Re: [PATCH 1/2] PCI: fix restoring resized BAR state on resume

["Christian König" <christian.koenig@amd.com>]

Am 02.07.2018 um 15:49 schrieb Bjorn Helgaas:
> On Mon, Jul 02, 2018 at 03:20:59PM +0200, Christian König wrote:
>> Am 30.06.2018 um 02:57 schrieb Bjorn Helgaas:
>>> On Thu, Jun 14, 2018 at 02:21:45PM +0200, Christian König wrote:
>>>> Resize BARs after resume to the expected size again.
>>>>
>>>> Signed-off-by: Christian König <christian.koenig@amd.com>
>>>> BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=199959
>>>> CC: stable@vger.kernel.org      # v4.15+
>>> I applied both of these to pci/resource for v4.19, thanks!  It'd be nice to
>>> have them in v4.18, but since it's not a regression and the problem has
>>> been there since v4.15, I'm not sure I could justify it.
>> Well it is a regression. The change to enable this in the upstream driver
>> was added in 4.16, but a good bunch of people also compile our backported
>> DKMS module.
>>
>> And now those people are complaining that the driver stopped working after
>> resume in 4.15 with the DKMS module and with 4.16 with the upstream module.
>>
>> Alternatively I can disable resizing from the driver from kernels older than
>> v4.19.
> Feel free to help justify including this in v4.18 by providing the
> details of the regression and the relevant commits.  If this is
> something that worked in v4.17, and some commit we merged for
> v4.18-rc1 broke it, it would be a slam dunk to include the fix for
> v4.18.  I took a quick look through the commits we added in v4.18-rc1
> but didn't see anything obvious.
>
> Obviously, putting a fix in v4.18 doesn't help the people running
> v4.15 or v4.16.  The stable backports *will* help with that, and we
> already have the tag for that.
>
> The current trajectory we're on is that these are headed for
> v4.19-rc1, and soon after they appear there, they will be backported
> to the v4.15-v4.18 stable kernels.

Ah! Thanks for the explanation on how fixes flow into stable kernels and 
you are right it's indeed not a regression in 4.18.

In this case everything is fine and the backports to stable kernels 
should handle that.

Thanks,
Christian.

>
> Bjorn
>
>>>> ---
>>>>    drivers/pci/pci.c | 28 ++++++++++++++++++++++++++++
>>>>    1 file changed, 28 insertions(+)
>>>>
>>>> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
>>>> index bd6f156dc3cf..d4685090378b 100644
>>>> --- a/drivers/pci/pci.c
>>>> +++ b/drivers/pci/pci.c
>>>> @@ -1159,6 +1159,33 @@ static void pci_restore_config_space(struct pci_dev *pdev)
>>>>    	}
>>>>    }
>>>> +static void pci_restore_rebar_state(struct pci_dev *pdev)
>>>> +{
>>>> +	unsigned int pos, nbars, i;
>>>> +	u32 ctrl;
>>>> +
>>>> +	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
>>>> +	if (!pos)
>>>> +		return;
>>>> +
>>>> +	pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
>>>> +	nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
>>>> +		    PCI_REBAR_CTRL_NBAR_SHIFT;
>>>> +
>>>> +	for (i = 0; i < nbars; i++, pos += 8) {
>>>> +		struct resource *res;
>>>> +		int bar_idx, size;
>>>> +
>>>> +		pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
>>>> +		bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
>>>> +		res = pdev->resource + bar_idx;
>>>> +		size = order_base_2((resource_size(res) >> 20) | 1) - 1;
>>>> +		ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
>>>> +		ctrl |= size << 8;
>>>> +		pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
>>>> +	}
>>>> +}
>>>> +
>>>>    /**
>>>>     * pci_restore_state - Restore the saved state of a PCI device
>>>>     * @dev: - PCI device that we're dealing with
>>>> @@ -1174,6 +1201,7 @@ void pci_restore_state(struct pci_dev *dev)
>>>>    	pci_restore_pri_state(dev);
>>>>    	pci_restore_ats_state(dev);
>>>>    	pci_restore_vc_state(dev);
>>>> +	pci_restore_rebar_state(dev);
>>>>    	pci_cleanup_aer_error_status_regs(dev);
>>>> -- 
>>>> 2.14.1
>>>>