hw/arm/virt-acpi-build: Fix SMMUv3 ACPI integration
PATCH 1/1 hw/arm/virt-acpi-build: Fix SMMUv3 ACPI integration
The AcpiIortSmmu3 misses 2 32b fields corresponding to the proximity domain and the device id mapping index.
Also let's report IO-coherent access is supported for translation table walks, descriptor fetches and queues by setting the COHACC override flag. Without that, we observe wrong command opcodes. The DT description also advertises the dma coherency.
Fixes a703b4f6c1ee ("hw/arm/virt-acpi-build: Add smmuv3 node in IORT table")
--- hw/arm/virt-acpi-build.c | 1 + include/hw/acpi/acpi-defs.h | 8 ++++++++ 2 files changed, 9 insertions(+)
hw/arm/virt-acpi-build.c +1 -0
@@ -448,6 +448,7 @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) smmu->mapping_count =cpu_to_le32 (1 ); smmu->mapping_offset =cpu_to_le32 (sizeof (*smmu)); smmu->base_address =cpu_to_le64 (vms->memmap[VIRT_SMMU ].base); + smmu->flags =ACPI_IORT_SMMU_V3_COHACC_OVERRIDE ; smmu->event_gsiv =cpu_to_le32 (irq); smmu->pri_gsiv =cpu_to_le32 (irq +1 ); smmu->gerr_gsiv =cpu_to_le32 (irq +2 );
include/hw/acpi/acpi-defs.h +8 -0
@@ -628,6 +628,12 @@ struct AcpiIortItsGroup { }QEMU_PACKED ;typedef struct AcpiIortItsGroup AcpiIortItsGroup ; +enum { +ACPI_IORT_SMMU_V3_COHACC_OVERRIDE =1 <<0 , +ACPI_IORT_SMMU_V3_HTTU_OVERRIDE =3 <<1 , +ACPI_IORT_SMMU_V3_PXM_VALID =1 <<3 +}; +struct AcpiIortSmmu3 {ACPI_IORT_NODE_HEADER_DEF uint64_t base_address;
@@ -639,6 +645,8 @@ struct AcpiIortSmmu3 {uint32_t pri_gsiv;uint32_t gerr_gsiv;uint32_t sync_gsiv; +uint32_t pxm; +uint32_t id_mapping_index;AcpiIortIdMapping id_mapping_array[0 ]; }QEMU_PACKED ;typedef struct AcpiIortSmmu3 AcpiIortSmmu3 ;
Signature · 1 line
2.17.2
Hi Eric,
Quoted 9 lines
-----Original Message-----
From: Eric Auger [mailto:[email protected]]
Sent: 26 November 2018 15:46
To: [email protected]; [email protected]; qemu- [email protected]; [email protected]; [email protected]; [email protected]
Cc: Shameerali Kolothum Thodi <[email protected]>
Subject: [PATCH for-3.1] hw/arm/virt-acpi-build: Fix SMMUv3 ACPI integration
Quoted 8 lines
The AcpiIortSmmu3 misses 2 32b fields corresponding to the proximity domain and the device id mapping index.
Also let's report IO-coherent access is supported for translation table walks, descriptor fetches and queues by setting the COHACC override flag. Without that, we observe wrong command opcodes. The DT description also advertises the dma coherency.
Ah..that explains the "IDR0.COHACC overridden" and "CMD_SYNC timeout " entries in the boot log. Thanks for the fix and I can confirm this fixes the issue reported earlier[1].
