Hi Lukas,
This issue is happening with multiple times device removal and
rescan from sysfs. Card is not removed physically.
Is_added bit is set after device attach which probe nvme driver.
NVMe driver starts one workqueue and that one is calling pci_set_master()
to set is_busmaster bit.
With multiple times device removal and rescan from sysfs, race
condition is observed and is_added bit is over-written to 0 from workqueue
started by NVMe driver.
Hope it clarifies concern.
Sequence 1:
pci_bus_add_device()
{
device_attach();
...
dev->is_added = 1;
}
Sequence 2:
nvme_probe()
{
...
INIT_WORK(&dev->ctrl.reset_work, nvme_reset_work);
...
}
nvme_reset_work()--->nvme_pci_enable()-->pci_set_master()-->__pci_set_mast
er(true)-->dev->is_busmaster = enable
Regards,
Hari
-----Original Message-----
From: Lukas Wunner [mailto:lukas@xxxxxxxxx]
Sent: Monday, June 25, 2018 4:08 PM
To: Hari Vyas <hari.vyas@xxxxxxxxxxxx>
Cc: bhelgaas@xxxxxxxxxx; linux-pci@xxxxxxxxxxxxxxx; ray.jui@xxxxxxxxxxxx
Subject: Re: [PATCH] PCI: Data corruption happening due to race condition
On Mon, Jun 25, 2018 at 03:40:46PM +0530, Hari Vyas wrote:
> When a pci device is detected, a variable is_added is set to
> 1 in pci device structure and proc, sys entries are created.
>
> When a pci device is removed, first is_added is checked for one and
> then device is detached with clearing of proc and sys entries and at
> end, is_added is set to 0.
>
> is_added and is_busmaster are bit fields in pci_dev structure sharing
> same memory location.
>
> A strange issue was observed with multiple times removal and rescan of
> a pcie nvme device using sysfs commands where is_added flag was
> observed as zero instead of one while removing device and proc,sys
> entries are not cleared.
Where exactly was is_added incorrectly observed as 0? Normally addition
and removal of devices are serialized using pci_lock_rescan_remove(),
maybe this is missing somewhere?
Thanks,
Lukas
