Hi, I need some advice on how to manage the development of our viapcib (lacks a man page) and viaenv (https://man.netbsd.org/viaenv.4) drivers.
These two drivers are somewhat related, as they both pertain to the bus control and power management device controller (with SMBus support), but they support different sets of chipsets. The viaenv(4) driver supports two older chipsets, the VT82C686A and VT8231. It maps the power management I/O space, which, as far as I currently understand, is used only to enable the device timecounter (though I see potential to hook in sysctl options). The VT8231 also features hardware monitoring, which is integrated through envstat. This hardware monitor appears to be specific to this chipset only; newer or older ones don't seem to have it, according to the datasheets. Since the VT8231 was used in the PowerPC-based Pegasos II, the driver is located in the architecture-independent dev/pci path. viapcib(4) is currently configured for the VT8235, VT8237(R), and VT8237A. It maps the SMBus host I/O space and allows its control over the iic framework. This driver is currently placed under the i386 architecture. Since it's supported across all newer VIA chipsets, my plan was to move it to x86 and add all of them to its identification. In reality, viaenv(4) chipsets also support SMBus under the same device. At the same time, newer chipsets do support the power management controller (with timecounter enabling control). The difference is that older chipsets would have a separate device function, while newer ones merged into function 0, the same as the PCI-ISA bus. Additionally, this device may have more features depending on the chipset (e.g., GPIO control), which could be controlled and potentially added in the future (though these devices are quite obsolete now, so it may never happen). Given all the above, I'd like to understand the best way forward: 1) Keep everything as is, just move viapcib(4), add a man page, and perhaps copy the power management controller code to it (this doesn't seem very relevant currently, as I believe the same timecounter is attached using ACPI). Advantage of this approach that there's no need to think how to attach device, same base address can be used between all devices (no conditions for older chipsets). Minimal work is required. viaenv(4) stays as is. 2) Move smbus code to viaenv(4) and attach to PCI or (via)pcib depending on the chipset. The problem is the different base addresses between old and "new" chipsets, conditionals or flags are needed, but otherwise, viaenv code works on newer chipsets with minimal adjustments. Also problem, that I don't have hardware for older chipsets, can't test them. The advantage are smbus support for all, code in one place. And it can be made as module I believe. 3) Separate drivers even more by function into "pseudo" devices such as viapm, viasmb, viahwm, and viagpio, and attach them all somehow to the same pci/viapcib? Handle them as required by the chipset. Advantages I would see that user can control more what he needs, also can be added as modules, driver are smaller and focused on "one thing". 4) Please suggest... Please see some data from my experiments below. Timecounters if viaenv is attached: kern.timecounter.choice = TSC(q=-100, f=1600059000 Hz) clockinterrupt(q=0, f=100 Hz) lapic(q=-100, f=200007000 Hz) viaenv0(q=1000, f=3579545 Hz) hpet0(q=2000, f=14318180 Hz) ACPI-Fast(q=1000, f=3579545 Hz) i8254(q=100, f=1193182 Hz) dummy(q=-1000000, f=1000000 Hz) (take notice that viaenv and ACPI-Fast timecounters seem to be the same ones, or at least they share same frequency and quality) viaenv from dmesg (adjusted for VX900 with a few changes (mainly adjusting base addresses) and disabled viapcib/pcib so it would attach to pci): viaenv0 at pci0 dev 17 function 0: VIA Technologies VX900 Power Management Controller timecounter: Timecounter "viaenv0" frequency 3579545 Hz quality 1000 viaenv0: 24-bit timer viapcib dmesg with adjustments (moving to x86, adding to amd64 kernel config, updating identification) and attaching spdmem (attaching spdmem slows down boot considerably, taking time to get memory info): ECS VX900-I viapcib0 at pci0 dev 17 function 0: VIA Technologies VX900 Bus Control and Power Management (rev. 0x00) viapcib0: SMBus found at 0x400 (revision 0x0) iic0 at viapcib0: I2C bus spdmem0 at iic0 addr 0x50: M0820031 spdmem0: DDR3 SDRAM, no ECC, no temp-sensor, 1GB, 1067MHz (PC3-8533) spdmem0: 13 rows, 10 cols, 8 log. banks, 2 phys. banks, 1.875ns cycle time spdmem0: tAA-tRCD-tRP-tRAS: 7-7-7-20 spdmem0: 1.5V operable spdmem1 at iic0 addr 0x51: M0820031 spdmem1: DDR3 SDRAM, no ECC, no temp-sensor, 1GB, 1067MHz (PC3-8533) spdmem1: 13 rows, 10 cols, 8 log. banks, 2 phys. banks, 1.875ns cycle time spdmem1: tAA-tRCD-tRP-tRAS: 7-7-7-20 spdmem1: 1.5V operable Current code for reference: https://nxr.netbsd.org/xref/src/sys/dev/pci/viaenv.c nxr.netbsd.org/xref/src/sys/arch/i386/pci/viapcib.c Side question, iic devices were not available by default in /dev/. Manual execution of MAKEDEV iic was required. Is it expected? Regards, Andrius V
