[PATCH v2] of/iommu: support multiple iommu-map entries per input ID
Robin Murphy
robin.murphy at arm.com
Mon Mar 16 04:48:58 PDT 2026
On 2026-03-15 5:07 pm, Vijayanand Jitta wrote:
> When multiple mappings are present for an input ID, linux matches
> just the first one. There is a usecase [1] where all the mappings
> are to be maintained in parallel for an iommu-map entry of a same
> input id.
>
> Add a next_offset iterator parameter to of_map_iommu_id() and
> refactor of_map_id() internals into a static helper to carry it.
> Update of_iommu_configure_dev_id() to loop over all matching
> entries to support this case. All other callers pass NULL and
> are unaffected.
>
> [1] https://lore.kernel.org/all/20250627-video_cb-v3-0-51e18c0ffbce@quicinc.com/
That series doesn't even reference "iommu-map" though?
Once again, NAK to this - the purpose of "iommu-map"/"msi-map" is to
describe a translation from one hardware ID space to another, where a
one-to-many mapping of IDs to a single target would never make sense.
While it is in principle possible for a single device to map to multiple
different *targets*, that is ambiguous as to what it means - is it
giving the OS a choice to use whichever one it prefers, or implying they
must all be configured identically at all times? - so depending on how
you prefer to look at it, we either do not support that, or we have
chosen the first option.
> Signed-off-by: Vijayanand Jitta <vijayanand.jitta at oss.qualcomm.com>
> ---
> On Qualcomm kaanapali, the VPU hardware has multiple internal blocks
> that generate different stream IDs for the same input ID. The device
> tree encodes them as separate iommu-map entries sharing the same input
> ID:
And all that means is that your invented notion of "input IDs" is
clearly wrong. Please stop this abuse of common bindings and just come
up with a proper way to describe your hardware appropriately for what it
actually is and does.
Thanks,
Robin.
> iommu-map = <0x100 &apps_smmu 0x1940 0x1>,
> <0x100 &apps_smmu 0x1a20 0x1>, ...
>
> This requires multiple iommu-map entries per device.
> of_iommu_configure_dev_id() currently stops at the first match,
> so only one stream ID gets registered with the IOMMU.
>
> The v1 series [1] addressed this with a callback threaded through
> of_map_id().
>
> This patch uses a next_offset iterator on of_map_iommu_id() instead,
> keeping of_map_id() unchanged, and updates of_iommu_configure_dev_id()
> to loop over all matching entries.
>
> This patch also depends on iommu-cells series [4].
>
> Changes since v1:
> - Split patches 2/7 [2] and 3/7 [3] out into this standalone series.
> - Dropped the callback (of_map_id_cb / of_map_id_arg) entirely.
> - Replaced with a next_offset iterator on of_map_iommu_id()
> only; of_map_id() public API is unchanged.
> - of_iommu_configure_dev_id() now loops explicitly; no
> bus-type heuristic (dev_is_platform()) needed.
>
> [1] https://lore.kernel.org/all/20260126-kaanapali-iris-v1-0-e2646246bfc1@oss.qualcomm.com/
> [2] https://lore.kernel.org/all/20260126-kaanapali-iris-v1-2-e2646246bfc1@oss.qualcomm.com/
> [3] https://lore.kernel.org/all/20260126-kaanapali-iris-v1-3-e2646246bfc1@oss.qualcomm.com/
> [4] https://lore.kernel.org/all/ce25b963-0e8e-4411-a406-7b466eadb1f9@oss.qualcomm.com/
>
> Signed-off-by: Vijayanand Jitta <vijayanand.jitta at oss.qualcomm.com>
> ---
> drivers/iommu/of_iommu.c | 20 +++++++----
> drivers/of/base.c | 65 ++++++++++++++++++++++++++++++-----
> drivers/pci/controller/dwc/pci-imx6.c | 2 +-
> drivers/pci/controller/pcie-apple.c | 2 +-
> drivers/xen/grant-dma-ops.c | 2 +-
> include/linux/of.h | 4 +--
> 6 files changed, 75 insertions(+), 20 deletions(-)
>
> diff --git a/drivers/iommu/of_iommu.c b/drivers/iommu/of_iommu.c
> index a18bb60f6f3d..947eedd9a88b 100644
> --- a/drivers/iommu/of_iommu.c
> +++ b/drivers/iommu/of_iommu.c
> @@ -46,14 +46,22 @@ static int of_iommu_configure_dev_id(struct device_node *master_np,
> const u32 *id)
> {
> struct of_phandle_args iommu_spec = {};
> - int err;
> + int offset = 0, err;
> + bool found = false;
>
> - err = of_map_iommu_id(master_np, *id, &iommu_spec);
> - if (err)
> - return err;
> + while (!(err = of_map_iommu_id(master_np, *id, &iommu_spec, &offset))) {
> + err = of_iommu_xlate(dev, &iommu_spec);
> + of_node_put(iommu_spec.np);
> + iommu_spec.np = NULL;
> + if (err)
> + return err;
> + found = true;
> + }
> +
> + /* -ENODEV means all entries exhausted; success if at least one was processed */
> + if (err == -ENODEV && found)
> + return 0;
>
> - err = of_iommu_xlate(dev, &iommu_spec);
> - of_node_put(iommu_spec.np);
> return err;
> }
>
> diff --git a/drivers/of/base.c b/drivers/of/base.c
> index 9c44eb6d445d..71175e670757 100644
> --- a/drivers/of/base.c
> +++ b/drivers/of/base.c
> @@ -2146,13 +2146,13 @@ static bool of_check_bad_map(const __be32 *map, int len)
> *
> * Return: 0 on success or a standard error code on failure.
