On Thu, Jun 12, 2025 at 11:58:21AM +0530, Dev Jain wrote:
On 11/06/25 6:00 pm, Pedro Falcato wrote:
On Wed, Jun 11, 2025 at 05:40:11PM +0530, Dev Jain wrote:
If CONFIG_UPROBES is not set, a merge subtest fails:
Failure log:
7151 12:46:54.627936 # # # RUN merge.handle_uprobe_upon_merged_vma ... 7152 12:46:54.639014 # # f /sys/bus/event_source/devices/uprobe/type 7153 12:46:54.639306 # # fopen: No such file or directory 7154 12:46:54.650451 # # # merge.c:473:handle_uprobe_upon_merged_vma:Expected read_sysfs("/sys/bus/event_source/devices/uprobe/type", &type) (1) == 0 (0) 7155 12:46:54.650730 # # # handle_uprobe_upon_merged_vma: Test terminated by assertion 7156 12:46:54.661750 # # # FAIL merge.handle_uprobe_upon_merged_vma 7157 12:46:54.662030 # # not ok 8 merge.handle_uprobe_upon_merged_vma
CONFIG_UPROBES is enabled by CONFIG_UPROBE_EVENTS, which gets enabled by CONFIG_FTRACE. Therefore add this config to selftests/mm/config so that CI systems can include this config in the kernel build.
While I did consider this before sending my patch, not every architecture supports uprobes :/ So this isn't a complete fix.
Also, does every architecture support ftrace? If not (and it does not seem so, e.g there's an ARM variant that does not), I would guess this config fragment blows up.
Sorry forgot to reply.
What do you mean by "config fragment blows up"? I think scripts will just pull these configs and build the kernel, and if the config is not supported, it will be thrown out of .config, and after applying your patch, the test will be skipped, so everyone is happy I guess?
Right, I spoke with Mark Brown off-list and apparently it should be safe to add unsupported config options there (and I wasn't sure about that, maybe it could fail, or maybe kbuild would fail, or<...>). So yeah, I'm fully supportive of a patch adding CONFIG_UPROBES.
- Pedro