Re: [PATCH V4] PCI/ASPM: Update saved buffers with latest ASPM

[Bjorn Helgaas <helgaas@kernel.org>]

On Tue, Mar 05, 2024 at 04:03:42PM -0600, Bjorn Helgaas wrote:
> [+to Sathy, David in case you want to update your Reviewed-by]
> 
> On Thu, Feb 22, 2024 at 11:14:36PM +0530, Vidya Sagar wrote:
> > Many PCIe device drivers save the configuration state of their respective
> > devices during probe and restore the same when their 'slot_reset' hook
> > is called through PCIe Error Recovery Handler.
> > 
> > If the system has a change in ASPM policy after the driver's probe is
> > called and before error event occurred, 'slot_reset' hook restores the
> > PCIe configuration state to what it was at the time of probe but not to
> > what it was just before the occurrence of the error event.
> > This effectively leads to a mismatch in the ASPM configuration between
> > the device and its upstream parent device.
> > 
> > Update the saved configuration state of the device with the latest info
> > whenever there is a change w.r.t ASPM policy.
> > 
> > Signed-off-by: Vidya Sagar <vidyas@nvidia.com>
> 
> > -void pci_save_aspm_state(struct pci_dev *pdev);
> > +void pci_save_aspm_l1ss_state(struct pci_dev *pdev);
> 
> I rebased this again on top of my pci/aspm updates to remove the need
> for the rename above.
> 
> > +static void pci_save_aspm_state(struct pci_dev *dev)
> > +{
> > +	struct pci_cap_saved_state *save_state;
> > +	u16 *cap;
> > +
> > +	if (!pci_is_pcie(dev))
> > +		return;
> > +
> > +	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
> > +	if (!save_state)
> > +		return;
> > +
> > +	cap = (u16 *)&save_state->cap.data[0];
> > +	pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[1]);
> 
> And I changed this part so it only updates the PCI_EXP_LNKCTL_ASPMC
> bits, not the entire LNKCTL.
> 
> Updating the entire saved register probably wouldn't *break* anything,
> but it could randomly hide other LNKCTL changes depending on whether
> or not ASPM configuration was changed in the interim.  For example:
> 
>   - driver .probe() saves LNKCTL
>   - LNKCTL changes some non-ASPMC thing via setpci or other mechanism
>   - save_state updated via pcie_config_aspm_link()
> 
> A restore in .slot_reset() would restore different LNKCTL values for
> the non-ASPMC change depending on whether pcie_config_aspm_link() was
> used.

Oops, I blew it here.  I think it's good to limit the LNKCTL changes
we put in save_state, but PCI_EXP_LNKCTL_ASPMC is not enough.  We
should include PCI_EXP_LNKCTL_CLKREQ_EN as well since that may be
updated in many of the same paths that update PCI_EXP_LNKCTL_ASPMC.

I updated it to the patch at the bottom; here's the interdiff first:

-------------------------------------------------------------------------

diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index 46352132bb14..10160d82c10a 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -304,18 +304,25 @@ static int policy_to_clkpm_state(struct pcie_link_state *link)
 static void pci_update_aspm_saved_state(struct pci_dev *dev)
 {
 	struct pci_cap_saved_state *save_state;
-	u16 *cap, lnkctl, aspmc;
+	u16 *cap, lnkctl, aspm_ctl;
 
 	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
 	if (!save_state)
 		return;
 
+	pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &lnkctl);
+
+	/*
+	 * Update ASPM and CLKREQ bits of LNKCTL in save_state. We only
+	 * write PCI_EXP_LNKCTL_CCC during enumeration, so it shouldn't
+	 * change after being captured in save_state.
+	 */
+	aspm_ctl = lnkctl & (PCI_EXP_LNKCTL_ASPMC | PCI_EXP_LNKCTL_CLKREQ_EN);
+	lnkctl &= ~(PCI_EXP_LNKCTL_ASPMC | PCI_EXP_LNKCTL_CLKREQ_EN);
+
 	/* Depends on pci_save_pcie_state(): cap[1] is LNKCTL */
 	cap = (u16 *)&save_state->cap.data[0];
-	pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &lnkctl);
-	aspmc = FIELD_GET(PCI_EXP_LNKCTL_ASPMC, lnkctl);
-	cap[1] = (lnkctl & ~PCI_EXP_LNKCTL_ASPMC) |
-		FIELD_PREP(PCI_EXP_LNKCTL_ASPMC, aspmc);
+	cap[1] = lnkctl | aspm_ctl;
 }
 
 static void pcie_set_clkpm_nocheck(struct pcie_link_state *link, int enable)

