[PATCH v2] PCI: hv: Disable/enable irq rather than bh in hv_compose_msi_msg()

[Dexuan Cui]

Commit de0aa7b2f97d ("PCI: hv: Fix 2 hang issues in hv_compose_msi_msg()")
uses local_bh_disable()/enable(), because hv_pci_onchannelcallback() can
also run in tasklet context as the channel event callback, and here we
want to avoid the race.

With CONFIG_PROVE_LOCKING=y in the recent mainline, or old kernels that
don't have commit f71b74bca637 ("irq/softirqs: Use lockdep to assert IRQs
are disabled/enabled"), when the upper layer irq code calls
hv_compose_msi_msg() with local irq DISABLED, we'll see a warning at the
beginning of __local_bh_enable_ip():

IRQs not enabled as expected
  WARNING: CPU: 0 PID: 408 at kernel/softirq.c:162 __local_bh_enable_ip

The warning exposes an issue in de0aa7b2f97d: local_bh_enable() can
potentially call do_softirq(), which is not supposed to run when local
irq is DISABLED. Let's fix this by using local_irq_save()/restore()
instead.

Note: hv_pci_onchannelcallback() is not a hot path because it's only
called when the PCI device is hot added and removed, which is infrequent.

Fixes: de0aa7b2f97d ("PCI: hv: Fix 2 hang issues in hv_compose_msi_msg()")
Signed-off-by: Dexuan Cui <decui@xxxxxxxxxxxxx>
Reviewed-by: Haiyang Zhang <haiyangz@xxxxxxxxxxxxx>
Cc: <stable@xxxxxxxxxxxxxxx>
Cc: Stephen Hemminger <sthemmin@xxxxxxxxxxxxx>
Cc: K. Y. Srinivasan <kys@xxxxxxxxxxxxx>
---

A trimmed version of the warning is:

IRQs not enabled as expected
WARNING: CPU: 0 PID: 408 at kernel/softirq.c:162 __local_bh_enable_ip+0xb0/0xe0
Call Trace:
 hv_compose_msi_msg+0x209/0x462 [pci_hyperv]
 irq_chip_compose_msi_msg+0x41/0x50
 msi_domain_activate+0x1a/0x40
 __irq_domain_activate_irq+0x59/0x90
 irq_domain_activate_irq+0x25/0x40
 __setup_irq+0x3ec/0x730
request_threaded_irq+0xfa/0x1a0
mlx4_init_eq_table+0x3c3/0x5f0 [mlx4_core]
mlx4_setup_hca+0x1db/0x750 [mlx4_core]
mlx4_load_one+0xad2/0x13b0 [mlx4_core]
mlx4_init_one+0x578/0x710 [mlx4_core]
local_pci_probe+0x1e/0x50
work_for_cpu_fn+0x10/0x20
process_one_work+0x1d4/0x5a0
worker_thread+0x1cb/0x3d0
kthread+0xf5/0x130


Changes since v1:
	Updated the changelog only (fixed typos and some inaccuracy)


 drivers/pci/controller/pci-hyperv.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
index ba1d4b5..eb20296 100644
--- a/drivers/pci/controller/pci-hyperv.c
+++ b/drivers/pci/controller/pci-hyperv.c
@@ -1073,6 +1073,7 @@ static void hv_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
 	struct pci_bus *pbus;
 	struct pci_dev *pdev;
 	struct cpumask *dest;
+	unsigned long flags;
 	struct compose_comp_ctxt comp;
 	struct tran_int_desc *int_desc;
 	struct {
@@ -1164,14 +1165,15 @@ static void hv_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
 		 * the channel callback directly when channel->target_cpu is
 		 * the current CPU. When the higher level interrupt code
 		 * calls us with interrupt enabled, let's add the
-		 * local_bh_disable()/enable() to avoid race.
+		 * local_irq_save()/restore() to avoid race:
+		 * hv_pci_onchannelcallback() can also run in tasklet.
 		 */
-		local_bh_disable();
+		local_irq_save(flags);
 
 		if (hbus->hdev->channel->target_cpu == smp_processor_id())
 			hv_pci_onchannelcallback(hbus);
 
-		local_bh_enable();
+		local_irq_restore(flags);
 
 		if (hpdev->state == hv_pcichild_ejecting) {
 			dev_err_once(&hbus->hdev->device,
-- 
2.7.4