[PATCH v3] drm/amdgpu: Save and restore switch state
Wang, Yang(Kevin)
KevinYang.Wang at amd.com
Fri Aug 8 00:20:47 UTC 2025
[AMD Official Use Only - AMD Internal Distribution Only]
> + do {
> + usleep_range(10000, 10500);
> + r = pci_read_config_word(link_dev, PCI_VENDOR_ID, &status);
> + } while ((status != PCI_VENDOR_ID_ATI) &&
> + (status != PCI_VENDOR_ID_AMD));
You'd better add another exit to avoid this becoming a dead loop. E.g: check response 'r' or add a try counter ?
Best Regards,
Kevin
-----Original Message-----
From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> On Behalf Of Lazar, Lijo
Sent: Thursday, August 7, 2025 3:33 PM
To: amd-gfx at lists.freedesktop.org
Cc: Zhang, Hawking <Hawking.Zhang at amd.com>; Deucher, Alexander <Alexander.Deucher at amd.com>; Sun, Ce(Overlord) <Ce.Sun at amd.com>
Subject: Re: [PATCH v3] drm/amdgpu: Save and restore switch state
<Ping>
On 8/1/2025 6:10 PM, Lijo Lazar wrote:
> During a DPC error kernel waits for the link to be active before
> notifying downstream devices. On certain platforms with Broadcom
> switch in synthetiic mode, switch responds with values even though the
> link is not fully ready. The config space restoration done by pcie
> port driver for SWUS/DS of dGPU is thus not effective as the switch is
> still doing internal enumeration.
>
> As a workaround, save state of SWUS/DS device in driver. Add
> additional check to see if link is active and restore the values
> during DPC error callbacks.
>
> Signed-off-by: Lijo Lazar <lijo.lazar at amd.com>
> ---
> v2: Use usleep_range as sleep is short. Remove dev_info logs.
> v3: remove redundant increment of 'i' in loop (Ce Sun).
>
> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 3 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 72
> +++++++++++++++++++++-
> 2 files changed, 73 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> index 3550c2fac184..96d772aadb04 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> @@ -904,6 +904,9 @@ struct amdgpu_pcie_reset_ctx {
> bool in_link_reset;
> bool occurs_dpc;
> bool audio_suspended;
> + struct pci_dev *swus;
> + struct pci_saved_state *swus_pcistate;
> + struct pci_saved_state *swds_pcistate;
> };
>
> /*
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index cfd72faec16e..e58f42531974 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -178,6 +178,8 @@ struct amdgpu_init_level amdgpu_init_minimal_xgmi = {
> BIT(AMD_IP_BLOCK_TYPE_PSP)
> };
>
> +static void amdgpu_device_load_switch_state(struct amdgpu_device
> +*adev);
> +
> static inline bool amdgpu_ip_member_of_hwini(struct amdgpu_device *adev,
> enum amd_ip_block_type block) { @@ -5006,7 +5008,8 @@ void
> amdgpu_device_fini_sw(struct amdgpu_device *adev)
> adev->reset_domain = NULL;
>
> kfree(adev->pci_state);
> -
> + kfree(adev->pcie_reset_ctx.swds_pcistate);
> + kfree(adev->pcie_reset_ctx.swus_pcistate);
> }
>
> /**
> @@ -6963,16 +6966,27 @@ pci_ers_result_t amdgpu_pci_slot_reset(struct pci_dev *pdev)
> struct amdgpu_device *tmp_adev;
> struct amdgpu_hive_info *hive;
> struct list_head device_list;
> + struct pci_dev *link_dev;
> int r = 0, i;
> u32 memsize;
> + u16 status;
>
> dev_info(adev->dev, "PCI error: slot reset callback!!\n");
>
> memset(&reset_context, 0, sizeof(reset_context));
>
> + if (adev->pcie_reset_ctx.swus)
> + link_dev = adev->pcie_reset_ctx.swus;
> + else
> + link_dev = adev->pdev;
> /* wait for asic to come out of reset */
> - msleep(700);
> + do {
> + usleep_range(10000, 10500);
> + r = pci_read_config_word(link_dev, PCI_VENDOR_ID, &status);
> + } while ((status != PCI_VENDOR_ID_ATI) &&
> + (status != PCI_VENDOR_ID_AMD));
>
> + amdgpu_device_load_switch_state(adev);
> /* Restore PCI confspace */
> amdgpu_device_load_pci_state(pdev);
>
> @@ -7074,6 +7088,58 @@ void amdgpu_pci_resume(struct pci_dev *pdev)
> }
> }
>
> +static void amdgpu_device_cache_switch_state(struct amdgpu_device
> +*adev) {
> + struct pci_dev *parent = pci_upstream_bridge(adev->pdev);
> + int r;
> +
> + if (parent->vendor != PCI_VENDOR_ID_ATI)
> + return;
> +
> + /* If already saved, return */
> + if (adev->pcie_reset_ctx.swus)
> + return;
> + /* Upstream bridge is ATI, assume it's SWUS/DS architecture */
> + r = pci_save_state(parent);
> + if (r)
> + return;
> + adev->pcie_reset_ctx.swds_pcistate = pci_store_saved_state(parent);
> +
> + parent = pci_upstream_bridge(parent);
> + r = pci_save_state(parent);
> + if (r)
> + return;
> + adev->pcie_reset_ctx.swus_pcistate = pci_store_saved_state(parent);
> +
> + adev->pcie_reset_ctx.swus = parent;
> +}
> +
> +static void amdgpu_device_load_switch_state(struct amdgpu_device
> +*adev) {
> + struct pci_dev *pdev;
> + int r;
> +
> + if (!adev->pcie_reset_ctx.swds_pcistate ||
> + !adev->pcie_reset_ctx.swus_pcistate)
> + return;
> +
> + pdev = adev->pcie_reset_ctx.swus;
> + r = pci_load_saved_state(pdev, adev->pcie_reset_ctx.swus_pcistate);
> + if (!r) {
> + pci_restore_state(pdev);
> + } else {
> + dev_warn(adev->dev, "Failed to load SWUS state, err:%d\n", r);
> + return;
> + }
> +
> + pdev = pci_upstream_bridge(adev->pdev);
> + r = pci_load_saved_state(pdev, adev->pcie_reset_ctx.swds_pcistate);
> + if (!r)
> + pci_restore_state(pdev);
> + else
> + dev_warn(adev->dev, "Failed to load SWDS state, err:%d\n", r); }
> +
> bool amdgpu_device_cache_pci_state(struct pci_dev *pdev) {
> struct drm_device *dev = pci_get_drvdata(pdev); @@ -7098,6 +7164,8
> @@ bool amdgpu_device_cache_pci_state(struct pci_dev *pdev)
> return false;
> }
>
> + amdgpu_device_cache_switch_state(adev);
> +
> return true;
> }
>
More information about the amd-gfx
mailing list