Re: [PATCH v6 1/1] vfio/nvgrace-gpu: Add Blackwell-Next GPU readiness check via CXL DVSEC

[Alex Williamson <alex@shazbot.org>]

Bjorn, please let us know if the pci_regs.h addition below is
acceptable.  Thanks!

Ankit, some comments below.

On Wed, 22 Apr 2026 13:49:26 +0000
Ankit Agrawal <ankita@nvidia.com> wrote:

> Add a CXL DVSEC-based readiness check for Blackwell-Next GPUs alongside
> the existing legacy BAR0 polling path. On probe and after reset, the
> driver reads the CXL Device DVSEC capability to determine whether the
> GPU memory is ready. A static inline wrapper dispatches to the
> appropriate readiness check (legacy v/s blackwell-next based on whether
> the CXL DVSEC capability is present.
> 
> The memory readiness is checked by polling on the Memory_Active bit
> based on the Memory_Active_Timeout. It also checks if MEM_INFO_VALID
> is set within 1 second. If not, return error. This is based on the
> CXL spec 4.0 Tables 8-13.
> 
> Add PCI_DVSEC_CXL_MEM_ACTIVE_TIMEOUT to pci_regs.h for the timeout
> field encoding.
> 
> Cc: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
> Cc: Kevin Tian <kevin.tian@intel.com>
> Suggested-by: Alex Williamson <alex@shazbot.org>
> Signed-off-by: Ankit Agrawal <ankita@nvidia.com>
> ---
>  drivers/vfio/pci/nvgrace-gpu/main.c | 107 +++++++++++++++++++++++++---
>  include/uapi/linux/pci_regs.h       |   1 +
>  2 files changed, 99 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/vfio/pci/nvgrace-gpu/main.c b/drivers/vfio/pci/nvgrace-gpu/main.c
> index fa056b69f899..4e1d20ad7510 100644
> --- a/drivers/vfio/pci/nvgrace-gpu/main.c
> +++ b/drivers/vfio/pci/nvgrace-gpu/main.c
> @@ -3,7 +3,9 @@
>   * Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved
>   */
>  
> +#include <linux/bitfield.h>
>  #include <linux/sizes.h>
> +#include <linux/time64.h>
>  #include <linux/vfio_pci_core.h>
>  #include <linux/delay.h>
>  #include <linux/jiffies.h>
> @@ -64,6 +66,8 @@ struct nvgrace_gpu_pci_core_device {
>  	bool has_mig_hw_bug;
>  	/* GPU has just been reset */
>  	bool reset_done;
> +	/* CXL Device DVSEC offset; 0 if not present (legacy GB path) */
> +	int cxl_dvsec;
>  };
>  
>  static void nvgrace_gpu_init_fake_bar_emu_regs(struct vfio_device *core_vdev)
> @@ -242,7 +246,7 @@ static void nvgrace_gpu_close_device(struct vfio_device *core_vdev)
>  	vfio_pci_core_close_device(core_vdev);
>  }
>  
> -static int nvgrace_gpu_wait_device_ready(void __iomem *io)
> +static int nvgrace_gpu_wait_device_ready_legacy(void __iomem *io)
>  {
>  	unsigned long timeout = jiffies + msecs_to_jiffies(POLL_TIMEOUT_MS);
>  
> @@ -256,6 +260,81 @@ static int nvgrace_gpu_wait_device_ready(void __iomem *io)
>  	return -ETIME;
>  }
>  
> +/*
> + * Decode the 3-bit Memory_Active_Timeout field from CXL DVSEC Range 1 Low
> + * (bits 15:13) into milliseconds. Encoding per CXL spec r4.0 sec 8.1.3.8.2:
> + * 000b = 1s, 001b = 4s, 010b = 16s, 011b = 64s, 100b = 256s,
> + * 101b-111b = reserved (clamped to 256s).
> + */
> +static inline unsigned long cxl_mem_active_timeout_ms(u8 timeout)
> +{
> +	return MSEC_PER_SEC << (2 * min_t(u8, timeout, 4));
> +}
> +
> +/*
> + * Check if CXL DVSEC reports memory as valid and active.
> + */
> +static inline bool cxl_dvsec_mem_is_active(u32 status)
> +{
> +	return (status & PCI_DVSEC_CXL_MEM_INFO_VALID) &&
> +	       (status & PCI_DVSEC_CXL_MEM_ACTIVE);
> +}
> +
> +static int nvgrace_gpu_wait_device_ready_cxl(struct nvgrace_gpu_pci_core_device *nvdev)
> +{
> +	struct pci_dev *pdev = nvdev->core_device.pdev;
> +	int cxl_dvsec = nvdev->cxl_dvsec;
> +	unsigned long mem_info_valid_deadline;
> +	unsigned long timeout = 0;
> +	u32 dvsec_memory_status;
> +
> +	mem_info_valid_deadline = jiffies + msecs_to_jiffies(POLL_QUANTUM_MS);
> +
> +	do {
> +		pci_read_config_dword(pdev,
> +				      cxl_dvsec + PCI_DVSEC_CXL_RANGE_SIZE_LOW(0),
> +				      &dvsec_memory_status);
> +
> +		if (dvsec_memory_status == ~0U)
> +			return -ENODEV;
> +
> +		if (cxl_dvsec_mem_is_active(dvsec_memory_status))
> +			return 0;
> +
> +		/*
> +		 * Once MEM_INFO_VALID is set, derive the MEM_ACTIVE timeout
> +		 * from the register.
> +		 */
> +		if (dvsec_memory_status & PCI_DVSEC_CXL_MEM_INFO_VALID) {
> +			if (!timeout) {
> +				u8 mem_active_timeout =
> +					FIELD_GET(PCI_DVSEC_CXL_MEM_ACTIVE_TIMEOUT,
> +						  dvsec_memory_status);
> +
> +				timeout = jiffies +
> +					  msecs_to_jiffies(cxl_mem_active_timeout_ms(mem_active_timeout));
> +			}
> +		}
> +
> +		/* Bail early if MEM_INFO_VALID is not set within 1 second */
> +		if (!(dvsec_memory_status & PCI_DVSEC_CXL_MEM_INFO_VALID) &&
> +		    time_after(jiffies, mem_info_valid_deadline))
> +			return -ETIME;
> +
> +		msleep(POLL_QUANTUM_MS);
> +	} while (!timeout || !time_after(jiffies, timeout));
> +
> +	return -ETIME;
> +}

