On 11/6/2018 2:23 PM, Matan Azrad wrote:
Hi Jeff Can you detail more in the commit log that we can understand the synchronization problematic scenario. And how does this commit fix it?
Please check my reply in the 1/3 mail. And explain more here is that, when device be hot-unplugged in vfio, the req notifier will invoked, then user space could release device resource in user space side,
then vfio check that the device be released out from the device group, it will take the device control again and trigger the device kernel release processing, at the mean time it will sent remove uevent to
user space. Here although the req handler seems will always process before uevent handler, but even for fast path and slow path protection of device accessing when device is removing , it should also be need.
what do you think about that?
-----Original Message----- From: Jeff Guo <jia....@intel.com> Sent: Tuesday, November 6, 2018 8:07 AM To: konstantin.anan...@intel.com; anatoly.bura...@intel.com; Thomas Monjalon <tho...@monjalon.net>; bernard.iremon...@intel.com; jingjing...@intel.com; wenzhuo...@intel.com Cc: ferruh.yi...@intel.com; dev@dpdk.org; jia....@intel.com; helin.zh...@intel.com; Matan Azrad <ma...@mellanox.com>; shaopeng...@intel.com Subject: [PATCH 2/3] vfio: fix to add handler lock for hot-unplug This patch add hot-unplug handler lock and unlock in device request handler when process bus and device resource, in order to avoid the synchronization issue when device be hot-unplugged. Fixes: c115fd000c32 ("vfio: handle hotplug request notifier") Signed-off-by: Jeff Guo <jia....@intel.com> --- drivers/bus/pci/linux/pci_vfio.c | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-) diff --git a/drivers/bus/pci/linux/pci_vfio.c b/drivers/bus/pci/linux/pci_vfio.c index 305cc06..d2c8410 100644 --- a/drivers/bus/pci/linux/pci_vfio.c +++ b/drivers/bus/pci/linux/pci_vfio.c @@ -19,6 +19,7 @@ #include <rte_vfio.h> #include <rte_eal.h> #include <rte_bus.h> +#include <rte_spinlock.h> #include "eal_filesystem.h" @@ -35,6 +36,14 @@ * This file is only compiled if CONFIG_RTE_EAL_VFIO is set to "y". */ +/* + * spinlock for device hot-unplug failure handling. If it try to access +bus or + * device, such as handle sigbus on bus or handle memory failure for +device + * just need to use this lock. It could protect the bus and the device +to avoid + * race condition. + */ +static rte_spinlock_t failure_handle_lock = RTE_SPINLOCK_INITIALIZER; + #ifdef VFIO_PRESENT #ifndef PAGE_SIZE @@ -289,11 +298,12 @@ pci_vfio_req_handler(void *param) int ret; struct rte_device *device = (struct rte_device *)param; + rte_spinlock_lock(&failure_handle_lock); bus = rte_bus_find_by_device(device); if (bus == NULL) { RTE_LOG(ERR, EAL, "Cannot find bus for device (%s)\n", device->name); - return; + goto handle_end; } /* @@ -306,6 +316,8 @@ pci_vfio_req_handler(void *param) RTE_LOG(ERR, EAL, "Can not handle hot-unplug for device (%s)\n", device->name); +handle_end: + rte_spinlock_unlock(&failure_handle_lock); } /* enable notifier (only enable req now) */ -- 2.7.4