FWIW:
Thanks,
Shameer
[1] https://patchwork.kernel.org/cover/10609261/
Quoted 52 lines
Fixes
a703b4f6c1ee("hw/arm/virt-acpi-build: Add smmuv3 node in IORT table")Signed-off-by Eric Auger <[email protected]>Reported-by Shameerali Kolothum Thodi--- hw/arm/virt-acpi-build.c | 1 + include/hw/acpi/acpi-defs.h | 8 ++++++++ 2 files changed, 9 insertions(+)hw/arm/virt-acpi-build.c +1
-0@@ -448,6 +448,7 @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) smmu->mapping_count = cpu_to_le32(1); smmu->mapping_offset =cpu_to_le32 (sizeof (*smmu)); smmu->base_address =cpu_to_le64 (vms->memmap[VIRT_SMMU ].base); + smmu->flags =ACPI_IORT_SMMU_V3_COHACC_OVERRIDE ; smmu->event_gsiv =cpu_to_le32 (irq); smmu->pri_gsiv =cpu_to_le32 (irq +1 );smmu->gerr_gsiv = cpu_to_le32(irq + 2);include/hw/acpi/acpi-defs.h +8
-1@@ -628,6 +628,12 @@ struct AcpiIortItsGroup { }QEMU_PACKED ;typedef struct AcpiIortItsGroup AcpiIortItsGroup ; +enum { +ACPI_IORT_SMMU_V3_COHACC_OVERRIDE =1 <<0 , +ACPI_IORT_SMMU_V3_HTTU_OVERRIDE =3 <<1 , +ACPI_IORT_SMMU_V3_PXM_VALID =1 <<3 +}; +struct AcpiIortSmmu3 {ACPI_IORT_NODE_HEADER_DEF uint64_t base_address;@@ -639,6 +645,8 @@ struct AcpiIortSmmu3 {uint32_t pri_gsiv;uint32_t gerr_gsiv;uint32_t sync_gsiv; +uint32_t pxm; +uint32_t id_mapping_index;AcpiIortIdMapping id_mapping_array[0 ]; }QEMU_PACKED ;typedef struct AcpiIortSmmu3 AcpiIortSmmu3 ; --2.17.2
Hi Shameer,
Quoted 20 lines · Shameerali Kolothum Thodi
Hi Eric,
Quoted 9 lines
-----Original Message-----
From: Eric Auger [mailto:[email protected]]
Sent: 26 November 2018 15:46
To: [email protected]; [email protected]; qemu- [email protected]; [email protected]; [email protected]; [email protected]
Cc: Shameerali Kolothum Thodi <[email protected]>
Subject: [PATCH for-3.1] hw/arm/virt-acpi-build: Fix SMMUv3 ACPI integrationQuoted 8 lines
The AcpiIortSmmu3 misses 2 32b fields corresponding to the proximity domain and the device id mapping index.
Also let's report IO-coherent access is supported for translation table walks, descriptor fetches and queues by setting the COHACC override flag. Without that, we observe wrong command opcodes. The DT description also advertises the dma coherency.
Quoted 6 lines
Ah..that explains the "IDR0.COHACC overridden" and "
CMD_SYNCtimeout " entries in the boot log. Thanks for the fix and I can confirm this fixes the issue reported earlier[1].FWIW:
Tested-by Shameer Kolothum <[email protected]>
Many thanks!
Best Regards
Eric
Quoted 60 lines
Thanks,
Shameer[1] https://patchwork.kernel.org/cover/10609261/
Quoted 52 lines
Fixes
a703b4f6c1ee("hw/arm/virt-acpi-build: Add smmuv3 node in IORT table")Signed-off-by Eric Auger <[email protected]>Reported-by Shameerali Kolothum Thodi--- hw/arm/virt-acpi-build.c | 1 + include/hw/acpi/acpi-defs.h | 8 ++++++++ 2 files changed, 9 insertions(+)hw/arm/virt-acpi-build.c +1
-0@@ -448,6 +448,7 @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) smmu->mapping_count = cpu_to_le32(1); smmu->mapping_offset =cpu_to_le32 (sizeof (*smmu)); smmu->base_address =cpu_to_le64 (vms->memmap[VIRT_SMMU ].base); + smmu->flags =ACPI_IORT_SMMU_V3_COHACC_OVERRIDE ; smmu->event_gsiv =cpu_to_le32 (irq); smmu->pri_gsiv =cpu_to_le32 (irq +1 );smmu->gerr_gsiv = cpu_to_le32(irq + 2);include/hw/acpi/acpi-defs.h +8
-1@@ -628,6 +628,12 @@ struct AcpiIortItsGroup { }QEMU_PACKED ;typedef struct AcpiIortItsGroup AcpiIortItsGroup ; +enum { +ACPI_IORT_SMMU_V3_COHACC_OVERRIDE =1 <<0 , +ACPI_IORT_SMMU_V3_HTTU_OVERRIDE =3 <<1 , +ACPI_IORT_SMMU_V3_PXM_VALID =1 <<3 +}; +struct AcpiIortSmmu3 {ACPI_IORT_NODE_HEADER_DEF uint64_t base_address;@@ -639,6 +645,8 @@ struct AcpiIortSmmu3 {uint32_t pri_gsiv;uint32_t gerr_gsiv;uint32_t sync_gsiv; +uint32_t pxm; +uint32_t id_mapping_index;AcpiIortIdMapping id_mapping_array[0 ]; }QEMU_PACKED ;typedef struct AcpiIortSmmu3 AcpiIortSmmu3 ; --2.17.2
Hi Shannon,
Quoted 2 lines · Eric Auger
The AcpiIortSmmu3 misses 2 32b fields corresponding to the proximity domain and the device id mapping index.
