Re: [PATCH 6/6] remoteproc: qcom: Enable map/unmap and SHM bridge support

From: neil . armstrong
Date: Thu Oct 10 2024 - 02:58:12 EST


On 08/10/2024 08:21, Mukesh Ojha wrote:
On Mon, Oct 07, 2024 at 08:22:39PM +0530, Mukesh Ojha wrote:
On Mon, Oct 07, 2024 at 10:05:08AM +0200, neil.armstrong@xxxxxxxxxx wrote:
On 04/10/2024 23:23, Mukesh Ojha wrote:
For Qualcomm SoCs runnning with Qualcomm EL2 hypervisor(QHEE), IOMMU
translation for remote processors is managed by QHEE and if the same SoC
run under KVM, remoteproc carveout and devmem region should be IOMMU
mapped from Linux PAS driver before remoteproc is brought up and
unmapped once it is tear down and apart from this, SHM bridge also need
to set up to enable memory protection on both remoteproc meta data
memory as well as for the carveout region.

Enable the support required to run Qualcomm remoteprocs on non-QHEE
hypervisors.

Signed-off-by: Mukesh Ojha <quic_mojha@xxxxxxxxxxx>
---
drivers/remoteproc/qcom_q6v5_pas.c | 41 +++++++++++++++++++++++++++++-
1 file changed, 40 insertions(+), 1 deletion(-)

diff --git a/drivers/remoteproc/qcom_q6v5_pas.c b/drivers/remoteproc/qcom_q6v5_pas.c
index ac339145e072..13bd13f1b989 100644
--- a/drivers/remoteproc/qcom_q6v5_pas.c
+++ b/drivers/remoteproc/qcom_q6v5_pas.c
@@ -122,6 +122,7 @@ struct qcom_adsp {
struct qcom_devmem_table *devmem;
struct qcom_tzmem_area *tzmem;
+ unsigned long sid;
};
static void adsp_segment_dump(struct rproc *rproc, struct rproc_dump_segment *segment,
@@ -310,9 +311,21 @@ static int adsp_start(struct rproc *rproc)
if (ret)
return ret;
+ ret = qcom_map_unmap_carveout(rproc, adsp->mem_phys, adsp->mem_size, true, true, adsp->sid);
+ if (ret) {
+ dev_err(adsp->dev, "iommu mapping failed, ret: %d\n", ret);
+ goto disable_irqs;
+ }
+
+ ret = qcom_map_devmem(rproc, adsp->devmem, true, adsp->sid);
+ if (ret) {
+ dev_err(adsp->dev, "devmem iommu mapping failed, ret: %d\n", ret);
+ goto unmap_carveout;
+ }
+
ret = adsp_pds_enable(adsp, adsp->proxy_pds, adsp->proxy_pd_count);
if (ret < 0)
- goto disable_irqs;
+ goto unmap_devmem;
ret = clk_prepare_enable(adsp->xo);
if (ret)
@@ -400,6 +413,10 @@ static int adsp_start(struct rproc *rproc)
clk_disable_unprepare(adsp->xo);
disable_proxy_pds:
adsp_pds_disable(adsp, adsp->proxy_pds, adsp->proxy_pd_count);
+unmap_devmem:
+ qcom_unmap_devmem(rproc, adsp->devmem, adsp->sid);
+unmap_carveout:
+ qcom_map_unmap_carveout(rproc, adsp->mem_phys, adsp->mem_size, false, true, adsp->sid);
disable_irqs:
qcom_q6v5_unprepare(&adsp->q6v5);
@@ -445,6 +462,9 @@ static int adsp_stop(struct rproc *rproc)
dev_err(adsp->dev, "failed to shutdown dtb: %d\n", ret);
}
+ qcom_unmap_devmem(rproc, adsp->devmem, adsp->sid);
+ qcom_map_unmap_carveout(rproc, adsp->mem_phys, adsp->mem_size, false, true, adsp->sid);
+
handover = qcom_q6v5_unprepare(&adsp->q6v5);
if (handover)
qcom_pas_handover(&adsp->q6v5);
@@ -844,6 +864,25 @@ static int adsp_probe(struct platform_device *pdev)
}
platform_set_drvdata(pdev, adsp);
+ if (of_property_present(pdev->dev.of_node, "iommus")) {
+ struct of_phandle_args args;
+
+ ret = of_parse_phandle_with_args(pdev->dev.of_node, "iommus", "#iommu-cells", 0, &args);
+ if (ret < 0)
+ return ret;
+
+ rproc->has_iommu = true;
+ adsp->sid = args.args[0];
+ of_node_put(args.np);
+ ret = adsp_devmem_init(adsp);
+ if (ret)
+ return ret;

Why don't you get this table from the firmware like presumably QHEE does ?

Well, AFAIK, QHEE(EL2) has this information statically present and does
not get it from anywhere., but will confirm this twice..

Double confirmed, device memory region required by remoteproc is
statically present with QHEE.

Right, in this case why those tables can't be embedded in the elf .resource_table
like it's done with qcom_q6v5_adsp.c by calling rproc_elf_load_rsc_table()
and let the remoteproc framework load the resource table and setup
the devmem ssmu_map ?

Neil


-Mukesh


-Mukesh