-------------------------------------------------------------------------

And here's the entire patch:

commit ff92b9348534 ("PCI/ASPM: Update save_state when configuration changes")
Author: Vidya Sagar <vidyas@nvidia.com>
Date:   Fri Feb 23 13:36:24 2024 -0600

    PCI/ASPM: Update save_state when configuration changes
    
    Many PCIe device drivers save the configuration state of their device
    during probe and restore it when their .slot_reset() hook is called during
    PCIe error recovery.
    
    If the ASPM configuration is changed after the driver's probe is called and
    before an error event occurs, .slot_reset() restores the ASPM configuration
    to what it was at the time of probe, not to what it was just before the
    occurrence of the error event.  This leads to a mismatch in ASPM
    configuration between the device and its upstream device.
    
    Update the saved configuration of the device when the ASPM configuration
    changes.
    
    Link: https://lore.kernel.org/r/20240222174436.3565146-1-vidyas@nvidia.com
    Signed-off-by: Vidya Sagar <vidyas@nvidia.com>
    [bhelgaas: commit log, rebase to pci/aspm, rename to
    pci_update_aspm_saved_state() since it updates only LNKCTL, update only
    ASPMC and CLKREQ_EN in LNKCTL]
    Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
    Reviewed-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
    Reviewed-by: David E. Box <david.e.box@linux.intel.com>


diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index 1379b8decdf1..10160d82c10a 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -301,16 +301,42 @@ static int policy_to_clkpm_state(struct pcie_link_state *link)
 	return 0;
 }
 
+static void pci_update_aspm_saved_state(struct pci_dev *dev)
+{
+	struct pci_cap_saved_state *save_state;
+	u16 *cap, lnkctl, aspm_ctl;
+
+	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
+	if (!save_state)
+		return;
+
+	pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &lnkctl);
+
+	/*
+	 * Update ASPM and CLKREQ bits of LNKCTL in save_state. We only
+	 * write PCI_EXP_LNKCTL_CCC during enumeration, so it shouldn't
+	 * change after being captured in save_state.
+	 */
+	aspm_ctl = lnkctl & (PCI_EXP_LNKCTL_ASPMC | PCI_EXP_LNKCTL_CLKREQ_EN);
+	lnkctl &= ~(PCI_EXP_LNKCTL_ASPMC | PCI_EXP_LNKCTL_CLKREQ_EN);
+
+	/* Depends on pci_save_pcie_state(): cap[1] is LNKCTL */
+	cap = (u16 *)&save_state->cap.data[0];
+	cap[1] = lnkctl | aspm_ctl;
+}
+
 static void pcie_set_clkpm_nocheck(struct pcie_link_state *link, int enable)
 {
 	struct pci_dev *child;
 	struct pci_bus *linkbus = link->pdev->subordinate;
 	u32 val = enable ? PCI_EXP_LNKCTL_CLKREQ_EN : 0;
 
-	list_for_each_entry(child, &linkbus->devices, bus_list)
+	list_for_each_entry(child, &linkbus->devices, bus_list) {
 		pcie_capability_clear_and_set_word(child, PCI_EXP_LNKCTL,
 						   PCI_EXP_LNKCTL_CLKREQ_EN,
 						   val);
+		pci_update_aspm_saved_state(child);
+	}
 	link->clkpm_enabled = !!enable;
 }
 
@@ -929,6 +955,12 @@ static void pcie_config_aspm_link(struct pcie_link_state *link, u32 state)
 		pcie_config_aspm_dev(parent, upstream);
 
 	link->aspm_enabled = state;
+
+	/* Update latest ASPM configuration in saved context */
+	pci_save_aspm_l1ss_state(link->downstream);
+	pci_update_aspm_saved_state(link->downstream);
+	pci_save_aspm_l1ss_state(parent);
+	pci_update_aspm_saved_state(parent);
 }
 
 static void pcie_config_aspm_path(struct pcie_link_state *link)