iommu: Clean up cookie and sw_msi in struct iommu_domain
Subsystems (3)
v4 cover iommu: Clean up cookie and sw_msi in struct iommu_domain
A few things in iommu_domain structure (such as iova_cookie, iommufd_hwpt, fault handler, and etc) are exclusive to certain types of iommu_domains. So, all of them can be seen as domain cookies in different cookie types.
Furthermore, iova_cookie is shared with the msi_cookie use case. It could be cleaner by decoupling these too.
Define an enum iommu_domain_cookie_type to fit all those cookies in.
Since there are only two sw_msi implementations, check domain->cookie_type in iommu_dma_prepare_msi() to call the corresponding sw_msi function.
This is a clean-up series for the previous sw_msi Part-1 core series. It's on github:
https://github.com/nicolinc/iommufd/commits/iommufd_msi_cleanup-v4
Changelog
v4
* Rebase on top of a bug fix for hwpt_iommufd cookie
https://lore.kernel.org/all/[email protected]/
* Add Reviewed-by from Robin and Kevin
* Add missing "break;" in iommu_domain_free()
* Drop "linux/msi.h" header from the iommufd/device code
* Set iommufd_sw_msi_install's SYMBOL to "IOMMUFD_INTERNAL"
* Update commit messages for typo fixing and size comparison
* Add a local msi_page_list to cache cookie_msi_pages(domain) return
* Add an inline iommufd_group_setup_msi() for !CONFIG_IRQ_MSI_IOMMU
v3
https://lore.kernel.org/all/[email protected]/
* Integrate Robin's patch
https://lore.kernel.org/linux-iommu/4e68d5820be06adc1b34fc0d1c9399481151daee.1740742271.git.robin.murphy@arm.com/
replacing private_data_owner with the new cookie_type
* Move iommu_sw_msi stub from the public header to iommu-priv.h
v2
https://lore.kernel.org/all/[email protected]/
* Drop sw_msi function pointer
* Add a new private_data_owner tag in iommu_domain
* Let iommu core call the two sw_msi implementations directly
v1
https://lore.kernel.org/all/[email protected]/
Thanks
Nicolin
Nicolin Chen (2):
iommufd: Move iommufd_sw_msi and related functions to driver.c
iommu: Drop sw_msi from iommu_domain
Robin Murphy (1):
iommu: Sort out domain user data
drivers/iommu/dma-iommu.h | 14 ++ drivers/iommu/iommu-priv.h | 13 ++ drivers/iommu/iommufd/iommufd_private.h | 7 +- include/linux/iommu.h | 35 ++-- drivers/iommu/dma-iommu.c | 208 ++++++++++++------------ drivers/iommu/iommu-sva.c | 1 + drivers/iommu/iommu.c | 35 +++- drivers/iommu/iommufd/device.c | 131 +-------------- drivers/iommu/iommufd/driver.c | 125 ++++++++++++++ drivers/iommu/iommufd/hw_pagetable.c | 6 +- 10 files changed, 320 insertions(+), 255 deletions(-)
base-commit: 4b022ff93d8b332891cae818a7331942db50610f
Signature · 1 line
2.43.0
PATCH v4 1/3 iommu: Sort out domain user data
From: Robin Murphy <[email protected]>
When DMA/MSI cookies were made first-class citizens back in commit 46983fcd67ac ("iommu: Pull IOVA cookie management into the core"), there was no real need to further expose the two different cookie types. However, now that IOMMUFD wants to add a third type of MSI-mapping cookie, we do have a nicely compelling reason to properly dismabiguate things at the domain level beyond just vaguely guessing from the domain type.
Meanwhile, we also effectively have another "cookie" in the form of the anonymous union for other user data, which isn't much better in terms of being vague and unenforced. The fact is that all these cookie types are mutually exclusive, in the sense that combining them makes zero sense and/or would be catastrophic (iommu_set_fault_handler() on an SVA domain, anyone?) - the only combination which *might* be reasonable is perhaps a fault handler and an MSI cookie, but nobody's doing that at the moment, so let's rule it out as well for the sake of being clear and robust. To that end, we pull DMA and MSI cookies apart a little more, mostly to clear up the ambiguity at domain teardown, then for clarity (and to save a little space), move them into the union, whose ownership we can then properly describe and enforce entirely unambiguously.
[nicolinc: rebase on latest tree; use prefix IOMMU_COOKIE_; merge unions
in iommu_domain; add IOMMU_COOKIE_IOMMUFD for iommufd_hwpt]
--- drivers/iommu/dma-iommu.h | 5 + include/linux/iommu.h | 20 ++- drivers/iommu/dma-iommu.c | 194 ++++++++++++++------------- drivers/iommu/iommu-sva.c | 1 + drivers/iommu/iommu.c | 18 ++- drivers/iommu/iommufd/hw_pagetable.c | 3 + 6 files changed, 143 insertions(+), 98 deletions(-)
drivers/iommu/dma-iommu.h +5 -0
@@ -13,6 +13,7 @@ void iommu_setup_dma_ops(struct device *dev);int iommu_get_dma_cookie (struct iommu_domain *domain);void iommu_put_dma_cookie (struct iommu_domain *domain); +void iommu_put_msi_cookie (struct iommu_domain *domain);int iommu_dma_init_fq (struct iommu_domain *domain);
@@ -40,6 +41,10 @@ static inline void iommu_put_dma_cookie(struct iommu_domain *domain) { } +static inline void iommu_put_msi_cookie (struct iommu_domain *domain) +{ +} +static inline void iommu_dma_get_resv_regions (struct device *dev,struct list_head *list) { }
include/linux/iommu.h +15 -5
@@ -41,6 +41,7 @@ struct iommu_dirty_ops;struct notifier_block ;struct iommu_sva ;struct iommu_dma_cookie ; +struct iommu_dma_msi_cookie ;struct iommu_fault_param ;struct iommufd_ctx ;struct iommufd_viommu ;
@@ -165,6 +166,15 @@ struct iommu_domain_geometry {bool force_aperture;/* DMA only allowed in mappable range? */ }; +enum iommu_domain_cookie_type { +IOMMU_COOKIE_NONE , +IOMMU_COOKIE_DMA_IOVA , +IOMMU_COOKIE_DMA_MSI , +IOMMU_COOKIE_FAULT_HANDLER , +IOMMU_COOKIE_SVA , +IOMMU_COOKIE_IOMMUFD , +}; +/* Domain feature flags */ #define __IOMMU_DOMAIN_PAGING (1U << 0)/* Support for iommu_map/unmap */ #define __IOMMU_DOMAIN_DMA_API (1U << 1)/* Domain for use in DMA-API
@@ -211,12 +221,12 @@ struct iommu_domain_geometry {struct iommu_domain {unsigned type; +enum iommu_domain_cookie_type cookie_type;const struct iommu_domain_ops *ops;const struct iommu_dirty_ops *dirty_ops;const struct iommu_ops *owner;/* Whose domain_alloc we came from */ unsigned long pgsize_bitmap;/* Bitmap of page sizes in use */ struct iommu_domain_geometry geometry; -struct iommu_dma_cookie *iova_cookie;int (*iopf_handler)(struct iopf_group *group);#if IS_ENABLED (CONFIG_IRQ_MSI_IOMMU )
@@ -224,10 +234,10 @@ struct iommu_domain { phys_addr_t msi_addr);#endif -union {/* Pointer usable by owner of the domain */ -struct iommufd_hw_pagetable *iommufd_hwpt;/* iommufd */ - }; - union { /* Fault handler */ +union {/* cookie */ +struct iommu_dma_cookie *iova_cookie; +struct iommu_dma_msi_cookie *msi_cookie; +struct iommufd_hw_pagetable *iommufd_hwpt;struct {iommu_fault_handler_t handler;void *handler_token;
drivers/iommu/dma-iommu.c +104 -90
343 lines not on this page · raw message
drivers/iommu/iommu-sva.c +1 -0
@@ -310,6 +310,7 @@ static struct iommu_domain *iommu_sva_domain_alloc(struct device *dev, } domain->type =IOMMU_DOMAIN_SVA ; + domain->cookie_type =IOMMU_COOKIE_SVA ;mmgrab (mm); domain->mm = mm; domain->owner = ops;
drivers/iommu/iommu.c +15 -3
@@ -1953,8 +1953,10 @@ void iommu_set_fault_handler(struct iommu_domain *domain, iommu_fault_handler_t handler, void *token) { - BUG_ON(!domain); +if (WARN_ON (!domain || domain->cookie_type !=IOMMU_COOKIE_NONE )) +return ; + domain->cookie_type =IOMMU_COOKIE_FAULT_HANDLER ; domain->handler = handler; domain->handler_token = token; }
