Re: [PATCH] drm/hisilicon: Fix rmmod hibmc_drm failed

From: Thomas Zimmermann
Date: Tue Dec 15 2020 - 02:55:08 EST


Hi Tian

Am 15.12.20 um 04:01 schrieb Tian Tao:
drm_irq_uninstall should be called before pci_disable_msi, if use
devm_drm_irq_install to register the interrupt, the system will
call pci_disable_msi first and then call drm_irq_uninstall, which
will result in the following callstack.

kernel BUG at drivers/pci/msi.c:376!
Internal error: Oops - BUG: 0 [#1] SMP
CPU: 93 PID: 173814 Comm: rmmod Tainted:
pstate: a0400009 (NzCv daif +PAN -UAO -TCO BTYPE=--)
pc : free_msi_irqs+0x17c/0x1a0
lr : free_msi_irqs+0x16c/0x1a0
sp : ffff2028157f7bd0
x29: ffff2028157f7bd0 x28: ffff202849edab00
x27: 0000000000000000 x26: 0000000000000000
x25: 0000000000000000 x24: 0000000000000000
x23: ffff0020851da000 x22: ffff0020851da2d8
x21: ffff0020cc829000 x20: 0000000000000000
x19: ffff0020d6714800 x18: 0000000000000010
x17: 0000000000000000 x16: 0000000000000000
x15: ffffffffffffffff x14: ffff2028957f77df
x13: ffff2028157f77ed x12: 0000000000000000
x11: 0000000000000040 x10: ffff800011b2f8e0
x9 : ffff800011b2f8d8 x8 : ffff2020203fc458
x7 : 0000000000000000 x6 : 0000000000000000
x5 : ffff2020203fc430 x4 : ffff2020203fc4a0
x3 : 0000000000000000 x2 : 0000000000000000
x1 : 00000000000002c9 x0 : ffff0020d6719500
Call trace:
free_msi_irqs+0x17c/0x1a0
pci_disable_msi+0xe4/0x118
hibmc_unload+0x44/0x80 [hibmc_drm]
hibmc_pci_remove+0x2c/0x38 [hibmc_drm]
pci_device_remove+0x48/0x108
device_release_driver_internal+0x118/0x1f0
driver_detach+0x6c/0xe0
bus_remove_driver+0x74/0x100
driver_unregister+0x34/0x60
pci_unregister_driver+0x24/0xd8
hibmc_pci_driver_exit+0x14/0xe768 [hibmc_drm]
__arm64_sys_delete_module+0x1fc/0x2d0
el0_svc_common.constprop.3+0xa8/0x188
do_el0_svc+0x80/0xa0
el0_sync_handler+0x8c/0xb0
el0_sync+0x15c/0x180
Code: f940b400 b4ffff00 a903e7b8 f90013b5 (d4210000)
---[ end trace 310d94ee8abef44f ]---
Kernel panic - not syncing: Oops - BUG: Fatal exception

Signed-off-by: Tian Tao <tiantao6@xxxxxxxxxxxxx>
---
drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.c b/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.c
index e3ab765b..02f3bd1 100644
--- a/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.c
+++ b/drivers/gpu/drm/hisilicon/hibmc/hibmc_drm_drv.c
@@ -251,6 +251,10 @@ static int hibmc_hw_init(struct hibmc_drm_private *priv)
static int hibmc_unload(struct drm_device *dev)
{
drm_atomic_helper_shutdown(dev);
+
+ if (dev->irq_enabled)
+ drm_irq_uninstall(dev);
+
pci_disable_msi(dev->pdev);

We're trying to move towards managed driver releases; and this feels like a step backwards.

I looked through the PCI-device release code in pcim_release() and it already disables MSI automatically. [1] You can enable the PCI device with pcim_enable_device() instead of pci_enable_device() to use the automatic release. No more need to disable MSI manually.

If this does not work, could you provide a managed version of pci_disable_msi() that solves the problem?

Best regards
Thomas

[1] https://elixir.bootlin.com/linux/latest/source/drivers/pci/pci.c#L1976


return 0;
@@ -286,7 +290,7 @@ static int hibmc_load(struct drm_device *dev)
if (ret) {
drm_warn(dev, "enabling MSI failed: %d\n", ret);
} else {
- ret = devm_drm_irq_install(dev, dev->pdev->irq);
+ ret = drm_irq_install(dev, dev->pdev->irq);
if (ret)
drm_warn(dev, "install irq failed: %d\n", ret);
}


--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer

Attachment: OpenPGP_signature
Description: OpenPGP digital signature