I fail to understand how we currently track the evolutions of the IORT structures:
Looking at the smmuv3 node in kernel include/acpi/actbl2.h
- the following fields were added in c944230064eb ACPICA: iasl: Update to IORT SMMUv3 disassembling (1 year, 4 months ago)
u8 pxm; u8 reserved1; u16 reserved2;
- id_mapping_index was added in 4c106aa411ee ACPICA: iasl: Add SMMUv3 device ID mapping index support (1 year ago)
- current u32 pxm was introduced in d87be0438e3d ACPICA: IORT: Update for revision D (6 months ago)
I would have expected each of those evolutions to be tagged by a revision increment in the acpi_iort_node.revision field but this is not done. Also I fail to see any enum for encoding the revision.
The smmuv3 struct that has been exposed until now corresponds to Rev C
https://lists.linuxfoundation.org/pipermail/iommu/2017-May/022000.html
(0c2021c047ba ACPICA: IORT: Update SMMU models for revision C (1 year, 4 months ago)
Also I noticed that in rev D, new fields were added in struct acpi_iort_root_complex. We miss them at the moment in the IORT description.
How does the guest kernel know which revision of the IORT is exposed? What do I miss?
Thanks
Eric
Quoted 55 lines
Also let's report IO-coherent access is supported for translation table walks, descriptor fetches and queues by setting the COHACC override flag. Without that, we observe wrong command opcodes. The DT description also advertises the dma coherency.
Fixes
a703b4f6c1ee("hw/arm/virt-acpi-build: Add smmuv3 node in IORT table")Signed-off-by Eric Auger <[email protected]>Reported-by Shameerali Kolothum Thodi <[email protected]>--- hw/arm/virt-acpi-build.c | 1 + include/hw/acpi/acpi-defs.h | 8 ++++++++ 2 files changed, 9 insertions(+)hw/arm/virt-acpi-build.c +1
-0@@ -448,6 +448,7 @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) smmu->mapping_count =cpu_to_le32 (1 ); smmu->mapping_offset =cpu_to_le32 (sizeof (*smmu)); smmu->base_address =cpu_to_le64 (vms->memmap[VIRT_SMMU ].base); + smmu->flags =ACPI_IORT_SMMU_V3_COHACC_OVERRIDE ; smmu->event_gsiv =cpu_to_le32 (irq); smmu->pri_gsiv =cpu_to_le32 (irq +1 ); smmu->gerr_gsiv =cpu_to_le32 (irq +2 );include/hw/acpi/acpi-defs.h +8
-0@@ -628,6 +628,12 @@ struct AcpiIortItsGroup { }QEMU_PACKED ;typedef struct AcpiIortItsGroup AcpiIortItsGroup ; +enum { +ACPI_IORT_SMMU_V3_COHACC_OVERRIDE =1 <<0 , +ACPI_IORT_SMMU_V3_HTTU_OVERRIDE =3 <<1 , +ACPI_IORT_SMMU_V3_PXM_VALID =1 <<3 +}; +struct AcpiIortSmmu3 {ACPI_IORT_NODE_HEADER_DEF uint64_t base_address;@@ -639,6 +645,8 @@ struct AcpiIortSmmu3 {uint32_t pri_gsiv;uint32_t gerr_gsiv;uint32_t sync_gsiv; +uint32_t pxm; +uint32_t id_mapping_index;AcpiIortIdMapping id_mapping_array[0 ]; }QEMU_PACKED ;typedef struct AcpiIortSmmu3 AcpiIortSmmu3 ;
Quoted 29 lines · Auger Eric
Hi Shannon,
Quoted 2 lines · Eric Auger
The AcpiIortSmmu3 misses 2 32b fields corresponding to the proximity domain and the device id mapping index.