@@ -2024,9 +2026,19 @@ EXPORT_SYMBOL_GPL(iommu_paging_domain_alloc_flags);void iommu_domain_free (struct iommu_domain *domain) { -if (domain->type ==IOMMU_DOMAIN_SVA ) +switch (domain->cookie_type) { +case IOMMU_COOKIE_DMA_IOVA : +iommu_put_dma_cookie (domain); +break ; +case IOMMU_COOKIE_DMA_MSI : +iommu_put_msi_cookie (domain); +break ; +case IOMMU_COOKIE_SVA :mmdrop (domain->mm); -iommu_put_dma_cookie (domain); +break ; +default : +break ; + }if (domain->ops->free) domain->ops->free (domain); }
drivers/iommu/iommufd/hw_pagetable.c +3 -0
@@ -160,6 +160,7 @@ iommufd_hwpt_paging_alloc(struct iommufd_ctx *ictx, struct iommufd_ioas *ioas, } } hwpt->domain->iommufd_hwpt = hwpt; + hwpt->domain->cookie_type =IOMMU_COOKIE_IOMMUFD ;iommu_domain_set_sw_msi (hwpt->domain, iommufd_sw_msi);/*
@@ -257,6 +258,7 @@ iommufd_hwpt_nested_alloc(struct iommufd_ctx *ictx, } hwpt->domain->owner = ops; hwpt->domain->iommufd_hwpt = hwpt; + hwpt->domain->cookie_type =IOMMU_COOKIE_IOMMUFD ;iommu_domain_set_sw_msi (hwpt->domain, iommufd_sw_msi);if (WARN_ON_ONCE (hwpt->domain->type !=IOMMU_DOMAIN_NESTED )) {
@@ -315,6 +317,7 @@ iommufd_viommu_alloc_hwpt_nested(struct iommufd_viommu *viommu, u32 flags, } hwpt->domain->iommufd_hwpt = hwpt; hwpt->domain->owner = viommu->iommu_dev->ops; + hwpt->domain->cookie_type =IOMMU_COOKIE_IOMMUFD ;iommu_domain_set_sw_msi (hwpt->domain, iommufd_sw_msi);if (WARN_ON_ONCE (hwpt->domain->type !=IOMMU_DOMAIN_NESTED )) {
Signature · 1 line
2.43.0
Re: [PATCH v4 1/3] iommu: Sort out domain user data
Quoted 107 lines · Nicolin Chen
From: Robin Murphy<[email protected]>
When DMA/MSI cookies were made first-class citizens back in commit
46983fcd67ac("iommu: Pull IOVA cookie management into the core"), there was no real need to further expose the two different cookie types. However, now that IOMMUFD wants to add a third type of MSI-mapping cookie, we do have a nicely compelling reason to properly dismabiguate things at the domain level beyond just vaguely guessing from the domain type.Meanwhile, we also effectively have another "cookie" in the form of the anonymous union for other user data, which isn't much better in terms of being vague and unenforced. The fact is that all these cookie types are mutually exclusive, in the sense that combining them makes zero sense and/or would be catastrophic (
iommu_set_fault_handler()on an SVA domain, anyone?) - the only combination which*might* be reasonable is perhaps a fault handler and an MSI cookie, but nobody's doing that at the moment, so let's rule it out as well for the sake of being clear and robust. To that end, we pull DMA and MSI cookies apart a little more, mostly to clear up the ambiguity at domain teardown, then for clarity (and to save a little space), move them into the union, whose ownership we can then properly describe and enforce entirely unambiguously.Signed-off-by Robin Murphy<[email protected]>Reviewed-by Kevin Tian<[email protected]>[nicolinc: rebase on latest tree; use prefix
IOMMU_COOKIE_; merge unions
iniommu_domain; addIOMMU_COOKIE_IOMMUFDforiommufd_hwpt]Signed-off-by Nicolin Chen<[email protected]>--- drivers/iommu/dma-iommu.h | 5 + include/linux/iommu.h | 20 ++- drivers/iommu/dma-iommu.c | 194 ++++++++++++++------------- drivers/iommu/iommu-sva.c | 1 + drivers/iommu/iommu.c | 18 ++- drivers/iommu/iommufd/hw_pagetable.c | 3 + 6 files changed, 143 insertions(+), 98 deletions(-)drivers/iommu/dma-iommu.h +5
-0@@ -13,6 +13,7 @@ void iommu_setup_dma_ops(struct device *dev);int iommu_get_dma_cookie (struct iommu_domain *domain);void iommu_put_dma_cookie (struct iommu_domain *domain); +void iommu_put_msi_cookie (struct iommu_domain *domain);int iommu_dma_init_fq (struct iommu_domain *domain);@@ -40,6 +41,10 @@ static inline void iommu_put_dma_cookie(struct iommu_domain *domain) { } +static inline void iommu_put_msi_cookie (struct iommu_domain *domain) +{ +} +static inline void iommu_dma_get_resv_regions (struct device *dev,struct list_head *list) { }include/linux/iommu.h +11
-2@@ -41,6 +41,7 @@ struct iommu_dirty_ops;struct notifier_block ;struct iommu_sva ;struct iommu_dma_cookie ; +struct iommu_dma_msi_cookie ;struct iommu_fault_param ;struct iommufd_ctx ;struct iommufd_viommu ;@@ -165,6 +166,15 @@ struct iommu_domain_geometry {bool force_aperture;/* DMA only allowed in mappable range? */ }; +enum iommu_domain_cookie_type { +IOMMU_COOKIE_NONE , +IOMMU_COOKIE_DMA_IOVA , +IOMMU_COOKIE_DMA_MSI , +IOMMU_COOKIE_FAULT_HANDLER , +IOMMU_COOKIE_SVA , +IOMMU_COOKIE_IOMMUFD , +}; +/* Domain feature flags */ #define __IOMMU_DOMAIN_PAGING (1U << 0)/* Support for iommu_map/unmap */ #define __IOMMU_DOMAIN_DMA_API (1U << 1)/* Domain for use in DMA-API @@ -211,12 +221,12 @@ struct iommu_domain_geometry {struct iommu_domain {unsigned type; +enum iommu_domain_cookie_type cookie_type;const struct iommu_domain_ops *ops;const struct iommu_dirty_ops *dirty_ops;const struct iommu_ops *owner;/* Whose domain_alloc we came from */ unsigned long pgsize_bitmap;/* Bitmap of page sizes in use */ struct iommu_domain_geometry geometry; -struct iommu_dma_cookie *iova_cookie;int (*iopf_handler)(struct iopf_group *group);#if IS_ENABLED (CONFIG_IRQ_MSI_IOMMU )@@ -224,10 +234,10 @@ struct iommu_domain { phys_addr_t msi_addr);#endif -union {/* Pointer usable by owner of the domain */
include/linux/iommu.h struct iommu_domain { · line 243
Quoted 10 lines
include/linux/iommu.h +4
-3-struct iommufd_hw_pagetable *iommufd_hwpt;/* iommufd */ - }; - union { /* Fault handler */ +union {/* cookie */ +struct iommu_dma_cookie *iova_cookie; +struct iommu_dma_msi_cookie *msi_cookie; +struct iommufd_hw_pagetable *iommufd_hwpt;struct {iommu_fault_handler_t handler;void *handler_token;
My feeling is that IOMMU_COOKIE_FAULT_HANDLER isn't exclusive to IOMMU_COOKIE_DMA_IOVA; both might be used for kernel DMA with a paging domain.
I am afraid that iommu_set_fault_handler() doesn't work anymore as the domain's cookie type has already been set to IOMMU_COOKIE_DMA_IOVA.
void iommu_set_fault_handler(struct iommu_domain *domain,
iommu_fault_handler_t handler, void *token)
{
if (WARN_ON(!domain || domain->cookie_type != IOMMU_COOKIE_NONE))
return;
domain->cookie_type = IOMMU_COOKIE_FAULT_HANDLER;
domain->handler = handler;
domain->handler_token = token;
}EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
Anybody has a chance to test whether above WARN_ON will be triggered?
Thanks,
baolu
Re: [PATCH v4 1/3] iommu: Sort out domain user data
Hi Baolu,
Quoted 2 lines · Baolu Lu
Quoted 1 line · Nicolin Chen
From: Robin Murphy<[email protected]>
Robin had remarks here, wrt iommu_set_fault_handler():
Quoted 7 lines
Quoted 7 lines
The fact is that all these cookie types are
mutually exclusive, in the sense that combining them makes zero sense and/or would be catastrophic (iommu_set_fault_handler()on an SVA domain, anyone?) - the only combination which*might* be reasonable is perhaps a fault handler and an MSI cookie, but nobody's doing that at the moment, so let's rule it out as well for the sake of being clear and robust.
[...]
