On Thu, Nov 19, 2026 at 03:57:19PM +0000, Conor Dooley wrote: > On Thu, Nov 19, 2026 at 09:03:56PM +0800, Zhouqi Jiang wrote: > > Linux queries the SBI specification and implementation identity during > > boot. It caches the specification version, but discards the > > implementation ID and version after logging them. Userspace therefore > > has no structured interface for this immutable firmware metadata. > > > > Cache the implementation ID and version and expose the available values > > as read-only attributes under /sys/firmware/sbi: > > > > spec_version > > impl_id > > impl_version > > > > SBI v0.1 systems expose only spec_version because the implementation > > fields were introduced by the SBI Base extension in v0.2. > > > > This gives tools such as lssbi a native source for basic SBI identity > > information and starts moving those queries away from out-of-tree > > kernel modules. > > > > Codex was prompted to research lssbi's out-of-tree SBI probe and design > > a minimal in-kernel replacement. It assisted with the sysfs interface, > > implementation, ABI documentation, changelog, and QEMU test setup. The > > result was built with a RISC-V defconfig and boot-tested under QEMU > > virt with: > > > > OpenSBI v1.3 > > RustSBI v0.4.1 > > I like this idea. FWIW I tested it with the polarfire hss. Tested-by: Conor Dooley > > > Assisted-by: Codex:GPT-5 > > FYI, these tags are now deprecated. > > > +static ssize_t impl_id_show(struct kobject *kobj, > > + struct kobj_attribute *attr, char *buf) > > +{ > > + return sysfs_emit(buf, "0x%lx\n", sbi_impl_id); > > +} > > Should this actually be printed as a decimal? It's a minor thing, but > the numbers in the spec are defined that way (see 4.9 SBI Implementation > IDs in the 3.0 spec). > > > + > > +static ssize_t impl_version_show(struct kobject *kobj, > > + struct kobj_attribute *attr, char *buf) > > +{ > > + return sysfs_emit(buf, "0x%lx\n", sbi_impl_version); > > +}