| Message ID | 20230504113509.184633-3-angelogioacchino.delregno@collabora.com |
|---|---|
| State | New |
| Headers | show |
| Series | MediaTek PCIe Gen3: Suspend fixes | expand |
On Thu, May 04, 2023 at 01:35:09PM +0200, AngeloGioacchino Del Regno wrote: > Some SoCs have two PCI-Express controllers: in the case of MT8195, > one of them is using a dedicated PHY, but the other uses a combo PHY > that is shared with USB and in that case the PHY cannot be reset > from the PCIe driver, or USB functionality will be unable to resume. > > Resetting the PCIe MAC without also resetting the PHY will result in > a full system lockup at PCIe resume time and the only option to > resume operation is to hard reboot the system (with a PMIC cut-off). > > To resolve this issue, check if we've got both a PHY and a MAC reset > and, if not, never assert resets at PM suspend time: in that case, > the link is still getting powered down as both the clocks and the > power domains will go down anyway. > > Fixes: d537dc125f07 ("PCI: mediatek-gen3: Add system PM support") > Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com> Hi Angelo, It seems this patch was forgotten but it's still very much needed. As you describe above, the Tomato Chromebook (MT8195-based) is currently unable to resume from suspend due to this issue. Upon resume, the following error is printed, and the system hangs: [ 67.018281] mtk-pcie-gen3 112f8000.pcie: PCIe link down, current LTSSM state: detect.quiet (0x0) [ 67.027162] mtk-pcie-gen3 112f8000.pcie: PM: dpm_run_callback(): genpd_resume_noirq+0x0/0x24 returns -110 [ 67.036791] mtk-pcie-gen3 112f8000.pcie: PM: failed to resume noirq: error -110 And further investigation showed that all PCIe registers return 0x0 when read in this situation. Commenting out the MAC reset in the PCIe DT node fixes the issue: the PCIe registers can be read correctly upon resume and resume proceeds succesfully. Your patch here essentially does the same as not providing the MAC reset, with the benefit of us still being able to describe the reset in DT and thus having a more complete HW description. But this patch no longer applies, so please rebase it so we can get working suspend/resume on MT8195-Tomato :). Thanks, Nícolas
diff --git a/drivers/pci/controller/pcie-mediatek-gen3.c b/drivers/pci/controller/pcie-mediatek-gen3.c index 52f52ca5db71..480621ca1450 100644 --- a/drivers/pci/controller/pcie-mediatek-gen3.c +++ b/drivers/pci/controller/pcie-mediatek-gen3.c @@ -859,17 +859,26 @@ static int mtk_pcie_power_up(struct mtk_gen3_pcie *pcie) return err; } -static void mtk_pcie_power_down(struct mtk_gen3_pcie *pcie) +static void mtk_pcie_power_down(struct mtk_gen3_pcie *pcie, bool is_suspend) { + bool suspend_reset_supported = pcie->mac_reset && pcie->phy_reset; + clk_bulk_disable_unprepare(pcie->num_clks, pcie->clks); pm_runtime_put_sync(pcie->dev); pm_runtime_disable(pcie->dev); - reset_control_assert(pcie->mac_reset); + + /* + * Assert MAC reset only if we also got a PHY reset, otherwise + * the system will lockup at PM resume time. + */ + if (is_suspend && suspend_reset_supported) + reset_control_assert(pcie->mac_reset); phy_power_off(pcie->phy); phy_exit(pcie->phy); - reset_control_assert(pcie->phy_reset); + if (is_suspend && suspend_reset_supported) + reset_control_assert(pcie->phy_reset); } static int mtk_pcie_setup(struct mtk_gen3_pcie *pcie) @@ -905,7 +914,7 @@ static int mtk_pcie_setup(struct mtk_gen3_pcie *pcie) return 0; err_setup: - mtk_pcie_power_down(pcie); + mtk_pcie_power_down(pcie, false); return err; } @@ -936,7 +945,7 @@ static int mtk_pcie_probe(struct platform_device *pdev) err = pci_host_probe(host); if (err) { mtk_pcie_irq_teardown(pcie); - mtk_pcie_power_down(pcie); + mtk_pcie_power_down(pcie, false); return err; } @@ -954,7 +963,7 @@ static int mtk_pcie_remove(struct platform_device *pdev) pci_unlock_rescan_remove(); mtk_pcie_irq_teardown(pcie); - mtk_pcie_power_down(pcie); + mtk_pcie_power_down(pcie, false); return 0; } @@ -1023,7 +1032,7 @@ static int mtk_pcie_suspend_noirq(struct device *dev) dev_dbg(pcie->dev, "entered L2 states successfully"); mtk_pcie_irq_save(pcie); - mtk_pcie_power_down(pcie); + mtk_pcie_power_down(pcie, true); return 0; } @@ -1039,7 +1048,7 @@ static int mtk_pcie_resume_noirq(struct device *dev) err = mtk_pcie_startup_port(pcie); if (err) { - mtk_pcie_power_down(pcie); + mtk_pcie_power_down(pcie, false); return err; }
Some SoCs have two PCI-Express controllers: in the case of MT8195, one of them is using a dedicated PHY, but the other uses a combo PHY that is shared with USB and in that case the PHY cannot be reset from the PCIe driver, or USB functionality will be unable to resume. Resetting the PCIe MAC without also resetting the PHY will result in a full system lockup at PCIe resume time and the only option to resume operation is to hard reboot the system (with a PMIC cut-off). To resolve this issue, check if we've got both a PHY and a MAC reset and, if not, never assert resets at PM suspend time: in that case, the link is still getting powered down as both the clocks and the power domains will go down anyway. Fixes: d537dc125f07 ("PCI: mediatek-gen3: Add system PM support") Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com> --- drivers/pci/controller/pcie-mediatek-gen3.c | 25 ++++++++++++++------- 1 file changed, 17 insertions(+), 8 deletions(-)