Quoted 19 lines
Quoted 4 lines
include/linux/iommu.h +0
-1@@ -224,10 +234,10 @@ struct iommu_domain { phys_addr_t msi_addr);#endif -union {/* Pointer usable by owner of the domain */ include/linux/iommu.h struct iommu_domain { · line 242
Quoted 10 lines
include/linux/iommu.h +4
-3-struct iommufd_hw_pagetable *iommufd_hwpt;/* iommufd */ - }; - union { /* Fault handler */ +union {/* cookie */ +struct iommu_dma_cookie *iova_cookie; +struct iommu_dma_msi_cookie *msi_cookie; +struct iommufd_hw_pagetable *iommufd_hwpt;struct {iommu_fault_handler_t handler;void *handler_token;My feeling is that
IOMMU_COOKIE_FAULT_HANDLERisn't exclusive toIOMMU_COOKIE_DMA_IOVA; both might be used for kernel DMA with a paging domain.
Quoted 2 lines
I am afraid that
iommu_set_fault_handler()doesn't work anymore as the domain's cookie type has already been set toIOMMU_COOKIE_DMA_IOVA.
All three existing iommu_set_fault_handler() callers in the tree are UNMANAGED domain users:
5 451 drivers/gpu/drm/msm/msm_iommu.c <<msm_iommu_gpu_new>>
iommu_set_fault_handler(iommu->domain, msm_fault_handler, iommu);
6 453 drivers/infiniband/hw/usnic/usnic_uiom.c <<usnic_uiom_alloc_pd>>
iommu_set_fault_handler(pd->domain, usnic_uiom_dma_fault, NULL);
8 118 drivers/remoteproc/remoteproc_core.c <<rproc_enable_iommu>>
iommu_set_fault_handler(domain, rproc_iommu_fault, rproc);On the other hand, IOMMU_COOKIE_DMA_IOVA is a private cookie for dma-iommu only.
So, I think we are probably fine?
Thanks
Nicolin
Re: [PATCH v4 1/3] iommu: Sort out domain user data
Quoted 42 lines · Nicolin Chen
Quoted 2 lines · Baolu Lu
Quoted 1 line · Nicolin Chen
From: Robin Murphy<[email protected]>
Robin had remarks here, wrt
iommu_set_fault_handler():Quoted 7 lines
Quoted 7 lines
The fact is that all these cookie types are
mutually exclusive, in the sense that combining them makes zero sense and/or would be catastrophic (iommu_set_fault_handler()on an SVA domain, anyone?) - the only combination which*might* be reasonable is perhaps a fault handler and an MSI cookie, but nobody's doing that at the moment, so let's rule it out as well for the sake of being clear and robust.[...]
Quoted 18 lines
Quoted 4 lines
include/linux/iommu.h +0
-1@@ -224,10 +234,10 @@ struct iommu_domain { phys_addr_t msi_addr);#endif -union {/* Pointer usable by owner of the domain */ include/linux/iommu.h struct iommu_domain { · line 242
Quoted 10 lines
include/linux/iommu.h +4
-3-struct iommufd_hw_pagetable *iommufd_hwpt;/* iommufd */ - }; - union { /* Fault handler */ +union {/* cookie */ +struct iommu_dma_cookie *iova_cookie; +struct iommu_dma_msi_cookie *msi_cookie; +struct iommufd_hw_pagetable *iommufd_hwpt;struct {iommu_fault_handler_t handler;void *handler_token;exs My feeling is that
IOMMU_COOKIE_FAULT_HANDLERisn't exclusive toIOMMU_COOKIE_DMA_IOVA; both might be used for kernel DMA with a paging domain.Quoted 2 lines
I am afraid that
iommu_set_fault_handler()doesn't work anymore as the domain's cookie type has already been set toIOMMU_COOKIE_DMA_IOVA.All three existing
iommu_set_fault_handler()callers in the tree are UNMANAGED domain users:5 451 drivers/gpu/drm/msm/msm_iommu.c <<msm_iommu_gpu_new>> iommu_set_fault_handler(iommu->domain, msm_fault_handler, iommu); 6 453 drivers/infiniband/hw/usnic/usnic_uiom.c <<usnic_uiom_alloc_pd>> iommu_set_fault_handler(pd->domain, usnic_uiom_dma_fault, NULL); 8 118 drivers/remoteproc/remoteproc_core.c <<rproc_enable_iommu>> iommu_set_fault_handler(domain, rproc_iommu_fault, rproc);
Quoted 4 lines
On the other hand,
IOMMU_COOKIE_DMA_IOVAis a private cookie for dma-iommu only.So, I think we are probably fine?
If all existing use cases are for UNMANAGED domains, that's fine. And when iommu_set_fault_handler() is miss-used, we already have a WARN_ON() there.
Thanks,
baolu
Robin Murphy
maintainerRe: [PATCH v4 1/3] iommu: Sort out domain user data
Quoted 43 lines · Baolu Lu
Quoted 42 lines · Nicolin Chen
Quoted 2 lines · Baolu Lu
Quoted 1 line · Nicolin Chen
From: Robin Murphy<[email protected]>
Robin had remarks here, wrt
iommu_set_fault_handler():Quoted 7 lines
Quoted 7 lines
The fact is that all these cookie types are
mutually exclusive, in the sense that combining them makes zero sense and/or would be catastrophic (iommu_set_fault_handler()on an SVA domain, anyone?) - the only combination which*might* be reasonable is perhaps a fault handler and an MSI cookie, but nobody's doing that at the moment, so let's rule it out as well for the sake of being clear and robust.[...]
Quoted 18 lines
Quoted 13 lines
include/linux/iommu.h +4
-4@@ -224,10 +234,10 @@ struct iommu_domain { phys_addr_t msi_addr); #endif - union { /* Pointer usable by owner of the domain */ - struct iommufd_hw_pagetable *iommufd_hwpt; /* iommufd */ - }; - union { /* Fault handler */ + union { /* cookie */ + struct iommu_dma_cookie *iova_cookie; + struct iommu_dma_msi_cookie *msi_cookie; + struct iommufd_hw_pagetable *iommufd_hwpt; struct { iommu_fault_handler_t handler;Quoted 1 line
void *handler_token;exs
My feeling is that
IOMMU_COOKIE_FAULT_HANDLERisn't exclusive toIOMMU_COOKIE_DMA_IOVA; both might be used for kernel DMA with a paging domain.Quoted 2 lines
I am afraid that
iommu_set_fault_handler()doesn't work anymore as the domain's cookie type has already been set toIOMMU_COOKIE_DMA_IOVA.All three existing
iommu_set_fault_handler()callers in the tree are UNMANAGED domain users:
5 451drivers/gpu/drm/msm/msm_iommu.c<<msm_iommu_gpu_new>>iommu_set_fault_handler(iommu->domain, msm_fault_handler, iommu); 6 453drivers/infiniband/hw/usnic/usnic_uiom.c<<usnic_uiom_alloc_pd>>iommu_set_fault_handler(pd->domain, usnic_uiom_dma_fault, NULL); 8 118drivers/remoteproc/remoteproc_core.c<<rproc_enable_iommu>>iommu_set_fault_handler(domain, rproc_iommu_fault, rproc);
Quoted 7 lines
Quoted 4 lines
On the other hand,
IOMMU_COOKIE_DMA_IOVAis a private cookie for dma-iommu only.So, I think we are probably fine?
If all existing use cases are for UNMANAGED domains, that's fine. And when
iommu_set_fault_handler()is miss-used, we already have aWARN_ON()there.
Right, it would be illogical for a driver to set a fault handler on a DMA domain since it doesn't control the IOVA space to be able to do any fault-handling, and iommu-dma itself isn't ever going to use a fault handler because it expects the DMA API to be used correctly and thus no faults to occur.
TBH at this point I view the fault_handler stuff as a legacy interface which we don't really want to encourage use of anyway - it's already proven not to be great for any true fault handling since many drivers can only call report_iommu_fault() in IRQ context. If some new case does come up in future where this mutual exclusion gets in the way, I would say that's the point where we then look at reworking the whole thing into a dedicated "fault notifier" mechanism instead, which could then logically be orthogonal to the IOVA-space-owner cookie.
Thanks,
Robin.
Jason Gunthorpe
maintainerRe: [PATCH v4 1/3] iommu: Sort out domain user data
Quoted 8 lines · Robin Murphy
TBH at this point I view the
fault_handlerstuff as a legacy interface which we don't really want to encourage use of anyway - it's already proven not to be great for any true fault handling since many drivers can only callreport_iommu_fault()in IRQ context. If some new case does come up in future where this mutual exclusion gets in the way, I would say that's the point where we then look at reworking the whole thing into a dedicated "fault notifier" mechanism instead, which could then logically be orthogonal to the IOVA-space-owner cookie.
I was under the impression we would go forward with the PRI focused interface we already have:
int (*iopf_handler)(struct iopf_group *group);
And this olld iommu_fault_handler_t is just because nobody wanted to go in any update the old drivers and DRM to use the new scheme?
Is there something fundamental that would prevent iommu drivers from using the new reporting path?
