pci: pcie-brcmstb: do not rely on CLKREQ# signal
Subsystems (2)
PATCH 1/1 pci: pcie-brcmstb: do not rely on CLKREQ# signal
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/[email protected]/
--- drivers/pci/pcie_brcmstb.c | 18 +++++++++++++----- 1 file changed, 13 insertions(+), 5 deletions(-)
drivers/pci/pcie_brcmstb.c +13 -5
@@ -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 ; }
Signature · 1 line
2.41.0
Quoted 12 lines · Sam Edwards
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).
Quoted 7 lines
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/[email protected]/
Signed-off-by Sam Edwards <[email protected]>
Applied to u-boot/next, thanks!
Signature · 1 line
Tom