (!vioc->del_window) {return ; } -return bcontainer->ops->del_window (bcontainer, section); +return vioc->del_window (bcontainer, section); }int vfio_container_set_dirty_page_tracking (VFIOContainerBase *bcontainer,bool start,Error **errp) { +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer); +if (!bcontainer->dirty_pages_supported) {return 0 ; } -g_assert (bcontainer->ops->set_dirty_page_tracking); -return bcontainer->ops->set_dirty_page_tracking (bcontainer, start, errp); +g_assert (vioc->set_dirty_page_tracking); +return vioc->set_dirty_page_tracking (bcontainer, start, errp); }int vfio_container_query_dirty_bitmap (const VFIOContainerBase *bcontainer,VFIOBitmap *vbmap,hwaddr iova,hwaddr size,Error **errp) { -g_assert (bcontainer->ops->query_dirty_bitmap); -return bcontainer->ops->query_dirty_bitmap (bcontainer, vbmap, iova, size, +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer); + +g_assert (vioc->query_dirty_bitmap); +return vioc->query_dirty_bitmap (bcontainer, vbmap, iova, size, errp); }void vfio_container_init (VFIOContainerBase *bcontainer,const VFIOIOMMUClass *ops) { - bcontainer->ops = ops; }void vfio_container_destroy (VFIOContainerBase *bcontainer)
hw/vfio/container.c +9 -6
@@ -548,6 +548,7 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as,VFIOContainerBase *bcontainer;int ret, fd;VFIOAddressSpace *space; +VFIOIOMMUClass *vioc; space =vfio_get_address_space (as);
@@ -632,9 +633,10 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as, goto unregister_container_exit; } -assert (bcontainer->ops->setup); + vioc =VFIO_IOMMU_GET_CLASS (bcontainer); +assert (vioc->setup); -if (!bcontainer->ops->setup (bcontainer, errp)) { +if (!vioc->setup (bcontainer, errp)) { goto enable_discards_exit; }
@@ -663,8 +665,8 @@ listener_release_exit:QLIST_REMOVE (bcontainer, next);vfio_kvm_device_del_group (group);memory_listener_unregister (&bcontainer->listener); -if (bcontainer->ops->release) { - bcontainer->ops->release (bcontainer); +if (vioc->release) { + vioc->release (bcontainer); } enable_discards_exit:
@@ -689,6 +691,7 @@ static void vfio_disconnect_container(VFIOGroup *group) {VFIOContainer *container = group->container;VFIOContainerBase *bcontainer = &container->bcontainer; +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer);QLIST_REMOVE (group, container_next); group->container =NULL ;
@@ -700,8 +703,8 @@ static void vfio_disconnect_container(VFIOGroup *group)*/ if (QLIST_EMPTY (&container->group_list)) {memory_listener_unregister (&bcontainer->listener); -if (bcontainer->ops->release) { - bcontainer->ops->release (bcontainer); +if (vioc->release) { + vioc->release (bcontainer); } }
hw/vfio/iommufd.c +2 -2
@@ -324,7 +324,7 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev,/* try to attach to an existing container in this space */ QLIST_FOREACH(bcontainer, &space->containers, next) { container =container_of (bcontainer, VFIOIOMMUFDContainer, bcontainer); -if (bcontainer->ops != iommufd_vioc || +if (VFIO_IOMMU_GET_CLASS (bcontainer) != iommufd_vioc || vbasedev->iommufd != container->be) {continue ; }
@@ -465,7 +465,7 @@ static VFIODevice *iommufd_cdev_pci_find_by_devid(__u32 devid)VFIO_IOMMU_CLASS (object_class_by_name (TYPE_VFIO_IOMMU_IOMMUFD )); QLIST_FOREACH(vbasedev_iter, &vfio_device_list, global_next) { -if (vbasedev_iter->bcontainer->ops != iommufd_vioc) { +if (VFIO_IOMMU_GET_CLASS (vbasedev_iter->bcontainer) != iommufd_vioc) {continue ; }if (devid == vbasedev_iter->devid) {
hw/vfio/pci.c +2 -2
@@ -2511,9 +2511,9 @@ int vfio_pci_get_pci_hot_reset_info(VFIOPCIDevice *vdev,static int vfio_pci_hot_reset (VFIOPCIDevice *vdev,bool single) {VFIODevice *vbasedev = &vdev->vbasedev; -const VFIOIOMMUClass *ops = vbasedev->bcontainer->ops; +const VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (vbasedev->bcontainer); -return ops->pci_hot_reset (vbasedev, single); +return vioc->pci_hot_reset (vbasedev, single); }/*
Signature · 1 line
2.45.2
Re: [PATCH v2 14/17] vfio/container: Remove VFIOContainerBase::ops
Quoted 4 lines · Cédric Le Goater
Instead, use