Jason
Re: [PATCH v4 1/3] iommu: Sort out domain user data
In reply to Nicolin ChenQuoted 28 lines · Nicolin Chen
From: Robin Murphy <[email protected]>
When DMA/MSI cookies were made first-class citizens back in commit
46983fcd67ac("iommu: Pull IOVA cookie management into the core"), there was no real need to further expose the two different cookie types. However, now that IOMMUFD wants to add a third type of MSI-mapping cookie, we do have a nicely compelling reason to properly dismabiguate things at the domain level beyond just vaguely guessing from the domain type.Meanwhile, we also effectively have another "cookie" in the form of the anonymous union for other user data, which isn't much better in terms of being vague and unenforced. The fact is that all these cookie types are mutually exclusive, in the sense that combining them makes zero sense and/or would be catastrophic (
iommu_set_fault_handler()on an SVA domain, anyone?) - the only combination which *might* be reasonable is perhaps a fault handler and an MSI cookie, but nobody's doing that at the moment, so let's rule it out as well for the sake of being clear and robust. To that end, we pull DMA and MSI cookies apart a little more, mostly to clear up the ambiguity at domain teardown, then for clarity (and to save a little space), move them into the union, whose ownership we can then properly describe and enforce entirely unambiguously.Signed-off-by Robin Murphy <[email protected]>Reviewed-by Kevin Tian <[email protected]>[nicolinc: rebase on latest tree; use prefix
IOMMU_COOKIE_; merge unions
iniommu_domain; addIOMMU_COOKIE_IOMMUFDforiommufd_hwpt]Signed-off-by Nicolin Chen <[email protected]>
Quoted 8 lines
--- drivers/iommu/dma-iommu.h | 5 + include/linux/iommu.h | 20 ++- drivers/iommu/dma-iommu.c | 194 ++++++++++++++------------- drivers/iommu/iommu-sva.c | 1 + drivers/iommu/iommu.c | 18 ++- drivers/iommu/iommufd/hw_pagetable.c | 3 + 6 files changed, 143 insertions(+), 98 deletions(-)
Jason
PATCH v4 2/3 iommufd: Move iommufd_sw_msi and related functions to driver.c
In reply to Nicolin ChenTo provide the iommufd_sw_msi() to the iommu core that is under a different Kconfig, move it and its related functions to driver.c. Then, stub it into the iommu-priv header. The iommufd_sw_msi_install() continues to be used by iommufd internal, so put it in the private header.
Since this affects the module size, here is before-n-after size comparison:
[Before]
text data bss dec hex filename
18797 848 56 19701 4cf5 drivers/iommu/iommufd/device.o
722 44 0 766 2fe drivers/iommu/iommufd/driver.o
[After]
text data bss dec hex filename
17671 808 56 18535 4867 drivers/iommu/iommufd/device.o
1900 100 0 2000 7d0 drivers/iommu/iommufd/driver.o
--- drivers/iommu/iommu-priv.h | 13 +++ drivers/iommu/iommufd/iommufd_private.h | 7 +- drivers/iommu/iommufd/device.c | 131 ++---------------------- drivers/iommu/iommufd/driver.c | 125 ++++++++++++++++++++++ 4 files changed, 151 insertions(+), 125 deletions(-)
drivers/iommu/iommu-priv.h +13 -0
@@ -5,6 +5,7 @@#define __LINUX_IOMMU_PRIV_H#include <linux/iommu.h> +#include <linux/msi.h> static inline const struct iommu_ops *dev_iommu_ops (struct device *dev) {
@@ -43,4 +44,16 @@ void iommu_detach_group_handle(struct iommu_domain *domain,int iommu_replace_group_handle (struct iommu_group *group,struct iommu_domain *new_domain,struct iommu_attach_handle *handle); + +#if IS_ENABLED (CONFIG_IOMMUFD_DRIVER_CORE ) &&IS_ENABLED (CONFIG_IRQ_MSI_IOMMU ) +int iommufd_sw_msi (struct iommu_domain *domain,struct msi_desc *desc, +phys_addr_t msi_addr); +#else /* !CONFIG_IOMMUFD_DRIVER_CORE || !CONFIG_IRQ_MSI_IOMMU */ +static inline int iommufd_sw_msi (struct iommu_domain *domain, +struct msi_desc *desc,phys_addr_t msi_addr) +{ +return -EOPNOTSUPP ; +} +#endif /* CONFIG_IOMMUFD_DRIVER_CORE && CONFIG_IRQ_MSI_IOMMU */ +#endif /* __LINUX_IOMMU_PRIV_H */
drivers/iommu/iommufd/iommufd_private.h +5 -2
@@ -32,8 +32,11 @@ struct iommufd_sw_msi_maps {DECLARE_BITMAP (bitmap,64 ); }; -int iommufd_sw_msi (struct iommu_domain *domain,struct msi_desc *desc, -phys_addr_t msi_addr); +#ifdef CONFIG_IRQ_MSI_IOMMU +int iommufd_sw_msi_install (struct iommufd_ctx *ictx, +struct iommufd_hwpt_paging *hwpt_paging, +struct iommufd_sw_msi_map *msi_map); +#endif struct iommufd_ctx {struct file *file;
drivers/iommu/iommufd/device.c +8 -123
150 lines not on this page · raw message
drivers/iommu/iommufd/driver.c +125 -0
130 lines not on this page · raw message
Signature · 1 line
2.43.0
RE: [PATCH v4 2/3] iommufd: Move iommufd_sw_msi and related functions to driver.c
Quoted 14 lines
From: Nicolin Chen <[email protected]>
Sent: Friday, March 7, 2025 5:01 AMTo provide the
iommufd_sw_msi()to the iommu core that is under a differentKconfig, move it and its related functions todriver.c. Then, stub it into the iommu-priv header. Theiommufd_sw_msi_install()continues to be used by
iommufd internal, so put it in the private header.Since this affects the module size, here is before-n-after size comparison:
[Before]
text data bss dec hex filename
18797 848 56 19701 4cf5drivers/iommu/iommufd/device.o
Quoted 7 lines
722 44 0 766 2fe drivers/iommu/iommufd/driver.o[After]
text data bss dec hex filename
17671 808 56 18535 4867drivers/iommu/iommufd/device.o
1900 100 0 2000 7d0drivers/iommu/iommufd/driver.oSigned-off-by Nicolin Chen <[email protected]>
Re: [PATCH v4 2/3] iommufd: Move iommufd_sw_msi and related functions to driver.c
In reply to Nicolin ChenQuoted 15 lines · Nicolin Chen
To provide the
iommufd_sw_msi()to the iommu core that is under a differentKconfig, move it and its related functions todriver.c. Then, stub it into the iommu-priv header. Theiommufd_sw_msi_install()continues to be used by iommufd internal, so put it in the private header.Since this affects the module size, here is before-n-after size comparison:
[Before]
text data bss dec hex filename
18797 848 56 19701 4cf5drivers/iommu/iommufd/device.o722 44 0 766 2fe drivers/iommu/iommufd/driver.o[After]
text data bss dec hex filename
17671 808 56 18535 4867drivers/iommu/iommufd/device.o
1900 100 0 2000 7d0drivers/iommu/iommufd/driver.o
Quoted 7 lines
Signed-off-by Nicolin Chen <[email protected]>--- drivers/iommu/iommu-priv.h | 13 +++ drivers/iommu/iommufd/iommufd_private.h | 7 +- drivers/iommu/iommufd/device.c | 131 ++---------------------- drivers/iommu/iommufd/driver.c | 125 ++++++++++++++++++++++ 4 files changed, 151 insertions(+), 125 deletions(-)
+1K is more than I would have liked, but OK
Jason
PATCH v4 3/3 iommu: Drop sw_msi from iommu_domain
In reply to Nicolin ChenThere are only two sw_msi implementations in the entire system, thus it's not very necessary to have an sw_msi pointer.
Instead, check domain->cookie_type to call the two sw_msi implementations directly from the core code.
