linux-kernel - Re: [PATCH v2 9/9] PCI: rockchip: Add parameter check for RK3399 PCIe endpoint core set

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <38ae72c9-0f0b-1a94-d2e0-f4ea80e94705@opensource.wdc.com>
Date:   Tue, 21 Feb 2023 19:55:26 +0900
From:   Damien Le Moal <damien.lemoal@...nsource.wdc.com>
To:     Rick Wertenbroek <rick.wertenbroek@...il.com>
Cc:     alberto.dassatti@...g-vd.ch, xxm@...k-chips.com,
        rick.wertenbroek@...g-vd.ch, Rob Herring <robh+dt@...nel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>,
        Heiko Stuebner <heiko@...ech.de>,
        Shawn Lin <shawn.lin@...k-chips.com>,
        Lorenzo Pieralisi <lpieralisi@...nel.org>,
        Krzysztof Wilczyński <kw@...ux.com>,
        Bjorn Helgaas <bhelgaas@...gle.com>,
        Jani Nikula <jani.nikula@...el.com>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Rodrigo Vivi <rodrigo.vivi@...el.com>,
        Mikko Kovanen <mikko.kovanen@...amobile.com>,
        devicetree@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
        linux-rockchip@...ts.infradead.org, linux-kernel@...r.kernel.org,
        linux-pci@...r.kernel.org
Subject: Re: [PATCH v2 9/9] PCI: rockchip: Add parameter check for RK3399 PCIe
 endpoint core set_msi()

On 2/21/23 19:47, Rick Wertenbroek wrote:
> On Wed, Feb 15, 2023 at 2:39 AM Damien Le Moal
> <damien.lemoal@...nsource.wdc.com> wrote:
>>
>> On 2/14/23 23:08, Rick Wertenbroek wrote:
>>> The RK3399 PCIe endpoint core supports only a single PCIe physcial
>>> function (function number 0), therefore return -EINVAL if set_msi() is
>>> called with a function number greater than 0.
>>> The PCIe standard only allows the multi message capability (MMC) value
>>> to be up to 0x5 (32 messages), therefore return -EINVAL if set_msi() is
>>> called with a MMC value of over 0x5.
>>>
>>> Signed-off-by: Rick Wertenbroek <rick.wertenbroek@...il.com>
>>> ---
>>>  drivers/pci/controller/pcie-rockchip-ep.c | 10 ++++++++++
>>>  1 file changed, 10 insertions(+)
>>>
>>> diff --git a/drivers/pci/controller/pcie-rockchip-ep.c b/drivers/pci/controller/pcie-rockchip-ep.c
>>> index b7865a94e..80634b690 100644
>>> --- a/drivers/pci/controller/pcie-rockchip-ep.c
>>> +++ b/drivers/pci/controller/pcie-rockchip-ep.c
>>> @@ -294,6 +294,16 @@ static int rockchip_pcie_ep_set_msi(struct pci_epc *epc, u8 fn, u8 vfn,
>>>       struct rockchip_pcie *rockchip = &ep->rockchip;
>>>       u32 flags;
>>>
>>> +     if (fn) {
>>> +             dev_err(&epc->dev, "This endpoint controller only supports a single physical function\n");
>>> +             return -EINVAL;
>>> +     }
>>
>> Checking this here is late... Given that at most only one physical
>> function is supported, the check should be in rockchip_pcie_parse_ep_dt().
>> Something like:
>>
>>         err = of_property_read_u8(dev->of_node, "max-functions",
>>                                   &ep->epc->max_functions);
>>
>>         if (err < 0 || ep->epc->max_functions > 1)
>>
>>                 ep->epc->max_functions = 1;
>>
> 
> Yes, this could be moved to the probe, thanks.
> 
>> And all the macros with the (fn) argument could also be simplified
>> (argument fn removed) since fn will always be 0.
> 
> These functions cannot be simplified because they have to follow the signature
> given by "pci_epc_ops" (include/linux/pci-epc.h). And this signature has the
> function number as a parameter. If we change the function signature we won't
> be able to assign these functions to the pc_epc_ops structure

I was not suggesting to change the functions signature. I was suggesting
dropping the fn argument for the *macros*, e.g.

ROCKCHIP_PCIE_EP_FUNC_BASE(fn) -> ROCKCHIP_PCIE_EP_FUNC_BASE

since fn is always 0.

That said, I am not entirely sure if the limit really is 1 function at most. The
TRM seems to be suggesting that up to 4 functions can be supported...

[...]

>> Another nice cleanup: define ROCKCHIP_PCIE_EP_MSI_CTRL_REG to include the
>> ROCKCHIP_PCIE_EP_FUNC_BASE(fn) addition so that we do not have to do it
>> here all the time.
> 
> Yes, this could be an improvement but this is the way it is written
> everywhere in this
> driver, I chose to keep it so as to remain coherent with the rest of the driver.
> Cleaning this is not so important since this code will not be
> rewritten / changed so
> often. But I agree that it might be nicer. But, on the other side if
> at some point
> support for virtual functions would be added, the offsets would need
> to be computed
> based on the virtual function number and the code would be written
> like it is now,
> so I suggest keeping this the way it is for now.

Yes, sure, this can be cleaned later.

A more pressing problem is the lack of support for MSIX despite the fact that
the controller supports that *and* advertize it as a capability. That is what
was causing my problem with the Linux nvme driver and my prototype nvme epf
function driver: the host driver was seeing MSIX support (1 vector supported by
default), and so was allocating one MSIX for the device probe. But on the EP
end, it is MSI or INTX only... Working on adding that to solve this issue.

-- 
Damien Le Moal
Western Digital Research