I fail to understand how we currently track the evolutions of the IORT structures:
Looking at the smmuv3 node in kernel
include/acpi/actbl2.h- the following fields were added in
c944230064ebACPICA: iasl: Update to IORT SMMUv3 disassembling (1 year, 4 months ago)u8 pxm; u8 reserved1; u16 reserved2;-
id_mapping_indexwas added in4c106aa411eeACPICA: iasl: Add SMMUv3 device ID mapping index support (1 year ago)
- current u32 pxm was introduced ind87be0438e3dACPICA: IORT: Update for revision D (6 months ago)I would have expected each of those evolutions to be tagged by a revision increment in the acpi_iort_node.revision field but this is not done. Also I fail to see any enum for encoding the revision.
The smmuv3 struct that has been exposed until now corresponds to Rev C
https://lists.linuxfoundation.org/pipermail/iommu/2017-May/022000.html
(0c2021c047baACPICA: IORT: Update SMMU models for revision C (1 year, 4 months ago)
Quoted 6 lines
Also I noticed that in rev D, new fields were added in struct
acpi_iort_root_complex. We miss them at the moment in the IORT description.How does the guest kernel know which revision of the IORT is exposed? What do I miss?
Look at the kernel code in iort.c it checks if the flags field has set ACPI_IORT_SMMU_V3_PXM_VALID bit.
if (smmu->flags & ACPI_IORT_SMMU_V3_PXM_VALID) {
set_dev_node(dev, acpi_map_pxm_to_node(smmu->pxm));
pr_info("SMMU-v3[%llx] Mapped to Proximity domain %d\n",
smmu->base_address,
smmu->pxm);
}But it doesn't check the revision I think the reason is that revision 1 & 2 just use the next reserved fields. No matter u8 or u32, kernel driver get the same value.
The code handling id_mapping_index does check the revision.
static int iort_get_id_mapping_index(struct acpi_iort_node *node)
{
struct acpi_iort_smmu_v3 *smmu;
switch (node->type) {
case ACPI_IORT_NODE_SMMU_V3:
/*
* SMMUv3 dev ID mapping index was introduced in revision 1
* table, not available in revision 0
*/
if (node->revision < 1)
return -EINVAL;Quoted 40 lines
Thanks
Eric
Quoted 36 lines
Also let's report IO-coherent access is supported for translation table walks, descriptor fetches and queues by setting the COHACC override flag. Without that, we observe wrong command opcodes. The DT description also advertises the dma coherency.
Fixes
a703b4f6c1ee("hw/arm/virt-acpi-build: Add smmuv3 node in IORT table")Signed-off-by Eric Auger <[email protected]>Reported-by Shameerali Kolothum Thodi <[email protected]>--- hw/arm/virt-acpi-build.c | 1 + include/hw/acpi/acpi-defs.h | 8 ++++++++ 2 files changed, 9 insertions(+)hw/arm/virt-acpi-build.c +1
-0@@ -448,6 +448,7 @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) smmu->mapping_count =cpu_to_le32 (1 ); smmu->mapping_offset =cpu_to_le32 (sizeof (*smmu)); smmu->base_address =cpu_to_le64 (vms->memmap[VIRT_SMMU ].base); + smmu->flags =ACPI_IORT_SMMU_V3_COHACC_OVERRIDE ; smmu->event_gsiv =cpu_to_le32 (irq); smmu->pri_gsiv =cpu_to_le32 (irq +1 ); smmu->gerr_gsiv =cpu_to_le32 (irq +2 );include/hw/acpi/acpi-defs.h +0
-0@@ -628,6 +628,12 @@ struct AcpiIortItsGroup { }QEMU_PACKED ;typedef struct AcpiIortItsGroup AcpiIortItsGroup ;
Quoted 14 lines
Quoted 10 lines
include/hw/acpi/acpi-defs.h +6
-0+enum { +ACPI_IORT_SMMU_V3_COHACC_OVERRIDE =1 <<0 , +ACPI_IORT_SMMU_V3_HTTU_OVERRIDE =3 <<1 , +ACPI_IORT_SMMU_V3_PXM_VALID =1 <<3 +}; +struct AcpiIortSmmu3 {ACPI_IORT_NODE_HEADER_DEF uint64_t base_address;include/hw/acpi/acpi-defs.h struct AcpiIortSmmu3 { · line 648
Quoted 4 lines
include/hw/acpi/acpi-defs.h +1
-0@@ -639,6 +645,8 @@ struct AcpiIortSmmu3 {uint32_t pri_gsiv;uint32_t gerr_gsiv;uint32_t sync_gsiv; +uint32_t pxm;
So if we use this field ,we need to set the flags with ACPI_IORT_SMMU_V3_PXM_VALID.