--- drivers/iommu/dma-iommu.h | 9 +++++++++ include/linux/iommu.h | 15 --------------- drivers/iommu/dma-iommu.c | 14 ++------------ drivers/iommu/iommu.c | 17 +++++++++++++++-- drivers/iommu/iommufd/hw_pagetable.c | 3 --- 5 files changed, 26 insertions(+), 32 deletions(-)
drivers/iommu/dma-iommu.h +9 -0
@@ -19,6 +19,9 @@ int iommu_dma_init_fq(struct iommu_domain *domain);void iommu_dma_get_resv_regions (struct device *dev,struct list_head *list); +int iommu_dma_sw_msi (struct iommu_domain *domain,struct msi_desc *desc, +phys_addr_t msi_addr); +extern bool iommu_dma_forcedac;#else /* CONFIG_IOMMU_DMA */
@@ -49,5 +52,11 @@ static inline void iommu_dma_get_resv_regions(struct device *dev, struct list_he { } +static inline int iommu_dma_sw_msi (struct iommu_domain *domain, +struct msi_desc *desc,phys_addr_t msi_addr) +{ +return -ENODEV ; +} +#endif /* CONFIG_IOMMU_DMA */ #endif /* __DMA_IOMMU_H */
include/linux/iommu.h +0 -15
@@ -229,11 +229,6 @@ struct iommu_domain {struct iommu_domain_geometry geometry;int (*iopf_handler)(struct iopf_group *group); -#if IS_ENABLED (CONFIG_IRQ_MSI_IOMMU ) -int (*sw_msi)(struct iommu_domain *domain,struct msi_desc *desc, -phys_addr_t msi_addr); -#endif -union {/* cookie */ struct iommu_dma_cookie *iova_cookie;struct iommu_dma_msi_cookie *msi_cookie;
@@ -254,16 +249,6 @@ struct iommu_domain { }; }; -static inline void iommu_domain_set_sw_msi ( -struct iommu_domain *domain, -int (*sw_msi)(struct iommu_domain *domain,struct msi_desc *desc, -phys_addr_t msi_addr)) -{ -#if IS_ENABLED (CONFIG_IRQ_MSI_IOMMU ) - domain->sw_msi = sw_msi; -#endif -} -static inline bool iommu_is_dma_domain (struct iommu_domain *domain) {return domain->type & __IOMMU_DOMAIN_DMA_API;
drivers/iommu/dma-iommu.c +2 -12
@@ -94,9 +94,6 @@ static int __init iommu_dma_forcedac_setup(char *str) }early_param ("iommu.forcedac" , iommu_dma_forcedac_setup); -static int iommu_dma_sw_msi (struct iommu_domain *domain,struct msi_desc *desc, -phys_addr_t msi_addr); -/* Number of entries per flush queue */ #define IOVA_DEFAULT_FQ_SIZE 256#define IOVA_SINGLE_FQ_SIZE 32768
@@ -377,7 +374,6 @@ int iommu_get_dma_cookie(struct iommu_domain *domain)mutex_init (&cookie->mutex);INIT_LIST_HEAD (&cookie->msi_page_list); -iommu_domain_set_sw_msi (domain, iommu_dma_sw_msi); domain->cookie_type =IOMMU_COOKIE_DMA_IOVA ; domain->iova_cookie = cookie;return 0 ;
@@ -411,7 +407,6 @@ int iommu_get_msi_cookie(struct iommu_domain *domain, dma_addr_t base) cookie->msi_iova = base;INIT_LIST_HEAD (&cookie->msi_page_list); -iommu_domain_set_sw_msi (domain, iommu_dma_sw_msi); domain->cookie_type =IOMMU_COOKIE_DMA_MSI ; domain->msi_cookie = cookie;return 0 ;
@@ -427,11 +422,6 @@ void iommu_put_dma_cookie(struct iommu_domain *domain)struct iommu_dma_cookie *cookie = domain->iova_cookie;struct iommu_dma_msi_page *msi, *tmp; -#if IS_ENABLED (CONFIG_IRQ_MSI_IOMMU ) -if (domain->sw_msi != iommu_dma_sw_msi) -return ; -#endif -if (cookie->iovad.granule) {iommu_dma_free_fq (cookie);put_iova_domain (&cookie->iovad);
@@ -1826,8 +1816,8 @@ static struct iommu_dma_msi_page *iommu_dma_get_msi_page(struct device *dev,return NULL ; } -static int iommu_dma_sw_msi (struct iommu_domain *domain,struct msi_desc *desc, -phys_addr_t msi_addr) +int iommu_dma_sw_msi (struct iommu_domain *domain,struct msi_desc *desc, +phys_addr_t msi_addr) {struct device *dev =msi_desc_to_dev (desc);const struct iommu_dma_msi_page *msi_page;
drivers/iommu/iommu.c +15 -2
@@ -18,6 +18,7 @@#include <linux/errno.h> #include <linux/host1x_context_bus.h> #include <linux/iommu.h> +#include <linux/iommufd.h> #include <linux/idr.h> #include <linux/err.h> #include <linux/pci.h>
@@ -3650,8 +3651,20 @@ int iommu_dma_prepare_msi(struct msi_desc *desc, phys_addr_t msi_addr)return 0 ;mutex_lock (&group->mutex); -if (group->domain && group->domain->sw_msi) - ret = group->domain->sw_msi (group->domain, desc, msi_addr); +if (group->domain) { +switch (group->domain->cookie_type) { +case IOMMU_COOKIE_DMA_MSI : +case IOMMU_COOKIE_DMA_IOVA : + ret =iommu_dma_sw_msi (group->domain, desc, msi_addr); +break ; +case IOMMU_COOKIE_IOMMUFD : + ret =iommufd_sw_msi (group->domain, desc, msi_addr); +break ; +default : + ret = -EOPNOTSUPP ; +break ; + } + }mutex_unlock (&group->mutex);return ret; }
drivers/iommu/iommufd/hw_pagetable.c +0 -3
@@ -161,7 +161,6 @@ iommufd_hwpt_paging_alloc(struct iommufd_ctx *ictx, struct iommufd_ioas *ioas, } hwpt->domain->iommufd_hwpt = hwpt; hwpt->domain->cookie_type =IOMMU_COOKIE_IOMMUFD ; -iommu_domain_set_sw_msi (hwpt->domain, iommufd_sw_msi);/* * Set the coherency mode before we do iopt_table_add_domain() as some
@@ -259,7 +258,6 @@ iommufd_hwpt_nested_alloc(struct iommufd_ctx *ictx, hwpt->domain->owner = ops; hwpt->domain->iommufd_hwpt = hwpt; hwpt->domain->cookie_type =IOMMU_COOKIE_IOMMUFD ; -iommu_domain_set_sw_msi (hwpt->domain, iommufd_sw_msi);if (WARN_ON_ONCE (hwpt->domain->type !=IOMMU_DOMAIN_NESTED )) { rc = -EINVAL ;
@@ -318,7 +316,6 @@ iommufd_viommu_alloc_hwpt_nested(struct iommufd_viommu *viommu, u32 flags, hwpt->domain->iommufd_hwpt = hwpt; hwpt->domain->owner = viommu->iommu_dev->ops; hwpt->domain->cookie_type =IOMMU_COOKIE_IOMMUFD ; -iommu_domain_set_sw_msi (hwpt->domain, iommufd_sw_msi);if (WARN_ON_ONCE (hwpt->domain->type !=IOMMU_DOMAIN_NESTED )) { rc = -EINVAL ;
Signature · 1 line
2.43.0
Re: [PATCH v4 3/3] iommu: Drop sw_msi from iommu_domain
Quoted 10 lines · Nicolin Chen
There are only two
sw_msiimplementations in the entire system, thus it's not very necessary to have answ_msipointer.Instead, check
domain->cookie_typeto call the twosw_msiimplementations directly from the core code.Suggested-by Robin Murphy <[email protected]>Reviewed-by Robin Murphy <[email protected]>Reviewed-by Kevin Tian <[email protected]>Signed-off-by Nicolin Chen <[email protected]>
Quoted 7 lines
--- drivers/iommu/dma-iommu.h | 9 +++++++++ include/linux/iommu.h | 15 --------------- drivers/iommu/dma-iommu.c | 14 ++------------ drivers/iommu/iommu.c | 17 +++++++++++++++-- drivers/iommu/iommufd/hw_pagetable.c | 3 --- 5 files changed, 26 insertions(+), 32 deletions(-)
Jason
Re: [PATCH v4 3/3] iommu: Drop sw_msi from iommu_domain
In reply to Nicolin ChenHi Nicolin,
Quoted 196 lines · Nicolin Chen
There are only two
sw_msiimplementations in the entire system, thus it's not very necessary to have answ_msipointer.Instead, check
domain->cookie_typeto call the twosw_msiimplementations directly from the core code.Suggested-by Robin Murphy <[email protected]>Reviewed-by Robin Murphy <[email protected]>Reviewed-by Kevin Tian <[email protected]>Signed-off-by Nicolin Chen <[email protected]>--- drivers/iommu/dma-iommu.h | 9 +++++++++ include/linux/iommu.h | 15 --------------- drivers/iommu/dma-iommu.c | 14 ++------------ drivers/iommu/iommu.c | 17 +++++++++++++++-- drivers/iommu/iommufd/hw_pagetable.c | 3 --- 5 files changed, 26 insertions(+), 32 deletions(-)drivers/iommu/dma-iommu.h +9