VFIO_IOMMU_GET_CLASS()to get the class pointer.Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>
Eric
Quoted 216 lines
--- include/hw/vfio/vfio-container-base.h | 1 - hw/vfio/common.c | 2 +- hw/vfio/container-base.c | 37 +++++++++++++++++---------- hw/vfio/container.c | 15 ++++++----- hw/vfio/iommufd.c | 4 +-- hw/vfio/pci.c | 4 +-- 6 files changed, 38 insertions(+), 25 deletions(-)include/hw/vfio/vfio-container-base.h +0
-1@@ -35,7 +35,6 @@ typedef struct VFIOAddressSpace {*/ typedef struct VFIOContainerBase { Object parent; - const VFIOIOMMUClass *ops; VFIOAddressSpace *space; MemoryListener listener; Error *error;hw/vfio/common.c +1
-1@@ -1573,5 +1573,5 @@ void vfio_detach_device(VFIODevice *vbasedev)return ; }object_unref (vbasedev->hiod); - vbasedev->bcontainer->ops->detach_device (vbasedev); +VFIO_IOMMU_GET_CLASS (vbasedev->bcontainer)->detach_device (vbasedev); }hw/vfio/container-base.c +24
-13@@ -19,62 +19,73 @@ int vfio_container_dma_map(VFIOContainerBase *bcontainer, hwaddr iova, ram_addr_t size, void *vaddr, bool readonly) { - g_assert(bcontainer->ops->dma_map); -return bcontainer->ops->dma_map (bcontainer, iova, size, vaddr, readonly); + VFIOIOMMUClass *vioc = VFIO_IOMMU_GET_CLASS(bcontainer); + +g_assert (vioc->dma_map); +return vioc->dma_map (bcontainer, iova, size, vaddr, readonly); }int vfio_container_dma_unmap (VFIOContainerBase *bcontainer,hwaddr iova,ram_addr_t size,IOMMUTLBEntry *iotlb) { -g_assert (bcontainer->ops->dma_unmap); -return bcontainer->ops->dma_unmap (bcontainer, iova, size, iotlb); +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer); + +g_assert (vioc->dma_unmap); +return vioc->dma_unmap (bcontainer, iova, size, iotlb); }bool vfio_container_add_section_window (VFIOContainerBase *bcontainer,MemoryRegionSection *section,Error **errp) { -if (!bcontainer->ops->add_window) { +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer); + +if (!vioc->add_window) {return true; } -return bcontainer->ops->add_window (bcontainer, section, errp); +return vioc->add_window (bcontainer, section, errp); }void vfio_container_del_section_window (VFIOContainerBase *bcontainer,MemoryRegionSection *section) { -if (!bcontainer->ops->del_window) { +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer); + +if (!vioc->del_window) {return ; } -return bcontainer->ops->del_window (bcontainer, section); +return vioc->del_window (bcontainer, section); }int vfio_container_set_dirty_page_tracking (VFIOContainerBase *bcontainer,bool start,Error **errp) { +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer); +if (!bcontainer->dirty_pages_supported) {return 0 ; } -g_assert (bcontainer->ops->set_dirty_page_tracking); -return bcontainer->ops->set_dirty_page_tracking (bcontainer, start, errp); +g_assert (vioc->set_dirty_page_tracking); +return vioc->set_dirty_page_tracking (bcontainer, start, errp); }int vfio_container_query_dirty_bitmap (const VFIOContainerBase *bcontainer,VFIOBitmap *vbmap,hwaddr iova,hwaddr size,Error **errp) { -g_assert (bcontainer->ops->query_dirty_bitmap); -return bcontainer->ops->query_dirty_bitmap (bcontainer, vbmap, iova, size, +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer); + +g_assert (vioc->query_dirty_bitmap); +return vioc->query_dirty_bitmap (bcontainer, vbmap, iova, size, errp); }void vfio_container_init (VFIOContainerBase *bcontainer,const VFIOIOMMUClass *ops) { - bcontainer->ops = ops; }void vfio_container_destroy (VFIOContainerBase *bcontainer)hw/vfio/container.c +9
-6@@ -548,6 +548,7 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as,VFIOContainerBase *bcontainer;int ret, fd;VFIOAddressSpace *space; +VFIOIOMMUClass *vioc; space =vfio_get_address_space (as);@@ -632,9 +633,10 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as, goto unregister_container_exit; } -assert (bcontainer->ops->setup); + vioc =VFIO_IOMMU_GET_CLASS (bcontainer); +assert (vioc->setup); -if (!bcontainer->ops->setup (bcontainer, errp)) { +if (!vioc->setup (bcontainer, errp)) { goto enable_discards_exit; }@@ -663,8 +665,8 @@ listener_release_exit:QLIST_REMOVE (bcontainer, next);vfio_kvm_device_del_group (group);memory_listener_unregister (&bcontainer->listener); -if (bcontainer->ops->release) { - bcontainer->ops->release (bcontainer); +if (vioc->release) { + vioc->release (bcontainer); } enable_discards_exit:@@ -689,6 +691,7 @@ static void vfio_disconnect_container(VFIOGroup *group) {VFIOContainer *container = group->container;VFIOContainerBase *bcontainer = &container->bcontainer; +VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (bcontainer);QLIST_REMOVE (group, container_next); group->container =NULL ;@@ -700,8 +703,8 @@ static void vfio_disconnect_container(VFIOGroup *group)*/ if (QLIST_EMPTY (&container->group_list)) {memory_listener_unregister (&bcontainer->listener); -if (bcontainer->ops->release) { - bcontainer->ops->release (bcontainer); +if (vioc->release) { + vioc->release (bcontainer); } }hw/vfio/iommufd.c +2
-2@@ -324,7 +324,7 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev,/* try to attach to an existing container in this space */ QLIST_FOREACH(bcontainer, &space->containers, next) { container =container_of (bcontainer, VFIOIOMMUFDContainer, bcontainer); -if (bcontainer->ops != iommufd_vioc || +if (VFIO_IOMMU_GET_CLASS (bcontainer) != iommufd_vioc || vbasedev->iommufd != container->be) {continue ; }@@ -465,7 +465,7 @@ static VFIODevice *iommufd_cdev_pci_find_by_devid(__u32 devid)VFIO_IOMMU_CLASS (object_class_by_name (TYPE_VFIO_IOMMU_IOMMUFD )); QLIST_FOREACH(vbasedev_iter, &vfio_device_list, global_next) { -if (vbasedev_iter->bcontainer->ops != iommufd_vioc) { +if (VFIO_IOMMU_GET_CLASS (vbasedev_iter->bcontainer) != iommufd_vioc) {continue ; }if (devid == vbasedev_iter->devid) {hw/vfio/pci.c +2
-2@@ -2511,9 +2511,9 @@ int vfio_pci_get_pci_hot_reset_info(VFIOPCIDevice *vdev,static int vfio_pci_hot_reset (VFIOPCIDevice *vdev,bool single) {VFIODevice *vbasedev = &vdev->vbasedev; -const VFIOIOMMUClass *ops = vbasedev->bcontainer->ops; +const VFIOIOMMUClass *vioc =VFIO_IOMMU_GET_CLASS (vbasedev->bcontainer); -return ops->pci_hot_reset (vbasedev, single); +return vioc->pci_hot_reset (vbasedev, single); }/*
PATCH v2 15/17 vfio/container: Remove vfio_container_init()
In reply to Cédric Le GoaterIt's now empty.
--- include/hw/vfio/vfio-container-base.h | 2 -- hw/vfio/container-base.c | 5 ----- hw/vfio/container.c | 3 --- hw/vfio/iommufd.c | 1 - 4 files changed, 11 deletions(-)
include/hw/vfio/vfio-container-base.h +0 -2
@@ -86,8 +86,6 @@ int vfio_container_set_dirty_page_tracking(VFIOContainerBase *bcontainer,int vfio_container_query_dirty_bitmap (const VFIOContainerBase *bcontainer,VFIOBitmap *vbmap,hwaddr iova,hwaddr size,Error **errp); -void vfio_container_init (VFIOContainerBase *bcontainer, -const VFIOIOMMUClass *ops);void vfio_container_destroy (VFIOContainerBase *bcontainer);
hw/vfio/container-base.c +0 -5
@@ -83,11 +83,6 @@ int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer, errp); } -void vfio_container_init (VFIOContainerBase *bcontainer, -const VFIOIOMMUClass *ops) -{ -} -void vfio_container_destroy (VFIOContainerBase *bcontainer) {VFIOGuestIOMMU *giommu, *tmp;
hw/vfio/container.c +0 -3
@@ -419,7 +419,6 @@ static VFIOContainer *vfio_create_container(int fd, VFIOGroup *group, Error **errp) {int iommu_type; -const VFIOIOMMUClass *vioc;const char *vioc_name;VFIOContainer *container;
@@ -433,12 +432,10 @@ static VFIOContainer *vfio_create_container(int fd, VFIOGroup *group, } vioc_name =vfio_get_iommu_class_name (iommu_type); - vioc =VFIO_IOMMU_CLASS (object_class_by_name (vioc_name)); container =VFIO_IOMMU_LEGACY (object_new (vioc_name)); container->fd = fd; container->iommu_type = iommu_type; -vfio_container_init (&container->bcontainer, vioc);return container; }
hw/vfio/iommufd.c +0 -1
@@ -357,7 +357,6 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev, container->ioas_id = ioas_id; bcontainer = &container->bcontainer; -vfio_container_init (bcontainer, iommufd_vioc);vfio_address_space_insert (space, bcontainer);if (!iommufd_cdev_attach_container (vbasedev, container, errp)) {
Signature · 1 line
2.45.2
Re: [PATCH v2 15/17] vfio/container: Remove vfio_container_init()
Quoted 4 lines · Cédric Le Goater
It's now empty.
Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>
Eric
Quoted 73 lines
--- include/hw/vfio/vfio-container-base.h | 2 -- hw/vfio/container-base.c | 5 ----- hw/vfio/container.c | 3 --- hw/vfio/iommufd.c | 1 - 4 files changed, 11 deletions(-)include/hw/vfio/vfio-container-base.h +0
-2@@ -86,8 +86,6 @@ int vfio_container_set_dirty_page_tracking(VFIOContainerBase *bcontainer,int vfio_container_query_dirty_bitmap (const VFIOContainerBase *bcontainer,VFIOBitmap *vbmap,hwaddr iova,hwaddr size,Error **errp); -void vfio_container_init (VFIOContainerBase *bcontainer, -const VFIOIOMMUClass *ops);void vfio_container_destroy (VFIOContainerBase *bcontainer);hw/vfio/container-base.c +0
-5@@ -83,11 +83,6 @@ int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer, errp); } -void vfio_container_init (VFIOContainerBase *bcontainer, -const VFIOIOMMUClass *ops) -{ -} -void vfio_container_destroy (VFIOContainerBase *bcontainer) {VFIOGuestIOMMU *giommu, *tmp;hw/vfio/container.c +0
-3@@ -419,7 +419,6 @@ static VFIOContainer *vfio_create_container(int fd, VFIOGroup *group, Error **errp) {int iommu_type; -const VFIOIOMMUClass *vioc;const char *vioc_name;VFIOContainer *container;@@ -433,12 +432,10 @@ static VFIOContainer *vfio_create_container(int fd, VFIOGroup *group, } vioc_name =vfio_get_iommu_class_name (iommu_type); - vioc =VFIO_IOMMU_CLASS (object_class_by_name (vioc_name)); container =VFIO_IOMMU_LEGACY (object_new (vioc_name)); container->fd = fd; container->iommu_type = iommu_type; -vfio_container_init (&container->bcontainer, vioc);return container; }hw/vfio/iommufd.c +0
-1@@ -357,7 +357,6 @@ static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev, container->ioas_id = ioas_id; bcontainer = &container->bcontainer; -vfio_container_init (bcontainer, iommufd_vioc);vfio_address_space_insert (space, bcontainer);if (!iommufd_cdev_attach_container (vbasedev, container, errp)) {
PATCH v2 16/17 vfio/container: Introduce vfio_iommu_legacy_instance_init()
In reply to Cédric Le GoaterJust as we did for the VFIOContainerBase object, introduce an instance_init() handler for the legacy VFIOContainer object and do the specific initialization there.