Quoted 1 line
include/hw/acpi/acpi-defs.h struct AcpiIortSmmu3 { · line 649
Quoted 1 line
include/hw/acpi/acpi-defs.h +1
-0+uint32_t id_mapping_index;
And if we use this field, it needs to set the revision to at least 1.
Thanks,
Shannon
Quoted 4 lines
Quoted 4 lines
include/hw/acpi/acpi-defs.h +0
-0AcpiIortIdMapping id_mapping_array[0 ]; }QEMU_PACKED ;typedef struct AcpiIortSmmu3 AcpiIortSmmu3 ;
Hi Shannon,
Quoted 126 lines · Shannon Zhao
Quoted 29 lines · Auger Eric
Hi Shannon,
Quoted 2 lines · Eric Auger
The AcpiIortSmmu3 misses 2 32b fields corresponding to the proximity domain and the device id mapping index.
I fail to understand how we currently track the evolutions of the IORT structures:
Looking at the smmuv3 node in kernel
include/acpi/actbl2.h- the following fields were added in c944230064eb ACPICA: iasl: Update to IORT SMMUv3 disassembling (1 year, 4 months ago)
u8 pxm;
u8 reserved1;
u16 reserved2;
-id_mapping_indexwas added in 4c106aa411ee ACPICA: iasl: Add SMMUv3 device ID mapping index support (1 year ago)
- current u32 pxm was introduced in d87be0438e3d ACPICA: IORT: Update for revision D (6 months ago)I would have expected each of those evolutions to be tagged by a revision increment in the acpi_iort_node.revision field but this is not done. Also I fail to see any enum for encoding the revision.
The smmuv3 struct that has been exposed until now corresponds to Rev C
https://lists.linuxfoundation.org/pipermail/iommu/2017-May/022000.html
(0c2021c047ba ACPICA: IORT: Update SMMU models for revision C (1 year, 4 months ago)Quoted 7 lines
Also I noticed that in rev D, new fields were added in struct
acpi_iort_root_complex. We miss them at the moment in the IORT description.How does the guest kernel know which revision of the IORT is exposed? What do I miss?
Look at the kernel code in
iort.cit checks if the flags field has setACPI_IORT_SMMU_V3_PXM_VALIDbit.if (
smmu->flags&ACPI_IORT_SMMU_V3_PXM_VALID) {set_dev_node(dev, acpi_map_pxm_to_node(smmu->pxm)); pr_info("SMMU-v3[%llx] Mapped to Proximity domain %d\n",smmu->base_address,
smmu->pxm);
}But it doesn't check the revision I think the reason is that revision 1 & 2 just use the next reserved fields. No matter u8 or u32, kernel driver get the same value.
The code handling
id_mapping_indexdoes check the revision.static int
iort_get_id_mapping_index(struct acpi_iort_node *node){
structacpi_iort_smmu_v3*smmu;switch (
node->type) {
caseACPI_IORT_NODE_SMMU_V3:
/*
* SMMUv3 dev ID mapping index was introduced in revision 1
* table, not available in revision 0
*/
if (node->revision< 1)
return-EINVAL;Quoted 48 lines
Thanks
Eric
Quoted 44 lines
Also let's report IO-coherent access is supported for translation table walks, descriptor fetches and queues by setting the COHACC override flag. Without that, we observe wrong command opcodes. The DT description also advertises the dma coherency.