-0@@ -19,6 +19,9 @@ int iommu_dma_init_fq(struct iommu_domain *domain);void iommu_dma_get_resv_regions (struct device *dev,struct list_head *list); +int iommu_dma_sw_msi (struct iommu_domain *domain,struct msi_desc *desc, +phys_addr_t msi_addr); +extern bool iommu_dma_forcedac;#else /* CONFIG_IOMMU_DMA */ @@ -49,5 +52,11 @@ static inline void iommu_dma_get_resv_regions(struct device *dev, struct list_he { } +static inline int iommu_dma_sw_msi (struct iommu_domain *domain, +struct msi_desc *desc,phys_addr_t msi_addr) +{ +return -ENODEV ; +} +#endif /* CONFIG_IOMMU_DMA */ #endif /* __DMA_IOMMU_H */ include/linux/iommu.h +0
-15@@ -229,11 +229,6 @@ struct iommu_domain {struct iommu_domain_geometry geometry;int (*iopf_handler)(struct iopf_group *group); -#if IS_ENABLED (CONFIG_IRQ_MSI_IOMMU ) -int (*sw_msi)(struct iommu_domain *domain,struct msi_desc *desc, -phys_addr_t msi_addr); -#endif -union {/* cookie */ struct iommu_dma_cookie *iova_cookie;struct iommu_dma_msi_cookie *msi_cookie;@@ -254,16 +249,6 @@ struct iommu_domain { }; }; -static inline void iommu_domain_set_sw_msi ( -struct iommu_domain *domain, -int (*sw_msi)(struct iommu_domain *domain,struct msi_desc *desc, -phys_addr_t msi_addr)) -{ -#if IS_ENABLED (CONFIG_IRQ_MSI_IOMMU ) - domain->sw_msi = sw_msi; -#endif -} -static inline bool iommu_is_dma_domain (struct iommu_domain *domain) {return domain->type & __IOMMU_DOMAIN_DMA_API;drivers/iommu/dma-iommu.c +2
-12@@ -94,9 +94,6 @@ static int __init iommu_dma_forcedac_setup(char *str) }early_param ("iommu.forcedac" , iommu_dma_forcedac_setup); -static int iommu_dma_sw_msi (struct iommu_domain *domain,struct msi_desc *desc, -phys_addr_t msi_addr); -/* Number of entries per flush queue */ #define IOVA_DEFAULT_FQ_SIZE 256#define IOVA_SINGLE_FQ_SIZE 32768@@ -377,7 +374,6 @@ int iommu_get_dma_cookie(struct iommu_domain *domain)mutex_init (&cookie->mutex);INIT_LIST_HEAD (&cookie->msi_page_list); -iommu_domain_set_sw_msi (domain, iommu_dma_sw_msi); domain->cookie_type =IOMMU_COOKIE_DMA_IOVA ; domain->iova_cookie = cookie;return 0 ;@@ -411,7 +407,6 @@ int iommu_get_msi_cookie(struct iommu_domain *domain, dma_addr_t base) cookie->msi_iova = base;INIT_LIST_HEAD (&cookie->msi_page_list); -iommu_domain_set_sw_msi (domain, iommu_dma_sw_msi); domain->cookie_type =IOMMU_COOKIE_DMA_MSI ; domain->msi_cookie = cookie;return 0 ;@@ -427,11 +422,6 @@ void iommu_put_dma_cookie(struct iommu_domain *domain)struct iommu_dma_cookie *cookie = domain->iova_cookie;struct iommu_dma_msi_page *msi, *tmp; -#if IS_ENABLED (CONFIG_IRQ_MSI_IOMMU ) -if (domain->sw_msi != iommu_dma_sw_msi) -return ; -#endif -if (cookie->iovad.granule) {iommu_dma_free_fq (cookie);put_iova_domain (&cookie->iovad);@@ -1826,8 +1816,8 @@ static struct iommu_dma_msi_page *iommu_dma_get_msi_page(struct device *dev,return NULL ; } -static int iommu_dma_sw_msi (struct iommu_domain *domain,struct msi_desc *desc, -phys_addr_t msi_addr) +int iommu_dma_sw_msi (struct iommu_domain *domain,struct msi_desc *desc, +phys_addr_t msi_addr) {struct device *dev =msi_desc_to_dev (desc);const struct iommu_dma_msi_page *msi_page;drivers/iommu/iommu.c +15
-2@@ -18,6 +18,7 @@#include <linux/errno.h> #include <linux/host1x_context_bus.h> #include <linux/iommu.h> +#include <linux/iommufd.h> #include <linux/idr.h> #include <linux/err.h> #include <linux/pci.h> @@ -3650,8 +3651,20 @@ int iommu_dma_prepare_msi(struct msi_desc *desc, phys_addr_t msi_addr)return 0 ;mutex_lock (&group->mutex); -if (group->domain && group->domain->sw_msi) - ret = group->domain->sw_msi (group->domain, desc, msi_addr); +if (group->domain) { +switch (group->domain->cookie_type) { +case IOMMU_COOKIE_DMA_MSI : +case IOMMU_COOKIE_DMA_IOVA : + ret =iommu_dma_sw_msi (group->domain, desc, msi_addr); +break ; +case IOMMU_COOKIE_IOMMUFD : + ret =iommufd_sw_msi (group->domain, desc, msi_addr); +break ; +default : + ret = -EOPNOTSUPP ; +break ; + } + }mutex_unlock (&group->mutex);return ret; }drivers/iommu/iommufd/hw_pagetable.c +0
-3@@ -161,7 +161,6 @@ iommufd_hwpt_paging_alloc(struct iommufd_ctx *ictx, struct iommufd_ioas *ioas, } hwpt->domain->iommufd_hwpt = hwpt; hwpt->domain->cookie_type =IOMMU_COOKIE_IOMMUFD ; -iommu_domain_set_sw_msi (hwpt->domain, iommufd_sw_msi);/* * Set the coherency mode before we do iopt_table_add_domain() as some @@ -259,7 +258,6 @@ iommufd_hwpt_nested_alloc(struct iommufd_ctx *ictx, hwpt->domain->owner = ops; hwpt->domain->iommufd_hwpt = hwpt; hwpt->domain->cookie_type =IOMMU_COOKIE_IOMMUFD ; -iommu_domain_set_sw_msi (hwpt->domain, iommufd_sw_msi);if (WARN_ON_ONCE (hwpt->domain->type !=IOMMU_DOMAIN_NESTED )) { rc = -EINVAL ;@@ -318,7 +316,6 @@ iommufd_viommu_alloc_hwpt_nested(struct iommufd_viommu *viommu, u32 flags, hwpt->domain->iommufd_hwpt = hwpt; hwpt->domain->owner = viommu->iommu_dev->ops; hwpt->domain->cookie_type =IOMMU_COOKIE_IOMMUFD ; -iommu_domain_set_sw_msi (hwpt->domain, iommufd_sw_msi);if (WARN_ON_ONCE (hwpt->domain->type !=IOMMU_DOMAIN_NESTED )) { rc = -EINVAL ;Signature · 1 line
2.43.0
I bisected a loss of networking on one of my machines to this change as commit e009e088d88e ("iommu: Drop sw_msi from iommu_domain") in -next.
With the parent change, I see:
[ +0.000000] Linux version 6.14.0-rc2-00032-g916a207692ce (nathan@ax162) (aarch64-linux-gcc (GCC) 14.2.0, GNU ld (GNU Binutils) 2.42) #1 SMP PREEMPT_DYNAMIC Mon Mar 24 09:13:34 MST 2025
...
[ +0.002375] fsl_mc_bus NXP0008:00: Adding to iommu group 0 [ +0.000532] fsl_mc_bus NXP0008:00: MC firmware version: 10.28.1 [ +0.002566] fsl_mc_dprc dprc.1: DMA mask not set [ +0.019213] fsl_mc_dprc dprc.1: Adding to iommu group 1 [ +0.050801] fsl_mc_allocator dpbp.0: Adding to iommu group 1 [ +0.006767] fsl_mc_allocator dpmcp.35: Adding to iommu group 1
...
At this change, I see:
[ +0.000000] Linux version 6.14.0-rc2-00033-ge009e088d88e (nathan@ax162) (aarch64-linux-gcc (GCC) 14.2.0, GNU ld (GNU Binutils) 2.42) #1 SMP PREEMPT_DYNAMIC Mon Mar 24 08:57:49 MST 2025
...
[ +0.002355] fsl_mc_bus NXP0008:00: Adding to iommu group 0 [ +0.000533] fsl_mc_bus NXP0008:00: MC firmware version: 10.28.1 [ +0.002565] fsl_mc_dprc dprc.1: DMA mask not set [ +0.019255] fsl_mc_dprc dprc.1: Adding to iommu group 1 [ +0.046820] fsl_mc_dprc dprc.1: Failed to allocate IRQs [ +0.005798] fsl_mc_dprc dprc.1: fsl_mc_driver_probe failed: -28 [ +0.005931] fsl_mc_dprc dprc.1: probe with driver fsl_mc_dprc failed with error -28
...
I know this is not much to go on initially but I am more than happy to provide any additional information and test patches as necessary.