> */
> -int of_map_id(const struct device_node *np, u32 id,
> - const char *map_name, const char *cells_name,
> - const char *map_mask_name,
> - struct of_phandle_args *arg)
> +static int of_map_id_next(const struct device_node *np, u32 id,
> + const char *map_name, const char *cells_name,
> + const char *map_mask_name,
> + struct of_phandle_args *arg, int *next_offset)
> {
> u32 map_mask, masked_id;
> - int map_bytes, map_len, offset = 0;
> + int map_bytes, map_len, offset = next_offset ? *next_offset : 0;
> bool bad_map = false;
> const __be32 *map = NULL;
>
> @@ -2161,7 +2161,7 @@ int of_map_id(const struct device_node *np, u32 id,
>
> map = of_get_property(np, map_name, &map_bytes);
> if (!map) {
> - if (arg->np)
> + if (arg->np || next_offset)
> return -ENODEV;
> /* Otherwise, no map implies no translation */
> arg->args[0] = id;
> @@ -2262,9 +2262,16 @@ int of_map_id(const struct device_node *np, u32 id,
> pr_debug("%pOF: %s, using mask %08x, id-base: %08x, out-base: %08x, length: %08x, id: %08x -> %08x\n",
> np, map_name, map_mask, id_base, be32_to_cpup(out_base),
> id_len, id, id_off + be32_to_cpup(out_base));
> +
> + if (next_offset)
> + *next_offset = offset; /* caller resumes from here */
> return 0;
> }
>
> + /* no (more) matches found in the map */
> + if (next_offset)
> + return -ENODEV;
> +
> pr_info("%pOF: no %s translation for id 0x%x on %pOF\n", np, map_name,
> id, arg->np);
>
> @@ -2276,6 +2283,38 @@ int of_map_id(const struct device_node *np, u32 id,
> pr_err("%pOF: Error: Bad %s length: %d\n", np, map_name, map_bytes);
> return -EINVAL;
> }
> +
> +/**
> + * of_map_id - Translate an ID through a downstream mapping.
> + * @np: root complex device node.
> + * @id: device ID to map.
> + * @map_name: property name of the map to use.
> + * @cells_name: property name of target specifier cells.
> + * @map_mask_name: optional property name of the mask to use.
> + * @arg: pointer to a &struct of_phandle_args. On input, @arg->np may be
> + * set to a target device node to match, or NULL to match any. On
> + * success, @arg->np will be set to the matched target node (with a
> + * reference held), @arg->args_count will be set to the number of
> + * output specifier cells as defined by @cells_name in the target node,
> + * and @arg->args[0..args_count-1] will contain the translated output
> + * specifier values.
> + *
> + * Given a device ID, look up the appropriate implementation-defined
> + * platform ID and/or the target device which receives transactions on that
> + * ID, as per the "iommu-map" and "msi-map" bindings. If @arg->np points to
> + * a non-NULL device node, only entries targeting that node will be matched;
> + * if it is NULL, it will receive the device node of the first matching
> + * target phandle, with a reference held.
> + *
> + * Return: 0 on success or a standard error code on failure.
> + */
> +int of_map_id(const struct device_node *np, u32 id,
> + const char *map_name, const char *cells_name,
> + const char *map_mask_name,
> + struct of_phandle_args *arg)
> +{
> + return of_map_id_next(np, id, map_name, cells_name, map_mask_name, arg, NULL);
> +}
> EXPORT_SYMBOL_GPL(of_map_id);
>
> /**
> @@ -2285,15 +2324,23 @@ EXPORT_SYMBOL_GPL(of_map_id);
> * @arg: pointer to a &struct of_phandle_args for the result. On success,
> * @arg->np holds a reference to the target node that the caller must
> * release with of_node_put().
