Re: [PATCH v5 -next 07/11] PCI: brcmstb: Adjust PHY PLL setup to use a 54MHz input refclk

From: Stanimir Varbanov
Date: Mon Feb 03 2025 - 06:27:35 EST


Hi Jim,

On 1/31/25 6:08 PM, Jim Quinlan wrote:
> On Mon, Jan 20, 2025 at 8:01 AM Stanimir Varbanov <svarbanov@xxxxxxx> wrote:
>>
>> The default input reference clock for the PHY PLL is 100Mhz, except for
>> some devices where it is 54Mhz like bcm2712C1 and bcm2712D0.
>>
>> To implement this adjustments introduce a new .post_setup op in
>> pcie_cfg_data and call it at the end of brcm_pcie_setup function.
>>
>> The bcm2712 .post_setup callback implements the required MDIO writes that
>> switch the PLL refclk and also change PHY PM clock period.
>>
>> Without this RPi5 PCIex1 is unable to enumerate endpoint devices on
>> the expansion connector.
>>
>> Signed-off-by: Stanimir Varbanov <svarbanov@xxxxxxx>
>> Reviewed-by: Florian Fainelli <florian.fainelli@xxxxxxxxxxxx>
>> ---
>> v4 -> v5:
>> - Updated a comment (Jim).
>>
>> drivers/pci/controller/pcie-brcmstb.c | 44 +++++++++++++++++++++++++++
>> 1 file changed, 44 insertions(+)
>>
>> diff --git a/drivers/pci/controller/pcie-brcmstb.c b/drivers/pci/controller/pcie-brcmstb.c
>> index 50607df34a66..03396a9d97be 100644
>> --- a/drivers/pci/controller/pcie-brcmstb.c
>> +++ b/drivers/pci/controller/pcie-brcmstb.c
>> @@ -55,6 +55,10 @@
>> #define PCIE_RC_DL_MDIO_WR_DATA 0x1104
>> #define PCIE_RC_DL_MDIO_RD_DATA 0x1108
>>
>> +#define PCIE_RC_PL_PHY_CTL_15 0x184c
>> +#define PCIE_RC_PL_PHY_CTL_15_DIS_PLL_PD_MASK 0x400000
>> +#define PCIE_RC_PL_PHY_CTL_15_PM_CLK_PERIOD_MASK 0xff
>> +
>> #define PCIE_MISC_MISC_CTRL 0x4008
>> #define PCIE_MISC_MISC_CTRL_PCIE_RCB_64B_MODE_MASK 0x80
>> #define PCIE_MISC_MISC_CTRL_PCIE_RCB_MPS_MODE_MASK 0x400
>> @@ -251,6 +255,7 @@ struct pcie_cfg_data {
>> u8 num_inbound_wins;
>> int (*perst_set)(struct brcm_pcie *pcie, u32 val);
>> int (*bridge_sw_init_set)(struct brcm_pcie *pcie, u32 val);
>> + int (*post_setup)(struct brcm_pcie *pcie);
>> };
>>
>> struct subdev_regulators {
>> @@ -826,6 +831,38 @@ static int brcm_pcie_perst_set_generic(struct brcm_pcie *pcie, u32 val)
>> return 0;
>> }
>>
>> +static int brcm_pcie_post_setup_bcm2712(struct brcm_pcie *pcie)
>> +{
>> + const u16 data[] = { 0x50b9, 0xbda1, 0x0094, 0x97b4, 0x5030, 0x5030, 0x0007 };
>> + const u8 regs[] = { 0x16, 0x17, 0x18, 0x19, 0x1b, 0x1c, 0x1e };
>> + int ret, i;
>> + u32 tmp;
>> +
>> + /* Allow a 54MHz (xosc) refclk source */
>> + ret = brcm_pcie_mdio_write(pcie->base, MDIO_PORT0, SET_ADDR_OFFSET, 0x1600);
>> + if (ret < 0)
>> + return ret;
>> +
>> + for (i = 0; i < ARRAY_SIZE(regs); i++) {
>> + ret = brcm_pcie_mdio_write(pcie->base, MDIO_PORT0, regs[i], data[i]);
>> + if (ret < 0)
>> + return ret;
>> + }
>> +
>> + usleep_range(100, 200);
>> +
>> + /*
>> + * Set L1SS sub-state timers to avoid lengthy state transitions,
>> + * PM clock period is 18.52ns (1/54MHz, round down).
>> + */
>> + tmp = readl(pcie->base + PCIE_RC_PL_PHY_CTL_15);
>> + tmp &= ~PCIE_RC_PL_PHY_CTL_15_PM_CLK_PERIOD_MASK;
>> + tmp |= 0x12;
>> + writel(tmp, pcie->base + PCIE_RC_PL_PHY_CTL_15);
>> +
>> + return 0;
>> +}
>> +
>> static void add_inbound_win(struct inbound_win *b, u8 *count, u64 size,
>> u64 cpu_addr, u64 pci_offset)
>> {
>> @@ -1189,6 +1226,12 @@ static int brcm_pcie_setup(struct brcm_pcie *pcie)
>> PCIE_RC_CFG_VENDOR_VENDOR_SPECIFIC_REG1_ENDIAN_MODE_BAR2_MASK);
>> writel(tmp, base + PCIE_RC_CFG_VENDOR_VENDOR_SPECIFIC_REG1);
>>
>> + if (pcie->cfg->post_setup) {
>> + ret = pcie->cfg->post_setup(pcie);
>> + if (ret < 0)
>> + return ret;
>> + }
>> +
>> return 0;
>> }
>>
>> @@ -1715,6 +1758,7 @@ static const struct pcie_cfg_data bcm2712_cfg = {
>> .soc_base = BCM7712,
>> .perst_set = brcm_pcie_perst_set_7278,
>> .bridge_sw_init_set = brcm_pcie_bridge_sw_init_set_generic,
>> + .post_setup = brcm_pcie_post_setup_bcm2712,
>> .quirks = CFG_QUIRK_AVOID_BRIDGE_SHUTDOWN,
>> .num_inbound_wins = 10,
>> };
>> --
>> 2.47.0
>>
> Hi Stan,
> Any reason you didn't make this a quirk like the other commit?

Not a special reason. I think that the code is more readable with
separate ops function which is executed after the generic .setup
function. The other SoC/variants could also benefit of such op.

> Reviewed-by: Jim Quinlan <james.quinlan@xxxxxxxxxxxx>

Thanks!

~Stan