diff mbox series

[7/8] core/pcie-slot: Better explain suprise_check

Message ID 20190318061058.3753-7-oohall@gmail.com
State Accepted
Headers show
Series [1/8] hdata/iohub: Look for IOVPD on P9 | expand

Checks

Context Check Description
snowpatch_ozlabs/apply_patch success Successfully applied on branch master (2ba5ce84a197ee61423355f443a3ff3eea185ff1)
snowpatch_ozlabs/snowpatch_job_snowpatch-skiboot fail Test snowpatch/job/snowpatch-skiboot on branch master
snowpatch_ozlabs/snowpatch_job_snowpatch-skiboot-dco success Signed-off-by present

Commit Message

Oliver O'Halloran March 18, 2019, 6:10 a.m. UTC 
Working out what was actually going on here took forever.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
---
 core/pcie-slot.c | 27 +++++++++++----------------
 1 file changed, 11 insertions(+), 16 deletions(-)
diff mbox series

Patch

diff --git a/core/pcie-slot.c b/core/pcie-slot.c
index e90fc5d3a932..8696129f3a23 100644
--- a/core/pcie-slot.c
+++ b/core/pcie-slot.c
@@ -215,24 +215,19 @@  static int64_t pcie_slot_set_power_state_ext(struct pci_slot *slot, uint8_t val,
 		return OPAL_SUCCESS;
 	}
 
-	/* The power supply to the slot should be always on when surprise
-	 * hotplug is claimed. For this case, update with the requested
-	 * power state and bail immediately.
+	/*
+	 * Suprise hotpluggable slots need to be handled with care since
+	 * many systems do not implement the presence detect side-band
+	 * signal. Instead, they rely on in-band presence to report the
+	 * existence of a hotplugged card.
 	 *
-	 * The PCIe link is likely down if we're powering on the slot upon
-	 * the detected presence. Nothing behind the slot will be probed if
-	 * we do it immediately even we do have PCI devices connected to the
-	 * slot. For this case, we force upper layer to wait for the PCIe
-	 * link to be up before probing the PCI devices behind the slot. It's
-	 * only concerned in surprise hotplug path. In managed hot-add path,
-	 * the PCIe link should have been ready before we power on the slot.
-	 * However, it's not harmful to do so in managed hot-add path.
+	 * This is problematic because:
+	 * a) When PERST is asserted in-band presence doesn't work, and
+	 * b) Switches assert PERST as a part of the "slot power down" sequence
 	 *
-	 * When flag PCI_SLOT_FLAG_FORCE_POWERON is set for the PCI slot, we
-	 * should turn on the slot's power supply on hardware on user's request
-	 * because that might have been lost. Otherwise, the PCIe link behind
-	 * the slot won't become ready for ever and PCI adapter behind the slot
-	 * can't be probed successfully.
+	 * To work around the problem we leave the slot physically powered on
+	 * and exit early here. This way when a new card is inserted, the switch
+	 * will raise an interrupt due to the PresDet status changing.
 	 */
 	if (surprise_check && slot->surprise_pluggable) {
 		slot->power_state = val;