Re: [PATCH v2 4/4] clk: siflower: Add SF19A2890 peripheral clock and reset manager driver

From: Philipp Zabel

Date: Thu Nov 12 2026 - 05:46:55 EST


On Mi, 2026-11-11 at 14:08 -0500, Joseph Telaak wrote:
> On Wed, Nov 11, 2026 at 10:56:47AM +0100, Philipp Zabel wrote:
> > > + spin_lock_irqsave(&crm->lock, flags);
> >
> > Could be simplified with guard(spinlock)(&crm->lock);
>
> Will do in v3, as guard(spinlock_irqsave)(), since the clock gates take
> the same lock from clk_enable(), which can run in interrupt context. I'll
> convert the gate enable and disable paths as well.
>
> > > +static const u32 sf19a2890_usb_resets[] = {
> > > + [RST_USB] = GENMASK(2, 0),
> >
> > Do these three bits control separate reset lines?
> > Does the hardware require them to be changed together?
>
> Yes, they are separate lines, one per clock domain of the DWC2
> controller. Bit 0 matches the controller's gate and bit 2 its AXI
> master's (usb_axim). Bit 1 has no gate, so it is probably the PHY clock
> domain, whose clock comes from the PHY rather than from this block.
>
> There is no public documentation for this block. Chuanhong explained in
> v1 that the lines of one IP's clock domains have to be asserted together
> [1], the same as the GMAC's.

Ah, I got confused by the commit message, which to me looked like the
bus lockup issue is limited to GMAC.

> The dwc2 binding also takes a single reset for the controller.

That is irrelevant from my PoV, the binding could be insufficient for
this integration.

>
> > > +static const u32 sf19a2890_uart_resets[] = {
> > > + [RST_UART0] = BIT(0) | BIT(4),
> > > + [RST_UART1] = BIT(1) | BIT(5),
> > > + [RST_UART2] = BIT(2) | BIT(6),
> >
> > Same question as above.
>
> Yes, they are separate. Going by the gates at the same bit positions,
> bits 0-2 are the UARTs' APB resets (PRESETn) and bits 4-6 their UART
> clock domain resets (nUARTRST). The PL011 TRM (DDI 0183, section 2.4.1)
> says the reset controller must use PRESETn to assert nUARTRST. This block
> leaves both to software, so the driver asserts them together. The pl011
> binding also allows a single reset.
>
> The same goes for the other two groups. The SPI block's bits 4 and 5 are
> the PL022's PRESETn and nSSPRST, which its TRM (DDI 0194) ties together in
> the same way. The WLAN blocks' bits 0-2 match their AXI slave, AXI master
> and platform gates.
>
> v3 will explain each multi-bit group in the commit message, not just the
> GMAC's.

Please do. With that,

Reviewed-by: Philipp Zabel <p.zabel@xxxxxxxxxxxxxx>

regards
Philipp