Hello,
there are several crash reports due to struct irq_data->chip_data being NULL.
I was under the impression all the "recent changes" to pci-hyperv.c
would fix them. But apparently this specific issue is still there.
What does serialize read and write access to struct irq_data->chip_data?
It seems hv_msi_free can run while other code paths still access at least
->chip_data.
The change below may reduce the window, but I'm not confident this would
actually resolve the concurrency issues.
Olaf
--- a/drivers/pci/controller/pci-hyperv.c
+++ b/drivers/pci/controller/pci-hyperv.c
@@ -1760,8 +1760,9 @@ static void hv_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
msi_desc->nvec_used > 1;
/* Reuse the previous allocation */
- if (data->chip_data && multi_msi) {
- int_desc = data->chip_data;
+ virt_rmb();
+ int_desc = READ_ONCE(data->chip_data);
+ if (int_desc && multi_msi) {
msg->address_hi = int_desc->address >> 32;
msg->address_lo = int_desc->address & 0xffffffff;
msg->data = int_desc->data;
@@ -1778,8 +1779,9 @@ static void hv_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
goto return_null_message;
/* Free any previous message that might have already been composed. */
- if (data->chip_data && !multi_msi) {
- int_desc = data->chip_data;
+ virt_rmb();
+ int_desc = READ_ONCE(data->chip_data);
+ if (int_desc && !multi_msi) {
data->chip_data = NULL;
hv_int_desc_free(hpdev, int_desc);
}
Attachment:
pgp2bGZflItVX.pgp
Description: Digitale Signatur von OpenPGP