Cheers,
Nathan
# bad: [9388ec571cb1adba59d1cded2300eeb11827679c] Add linux-next specific files for 20250321
# good: [b5329d5a35582abbef57562f9fb6cb26a643f252] Merge tag 'vfs-6.14-final.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
git bisect start '9388ec571cb1adba59d1cded2300eeb11827679c' 'b5329d5a35582abbef57562f9fb6cb26a643f252'
# good: [81a405abf1b06a6088197fe1d039ec5dbfd17989] Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next.git
git bisect good 81a405abf1b06a6088197fe1d039ec5dbfd17989
# good: [1afe6ad5ffa395d0809210a3b609751109de3f44] Merge branch 'apparmor-next' of git://git.kernel.org/pub/scm/linux/kernel/git/jj/linux-apparmor
git bisect good 1afe6ad5ffa395d0809210a3b609751109de3f44
# good: [b97539c1de91aa73b589e145e3c804b9ba5178f2] Merge branch 'driver-core-next' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core.git
git bisect good b97539c1de91aa73b589e145e3c804b9ba5178f2
# good: [db08813487b74820cb40f507fc6ea38994a44776] Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/remoteproc/linux.git
git bisect good db08813487b74820cb40f507fc6ea38994a44776
# good: [8dc029f217de83114fe7a59803ea0ac398db414c] Merge branch 'hyperv-next' of git://git.kernel.org/pub/scm/linux/kernel/git/hyperv/linux.git
git bisect good 8dc029f217de83114fe7a59803ea0ac398db414c
# good: [33239be7f682bf0ef9293c0b8ad53e691275e3a2] Merge branch 'rust-next' of https://github.com/Rust-for-Linux/linux.git
git bisect good 33239be7f682bf0ef9293c0b8ad53e691275e3a2
# good: [f57f2954197b265b6d793ac89d774aae1473854d] Merge branch 'for-next/kspp' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git
git bisect good f57f2954197b265b6d793ac89d774aae1473854d
# good: [acf9f8da5e19fc1cbf26f2ecb749369e13e7cd85] x86/crc: drop the avx10_256 functions and rename avx10_512 to avx512
git bisect good acf9f8da5e19fc1cbf26f2ecb749369e13e7cd85
# good: [da0c56520e880441d0503d0cf0d6853dcfb5f1a4] iommu/arm-smmu-v3: Set MEV bit in nested STE for DoS mitigations
git bisect good da0c56520e880441d0503d0cf0d6853dcfb5f1a4
# bad: [fe0cdd47410f7d5fc26e81c958cde3fe1d57d0a0] Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux.git
git bisect bad fe0cdd47410f7d5fc26e81c958cde3fe1d57d0a0
# good: [447c98c1ca4a4b0d43be99f76c558c09956484f3] tools/power turbostat: Add idle governor statistics reporting
git bisect good 447c98c1ca4a4b0d43be99f76c558c09956484f3
# good: [916a207692ce0a6f82ced6b37ca895d5219efb17] iommufd: Move iommufd_sw_msi and related functions to driver.c
git bisect good 916a207692ce0a6f82ced6b37ca895d5219efb17
# bad: [71c4c1a2a057bd800e08b752bac660f5426bf6b5] Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/jgg/iommufd.git
git bisect bad 71c4c1a2a057bd800e08b752bac660f5426bf6b5
# bad: [e009e088d88e8402539f9595b10c0014125a70c1] iommu: Drop sw_msi from iommu_domain
git bisect bad e009e088d88e8402539f9595b10c0014125a70c1
# first bad commit: [e009e088d88e8402539f9595b10c0014125a70c1] iommu: Drop sw_msi from iommu_domain
Jason Gunthorpe
maintainerRe: [PATCH v4 3/3] iommu: Drop sw_msi from iommu_domain
Quoted 2 lines · Nathan Chancellor
I bisected a loss of networking on one of my machines to this change as commit
e009e088d88e("iommu: Dropsw_msifromiommu_domain") in -next.
Okay wow, I will drop this series from the tree if I don't see a resolution in a few days. We can try again next cycle, thank you for testing and bisect!
Quoted 9 lines
At this change, I see:
[ +0.000000] Linux version 6.14.0-rc2-00033-ge009e088d88e (nathan@ax162) (aarch64-linux-gcc (GCC) 14.2.0, GNU ld (GNU Binutils) 2.42) #1 SMP
PREEMPT_DYNAMICMon Mar 24 08:57:49 MST 2025
...[ +0.002355] fsl_mc_bus NXP0008:00: Adding to iommu group 0 [ +0.000533] fsl_mc_bus NXP0008:00: MC firmware version: 10.28.1 [ +0.002565] fsl_mc_dprc dprc.1: DMA mask not set [ +0.019255] fsl_mc_dprc dprc.1: Adding to iommu group 1 [ +0.046820] fsl_mc_dprc dprc.1: Failed to allocate IRQs
I guess it is tripping up going through iommu_dma_prepare_msi() somehow?
Maybe fsl bus is special and doesn't manage to set IOMMU_COOKIE_DMA_IOVA for some reason?
I wonder if this is not right:
+ default:
+ ret = -EOPNOTSUPP;
+ break;
And it should be just break instead (return 0) which is what was happening before?
Jason
Re: [PATCH v4 3/3] iommu: Drop sw_msi from iommu_domain
Quoted 31 lines · Jason Gunthorpe
Quoted 2 lines · Nathan Chancellor
I bisected a loss of networking on one of my machines to this change as commit
e009e088d88e("iommu: Dropsw_msifromiommu_domain") in -next.Okay wow, I will drop this series from the tree if I don't see a resolution in a few days. We can try again next cycle, thank you for testing and bisect!
Quoted 9 lines
At this change, I see:
[ +0.000000] Linux version 6.14.0-rc2-00033-ge009e088d88e (nathan@ax162) (aarch64-linux-gcc (GCC) 14.2.0, GNU ld (GNU Binutils) 2.42) #1 SMP
PREEMPT_DYNAMICMon Mar 24 08:57:49 MST 2025
...[ +0.002355] fsl_mc_bus NXP0008:00: Adding to iommu group 0 [ +0.000533] fsl_mc_bus NXP0008:00: MC firmware version: 10.28.1 [ +0.002565] fsl_mc_dprc dprc.1: DMA mask not set [ +0.019255] fsl_mc_dprc dprc.1: Adding to iommu group 1 [ +0.046820] fsl_mc_dprc dprc.1: Failed to allocate IRQsI guess it is tripping up going through
iommu_dma_prepare_msi()somehow?Maybe fsl bus is special and doesn't manage to set
IOMMU_COOKIE_DMA_IOVAfor some reason?I wonder if this is not right:
drivers/iommu/iommu.c +3
-0+default : + ret = -EOPNOTSUPP ; +break ;
Quoted 2 lines
And it should be just break instead (return 0) which is what was happening before?
Yea, I found the diff here too.
Nathan, would you please give it a try by removing this
ret = -EOPNOTSUPP;
?
Thanks
Nicolin
Re: [PATCH v4 3/3] iommu: Drop sw_msi from iommu_domain
Quoted 28 lines · Nicolin Chen
Quoted 27 lines · Jason Gunthorpe
Quoted 2 lines · Nathan Chancellor
I bisected a loss of networking on one of my machines to this change as commit
e009e088d88e("iommu: Dropsw_msifromiommu_domain") in -next.Okay wow, I will drop this series from the tree if I don't see a resolution in a few days. We can try again next cycle, thank you for testing and bisect!
Quoted 9 lines
At this change, I see:
[ +0.000000] Linux version 6.14.0-rc2-00033-ge009e088d88e (nathan@ax162) (aarch64-linux-gcc (GCC) 14.2.0, GNU ld (GNU Binutils) 2.42) #1 SMP
PREEMPT_DYNAMICMon Mar 24 08:57:49 MST 2025
...[ +0.002355] fsl_mc_bus NXP0008:00: Adding to iommu group 0 [ +0.000533] fsl_mc_bus NXP0008:00: MC firmware version: 10.28.1 [ +0.002565] fsl_mc_dprc dprc.1: DMA mask not set [ +0.019255] fsl_mc_dprc dprc.1: Adding to iommu group 1 [ +0.046820] fsl_mc_dprc dprc.1: Failed to allocate IRQsI guess it is tripping up going through
iommu_dma_prepare_msi()somehow?Maybe fsl bus is special and doesn't manage to set
IOMMU_COOKIE_DMA_IOVAfor some reason?I wonder if this is not right:
Quoted 8 lines
Quoted 6 lines
drivers/iommu/iommu.c +3
-0+default : + ret = -EOPNOTSUPP ; +break ;And it should be just break instead (return 0) which is what was happening before?
Yea, I found the diff here too.
Or it should return 0 just for "case IOMMU_COOKIE_NONE" doing a passthrough for !CONFIG_IRQ_MSI_IOMMU configuration.
Thanks
Nicolin
Nathan Chancellor
Tested-byRe: [PATCH v4 3/3] iommu: Drop sw_msi from iommu_domain
In reply to Nicolin ChenQuoted 32 lines · Nicolin Chen
Quoted 31 lines · Jason Gunthorpe
Quoted 2 lines · Nathan Chancellor
I bisected a loss of networking on one of my machines to this change as commit
e009e088d88e("iommu: Dropsw_msifromiommu_domain") in -next.Okay wow, I will drop this series from the tree if I don't see a resolution in a few days. We can try again next cycle, thank you for testing and bisect!
