As PCI_EXP_LNKCAP is never masked when loading, this patch does affect
cross version migration. It seems we need machine type compat to deal
with that. What do you suggest, Michael?
On 5/20/22 12:49 AM, Michael S. Tsirkin wrote:
On Thu, May 19, 2022 at 10:45:59PM +0800, Wenliang Wang wrote:
pcie_cap_fill_slot_lnk() wrongly set PCI_EXP_LNKCAP when slot speed
and width is not set, causing strange downstream port link cap
(Speed unknown, Width x0) and pcie devices native hotplug error on Linux:
[ 3.545654] pcieport 0000:02:00.0: pciehp: link training error: status 0x2000
[ 3.547143] pcieport 0000:02:00.0: pciehp: Failed to check link status
We do not touch PCI_EXP_LNKCAP when speed=0 or width=0, as pcie_cap_v1_fill()
already do the default setting for us.
Signed-off-by: Wenliang Wang <wangwenliang.1...@bytedance.com>
do we need machine type compat dance with this?
can you check whether this affects cross version
migration please?
---
hw/pci/pcie.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/hw/pci/pcie.c b/hw/pci/pcie.c
index 68a62da..c82e7fc 100644
--- a/hw/pci/pcie.c
+++ b/hw/pci/pcie.c
@@ -92,6 +92,11 @@ static void pcie_cap_fill_slot_lnk(PCIDevice *dev)
return;
}
+ /* Use default LNKCAP setting */
+ if (s->speed == 0 || s->width == 0) {
+ return;
+ }
+
/* Clear and fill LNKCAP from what was configured above */
pci_long_test_and_clear_mask(exp_cap + PCI_EXP_LNKCAP,
PCI_EXP_LNKCAP_MLW | PCI_EXP_LNKCAP_SLS);
--
2.7.4
