Re: [PATCH v3 1/1] PCI/ERR: Fix reset logic in pcie_do_recovery() call

[Bjorn Helgaas <helgaas@kernel.org>]

On Fri, Jul 24, 2020 at 12:07:55PM -0700, sathyanarayanan.kuppuswamy@linux.intel.com wrote:
> From: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
> 
> Current pcie_do_recovery() implementation has following two issues:
> 

I'm having trouble parsing this out, probably just lack of my
understanding...

> 1. Fatal (DPC) error recovery is currently broken for non-hotplug
> capable devices. Current fatal error recovery implementation relies
> on PCIe hotplug (pciehp) handler for detaching and re-enumerating
> the affected devices/drivers. pciehp handler listens for DLLSC state
> changes and handles device/driver detachment on DLLSC_LINK_DOWN event
> and re-enumeration on DLLSC_LINK_UP event. So when dealing with
> non-hotplug capable devices, recovery code does not restore the state
> of the affected devices correctly. 

Apparently in the hotplug case, something *does* restore the state of
affected devices?

> Correct implementation should
> restore the device state and call report_slot_reset() function after
> resetting the link to restore the state of the device/driver.
> 
> You can find fatal non-hotplug related issues reported in following links:
> 
> https://lore.kernel.org/linux-pci/20200527083130.4137-1-Zhiqiang.Hou@nxp.com/
> https://lore.kernel.org/linux-pci/12115.1588207324@famine/
> https://lore.kernel.org/linux-pci/0e6f89cd6b9e4a72293cc90fafe93487d7c2d295.1585000084.git.sathyanarayanan.kuppuswamy@linux.intel.com/
> 
> 2. For non-fatal errors if report_error_detected() or
> report_mmio_enabled() functions requests PCI_ERS_RESULT_NEED_RESET then
> current pcie_do_recovery() implementation does not do the requested
> explicit device reset, instead just calls the report_slot_reset() on all
> affected devices. Notifying about the reset via report_slot_reset()
> without doing the actual device reset is incorrect.

Is it possible to fix these two issues separately?

> To fix above issues, use PCI_ERS_RESULT_NEED_RESET as error state after
> successful reset_link() operation. This will ensure ->slot_reset() be
> called after reset_link() operation for fatal errors. Also call
> pci_bus_reset() to do slot/bus reset() before triggering device specific
> ->slot_reset() callback. Also, using pci_bus_reset() will restore the state
> of the devices after performing the reset operation.
> 
> Even though using pci_bus_reset() will do redundant reset operation after
> ->reset_link() for fatal errors, it should should affect the functional
> behavior.
> 
> [original patch is from jay.vosburgh@canonical.com]
> [original patch link https://lore.kernel.org/linux-pci/12115.1588207324@famine/]
> Fixes: 6d2c89441571 ("PCI/ERR: Update error status after reset_link()")
> Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
> Signed-off-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
> ---
> 
> Changes since v2:
>  * Changed the subject of patch to "PCI/ERR: Fix reset logic in
>    pcie_do_recovery() call". v2 patch link is,
>    https://lore.kernel.org/linux-pci/ce417fbf81a8a46a89535f44b9224ee9fbb55a29.1591307288.git.sathyanarayanan.kuppuswamy@linux.intel.com/
>  * Squashed "PCI/ERR: Add reset support for non fatal errors" patch.
> 
>  drivers/pci/pcie/err.c | 41 +++++++++++++++++++++++++++++++++++++----
>  1 file changed, 37 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
> index 14bb8f54723e..b5eb6ba65be1 100644
> --- a/drivers/pci/pcie/err.c
> +++ b/drivers/pci/pcie/err.c
> @@ -165,8 +165,29 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
>  	pci_dbg(dev, "broadcast error_detected message\n");
>  	if (state == pci_channel_io_frozen) {
>  		pci_walk_bus(bus, report_frozen_detected, &status);
> +		/*
> +		 * After resetting the link using reset_link() call, the
> +		 * possible value of error status is either
> +		 * PCI_ERS_RESULT_DISCONNECT (failure case) or
> +		 * PCI_ERS_RESULT_NEED_RESET (success case).
> +		 * So ignore the return value of report_error_detected()
> +		 * call for fatal errors.
> +		 *
> +		 * In EDR mode, since AER and DPC Capabilities are owned by
> +		 * firmware, reported_error_detected() will return error
> +		 * status PCI_ERS_RESULT_NO_AER_DRIVER. Continuing
> +		 * pcie_do_recovery() with error status as
> +		 * PCI_ERS_RESULT_NO_AER_DRIVER will report recovery failure
> +		 * irrespective of recovery status. But successful reset_link()
> +		 * call usually recovers all fatal errors. So ignoring the
> +		 * status result of report_error_detected() also helps EDR based
> +		 * error recovery.

This chain of connections is too long and complicated to be
maintainable: EDR, AER/DPC ownership, NO_AER_DRIVER, etc.  It's always
a bad sign when code needs this much explanation.

I don't know how to simplify this, but it does need to be simplified
somehow.  I think it might have been my idea to feed all these paths
(AER, DPC, EDR) through the same recovery function, but I'm starting
to think it was a bad idea.

Or maybe it just isn't factored correctly.  IIUC for the DPC and EDR
paths, but not for AER, the device (actually the whole subtree) has
been reset before we even get here.  So it might help to separate out
the reset part.

> +		 */
>  		status = reset_link(dev);
> -		if (status != PCI_ERS_RESULT_RECOVERED) {
> +		if (status == PCI_ERS_RESULT_RECOVERED) {
> +			status = PCI_ERS_RESULT_NEED_RESET;
> +		} else {
> +			status = PCI_ERS_RESULT_DISCONNECT;
>  			pci_warn(dev, "link reset failed\n");
>  			goto failed;
>  		}
> @@ -182,10 +203,22 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
>  
>  	if (status == PCI_ERS_RESULT_NEED_RESET) {
>  		/*
> -		 * TODO: Should call platform-specific
> -		 * functions to reset slot before calling
> -		 * drivers' slot_reset callbacks?
> +		 * TODO: Optimize the call to pci_reset_bus()
> +		 *
> +		 * There are two components to pci_reset_bus().
> +		 *
> +		 * 1. Do platform specific slot/bus reset.
> +		 * 2. Save/Restore all devices in the bus.
> +		 *
> +		 * For hotplug capable devices and fatal errors,
> +		 * device is already in reset state due to link
> +		 * reset. So repeating platform specific slot/bus
> +		 * reset via pci_reset_bus() call is redundant. So
> +		 * can optimize this logic and conditionally call
> +		 * pci_reset_bus().
>  		 */
> +		pci_reset_bus(dev);
> +
>  		status = PCI_ERS_RESULT_RECOVERED;
>  		pci_dbg(dev, "broadcast slot_reset message\n");
>  		pci_walk_bus(bus, report_slot_reset, &status);
> -- 
> 2.17.1
>