Quoted 9 lines
At this change, I see:
[ +0.000000] Linux version 6.14.0-rc2-00033-ge009e088d88e (nathan@ax162) (aarch64-linux-gcc (GCC) 14.2.0, GNU ld (GNU Binutils) 2.42) #1 SMP
PREEMPT_DYNAMICMon Mar 24 08:57:49 MST 2025
...[ +0.002355] fsl_mc_bus NXP0008:00: Adding to iommu group 0 [ +0.000533] fsl_mc_bus NXP0008:00: MC firmware version: 10.28.1 [ +0.002565] fsl_mc_dprc dprc.1: DMA mask not set [ +0.019255] fsl_mc_dprc dprc.1: Adding to iommu group 1 [ +0.046820] fsl_mc_dprc dprc.1: Failed to allocate IRQsI guess it is tripping up going through
iommu_dma_prepare_msi()somehow?Maybe fsl bus is special and doesn't manage to set
IOMMU_COOKIE_DMA_IOVAfor some reason?I wonder if this is not right:
(quoted diff) +3
-0+ default: + ret = -EOPNOTSUPP; + break;
Quoted 8 lines
Quoted 2 lines
And it should be just break instead (return 0) which is what was happening before?
Yea, I found the diff here too.
Nathan, would you please give it a try by removing this
ret = -EOPNOTSUPP;?
Can confirm, I tested the following diff and it resolved my issue. If it goes as a standalone patch:
Thanks for the quick triage!
drivers/iommu/iommu.c +0 -1
@@ -3661,7 +3661,6 @@ int iommu_dma_prepare_msi(struct msi_desc *desc, phys_addr_t msi_addr) ret =iommufd_sw_msi (group->domain, desc, msi_addr);break ;default : - ret = -EOPNOTSUPP ;break ; } }
Jason Gunthorpe
maintainerRe: [PATCH v4 3/3] iommu: Drop sw_msi from iommu_domain
drivers/iommu/iommu.c int iommu_dma_prepare_msi(struct msi_desc *desc, phys_addr_t msi_addr) · line 3666
Quoted 12 lines · Nathan Chancellor
drivers/iommu/iommu.c +0
-1@@ -3661,7 +3661,6 @@ int iommu_dma_prepare_msi(struct msi_desc *desc, phys_addr_t msi_addr) ret =iommufd_sw_msi (group->domain, desc, msi_addr);break ;default : - ret = -EOPNOTSUPP ;break ; } }
Can we explain why this scenario has a 0 cookie_type?
Actually.. Is it just an identity domain? Nicolin did you test this on your arm system with a device using identity (iommu=pt kernel param)? I would expect identity to end up with a 0 cookie because we never setup dma-iommu.c code on it.
Should we be testing for identity to return 0 instead?
Jason
Re: [PATCH v4 3/3] iommu: Drop sw_msi from iommu_domain
Quoted 17 lines · Jason Gunthorpe
drivers/iommu/iommu.c int iommu_dma_prepare_msi(struct msi_desc *desc, phys_addr_t msi_addr) · line 3666
Quoted 12 lines · Nathan Chancellor
drivers/iommu/iommu.c +0
-1@@ -3661,7 +3661,6 @@ int iommu_dma_prepare_msi(struct msi_desc *desc, phys_addr_t msi_addr) ret =iommufd_sw_msi (group->domain, desc, msi_addr);break ;default : - ret = -EOPNOTSUPP ;break ; } }Can we explain why this scenario has a 0
cookie_type?
Quoted 6 lines
Actually.. Is it just an identity domain? Nicolin did you test this on your arm system with a device using identity (iommu=pt kernel param)? I would expect identity to end up with a 0 cookie because we never setup
dma-iommu.ccode on it.Should we be testing for identity to return 0 instead?
For the record, the particular system I noticed the issue on does need "iommu.passthrough=1" (which is the ARM equivalent to "iommu=pt" IIUC?) to boot due to a lack of firmware support for IORT RMR, so I think the answer to your first question is probably yes?
Cheers,
Nathan
Re: [PATCH v4 3/3] iommu: Drop sw_msi from iommu_domain
Quoted 18 lines · Nathan Chancellor
Quoted 17 lines · Jason Gunthorpe
drivers/iommu/iommu.c int iommu_dma_prepare_msi(struct msi_desc *desc, phys_addr_t msi_addr) · line 3666
Quoted 12 lines · Nathan Chancellor
drivers/iommu/iommu.c +0
-1@@ -3661,7 +3661,6 @@ int iommu_dma_prepare_msi(struct msi_desc *desc, phys_addr_t msi_addr) ret =iommufd_sw_msi (group->domain, desc, msi_addr);break ;default : - ret = -EOPNOTSUPP ;break ; } }Can we explain why this scenario has a 0
cookie_type?
Quoted 6 lines
Quoted 6 lines
Actually.. Is it just an identity domain? Nicolin did you test this on your arm system with a device using identity (iommu=pt kernel param)? I would expect identity to end up with a 0 cookie because we never setup
dma-iommu.ccode on it.Should we be testing for identity to return 0 instead?
My feeling is that we should just let all other cases return 0 like the previous function did, as this seems to be commonly on the IRQ allocation path that shouldn't fail like this. E.g. if we fail a blocked domain, would it retry switching domains?
Quoted 4 lines
For the record, the particular system I noticed the issue on does need "iommu.passthrough=1" (which is the ARM equivalent to "iommu=pt" IIUC?) to boot due to a lack of firmware support for IORT RMR, so I think the answer to your first question is probably yes?
Yea, I confirmed that identity domain on ARM fails too. My bad that I didn't catch this case.
Thanks
Nicolin
Jason Gunthorpe
maintainerRe: [PATCH v4 3/3] iommu: Drop sw_msi from iommu_domain
Quoted 4 lines · Nicolin Chen
My feeling is that we should just let all other cases return 0 like the previous function did, as this seems to be commonly on the IRQ allocation path that shouldn't fail like this. E.g. if we fail a blocked domain, would it retry switching domains?
I don't know, I think if the interrupt driver is calling this function it knows the MSI is translated by the iommu and if the iommu is set to something like blocked then it really should fail rather than silently not work. Same for a paging domain without any kind of sw_msi route.
So, I'm feeling like we should check for identity and still fail the other cases?
Can you send another version of the series with this and Arnd's two fixes integrated?
Thanks,
Jason
Re: [PATCH v4 3/3] iommu: Drop sw_msi from iommu_domain
Quoted 11 lines · Jason Gunthorpe
Quoted 4 lines · Nicolin Chen
My feeling is that we should just let all other cases return 0 like the previous function did, as this seems to be commonly on the IRQ allocation path that shouldn't fail like this. E.g. if we fail a blocked domain, would it retry switching domains?
I don't know, I think if the interrupt driver is calling this function it knows the MSI is translated by the iommu and if the iommu is set to something like blocked then it really should fail rather than silently not work. Same for a paging domain without any kind of
sw_msiroute.
OK. That sounds correct to me.
I will give the default blocked domain case a try. If later the irqchip driver can still allocate IRQ after it switches to some normal working domain, then we will be safe to do so.
Quoted 5 lines
So, I'm feeling like we should check for identity and still fail the other cases?
Can you send another version of the series with this and Arnd's two fixes integrated?
Yes, I will do that. And I assume they will be still rebased on:da0c56520e88 iommu/arm-smmu-v3: Set MEV bit in nested STE for DoS mitigations
Will take some time run some tests with different combinations this time. Sorry for the trouble. Should have been more careful.
Nicolin
Jason Gunthorpe
maintainerQuoted 4 lines · Nicolin Chen
Changelog
v4
* Rebase on top of a bug fix forhwpt_iommufdcookie
https://lore.kernel.org/all/[email protected]/
I suppose this means it needs to go to the iommufd tree?
Joerg OK?
Thanks,
Jason
Quoted 7 lines · Nicolin Chen
Nicolin Chen (2):
iommufd: Moveiommufd_sw_msiand related functions todriver.c
iommu: Dropsw_msifromiommu_domainRobin Murphy (1):
iommu: Sort out domain user data
Quoted 11 lines
drivers/iommu/dma-iommu.h | 14 ++ drivers/iommu/iommu-priv.h | 13 ++ drivers/iommu/iommufd/iommufd_private.h | 7 +- include/linux/iommu.h | 35 ++-- drivers/iommu/dma-iommu.c | 208 ++++++++++++------------ drivers/iommu/iommu-sva.c | 1 + drivers/iommu/iommu.c | 35 +++- drivers/iommu/iommufd/device.c | 131 +-------------- drivers/iommu/iommufd/driver.c | 125 ++++++++++++++ drivers/iommu/iommufd/hw_pagetable.c | 6 +- 10 files changed, 320 insertions(+), 255 deletions(-)
Applied
Thanks,
Jason