From 12404e04c8d2b3ac925e6967043f3e9fe750fa83 Mon Sep 17 00:00:00 2001 From: Angel Pons Date: Tue, 4 Aug 2020 00:26:45 +0200 Subject: sb/intel/lynxpoint: Consider root ports being disabled by strap PCIe RPC (Root Port Configuration) straps will force-disable some root port functions if some root ports have a width greater than x1. In two cases, this affects the last function. The PCIe init code will never finish configuring the root ports if that is the case: it assumes that the last function will eventually run through the code, but it doesn't. If PCIe initialization does not complete, pressing the power button will not power off the board, unless it is held for about five seconds. Also, Windows 10 will show a BSOD about MACHINE CHECK EXCEPTION, and lock up instead of rebooting. Depending on the microcode version, the BSOD may not be visible. This happens even when the root port is not populated. Use the strap fuse configuration value to know which configuration the PCH is strapped to. If needed, update the number of ports accordingly. In addition, print the updated value to ease debugging PCIe init code. Existing code in coreboot disagrees with public documentation about the root port width straps. Assume existing code is correct and document these assumptions in a table, as an explanation for the added code. Tested on Asrock B85M Pro4, PCIe initialization completes successfully. Change-Id: Id6da3a1f45467f00002a5ed41df8650f4a74eeba Signed-off-by: Angel Pons Reviewed-on: https://review.coreboot.org/c/coreboot/+/44155 Reviewed-by: Arthur Heymans Tested-by: build bot (Jenkins) --- src/southbridge/intel/lynxpoint/pcie.c | 36 ++++++++++++++++++++++++++++++++++ 1 file changed, 36 insertions(+) (limited to 'src/southbridge') diff --git a/src/southbridge/intel/lynxpoint/pcie.c b/src/southbridge/intel/lynxpoint/pcie.c index d2950e7916..4e94f121c0 100644 --- a/src/southbridge/intel/lynxpoint/pcie.c +++ b/src/southbridge/intel/lynxpoint/pcie.c @@ -100,6 +100,38 @@ static void root_port_config_update_gbe_port(void) } } +static void update_num_ports(void) +{ + /* + * According to existing code in 'root_port_check_disable()', which does + * not agree with the confusing information on the datasheets, the last + * visible function depends on the strapped root port width as follows: + * + * +-----+----+----+----+----+ + * | RPC | #5 | #6 | #7 | #8 | + * +-----+----+----+----+----+ + * | 0 | x1 | x1 | x1 | x1 | + * | 1 | x2 | | x1 | x1 | + * | 2 | x2 | | x2 | | + * | 3 | x4 | | | | + * +-----+----+----+----+----+ + */ + switch ((rpc.strpfusecfg2 >> 14) & 0x3) { + case 0: + case 1: + break; + case 2: + rpc.num_ports = MIN(rpc.num_ports, 7); + break; + case 3: + rpc.num_ports = MIN(rpc.num_ports, 5); + break; + } + + printk(BIOS_DEBUG, "Adjusted number of PCIe root ports to %d as per strpfusecfg2\n", + rpc.num_ports); +} + static void root_port_init_config(struct device *dev) { int rp; @@ -137,6 +169,10 @@ static void root_port_init_config(struct device *dev) case 5: rpc.strpfusecfg2 = pci_read_config32(dev, 0xfc); rpc.b0d28f4_32c = pci_read_config32(dev, 0x32c); + + if (!pch_is_lp()) + update_num_ports(); + break; case 6: rpc.b0d28f5_32c = pci_read_config32(dev, 0x32c); -- cgit v1.2.3