--- hw/vfio/container.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
hw/vfio/container.c +8 -1
@@ -639,7 +639,6 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as,vfio_kvm_device_add_group (group); -QLIST_INIT (&container->group_list);vfio_address_space_insert (space, bcontainer); group->container = container;
@@ -1183,6 +1182,13 @@ hiod_legacy_vfio_get_iova_ranges(HostIOMMUDevice *hiod, Error **errp)return l; } +static void vfio_iommu_legacy_instance_init (Object *obj) +{ +VFIOContainer *container =VFIO_IOMMU_LEGACY (obj); + +QLIST_INIT (&container->group_list); +} +static void hiod_legacy_vfio_class_init (ObjectClass *oc,void *data) {HostIOMMUDeviceClass *hioc =HOST_IOMMU_DEVICE_CLASS (oc);
@@ -1196,6 +1202,7 @@ static const TypeInfo types[] = { { .name = TYPE_VFIO_IOMMU_LEGACY, .parent = TYPE_VFIO_IOMMU, + .instance_init = vfio_iommu_legacy_instance_init, .instance_size = sizeof(VFIOContainer), .class_init = vfio_iommu_legacy_class_init, }, {
Signature · 1 line
2.45.2
Re: [PATCH v2 16/17] vfio/container: Introduce vfio_iommu_legacy_instance_init()
Quoted 6 lines · Cédric Le Goater
Just as we did for the VFIOContainerBase object, introduce an
instance_init()handler for the legacy VFIOContainer object and do the specific initialization there.Signed-off-by Cédric Le Goater <[email protected]>Reviewed-by Zhenzhong Duan <[email protected]>
Eric
Quoted 38 lines
--- hw/vfio/container.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)hw/vfio/container.c +8
-1@@ -639,7 +639,6 @@ static bool vfio_connect_container(VFIOGroup *group, AddressSpace *as,vfio_kvm_device_add_group (group); -QLIST_INIT (&container->group_list);vfio_address_space_insert (space, bcontainer); group->container = container;@@ -1183,6 +1182,13 @@ hiod_legacy_vfio_get_iova_ranges(HostIOMMUDevice *hiod, Error **errp)return l; } +static void vfio_iommu_legacy_instance_init (Object *obj) +{ +VFIOContainer *container =VFIO_IOMMU_LEGACY (obj); + +QLIST_INIT (&container->group_list); +} +static void hiod_legacy_vfio_class_init (ObjectClass *oc,void *data) {HostIOMMUDeviceClass *hioc =HOST_IOMMU_DEVICE_CLASS (oc);@@ -1196,6 +1202,7 @@ static const TypeInfo types[] = { { .name = TYPE_VFIO_IOMMU_LEGACY, .parent = TYPE_VFIO_IOMMU, + .instance_init = vfio_iommu_legacy_instance_init, .instance_size = sizeof(VFIOContainer), .class_init = vfio_iommu_legacy_class_init, }, {
PATCH v2 17/17 vfio/container: Move vfio_container_destroy() to an instance_finalize() handler
In reply to Cédric Le Goatervfio_container_destroy() clears the resources allocated VFIOContainerBase object. Now that VFIOContainerBase is a QOM object, add an instance_finalize() handler to do the cleanup. It will be called through object_unref().
--- include/hw/vfio/vfio-container-base.h | 3 --- hw/vfio/container-base.c | 4 +++- hw/vfio/container.c | 2 -- hw/vfio/iommufd.c | 1 - 4 files changed, 3 insertions(+), 7 deletions(-)
include/hw/vfio/vfio-container-base.h +0 -3
@@ -86,9 +86,6 @@ int vfio_container_set_dirty_page_tracking(VFIOContainerBase *bcontainer,int vfio_container_query_dirty_bitmap (const VFIOContainerBase *bcontainer,VFIOBitmap *vbmap,hwaddr iova,hwaddr size,Error **errp); -void vfio_container_destroy (VFIOContainerBase *bcontainer); - -#define TYPE_VFIO_IOMMU "vfio-iommu"#define TYPE_VFIO_IOMMU_LEGACY TYPE_VFIO_IOMMU "-legacy"#define TYPE_VFIO_IOMMU_SPAPR TYPE_VFIO_IOMMU "-spapr"
hw/vfio/container-base.c +3 -1
@@ -83,8 +83,9 @@ int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer, errp); } -void vfio_container_destroy (VFIOContainerBase *bcontainer) +static void vfio_container_instance_finalize (Object *obj) { +VFIOContainerBase *bcontainer =VFIO_IOMMU (obj);VFIOGuestIOMMU *giommu, *tmp;QLIST_REMOVE (bcontainer, next);