Fixes
a703b4f6c1ee("hw/arm/virt-acpi-build: Add smmuv3 node in IORT table")Signed-off-by Eric Auger <[email protected]>Reported-by Shameerali Kolothum Thodi<[email protected]>
---
hw/arm/virt-acpi-build.c | 1 +
include/hw/acpi/acpi-defs.h| 8 ++++++++2 files changed, 9 insertions(+)hw/arm/virt-acpi-build.c +1
-0@@ -448,6 +448,7 @@ build_iort(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) smmu->mapping_count = cpu_to_le32(1); smmu->mapping_offset = cpu_to_le32(sizeof(*smmu)); smmu->base_address = cpu_to_le64(vms->memmap[VIRT_SMMU].base); + smmu->flags = ACPI_IORT_SMMU_V3_COHACC_OVERRIDE; smmu->event_gsiv = cpu_to_le32(irq); smmu->pri_gsiv = cpu_to_le32(irq + 1);
smmu->gerr_gsiv=cpu_to_le32(irq + 2);include/hw/acpi/acpi-defs.h +5
-0@@ -628,6 +628,12 @@ struct AcpiIortItsGroup { } QEMU_PACKED; typedef struct AcpiIortItsGroup AcpiIortItsGroup; +enum { + ACPI_IORT_SMMU_V3_COHACC_OVERRIDE = 1 << 0, + ACPI_IORT_SMMU_V3_HTTU_OVERRIDE = 3 << 1, + ACPI_IORT_SMMU_V3_PXM_VALID = 1 << 3 +}; +Quoted 8 lines
include/hw/acpi/acpi-defs.h struct AcpiIortSmmu3 { · line 648
Quoted 8 lines
include/hw/acpi/acpi-defs.h +0
-0struct AcpiIortSmmu3 {
ACPI_IORT_NODE_HEADER_DEF
uint64_tbase_address;include/hw/acpi/acpi-defs.h +1
-0@@ -639,6 +645,8 @@ struct AcpiIortSmmu3 { uint32_t pri_gsiv; uint32_t gerr_gsiv; uint32_t sync_gsiv; + uint32_t pxm;So if we use this field ,we need to set the flags with
ACPI_IORT_SMMU_V3_PXM_VALID
Quoted 2 lines
Quoted 1 line
include/hw/acpi/acpi-defs.h struct AcpiIortSmmu3 { · line 649
Quoted 1 line
include/hw/acpi/acpi-defs.h +1
-0+ uint32_t id_mapping_index;And if we use this field, it needs to set the revision to at least 1.
But is it harmful to add those fields in the struct as this patch does?
- in our case ACPI_IORT_SMMU_V3_PXM_VALID flag is not set so the field value is ignored according to the spec and arm_smmu_v3_set_proximity() will not do anything.
- SMMU control interrupts are all GSIV based so spec says that deviceID index is ignored.
So eventually the fact the struct size is changing over the revisions does not look a problem because the node length is part of the struct and the offset to the ID array also is part of the structure.
So I could have left the structure as before (because we don't use the functionalities associated to those fields). But on the other hand it's good to upgrade the struct according to Rev D now.
So I think the patch is correct, isn't?
Thanks
Eric
Quoted 7 lines
Thanks,
ShannonQuoted 4 lines
Quoted 4 lines
include/hw/acpi/acpi-defs.h +0
-0AcpiIortIdMapping id_mapping_array[0]; } QEMU_PACKED; typedef struct AcpiIortSmmu3 AcpiIortSmmu3;
Quoted 25 lines · Auger Eric
Quoted 10 lines
Quoted 8 lines
include/hw/acpi/acpi-defs.h struct AcpiIortSmmu3 { · line 648
Quoted 8 lines
struct AcpiIortSmmu3 {
ACPI_IORT_NODE_HEADER_DEF
uint64_tbase_address;include/hw/acpi/acpi-defs.h +1
-0@@ -639,6 +645,8 @@ struct AcpiIortSmmu3 { uint32_t pri_gsiv; uint32_t gerr_gsiv; uint32_t sync_gsiv; + uint32_t pxm;So if we use this field ,we need to set the flags with
ACPI_IORT_SMMU_V3_PXM_VALIDQuoted 2 lines
Quoted 1 line
include/hw/acpi/acpi-defs.h struct AcpiIortSmmu3 { · line 649
Quoted 1 line
include/hw/acpi/acpi-defs.h +1
-0+ uint32_t id_mapping_index;And if we use this field, it needs to set the revision to at least 1.
But is it harmful to add those fields in the struct as this patch does?
- in our case
ACPI_IORT_SMMU_V3_PXM_VALIDflag is not set so the field value is ignored according to the spec andarm_smmu_v3_set_proximity()will not do anything.- SMMU control interrupts are all GSIV based so spec says that deviceID index is ignored.
So eventually the fact the struct size is changing over the revisions does not look a problem because the node length is part of the struct and the offset to the ID array also is part of the structure.
Quoted 5 lines
So I could have left the structure as before (because we don't use the functionalities associated to those fields). But on the other hand it's good to upgrade the struct according to Rev D now.
