Skip to contentSkip to conversation

thunderbolt: Fixes for v4.19-rc6

PATCH New Mika Westerberg 1 participant 3 messages Sep 24, 2018 Thunderbolt driver

lore
Share & tools
lore
#0 lore raw

cover thunderbolt: Fixes for v4.19-rc6

Hi Greg,

This includes two fixes:

- Stop handling ICM events when the domain is already removed upon module removal.

- If the module is built into the kernel image, initialize after IOMMUs to keep the driver working when IOMMUs are enabled.

I've included both patches as well in case prefer to apply them directely instead of pulling the signed tag.

Thanks!

The following changes since commit 11da3a7f84f19c26da6f86af878298694ede0804:

Linux 4.19-rc3 (2018-09-09 17:26:43 -0700)

are available in the Git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/westeri/thunderbolt.git tags/thunderbolt-fixes-for-v4.19-rc6

for you to fetch changes up to b3cea5ec12adc0fe0c0d264e56aa97ab617a51d6:

thunderbolt: Initialize after IOMMUs (2018-09-14 10:45:22 +0300)

----------------------------------------------------------------
thunderbolt: Fixes for v4.19-rc6

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

Mika Westerberg (2):
thunderbolt: Do not handle ICM events after domain is stopped
thunderbolt: Initialize after IOMMUs

 drivers/thunderbolt/icm.c | 49 ++++++++++++++++-----------------------
 drivers/thunderbolt/nhi.c |  2 +-
 2 files changed, 21 insertions(+), 30 deletions(-)
Signature · 1 line
2.18.0
#1 lore raw

PATCH 1/2 thunderbolt: Do not handle ICM events after domain is stopped

If there is a long chain of devices connected when the driver is loaded ICM sends device connected event for each and those are put to tb->wq for later processing. Now if the driver gets unloaded in the middle, so that the work queue is not yet empty it gets flushed by tb_domain_stop(). However, by that time the root switch is already removed so the driver crashes when it tries to dereference it in ICM event handling callbacks.

Fix this by checking whether the root switch is already removed. If it is we know that the domain is stopped and we should merely skip handling the event.

---
 drivers/thunderbolt/icm.c | 49 ++++++++++++++++-----------------------
 1 file changed, 20 insertions(+), 29 deletions(-)
drivers/thunderbolt/icm.c +20 -29
index e1e264a9a4c7..28fc4ce75edb 100644
@@ -738,14 +738,6 @@ icm_fr_xdomain_connected(struct tb *tb, const struct icm_pkg_header *hdr)
 	u8 link, depth;
 	u64 route;
 
-	/*
-	 * After NVM upgrade adding root switch device fails because we
-	 * initiated reset. During that time ICM might still send
-	 * XDomain connected message which we ignore here.
-	 */
-	if (!tb->root_switch)
-		return;
-
 	link = pkg->link_info & ICM_LINK_INFO_LINK_MASK;
 	depth = (pkg->link_info & ICM_LINK_INFO_DEPTH_MASK) >>
 		ICM_LINK_INFO_DEPTH_SHIFT;
@@ -1037,14 +1029,6 @@ icm_tr_device_connected(struct tb *tb, const struct icm_pkg_header *hdr)
 	if (pkg->hdr.packet_id)
 		return;
 
-	/*
-	 * After NVM upgrade adding root switch device fails because we
-	 * initiated reset. During that time ICM might still send device
-	 * connected message which we ignore here.
-	 */
-	if (!tb->root_switch)
-		return;
-
 	route = get_route(pkg->route_hi, pkg->route_lo);
 	authorized = pkg->link_info & ICM_LINK_INFO_APPROVED;
 	security_level = (pkg->hdr.flags & ICM_FLAGS_SLEVEL_MASK) >>
@@ -1408,19 +1392,26 @@ static void icm_handle_notification(struct work_struct *work)
 
 	mutex_lock(&tb->lock);
 
-	switch (n->pkg->code) {
-	case ICM_EVENT_DEVICE_CONNECTED:
-		icm->device_connected(tb, n->pkg);
-		break;
-	case ICM_EVENT_DEVICE_DISCONNECTED:
-		icm->device_disconnected(tb, n->pkg);
-		break;
-	case ICM_EVENT_XDOMAIN_CONNECTED:
-		icm->xdomain_connected(tb, n->pkg);
-		break;
-	case ICM_EVENT_XDOMAIN_DISCONNECTED:
-		icm->xdomain_disconnected(tb, n->pkg);
-		break;
+	/*
+	 * When the domain is stopped we flush its workqueue but before
+	 * that the root switch is removed. In that case we should treat
+	 * the queued events as being canceled.
+	 */
+	if (tb->root_switch) {
+		switch (n->pkg->code) {
+		case ICM_EVENT_DEVICE_CONNECTED:
+			icm->device_connected(tb, n->pkg);
+			break;
+		case ICM_EVENT_DEVICE_DISCONNECTED:
+			icm->device_disconnected(tb, n->pkg);
+			break;
+		case ICM_EVENT_XDOMAIN_CONNECTED:
+			icm->xdomain_connected(tb, n->pkg);
+			break;
+		case ICM_EVENT_XDOMAIN_DISCONNECTED:
+			icm->xdomain_disconnected(tb, n->pkg);
+			break;
+		}
 	}
 
 	mutex_unlock(&tb->lock);
Signature · 1 line
2.18.0
#2 lore raw

PATCH 2/2 thunderbolt: Initialize after IOMMUs

In reply to Mika Westerberg

If IOMMU is enabled and Thunderbolt driver is built into the kernel image, it will be probed before IOMMUs are attached to the PCI bus. Because of this DMA mappings the driver does will not go through IOMMU and start failing right after IOMMUs are enabled.

For this reason move the Thunderbolt driver initialization happen at rootfs level.

---
 drivers/thunderbolt/nhi.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/thunderbolt/nhi.c +1 -1
index 88cff05a1808..5cd6bdfa068f 100644
@@ -1191,5 +1191,5 @@ static void __exit nhi_unload(void)
 	tb_domain_exit();
 }
 
-fs_initcall(nhi_init);
+rootfs_initcall(nhi_init);
 module_exit(nhi_unload);
Signature · 1 line
2.18.0