Re: [PATCH v1 1/6] PCI: brcmstb: Refactor max speed limit functionality
From: Bjorn Helgaas <helgaas@kernel.org>
Date: 2025-02-06 17:04:19
Also in:
linux-pci, lkml
On Wed, Feb 05, 2025 at 02:12:01PM -0500, Jim Quinlan wrote:
Make changes to the code that limits the PCIe max speed.
(1) Do the changes before link-up, not after. We do not want
to temporarily rise to a higher speed than desired.This is a functional change that should be split into its own patch. That will also make it obvious that this is not simple refactoring as the subject line advertises.
(2) Use constants from pci_reg.h when possible (3) Use uXX_replace_bits(...) for setting a register field.
(4) Use the internal link capabilities register for writing
the max speed, not the official config space register
where the speed field is RO. Updating this field is
not necessary to limit the speed so this mistake was
harmless.Also a bug fix (though harmless in this case) that deserves to be split out so the distinction between the internal and the architected paths to the register is highlighted and may help prevent the same mistake in the future.
quoted hunk ↗ jump to hunk
Signed-off-by: Jim Quinlan <redacted> --- drivers/pci/controller/pcie-brcmstb.c | 16 +++++++++------- 1 file changed, 9 insertions(+), 7 deletions(-)diff --git a/drivers/pci/controller/pcie-brcmstb.c b/drivers/pci/controller/pcie-brcmstb.c index 546056f7f0d3..f8fc3d620ee2 100644 --- a/drivers/pci/controller/pcie-brcmstb.c +++ b/drivers/pci/controller/pcie-brcmstb.c@@ -47,6 +47,7 @@ #define PCIE_RC_CFG_PRIV1_LINK_CAPABILITY 0x04dc #define PCIE_RC_CFG_PRIV1_LINK_CAPABILITY_ASPM_SUPPORT_MASK 0xc00 +#define PCIE_RC_CFG_PRIV1_LINK_CAPABILITY_MAX_LINK_SPEED_MASK 0xf
If the format of this internal register is different, of course we need a new #define for this. But if this is just a different path to LNKCAP, and both paths read the same bits in the same format, I don't see the point of a new #define.
quoted hunk ↗ jump to hunk
#define PCIE_RC_CFG_PRIV1_ROOT_CAP 0x4f8 #define PCIE_RC_CFG_PRIV1_ROOT_CAP_L1SS_MODE_MASK 0xf8@@ -413,12 +414,12 @@ static int brcm_pcie_set_ssc(struct brcm_pcie *pcie) static void brcm_pcie_set_gen(struct brcm_pcie *pcie, int gen) { u16 lnkctl2 = readw(pcie->base + BRCM_PCIE_CAP_REGS + PCI_EXP_LNKCTL2); - u32 lnkcap = readl(pcie->base + BRCM_PCIE_CAP_REGS + PCI_EXP_LNKCAP); + u32 lnkcap = readl(pcie->base + PCIE_RC_CFG_PRIV1_LINK_CAPABILITY); - lnkcap = (lnkcap & ~PCI_EXP_LNKCAP_SLS) | gen; - writel(lnkcap, pcie->base + BRCM_PCIE_CAP_REGS + PCI_EXP_LNKCAP); + u32p_replace_bits(&lnkcap, gen, PCIE_RC_CFG_PRIV1_LINK_CAPABILITY_MAX_LINK_SPEED_MASK); + writel(lnkcap, pcie->base + PCIE_RC_CFG_PRIV1_LINK_CAPABILITY); - lnkctl2 = (lnkctl2 & ~0xf) | gen; + u16p_replace_bits(&lnkctl2, gen, PCI_EXP_LNKCTL2_TLS);
OK. I am not really a fan of the uXX_replace_bits() thing because it's not widely used (I found 341 instances tree-wide, compared to 14000+ for FIELD_PREP()), grep can't find the definition easily, and stylistically it doesn't fit with GENMASK(), FIELD_PREP(), etc. But it's already used throughout brcmstb.c, so we should be consistent.