So I think the patch is correct, isn't?
Yes, I think it's not harmful but it would be better to add some comments to explain why we don't increase the revision number ATM.
Thanks,
Shannon
Hi Shannon,
Quoted 28 lines · Shannon Zhao
Quoted 25 lines · Auger Eric
Quoted 10 lines
Quoted 8 lines
include/hw/acpi/acpi-defs.h struct AcpiIortSmmu3 { · line 648
Quoted 8 lines
struct AcpiIortSmmu3 {
ACPI_IORT_NODE_HEADER_DEF
uint64_tbase_address;include/hw/acpi/acpi-defs.h +1
-0@@ -639,6 +645,8 @@ struct AcpiIortSmmu3 { uint32_t pri_gsiv; uint32_t gerr_gsiv; uint32_t sync_gsiv; + uint32_t pxm;So if we use this field ,we need to set the flags with
ACPI_IORT_SMMU_V3_PXM_VALIDQuoted 2 lines
Quoted 1 line
include/hw/acpi/acpi-defs.h struct AcpiIortSmmu3 { · line 649
Quoted 1 line
include/hw/acpi/acpi-defs.h +1
-0+ uint32_t id_mapping_index;And if we use this field, it needs to set the revision to at least 1.
But is it harmful to add those fields in the struct as this patch does?
- in our case
ACPI_IORT_SMMU_V3_PXM_VALIDflag is not set so the field value is ignored according to the spec andarm_smmu_v3_set_proximity()will not do anything.- SMMU control interrupts are all GSIV based so spec says that deviceID index is ignored.
So eventually the fact the struct size is changing over the revisions does not look a problem because the node length is part of the struct and the offset to the ID array also is part of the structure.
Quoted 7 lines
Quoted 5 lines
So I could have left the structure as before (because we don't use the functionalities associated to those fields). But on the other hand it's good to upgrade the struct according to Rev D now.
So I think the patch is correct, isn't?
Yes, I think it's not harmful but it would be better to add some comments to explain why we don't increase the revision number ATM.
OK thank you for your input. I will repost for 3.2 and will upgrade the table structs and their associated revision fields.
Thanks
Eric
Quoted 4 lines
Thanks,
Shannon
Quoted 10 lines · Eric Auger
The AcpiIortSmmu3 misses 2 32b fields corresponding to the proximity domain and the device id mapping index.
Also let's report IO-coherent access is supported for translation table walks, descriptor fetches and queues by setting the COHACC override flag. Without that, we observe wrong command opcodes. The DT description also advertises the dma coherency.
Quoted 4 lines
Fixes
a703b4f6c1ee("hw/arm/virt-acpi-build: Add smmuv3 node in IORT table")Signed-off-by Eric Auger <[email protected]>Reported-by Shameerali Kolothum Thodi <[email protected]>
Hi; it looks (from looking at the code) like this isn't a regression from 3.0 -- is that right?
I think this fix has missed rc3 (discussion seems to be ongoing and it needs more time on-list for review), which means it's likely to miss the 3.1 release. So at the moment I'm expecting it to go into 4.0. (Maybe if we need an rc4 for some other reason we might be able to put it in.)
thanks
-- PMM
Hi Peter,
Quoted 11 lines · Peter Maydell
Quoted 10 lines · Eric Auger
The AcpiIortSmmu3 misses 2 32b fields corresponding to the proximity domain and the device id mapping index.
Also let's report IO-coherent access is supported for translation table walks, descriptor fetches and queues by setting the COHACC override flag. Without that, we observe wrong command opcodes. The DT description also advertises the dma coherency.
Quoted 7 lines
Quoted 4 lines
Fixes
a703b4f6c1ee("hw/arm/virt-acpi-build: Add smmuv3 node in IORT table")Signed-off-by Eric Auger <[email protected]>Reported-by Shameerali Kolothum Thodi <[email protected]>Hi; it looks (from looking at the code) like this isn't a regression from 3.0 -- is that right?
Correct, the lack of COHACC override flag has been there from the very beginning.
Quoted 6 lines
I think this fix has missed rc3 (discussion seems to be ongoing and it needs more time on-list for review), which means it's likely to miss the 3.1 release. So at the moment I'm expecting it to go into 4.0. (Maybe if we need an rc4 for some other reason we might be able to put it in.)
OK
Thanks
Eric
Quoted 4 lines
thanks
-- PMM