On Tue, Sep 08, 2020 at 02:37:22PM -0400, Mark Salter wrote: > On Mon, 2020-09-07 at 14:50 +0100, Will Deacon wrote: > > On Wed, Sep 02, 2020 at 02:27:29PM -0400, Mark Salter wrote: > > > diff --git a/drivers/perf/xgene_pmu.c b/drivers/perf/xgene_pmu.c > > > index edac28cd25dd..fdbbd0804b92 100644 > > > --- a/drivers/perf/xgene_pmu.c > > > +++ b/drivers/perf/xgene_pmu.c > > > @@ -1483,6 +1483,7 @@ xgene_pmu_dev_ctx *acpi_get_pmu_hw_inf(struct > > > xgene_pmu *xgene_pmu, > > > return NULL; > > > > > > INIT_LIST_HEAD(&resource_list); > > > + memset(&res, 0, sizeof(res)); > > > rc = acpi_dev_get_resources(adev, &resource_list, > > > acpi_pmu_dev_add_resource, &res); > > > acpi_dev_free_resource_list(&resource_list); > > > > Hmm, to be honest, I'm not sure we should be calling devm_ioremap_resource() > > at all here. The resource is clearly bogus, even with this change: the name > > and the resource hierarchy pointers will all be NULL. I think it would be > > better to follow the TX2 PMU driver (drivers/perf/thunderx2_pmu.c) which > > appears to assign the resource directly in tx2_uncore_pmu_init_dev(). > > > > Is there a reason we can't do that? > > > > Will > > > > There's no difference between xgene and tx2 wrt resouce name/hierarchy. > They both call devm_ioresource_remap() which ends up setting the name > and hierarchy. The difference is that xgene calls acpi_dev_resource_memory() > directly from the acpi_dev_get_resources() callback. TX2 doesn't use such a > callback and in that case, acpi_dev_resource_memory() gets called with a > zeroed struct. > > That said, changing xgene to avoid the callback like TX2 does would fix the > problem as well. If you'd rather go that route, I can send a patch for it.
Yes please Will