Sashiko has some comments[1] on this.  The first two could be
eliminated with some restructuring of this looop as:

  unsigned long deadline = jiffies + msecs_to_jiffies(POLL_QUANTUM_MS);
  bool active_phase = false;
  u32 status;                                                                                                          
  
  for (;;) {                                                                                                           
          pci_read_config_dword(pdev,                       
                                cxl_dvsec + PCI_DVSEC_CXL_RANGE_SIZE_LOW(0),                                           
                                &status);                                                                              
          if (status == ~0U)                                                                                           
                  return -ENODEV;                                                                                      
          if (cxl_dvsec_mem_is_active(status))              
                  return 0;                                                                                            
                                                            
          if (!active_phase && (status & PCI_DVSEC_CXL_MEM_INFO_VALID)) {                                              
                  u8 t = FIELD_GET(PCI_DVSEC_CXL_MEM_ACTIVE_TIMEOUT, status);
                                                                                                                       
                  deadline = jiffies +
                             msecs_to_jiffies(cxl_mem_active_timeout_ms(t));                                           
                  active_phase = true;                                                                                 
          }
                                                                                                                       
          if (time_after(jiffies, deadline))                
                  return -ETIME;

          msleep(POLL_QUANTUM_MS);                                                                                     
  }


The third point sashiko makes is whether it's really viable to hold the
memory_lock read-semaphore for the full extent specified by the CXL
spec.  In fact, does the CXL wait function need memory_lock at all?

