| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20240108162516.GH1315173@hu-bjorande-lv.qualcomm.com>
Date: Mon, 8 Jan 2024 08:25:16 -0800
From: Bjorn Andersson <quic_bjorande@...cinc.com>
To: Johan Hovold <johan@...nel.org>
CC: Bjorn Andersson <andersson@...nel.org>,
Konrad Dybcio
<konrad.dybcio@...aro.org>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
Rob Herring <robh+dt@...nel.org>,
Krzysztof Kozlowski
<krzysztof.kozlowski+dt@...aro.org>,
Conor Dooley <conor+dt@...nel.org>,
Wesley Cheng <quic_wcheng@...cinc.com>,
Thinh Nguyen
<Thinh.Nguyen@...opsys.com>,
Felipe Balbi <balbi@...nel.org>,
Philipp Zabel
<p.zabel@...gutronix.de>,
<linux-arm-msm@...r.kernel.org>, <linux-usb@...r.kernel.org>,
<devicetree@...r.kernel.org>, <linux-kernel@...r.kernel.org>,
Krishna Kurapati PSSNV
<quic_kriskura@...cinc.com>
Subject: Re: [PATCH 03/12] usb: dwc3: qcom: Merge resources from urs_usb
device
On Wed, Nov 22, 2023 at 11:24:07AM +0100, Johan Hovold wrote:
> On Mon, Oct 16, 2023 at 08:11:11PM -0700, Bjorn Andersson wrote:
> > With some ACPI DSDT tables, such as the one found in SC8180X devices,
> > the USB resources are split between the URSn and it's child USBn device
> > nodes, in particular the interrupts are placed in the child nodes.
> >
> > The solution that was chosen for handling this is to allocate a
> > platform_device from the child node and selectively pick interrupts
> > from the main platform_device, or from this created child device, when
> > creating the platform_device for the DWC3 core.
> >
> > This does however not work with the upcoming change where the DWC3 core
> > is instantiated from the same platform_device as the glue, as the DRD
> > and host code will attempt to resolve their interrupts from the shared
> > device, and not the child device.
> >
> > Work around this by merging the resources of the child device into the
> > glue device node, to present a single platform_device with all the
> > resources necessary.
>
> Nice approach.
>
> An alternative would be to drop ACPI support completely as Konrad
> suggested. Should simplify both this series and the multiport one.
>
> Is anyone really using the ACPI support here anymore?
>
At the introduction of SC8180X and the Lenovo Flex 5G we where able to
run the Debian installer off the ACPI support in the kernel.
Since then, at least the UFS support has regressed to the point that
this would no longer be possible - without anyone noticing.
I would like to see ACPI supported again in the future, but I can't
really argue for its existence currently. In the end the new flattened
code path is mostly shared with the ACPI path, so perhaps it makes sense
to drop the support after this refactor, perhaps not. I will re-evaluate
this.