@@ -116,6 +117,7 @@ static const TypeInfo types[] = { .name = TYPE_VFIO_IOMMU, .parent = TYPE_OBJECT, .instance_init = vfio_container_instance_init, + .instance_finalize = vfio_container_instance_finalize, .instance_size = sizeof(VFIOContainerBase), .class_size = sizeof(VFIOIOMMUClass), .abstract = true,
hw/vfio/container.c +0 -2
@@ -712,8 +712,6 @@ static void vfio_disconnect_container(VFIOGroup *group)if (QLIST_EMPTY (&container->group_list)) {VFIOAddressSpace *space = bcontainer->space; -vfio_container_destroy (bcontainer); -trace_vfio_disconnect_container (container->fd);vfio_cpr_unregister_container (bcontainer);close (container->fd);
hw/vfio/iommufd.c +0 -1
@@ -237,7 +237,6 @@ static void iommufd_cdev_container_destroy(VFIOIOMMUFDContainer *container)return ; }memory_listener_unregister (&bcontainer->listener); -vfio_container_destroy (bcontainer);iommufd_backend_free_id (container->be, container->ioas_id);object_unref (container); }
Signature · 1 line
2.45.2
RE: [PATCH v2 17/17] vfio/container: Move vfio_container_destroy() to an instance_finalize() handler
Quoted 7 lines
-----Original Message-----
From: Cédric Le Goater <[email protected]>
Subject: [PATCH v2 17/17] vfio/container: Movevfio_container_destroy()to aninstance_finalize()handler
Quoted 7 lines
vfio_container_destroy()clears the resources allocated VFIOContainerBase object. Now that VFIOContainerBase is a QOM object, add aninstance_finalize()handler to do the cleanup. It will be called throughobject_unref().Suggested-by Zhenzhong Duan <[email protected]>Signed-off-by Cédric Le Goater <[email protected]>
Thanks
Zhenzhong
Quoted 85 lines
---
include/hw/vfio/vfio-container-base.h | 3 --- hw/vfio/container-base.c | 4 +++- hw/vfio/container.c | 2 -- hw/vfio/iommufd.c | 1 - 4 files changed, 3 insertions(+), 7 deletions(-)include/hw/vfio/vfio- renamed from include/hw/vfio/vfio-container-base.h +0
-0
container-base.h
index6242a62771caa8cf19440a53ad6f4db862ca12d7..419e45ee7a5ac960dae4a993127fc9ee66d48db2100644include/hw/vfio/vfio-container-base.h +0
-0@@ -86,9 +86,6 @@ int vfio_container_set_dirty_page_tracking(VFIOContainerBase *bcontainer,int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer,
VFIOBitmap *vbmap, hwaddr iova, hwaddr size, Error **errp);include/hw/vfio/vfio-container-base.h +0
-3-void vfio_container_destroy (VFIOContainerBase *bcontainer); - -#define
TYPE_VFIO_IOMMU"vfio-iommu"
#defineTYPE_VFIO_IOMMU_LEGACYTYPE_VFIO_IOMMU"-legacy" #defineTYPE_VFIO_IOMMU_SPAPRTYPE_VFIO_IOMMU"-spapr"hw/vfio/container-base.c +0
-0index
970ae2356a92f87df44e1dd58ff8c67045a24ef1..50b1664f89a8192cf4021498e59f2a92cd2f6e89100644hw/vfio/container-base.c +2
-1@@ -83,8 +83,9 @@ int vfio_container_query_dirty_bitmap(constVFIOContainerBase *bcontainer, errp); } -void vfio_container_destroy (VFIOContainerBase *bcontainer) +static void vfio_container_instance_finalize (Object *obj) { +VFIOContainerBase *bcontainer =VFIO_IOMMU (obj);VFIOGuestIOMMU *giommu, *tmp;QLIST_REMOVE(bcontainer, next);hw/vfio/container-base.c +1
-0@@ -116,6 +117,7 @@ static const TypeInfo types[] = { .name = TYPE_VFIO_IOMMU, .parent = TYPE_OBJECT, .instance_init = vfio_container_instance_init, + .instance_finalize = vfio_container_instance_finalize, .instance_size = sizeof(VFIOContainerBase), .class_size = sizeof(VFIOIOMMUClass), .abstract = true,hw/vfio/container.c +0
-0index
45123acbdd6a681f4ce7cae7aa2509100ea225ab..2e7ecdf10edc4d84963a45ae9507096965da64fc100644hw/vfio/container.c +0
-2@@ -712,8 +712,6 @@ static void vfio_disconnect_container(VFIOGroup *group)if (QLIST_EMPTY (&container->group_list)) {VFIOAddressSpace *space = bcontainer->space; -vfio_container_destroy (bcontainer); -trace_vfio_disconnect_container (container->fd);vfio_cpr_unregister_container (bcontainer);close(container->fd);hw/vfio/iommufd.c +0
