lkml.org 
[lkml]   [2023]   [Sep]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH V13 2/5] PCI: Create device tree node for bridge
From

On 9/11/23 07:48, Jonathan Cameron wrote:
> On Tue, 15 Aug 2023 10:19:57 -0700
> Lizhi Hou <lizhi.hou@amd.com> wrote:
>
>> The PCI endpoint device such as Xilinx Alveo PCI card maps the register
>> spaces from multiple hardware peripherals to its PCI BAR. Normally,
>> the PCI core discovers devices and BARs using the PCI enumeration process.
>> There is no infrastructure to discover the hardware peripherals that are
>> present in a PCI device, and which can be accessed through the PCI BARs.
>>
>> Apparently, the device tree framework requires a device tree node for the
>> PCI device. Thus, it can generate the device tree nodes for hardware
>> peripherals underneath. Because PCI is self discoverable bus, there might
>> not be a device tree node created for PCI devices. Furthermore, if the PCI
>> device is hot pluggable, when it is plugged in, the device tree nodes for
>> its parent bridges are required. Add support to generate device tree node
>> for PCI bridges.
>>
>> Add an of_pci_make_dev_node() interface that can be used to create device
>> tree node for PCI devices.
>>
>> Add a PCI_DYNAMIC_OF_NODES config option. When the option is turned on,
>> the kernel will generate device tree nodes for PCI bridges unconditionally.
>>
>> Initially, add the basic properties for the dynamically generated device
>> tree nodes which include #address-cells, #size-cells, device_type,
>> compatible, ranges, reg.
>>
>> Acked-by: Bjorn Helgaas <bhelgaas@google.com>
>> Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
> I tried to bring this up for a custom PCIe card emulated in QEMU on an ARM ACPI
> machine.
>
> There are some missing parts that were present in Clements series, but not this
> one, particularly creation of the root pci object.
Thanks for trying this. The entire effort was separated in different
phases. That is why this patchset does not include creating of_root.
>
> Anyhow, hit an intermittent crash...
>
>
>> ---
>> +static int of_pci_prop_intr_map(struct pci_dev *pdev, struct of_changeset *ocs,
>> + struct device_node *np)
>> +{
>> + struct of_phandle_args out_irq[OF_PCI_MAX_INT_PIN];
>> + u32 i, addr_sz[OF_PCI_MAX_INT_PIN], map_sz = 0;
>> + __be32 laddr[OF_PCI_ADDRESS_CELLS] = { 0 };
>> + u32 int_map_mask[] = { 0xffff00, 0, 0, 7 };
>> + struct device_node *pnode;
>> + struct pci_dev *child;
>> + u32 *int_map, *mapp;
>> + int ret;
>> + u8 pin;
>> +
>> + pnode = pci_device_to_OF_node(pdev->bus->self);
>> + if (!pnode)
>> + pnode = pci_bus_to_OF_node(pdev->bus);
>> +
>> + if (!pnode) {
>> + pci_err(pdev, "failed to get parent device node");
>> + return -EINVAL;
>> + }
>> +
>> + laddr[0] = cpu_to_be32((pdev->bus->number << 16) | (pdev->devfn << 8));
>> + for (pin = 1; pin <= OF_PCI_MAX_INT_PIN; pin++) {
>> + i = pin - 1;
>> + out_irq[i].np = pnode;
>> + out_irq[i].args_count = 1;
>> + out_irq[i].args[0] = pin;
>> + ret = of_irq_parse_raw(laddr, &out_irq[i]);
>> + if (ret) {
>> + pci_err(pdev, "parse irq %d failed, ret %d", pin, ret);
>> + continue;
> If all the interrupt parsing fails we continue ever time...

Did you use Clement's patch to create of_root? I am just wondering if
parsing irq could fail on a dt-based system.

And yes, the failure case should be handled without crash. I think if
irq parsing failed,  the interrupt-map pair generation should be skipped.


Thanks,

Lizhi

>
>> + }
>> + ret = of_property_read_u32(out_irq[i].np, "#address-cells",
>> + &addr_sz[i]);
>> + if (ret)
>> + addr_sz[i] = 0;
> This never happens.
>
>> + }
>> +
>> + list_for_each_entry(child, &pdev->subordinate->devices, bus_list) {
>> + for (pin = 1; pin <= OF_PCI_MAX_INT_PIN; pin++) {
>> + i = pci_swizzle_interrupt_pin(child, pin) - 1;
>> + map_sz += 5 + addr_sz[i] + out_irq[i].args_count;
> and here we end up derefencing random memory which happens in my case to cause
> a massive allocation sometimes and that fails one of the assertions in the
> allocator.
>
> I'd suggest just setting addr_sz[xxx] = {}; above
> to ensure it's initialized. Then the if(ret) handling should not be needed
> as well as of_property_read_u32 should be side effect free I hope!
>
>> + }
>> + }
>> +
>> + int_map = kcalloc(map_sz, sizeof(u32), GFP_KERNEL);
>> + mapp = int_map;

\
 
 \ /
  Last update: 2023-09-12 00:12    [W:0.432 / U:0.800 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site