> > -static struct platform_device *
> > -dwc3_qcom_create_urs_usb_platdev(struct device *dev)
> > +static int dwc3_qcom_acpi_merge_urs_resources(struct platform_device *pdev)
> > {
> > + struct device *dev = &pdev->dev;
> > + struct list_head resource_list;
> > + struct resource_entry *rentry;
> > + struct resource *resources;
> > struct fwnode_handle *fwh;
> > struct acpi_device *adev;
> > char name[8];
> > + int count;
> > int ret;
> > int id;
> > + int i;
> >
> > /* Figure out device id */
> > ret = sscanf(fwnode_get_name(dev->fwnode), "URS%d", &id);
> > if (!ret)
> > - return NULL;
> > + return -EINVAL;
> >
> > /* Find the child using name */
> > snprintf(name, sizeof(name), "USB%d", id);
> > fwh = fwnode_get_named_child_node(dev->fwnode, name);
> > if (!fwh)
> > - return NULL;
> > + return 0;
> >
> > adev = to_acpi_device_node(fwh);
> > if (!adev)
> > - return NULL;
> > + return -EINVAL;
>
> This is currently leaking a reference to the fwnode, I fixed that up
> here:
>
> https://lore.kernel.org/linux-usb/20231117173650.21161-4-johan+linaro@kernel.org/
>
> > + INIT_LIST_HEAD(&resource_list);
> > +
> > + count = acpi_dev_get_resources(adev, &resource_list, NULL, NULL);
> > + if (count <= 0)
> > + return count;
> > +
> > + count += pdev->num_resources;
> > +
> > + resources = kcalloc(count, sizeof(*resources), GFP_KERNEL);
> > + if (!resources) {
> > + acpi_dev_free_resource_list(&resource_list);
> > + return -ENOMEM;
> > + }
> > +
> > + memcpy(resources, pdev->resource, sizeof(struct resource) * pdev->num_resources);
> > + count = pdev->num_resources;
> > + list_for_each_entry(rentry, &resource_list, node) {
> > + /* Avoid inserting duplicate entries, in case this is called more than once */
>
> Either shorten this one or make it a multiline comment to stay within 80
> chars.
>
> > + for (i = 0; i < count; i++) {
>
> Should this not be pdev->num_resources?
>
count is first used to denote the number of entries to allocate in the
new list, it's then reset to pdev->num_resources 3 lines above this and
after this list_for_each_entry() it would be the total number of
resources in the new list (which could be less than the allocated number
of items).
I can avoid reusing the variable, to clarify this - if I choose to keep
the ACPI support through the series.
> > + if (resource_type(&resources[i]) == resource_type(rentry->res) &&
> > + resources[i].start == rentry->res->start &&
> > + resources[i].end == rentry->res->end)
> > + break;
> > + }
> > +
> > + if (i == count)
>
> Same here.
>
> > + resources[count++] = *rentry->res;
> > + }
> >
> > - return acpi_create_platform_device(adev, NULL);
> > + ret = platform_device_add_resources(pdev, resources, count);
> > + if (ret)
> > + dev_err(&pdev->dev, "failed to add resources\n");
> > +
> > + acpi_dev_free_resource_list(&resource_list);
> > + kfree(resources);
> > +
> > + return ret;
> > }
> >
> > static int dwc3_qcom_probe(struct platform_device *pdev)
> > @@ -817,6 +853,12 @@ static int dwc3_qcom_probe(struct platform_device *pdev)
> > dev_err(&pdev->dev, "no supporting ACPI device data\n");
> > return -EINVAL;
> > }
> > +
> > + if (qcom->acpi_pdata->is_urs) {
> > + ret = dwc3_qcom_acpi_merge_urs_resources(pdev);
> > + if (ret < 0)
> > + goto clk_disable;
>
> The clocks have not been enabled here, just return ret.
>
Right.
Thanks,
Bjorn
> > + }
> > }
> >
> > qcom->resets = devm_reset_control_array_get_optional_exclusive(dev);
> > @@ -857,18 +899,6 @@ static int dwc3_qcom_probe(struct platform_device *pdev)
> > qcom->acpi_pdata->qscratch_base_offset;
> > parent_res->end = parent_res->start +
> > qcom->acpi_pdata->qscratch_base_size;
> > -
> > - if (qcom->acpi_pdata->is_urs) {
> > - qcom->urs_usb = dwc3_qcom_create_urs_usb_platdev(dev);
> > - if (IS_ERR_OR_NULL(qcom->urs_usb)) {
> > - dev_err(dev, "failed to create URS USB platdev\n");
> > - if (!qcom->urs_usb)
> > - ret = -ENODEV;
> > - else
> > - ret = PTR_ERR(qcom->urs_usb);
> > - goto clk_disable;
> > - }
> > - }
> > }
> >
> > qcom->qscratch_base = devm_ioremap_resource(dev, parent_res);
>
> Johan
Powered by blists - more mailing lists