net/mlx5: Free IRQ rmap and notifier on kernel shutdown

commit 314ded538e5f22e7610b1bf621402024a180ec80 upstream.

The kernel IRQ system needs the irq affinity notifier to be clear
before attempting to free the irq, see WARN_ON log below.

On a normal driver unload we don't have this issue since we do the
complete cleanup of the irq resources.

To fix this, put the important resources cleanup in a helper function
and use it in both normal driver unload and shutdown flows.

[ 4497.498434] ------------[ cut here ]------------
[ 4497.498726] WARNING: CPU: 0 PID: 9 at kernel/irq/manage.c:2034 free_irq+0x295/0x340
[ 4497.499193] Modules linked in:
[ 4497.499386] CPU: 0 PID: 9 Comm: kworker/0:1 Tainted: G        W          6.4.0-rc4+ #10
[ 4497.499876] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.2-1.fc38 04/01/2014
[ 4497.500518] Workqueue: events do_poweroff
[ 4497.500849] RIP: 0010:free_irq+0x295/0x340
[ 4497.501132] Code: 85 c0 0f 84 1d ff ff ff 48 89 ef ff d0 0f 1f 00 e9 10 ff ff ff 0f 0b e9 72 ff ff ff 49 8d 7f 28 ff d0 0f 1f 00 e9 df fd ff ff <0f> 0b 48 c7 80 c0 008
[ 4497.502269] RSP: 0018:ffffc90000053da0 EFLAGS: 00010282
[ 4497.502589] RAX: ffff888100949600 RBX: ffff88810330b948 RCX: 0000000000000000
[ 4497.503035] RDX: ffff888100949600 RSI: ffff888100400490 RDI: 0000000000000023
[ 4497.503472] RBP: ffff88810330c7e0 R08: ffff8881004005d0 R09: ffffffff8273a260
[ 4497.503923] R10: 0000000000000000 R11: 0000000000000000 R12: ffff8881009ae000
[ 4497.504359] R13: ffff8881009ae148 R14: 0000000000000000 R15: ffff888100949600
[ 4497.504804] FS:  0000000000000000(0000) GS:ffff88813bc00000(0000) knlGS:0000000000000000
[ 4497.505302] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 4497.505671] CR2: 00007fce98806298 CR3: 000000000262e005 CR4: 0000000000370ef0
[ 4497.506104] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 4497.506540] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 4497.507002] Call Trace:
[ 4497.507158]  <TASK>
[ 4497.507299]  ? free_irq+0x295/0x340
[ 4497.507522]  ? __warn+0x7c/0x130
[ 4497.507740]  ? free_irq+0x295/0x340
[ 4497.507963]  ? report_bug+0x171/0x1a0
[ 4497.508197]  ? handle_bug+0x3c/0x70
[ 4497.508417]  ? exc_invalid_op+0x17/0x70
[ 4497.508662]  ? asm_exc_invalid_op+0x1a/0x20
[ 4497.508926]  ? free_irq+0x295/0x340
[ 4497.509146]  mlx5_irq_pool_free_irqs+0x48/0x90
[ 4497.509421]  mlx5_irq_table_free_irqs+0x38/0x50
[ 4497.509714]  mlx5_core_eq_free_irqs+0x27/0x40
[ 4497.509984]  shutdown+0x7b/0x100
[ 4497.510184]  pci_device_shutdown+0x30/0x60
[ 4497.510440]  device_shutdown+0x14d/0x240
[ 4497.510698]  kernel_power_off+0x30/0x70
[ 4497.510938]  process_one_work+0x1e6/0x3e0
[ 4497.511183]  worker_thread+0x49/0x3b0
[ 4497.511407]  ? __pfx_worker_thread+0x10/0x10
[ 4497.511679]  kthread+0xe0/0x110
[ 4497.511879]  ? __pfx_kthread+0x10/0x10
[ 4497.512114]  ret_from_fork+0x29/0x50
[ 4497.512342]  </TASK>

Fixes: 9c2d08010963 ("net/mlx5: Free irqs only on shutdown callback")
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Mathieu Tortuyaux <mtortuyaux@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/pci_irq.c b/drivers/net/ethernet/mellanox/mlx5/core/pci_irq.c
index 5e0f7d9..d136360 100644 (file)
--- a/drivers/net/ethernet/mellanox/mlx5/core/pci_irq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/pci_irq.c
@@ -123,18 +123,32 @@ out:
        return ret;
 }
 
-static void irq_release(struct mlx5_irq *irq)
+/* mlx5_system_free_irq - Free an IRQ
+ * @irq: IRQ to free
+ *
+ * Free the IRQ and other resources such as rmap from the system.
+ * BUT doesn't free or remove reference from mlx5.
+ * This function is very important for the shutdown flow, where we need to
+ * cleanup system resoruces but keep mlx5 objects alive,
+ * see mlx5_irq_table_free_irqs().
+ */
+static void mlx5_system_free_irq(struct mlx5_irq *irq)
 {
-       struct mlx5_irq_pool *pool = irq->pool;
-
-       xa_erase(&pool->irqs, irq->index);
        /* free_irq requires that affinity_hint and rmap will be cleared
         * before calling it. This is why there is asymmetry with set_rmap
         * which should be called after alloc_irq but before request_irq.
         */
        irq_update_affinity_hint(irq->irqn, NULL);
-       free_cpumask_var(irq->mask);
        free_irq(irq->irqn, &irq->nh);
+}
+
+static void irq_release(struct mlx5_irq *irq)
+{
+       struct mlx5_irq_pool *pool = irq->pool;
+
+       xa_erase(&pool->irqs, irq->index);
+       mlx5_system_free_irq(irq);
+       free_cpumask_var(irq->mask);
        kfree(irq);
 }
 
@@ -597,7 +611,7 @@ static void mlx5_irq_pool_free_irqs(struct mlx5_irq_pool *pool)
        unsigned long index;
 
        xa_for_each(&pool->irqs, index, irq)
-               free_irq(irq->irqn, &irq->nh);
+               mlx5_system_free_irq(irq);
 }
 
 static void mlx5_irq_pools_free_irqs(struct mlx5_irq_table *table)