-0index
09b71a6617807c621275c74b924cfd39eb643961..c2f158e60386502eef267769ac9bce1effb67033100644hw/vfio/iommufd.c +0
-1@@ -237,7 +237,6 @@ static voidiommufd_cdev_container_destroy (VFIOIOMMUFDContainer *container) return; }memory_listener_unregister (&bcontainer->listener); -vfio_container_destroy (bcontainer);iommufd_backend_free_id (container->be, container->ioas_id);object_unref (container);}
--
2.45.2
Re: [PATCH v2 17/17] vfio/container: Move vfio_container_destroy() to an instance_finalize() handler
In reply to Cédric Le GoaterHi Cédric,
Quoted 8 lines · Cédric Le Goater
vfio_container_destroy()clears the resources allocated VFIOContainerBase object. Now that VFIOContainerBase is a QOM object, add aninstance_finalize()handler to do the cleanup. It will be called throughobject_unref().Suggested-by Zhenzhong Duan <[email protected]>Signed-off-by Cédric Le Goater <[email protected]>---
Thanks
Eric
Quoted 68 lines
include/hw/vfio/vfio-container-base.h | 3 --- hw/vfio/container-base.c | 4 +++- hw/vfio/container.c | 2 -- hw/vfio/iommufd.c | 1 - 4 files changed, 3 insertions(+), 7 deletions(-)include/hw/vfio/vfio-container-base.h +0
-3@@ -86,9 +86,6 @@ int vfio_container_set_dirty_page_tracking(VFIOContainerBase *bcontainer,int vfio_container_query_dirty_bitmap (const VFIOContainerBase *bcontainer,VFIOBitmap *vbmap,hwaddr iova,hwaddr size,Error **errp); -void vfio_container_destroy (VFIOContainerBase *bcontainer); - -#define TYPE_VFIO_IOMMU "vfio-iommu"#define TYPE_VFIO_IOMMU_LEGACY TYPE_VFIO_IOMMU "-legacy"#define TYPE_VFIO_IOMMU_SPAPR TYPE_VFIO_IOMMU "-spapr"hw/vfio/container-base.c +3
-1@@ -83,8 +83,9 @@ int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer, errp); } -void vfio_container_destroy (VFIOContainerBase *bcontainer) +static void vfio_container_instance_finalize (Object *obj) { +VFIOContainerBase *bcontainer =VFIO_IOMMU (obj);VFIOGuestIOMMU *giommu, *tmp;QLIST_REMOVE (bcontainer, next);@@ -116,6 +117,7 @@ static const TypeInfo types[] = { .name = TYPE_VFIO_IOMMU, .parent = TYPE_OBJECT, .instance_init = vfio_container_instance_init, + .instance_finalize = vfio_container_instance_finalize, .instance_size = sizeof(VFIOContainerBase), .class_size = sizeof(VFIOIOMMUClass), .abstract = true,hw/vfio/container.c +0
-2@@ -712,8 +712,6 @@ static void vfio_disconnect_container(VFIOGroup *group)if (QLIST_EMPTY (&container->group_list)) {VFIOAddressSpace *space = bcontainer->space; -vfio_container_destroy (bcontainer); -trace_vfio_disconnect_container (container->fd);vfio_cpr_unregister_container (bcontainer);close (container->fd);hw/vfio/iommufd.c +0
-1@@ -237,7 +237,6 @@ static void iommufd_cdev_container_destroy(VFIOIOMMUFDContainer *container)return ; }memory_listener_unregister (&bcontainer->listener); -vfio_container_destroy (bcontainer);iommufd_backend_free_id (container->be, container->ioas_id);object_unref (container); }
Hi Cédric,
Quoted 9 lines · Cédric Le Goater
Hello,
The series starts with simple changes (patch 1-4). Two of which were initially sent by Joao in a series adding VFIO migration support with vIOMMU [1].
The changes following prepare VFIOContainer for QOMification, switch the container models to QOM when ready and add some final cleanups.
Quoted 6 lines
Applies on top of :
* [v7] Add a host IOMMU device abstraction to check with vIOMMU
https://lore.kernel.org/all/[email protected]
* [v4] VIRTIO-IOMMU/VFIO: Fix host iommu geometry
https://lore.kernel.org/all/[email protected]
For the whole series, feel free to add my
I tested with legacy/iommufd BE and with/without virtio-iommu
Eric
Quoted 46 lines
Thanks,
C.
