On Sat, Apr 13, 2019 at 12:39:40AM +0800, Liu, Changcheng wrote: > Hi Leon, > Reply to you interleaved. > > For your first worry about madness looping to query device, what's > about changing pr_err to pr_debug with using dynamic debug feature? You can get similar functionality to pr_debug by using tracing infrastructure. > > Thanks > Changcheng > > On 18:51 Fri 12 Apr, Leon Romanovsky wrote: > > On Fri, Apr 12, 2019 at 10:41:54PM +0800, Liu, Changcheng wrote: > > > Hi Jason, > > > > Please don't do top-posting. > > > > > 1. When userspace get "-EACCES" value, it shouldn't continue query > > > within limited times. There's not much log. > > > > What will prevent from me to call such bad application in the loop? > No good method. Currently, ceph will abort application in this case. > > > > > > > > 2. It's helpful to know the permission problem through kernel logs. > > > > It is not kernel problem, but application. Kernel logs is not right > > place for it. > rdma_cm libary only report "query device failed and returned status > code: 13". dmesg doesn't show why it's permissioned denied after long > time. > > > > > > > > 3. I hit one userspace app crash and checked several aspects without > > > any finding until dynamic trace the kernel to know that why > > > "permission" is denied in ib_uverbs_write function. > > > > It is not different from many other errors returned from the kernel. > When hitting the "permission" problem, I swith to use root user to run > the app. But still hit the problem without any info. > > > > Thanks > >
Attachment:
signature.asc
Description: PGP signature