> + * @next_offset: if non-NULL, on success it is set to the map offset just
> + * past the matched entry. Pass this value back on the next call to
> + * resume scanning from where the previous call left off, allowing all
> + * matching entries for the same @id to be iterated. Pass NULL (or a
> + * pointer to 0) to find only the first match.
> *
> - * Convenience wrapper around of_map_id() using "iommu-map" and "iommu-map-mask".
> + * Wrapper around the internal iommu-map scanner using "iommu-map" and
> + * "iommu-map-mask". When @next_offset is non-NULL, returns -ENODEV once
> + * all matching entries have been exhausted.
> *
> * Return: 0 on success or a standard error code on failure.
> */
> int of_map_iommu_id(const struct device_node *np, u32 id,
> - struct of_phandle_args *arg)
> + struct of_phandle_args *arg, int *next_offset)
> {
> - return of_map_id(np, id, "iommu-map", "#iommu-cells", "iommu-map-mask", arg);
> + return of_map_id_next(np, id, "iommu-map", "#iommu-cells",
> + "iommu-map-mask", arg, next_offset);
> }
> EXPORT_SYMBOL_GPL(of_map_iommu_id);
>
> diff --git a/drivers/pci/controller/dwc/pci-imx6.c b/drivers/pci/controller/dwc/pci-imx6.c
> index 5f8a11774eb5..0d7f5e6d037a 100644
> --- a/drivers/pci/controller/dwc/pci-imx6.c
> +++ b/drivers/pci/controller/dwc/pci-imx6.c
> @@ -1146,7 +1146,7 @@ static int imx_pcie_add_lut_by_rid(struct imx_pcie *imx_pcie, u32 rid)
> u32 sid = 0;
>
> target = NULL;
> - err_i = of_map_iommu_id(dev->of_node, rid, &iommu_spec);
> + err_i = of_map_iommu_id(dev->of_node, rid, &iommu_spec, NULL);
> if (!err_i) {
> target = iommu_spec.np;
> sid_i = iommu_spec.args[0];
> diff --git a/drivers/pci/controller/pcie-apple.c b/drivers/pci/controller/pcie-apple.c
> index 2e86f8fd300b..c780e3f9d14d 100644
> --- a/drivers/pci/controller/pcie-apple.c
> +++ b/drivers/pci/controller/pcie-apple.c
> @@ -765,7 +765,7 @@ static int apple_pcie_enable_device(struct pci_host_bridge *bridge, struct pci_d
> dev_dbg(&pdev->dev, "added to bus %s, index %d\n",
> pci_name(pdev->bus->self), port->idx);
>
> - err = of_map_iommu_id(port->pcie->dev->of_node, rid, &iommu_spec);
> + err = of_map_iommu_id(port->pcie->dev->of_node, rid, &iommu_spec, NULL);
> if (err)
> return err;
>
> diff --git a/drivers/xen/grant-dma-ops.c b/drivers/xen/grant-dma-ops.c
> index 36547d7cf1d6..062beb5825f5 100644
> --- a/drivers/xen/grant-dma-ops.c
> +++ b/drivers/xen/grant-dma-ops.c
> @@ -325,7 +325,7 @@ static int xen_dt_grant_init_backend_domid(struct device *dev,
> struct pci_dev *pdev = to_pci_dev(dev);
> u32 rid = PCI_DEVID(pdev->bus->number, pdev->devfn);
>
> - if (of_map_iommu_id(np, rid, &iommu_spec)) {
> + if (of_map_iommu_id(np, rid, &iommu_spec, NULL)) {
> dev_dbg(dev, "Cannot translate ID\n");
> return -ESRCH;
> }
> diff --git a/include/linux/of.h b/include/linux/of.h
> index 953f2dbe0e86..990849f00e74 100644
> --- a/include/linux/of.h
> +++ b/include/linux/of.h
> @@ -467,7 +467,7 @@ int of_map_id(const struct device_node *np, u32 id,
> struct of_phandle_args *arg);
>
> int of_map_iommu_id(const struct device_node *np, u32 id,
> - struct of_phandle_args *arg);
> + struct of_phandle_args *arg, int *next_offset);
>
> int of_map_msi_id(const struct device_node *np, u32 id,
> struct of_phandle_args *arg);
> @@ -943,7 +943,7 @@ static inline int of_map_id(const struct device_node *np, u32 id,
> }
>
> static inline int of_map_iommu_id(const struct device_node *np, u32 id,
> - struct of_phandle_args *arg)
> + struct of_phandle_args *arg, int *next_offset)
> {
> return -EINVAL;
> }
>
> ---
> base-commit: 9e94742cffb7541f55fa904a40c1ca9d836d303d
> change-id: 20260315-iommu_multi_map-8c9b78490ace
>
> Best regards,
More information about the linux-arm-kernel
mailing list