[1] https://lore.kernel.org/qemu-devel/[email protected]/
Changes in v2:
- UsedOBJECT_DECLARE_SIMPLE_TYPE
- Introduced ainstance_finalize()handlerAvihai Horon (1):
vfio/common: Extract vIOMMU code fromvfio_sync_dirty_bitmap()Cédric Le Goater (15):
vfio: Makevfio_devices_dma_logging_start()return bool
vfio: Remove unused declarations fromvfio-common.hvfio/container: Introducevfio_address_space_insert()vfio/container: Simplifyvfio_container_init()vfio/container: Modifyvfio_get_iommu_type()to use a container fd vfio/container: Introducevfio_get_iommu_class_name()vfio/container: Introducevfio_create_container()vfio/container: Discover IOMMU type before creating the container vfio/container: Change VFIOContainerBase to use QOM vfio/container: Switch to QOM
vfio/container: Introduce aninstance_init()handler vfio/container: RemoveVFIOContainerBase::opsvfio/container: Removevfio_container_init()vfio/container: Introducevfio_iommu_legacy_instance_init()vfio/container: Movevfio_container_destroy()to an
instance_finalize()handlerJoao Martins (1):
vfio/common: Move dirty tracking ranges update to helperinclude/hw/vfio/vfio-common.h | 10 ++- include/hw/vfio/vfio-container-base.h | 19 +--- hw/vfio/common.c | 124 ++++++++++++++++---------- hw/vfio/container-base.c | 70 +++++++++------ hw/vfio/container.c | 106 ++++++++++++---------- hw/vfio/iommufd.c | 13 ++- hw/vfio/pci.c | 4 +- hw/vfio/spapr.c | 3 + 8 files changed, 196 insertions(+), 153 deletions(-)
Quoted 19 lines · Eric Auger
Hi Cédric,
Quoted 9 lines · Cédric Le Goater
Hello,
The series starts with simple changes (patch 1-4). Two of which were initially sent by Joao in a series adding VFIO migration support with vIOMMU [1].
The changes following prepare VFIOContainer for QOMification, switch the container models to QOM when ready and add some final cleanups.
Quoted 6 lines
Applies on top of :
* [v7] Add a host IOMMU device abstraction to check with vIOMMU
https://lore.kernel.org/all/[email protected] * [v4] VIRTIO-IOMMU/VFIO: Fix host iommu geometry https://lore.kernel.org/all/[email protected]
Quoted 5 lines
For the whole series, feel free to add my
Tested-by Eric Auger <[email protected]>I tested with legacy/iommufd BE and with/without virtio-iommu
Thanks !
Here is an updated version, with the new commit logs :
https://github.com/legoater/qemu/commits/vfio-9.1
I hope to push on vfio-next soon, without the series "vfio: VFIO migration support with vIOMMU".
C.
Cédric Le Goater
AppliedQuoted 51 lines · Cédric Le Goater
Hello,
The series starts with simple changes (patch 1-4). Two of which were initially sent by Joao in a series adding VFIO migration support with vIOMMU [1].
The changes following prepare VFIOContainer for QOMification, switch the container models to QOM when ready and add some final cleanups.
Applies on top of :
* [v7] Add a host IOMMU device abstraction to check with vIOMMU
https://lore.kernel.org/all/[email protected] * [v4] VIRTIO-IOMMU/VFIO: Fix host iommu geometry https://lore.kernel.org/all/[email protected]Thanks,
C.
[1] https://lore.kernel.org/qemu-devel/[email protected]/
Changes in v2:
- UsedOBJECT_DECLARE_SIMPLE_TYPE
- Introduced ainstance_finalize()handlerAvihai Horon (1):
vfio/common: Extract vIOMMU code fromvfio_sync_dirty_bitmap()Cédric Le Goater (15):
vfio: Makevfio_devices_dma_logging_start()return bool
vfio: Remove unused declarations fromvfio-common.hvfio/container: Introducevfio_address_space_insert()vfio/container: Simplifyvfio_container_init()vfio/container: Modifyvfio_get_iommu_type()to use a container fd vfio/container: Introducevfio_get_iommu_class_name()vfio/container: Introducevfio_create_container()vfio/container: Discover IOMMU type before creating the container vfio/container: Change VFIOContainerBase to use QOM vfio/container: Switch to QOM
vfio/container: Introduce aninstance_init()handler vfio/container: RemoveVFIOContainerBase::opsvfio/container: Removevfio_container_init()vfio/container: Introducevfio_iommu_legacy_instance_init()vfio/container: Movevfio_container_destroy()to an
instance_finalize()handlerJoao Martins (1):
vfio/common: Move dirty tracking ranges update to helper
Quoted 10 lines
include/hw/vfio/vfio-common.h | 10 ++- include/hw/vfio/vfio-container-base.h | 19 +--- hw/vfio/common.c | 124 ++++++++++++++++---------- hw/vfio/container-base.c | 70 +++++++++------ hw/vfio/container.c | 106 ++++++++++++---------- hw/vfio/iommufd.c | 13 ++- hw/vfio/pci.c | 4 +- hw/vfio/spapr.c | 3 + 8 files changed, 196 insertions(+), 153 deletions(-)
Applied to vfio-next.
Thanks,
C.