On 06/12/2015 04:24 PM, Carol Soto wrote: > > > On 6/12/2015 3:08 PM, Doug Ledford wrote: >> On 06/11/2015 12:06 PM, clsoto@xxxxxxxxxxxxxxxxxx wrote: >>> From: Carol L Soto <clsoto@xxxxxxxxxxxxxxxxxx> >>> >>> ib_ucm_release_dev clears wrong bit if devnum is greater than >>> IB_UCM_MAX_DEVICES. >>> >>> Signed-off-by: Carol L Soto <clsoto@xxxxxxxxxxxxxxxxxx> >>> --- >>> drivers/infiniband/core/ucm.c | 4 ++-- >>> 1 file changed, 2 insertions(+), 2 deletions(-) >>> >>> diff --git a/drivers/infiniband/core/ucm.c >>> b/drivers/infiniband/core/ucm.c >>> index f2f6393..e2fd085 100644 >>> --- a/drivers/infiniband/core/ucm.c >>> +++ b/drivers/infiniband/core/ucm.c >>> @@ -1193,6 +1193,7 @@ static int ib_ucm_close(struct inode *inode, >>> struct file *filp) >>> return 0; >>> } >>> +static DECLARE_BITMAP(overflow_map, IB_UCM_MAX_DEVICES); >>> static void ib_ucm_release_dev(struct device *dev) >>> { >>> struct ib_ucm_device *ucm_dev; >>> @@ -1202,7 +1203,7 @@ static void ib_ucm_release_dev(struct device *dev) >>> if (ucm_dev->devnum < IB_UCM_MAX_DEVICES) >>> clear_bit(ucm_dev->devnum, dev_map); >>> else >>> - clear_bit(ucm_dev->devnum - IB_UCM_MAX_DEVICES, dev_map); >>> + clear_bit(ucm_dev->devnum - IB_UCM_MAX_DEVICES, overflow_map); >>> kfree(ucm_dev); >>> } >>> @@ -1226,7 +1227,6 @@ static ssize_t show_ibdev(struct device *dev, >>> struct device_attribute *attr, >>> static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL); >>> static dev_t overflow_maj; >>> -static DECLARE_BITMAP(overflow_map, IB_UCM_MAX_DEVICES); >>> static int find_overflow_devnum(void) >>> { >>> int ret; >>> >> This doesn't look right to me. In particular, you are creating a bitmap >> and clearing bits, but I never see that bitmap get any bits set. So, >> are you overflowing on the set routine and you just didn't bother to fix >> that? > I just moved the declaration of the bitmap before ib_ucm_release_dev to > be able to compile. This bitmap is used in ib_ucm_add_one > in the case of devnum > IB_UCM_MAX_DEVICES. The error path in > ib_ucm_add_one did the clear bit correctly but not the > ib_ucm_release_dev function. We saw an issue when using SRIOV. The case > was that user has 2 Mellanox SRIOV cards and just did an > unbind/bind a VF that was using a ib ucm device with devnum greater than > 32. We saw this warning Got it. For some reason I saw the add of the bitmap but missed the removal of the original.
Attachment:
signature.asc
Description: OpenPGP digital signature