netdev - Re: [PATCH] qlcnic: check pci_reset

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: <32f18da1-eeb9-3cd6-398d-77f76596b7c3@yandex-team.ru>
Date:   Thu, 6 Apr 2023 12:23:49 +0300
From:   Denis Plotnikov <den-plotnikov@...dex-team.ru>
To:     Simon Horman <simon.horman@...igine.com>,
        Bjorn Helgaas <helgaas@...nel.org>
Cc:     netdev@...r.kernel.org, linux-kernel@...r.kernel.org,
        shshaikh@...vell.com, manishc@...vell.com,
        GR-Linux-NIC-Dev@...vell.com, davem@...emloft.net,
        edumazet@...gle.com, kuba@...nel.org, pabeni@...hat.com,
        Bjorn Helgaas <bhelgaas@...gle.com>, linux-pci@...r.kernel.org
Subject: Re: [PATCH] qlcnic: check pci_reset_function result


On 06.04.2023 10:03, Simon Horman wrote:
> On Wed, Apr 05, 2023 at 02:37:08PM -0500, Bjorn Helgaas wrote:
>> On Wed, Apr 05, 2023 at 03:04:39PM +0200, Simon Horman wrote:
>>> On Mon, Apr 03, 2023 at 01:58:49PM +0300, Denis Plotnikov wrote:
>>>> On 31.03.2023 20:52, Simon Horman wrote:
>>>>> On Fri, Mar 31, 2023 at 11:06:05AM +0300, Denis Plotnikov wrote:
>>>>>> Static code analyzer complains to unchecked return value.
>>>>>> It seems that pci_reset_function return something meaningful
>>>>>> only if "reset_methods" is set.
>>>>>> Even if reset_methods isn't used check the return value to avoid
>>>>>> possible bugs leading to undefined behavior in the future.
>>>>>>
>>>>>> Signed-off-by: Denis Plotnikov <den-plotnikov@...dex-team.ru>
>>>>> nit: The tree this patch is targeted at should be designated, probably
>>>>>        net-next, so the '[PATCH net-next]' in the subject.
>>>>>
>>>>>> ---
>>>>>>    drivers/net/ethernet/qlogic/qlcnic/qlcnic_ctx.c | 4 +++-
>>>>>>    1 file changed, 3 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/drivers/net/ethernet/qlogic/qlcnic/qlcnic_ctx.c b/drivers/net/ethernet/qlogic/qlcnic/qlcnic_ctx.c
>>>>>> index 87f76bac2e463..39ecfc1a1dbd0 100644
>>>>>> --- a/drivers/net/ethernet/qlogic/qlcnic/qlcnic_ctx.c
>>>>>> +++ b/drivers/net/ethernet/qlogic/qlcnic/qlcnic_ctx.c
>>>>>> @@ -628,7 +628,9 @@ int qlcnic_fw_create_ctx(struct qlcnic_adapter *dev)
>>>>>>    	int i, err, ring;
>>>>>>    	if (dev->flags & QLCNIC_NEED_FLR) {
>>>>>> -		pci_reset_function(dev->pdev);
>>>>>> +		err = pci_reset_function(dev->pdev);
>>>>>> +		if (err && err != -ENOTTY)
>>>>> Are you sure about the -ENOTTY part?
>>>>>
>>>>> It seems odd to me that an FLR would be required but reset is not supported.
>>>> No, I'm not sure. My logic is: if the reset method isn't set than
>>>> pci_reset_function() returns -ENOTTY so treat that result as ok.
>>>> pci_reset_function may return something different than -ENOTTY only if
>>>> pci_reset_fn_methods[m].reset_fn is set.
>>> I see your reasoning: -ENOTTY means nothing happened, and probably that is ok.
>>> I think my main question is if that can ever happen.
>>> If that is unknown, then I think this conservative approach makes sense.
>> The commit log mentions "reset_methods", which I don't think is really
>> relevant here because reset_methods is an internal implementation
>> detail.  The point is that pci_reset_function() returns 0 if it was
>> successful and a negative value if it failed.
>>
>> If the driver thinks the device needs to be reset, ignoring any
>> negative return value seems like a mistake because the device was not
>> reset.
>>
>> If the reset is required for a firmware update to take effect, maybe a
>> diagnostic would be helpful if it fails, e.g., the other "Adapter
>> initialization failed.  Please reboot" messages.
>>
>> "QLCNIC_NEED_FLR" suggests that the driver expects an FLR (as opposed
>> to other kinds of reset).  If the driver knows that all qlcnic devices
>> support FLR, it could use pcie_flr() directly.
>>
>> pci_reset_function() does have the possibility that the reset works on
>> some devices but not all.  Secondary Bus Reset fails if there are
>> other functions on the same bus, e.g., a multi-function device.  And
>> there's some value in doing the reset the same way in all cases.
>>
>> So I would suggest something like:
>>
>>    if (dev->flags & QLCNIC_NEED_FLR) {
>>      err = pcie_flr(dev->pdev);
>>      if (err) {
>>        dev_err(&pdev->dev, "Adapter reset failed (%d). Please reboot\n", err);
>>        return err;
>>      }
>>      dev->flags &= ~QLCNIC_NEED_FLR;
>>    }
>>
>> Or, if there are qlcnic devices that don't support FLR:
>>
>>    if (dev->flags & QLCNIC_NEED_FLR) {
>>      err = pci_reset_function(dev->pdev);
>>      if (err) {
>>        dev_err(&pdev->dev, "Adapter reset failed (%d). Please reboot\n", err);
>>        return err;
>>      }
>>      dev->flags &= ~QLCNIC_NEED_FLR;
>>    }
> Thanks Bjorn,
>
> that is very helpful.
>
> I think that in order to move to option #1 some information would be needed
> from those familiar with the device(s). As it is a more invasive change -
> pci_reset_function -> pcie_flr.
>
> So my feeling is that, in lieu of such feedback, option #2 is a good
> improvement on the current code.
>
> OTOH, this driver is 'Supported' as opposed to 'Maintained'.
> So perhaps we can just use our best judgement and go for option #1.

So, it looks like option #2 is the safest choice as we do reset only if 
FLR is needed (when pci_reset_function() makes sense)

If all agree with that I'll re-send the path