Hi Jerry, On Mon, 2024-03-18 at 13:35 -0700, Jerry Snitselaar wrote: > On Mon, Mar 18, 2024 at 01:18:58AM -0700, Jerry Snitselaar wrote: > > With adding the support for loading external drivers like iaa, > > autoloading, and default configs, systems with IAA that are booted > > in > > legacy mode get a number of probe failing messages from the user > > driver for the iax wqs before it probes with the iaa_crypto > > driver. Should the name match check occur prior to checking if user > > pasid is enabled in idxd_user_drv_probe? On a GNR system this will > > generate over 100 log messages at boot like the following: > > > > [ 56.885504] user: probe of wq15.0 failed with error -95 > > > > Regards, > > Jerry > > > > Hi Tom, > > A couple more iaa questions I had: > > - Are you supposed to disable all iax workqueues/devices to > reconfigure a workqueue? It seems perfectly happy to let you > disable, reconfigure, and enable just one. I know for idxd in > general the intent is to be able to disable, configure, and enable > workqueues/devices as needed for different users. I'm wondering if > that is the case for iaa as well since it talks about unloading and > loading iaa_crypto for new configurations. > In general the idea is that you set up your workqueues/devices, which registers the iaa-crypto algorithm and makes it available as a plugin to e.g. zswap. The register happens on the probe of the first wq, subsequent wqs are added after that and rebalance the wq table, so yeah, you can also reconfigure wqs in the same way. But you can't remove and reconfigure everything and re-register the algorithm, see below. > > - Is there a reason that iaa_crypto needs to be reloaded beyond the > compression algorithm registration? I tried moving the unregister > into iaa_crypto_remove with a check that the iaa_devices list is > empty, and it seemed to work, but I wasn't sure if there some other > reason for it being in iaa_crypto_cleanup_module instead of > iaa_crypto_remove similar to the register call in iaa_crypto_probe. > The requirement to only allow the algorithm to be unregistered on module unload came from the crypto maintainer during review [1]. Specifically, this part: 1) Never unregister your crypto algorithms, even after the last piece of hardware has been unplugged. The algorithms should only be unregistered (if they have been registered through the first successful probe call) in the module unload function. hth, Tom [1] https://lore.kernel.org/lkml/ZC58JggIXgpJ1tpD@xxxxxxxxxxxxxxxxxxx/ > Regards, > Jerry >
