When the Broadcom STB PCIe controller is initialized, it must be set into one of three CLKREQ# modes: "none"/"aspm"/"l1ss". The Linux driver, through today, hard-codes "aspm" since the vast majority of boards using this driver have a fixed PCIe bus with the CLKREQ# signal wired up.
The Raspberry Pi CM4, however, can be connected to a plethora of PCIe devices, some of which do not connect the CLKREQ# line (they just leave it floating). So "aspm" mode is no longer appropriate in all cases. In Linux, there is a proposed patchset [1] to determine the proper mode. This doesn't really make sense in U-Boot's case, so we just change the assumption from "aspm" to "none" (which is always safe). This patch DOES resolve a real-world crash that occurs when U-Boot is running on a Raspberry Pi CM4 installed in slot 3 of a Turing Pi 2 cluster board. [1]: https://lore.kernel.org/all/20230428223500.23337-1-jim2101...@gmail.com/ Signed-off-by: Sam Edwards <cfswo...@gmail.com> --- drivers/pci/pcie_brcmstb.c | 18 +++++++++++++----- 1 file changed, 13 insertions(+), 5 deletions(-) diff --git a/drivers/pci/pcie_brcmstb.c b/drivers/pci/pcie_brcmstb.c index 1de2802113..b8105654c5 100644 --- a/drivers/pci/pcie_brcmstb.c +++ b/drivers/pci/pcie_brcmstb.c @@ -33,6 +33,9 @@ #define PCIE_RC_CFG_PRIV1_ID_VAL3 0x043c #define CFG_PRIV1_ID_VAL3_CLASS_CODE_MASK 0xffffff +#define PCIE_RC_CFG_PRIV1_LINK_CAPABILITY 0x04dc +#define PCIE_RC_CFG_PRIV1_LINK_CAPABILITY_ASPM_SUPPORT_MASK 0xc00 + #define PCIE_RC_DL_MDIO_ADDR 0x1100 #define PCIE_RC_DL_MDIO_WR_DATA 0x1104 #define PCIE_RC_DL_MDIO_RD_DATA 0x1108 @@ -88,7 +91,6 @@ PCIE_MISC_CPU_2_PCIE_MEM_WIN0_LIMIT_HI + ((win) * 8) #define PCIE_MISC_HARD_PCIE_HARD_DEBUG 0x4204 -#define PCIE_HARD_DEBUG_CLKREQ_DEBUG_ENABLE_MASK 0x2 #define PCIE_HARD_DEBUG_SERDES_IDDQ_MASK 0x08000000 #define PCIE_MSI_INTR2_CLR 0x4508 @@ -562,12 +564,18 @@ static int brcm_pcie_probe(struct udevice *dev) clrsetbits_le32(base + PCIE_RC_CFG_VENDOR_SPECIFIC_REG1, VENDOR_SPECIFIC_REG1_ENDIAN_MODE_BAR2_MASK, VENDOR_SPECIFIC_REG1_LITTLE_ENDIAN); + /* - * Refclk from RC should be gated with CLKREQ# input when ASPM L0s,L1 - * is enabled => setting the CLKREQ_DEBUG_ENABLE field to 1. + * We used to enable the CLKREQ# input here, but a few PCIe cards don't + * attach anything to the CLKREQ# line, so we shouldn't assume that + * it's connected and working. The controller does allow detecting + * whether the port on the other side of our link is/was driving this + * signal, so we could check before we assume. But because this signal + * is for power management, which doesn't make sense in a bootloader, + * let's instead just unadvertise ASPM support. */ - setbits_le32(base + PCIE_MISC_HARD_PCIE_HARD_DEBUG, - PCIE_HARD_DEBUG_CLKREQ_DEBUG_ENABLE_MASK); + clrbits_le32(base + PCIE_RC_CFG_PRIV1_LINK_CAPABILITY, + PCIE_RC_CFG_PRIV1_LINK_CAPABILITY_ASPM_SUPPORT_MASK); return 0; } -- 2.41.0