I think the answer is no, it can be pushed into the legacy device-ready
check, but we do need to re-acquire and verify memory-enable after the
device-ready checks for each code path, fault, read, and write.  Thanks,

Alex

[1]https://sashiko.dev/#/patchset/20260422134926.653211-1-ankita@nvidia.com

> +
> +static inline int nvgrace_gpu_wait_device_ready(struct nvgrace_gpu_pci_core_device *nvdev,
> +						void __iomem *io)
> +{
> +	return nvdev->cxl_dvsec ?
> +		nvgrace_gpu_wait_device_ready_cxl(nvdev) :
> +		nvgrace_gpu_wait_device_ready_legacy(io);
> +}
> +
>  /*
>   * If the GPU memory is accessed by the CPU while the GPU is not ready
>   * after reset, it can cause harmless corrected RAS events to be logged.
> @@ -275,7 +354,7 @@ nvgrace_gpu_check_device_ready(struct nvgrace_gpu_pci_core_device *nvdev)
>  	if (!__vfio_pci_memory_enabled(vdev))
>  		return -EIO;
>  
> -	ret = nvgrace_gpu_wait_device_ready(vdev->barmap[0]);
> +	ret = nvgrace_gpu_wait_device_ready(nvdev, vdev->barmap[0]);
>  	if (ret)
>  		return ret;
>  
> @@ -1143,14 +1222,21 @@ static bool nvgrace_gpu_has_mig_hw_bug(struct pci_dev *pdev)
>   * is beneficial to make the check to ensure the device is in an
>   * expected state.
>   *
> - * Ensure that the BAR0 region is enabled before accessing the
> + * On Blackwell-Next systems, memory readiness is determined via the
> + * CXL Device DVSEC in PCI config space and does not require BAR0.
> + * For the legacy path, ensure BAR0 is enabled before accessing the
>   * registers.
>   */
> -static int nvgrace_gpu_probe_check_device_ready(struct pci_dev *pdev)
> +static int nvgrace_gpu_probe_check_device_ready(struct nvgrace_gpu_pci_core_device *nvdev)
>  {
> +	struct pci_dev *pdev = nvdev->core_device.pdev;
>  	void __iomem *io;
>  	int ret;
>  
> +	/* CXL path only reads PCI config space; no need to map BAR0. */
> +	if (nvdev->cxl_dvsec)
> +		return nvgrace_gpu_wait_device_ready_cxl(nvdev);
> +
>  	ret = pci_enable_device(pdev);
>  	if (ret)
>  		return ret;
> @@ -1165,7 +1251,7 @@ static int nvgrace_gpu_probe_check_device_ready(struct pci_dev *pdev)
>  		goto iomap_exit;
>  	}
>  
> -	ret = nvgrace_gpu_wait_device_ready(io);
> +	ret = nvgrace_gpu_wait_device_ready_legacy(io);
>  
>  	pci_iounmap(pdev, io);
>  iomap_exit:
> @@ -1183,10 +1269,6 @@ static int nvgrace_gpu_probe(struct pci_dev *pdev,
>  	u64 memphys, memlength;
>  	int ret;
>  
> -	ret = nvgrace_gpu_probe_check_device_ready(pdev);
> -	if (ret)
> -		return ret;
> -
>  	ret = nvgrace_gpu_fetch_memory_property(pdev, &memphys, &memlength);
>  	if (!ret)
>  		ops = &nvgrace_gpu_pci_ops;
> @@ -1198,6 +1280,13 @@ static int nvgrace_gpu_probe(struct pci_dev *pdev,
>  
>  	dev_set_drvdata(&pdev->dev, &nvdev->core_device);
>  
> +	nvdev->cxl_dvsec = pci_find_dvsec_capability(pdev, PCI_VENDOR_ID_CXL,
> +						     PCI_DVSEC_CXL_DEVICE);
> +
> +	ret = nvgrace_gpu_probe_check_device_ready(nvdev);
> +	if (ret)
> +		goto out_put_vdev;
> +
>  	if (ops == &nvgrace_gpu_pci_ops) {
>  		nvdev->has_mig_hw_bug = nvgrace_gpu_has_mig_hw_bug(pdev);
>  
> diff --git a/include/uapi/linux/pci_regs.h b/include/uapi/linux/pci_regs.h
> index 14f634ab9350..718fb630f5bb 100644
> --- a/include/uapi/linux/pci_regs.h
> +++ b/include/uapi/linux/pci_regs.h
> @@ -1357,6 +1357,7 @@
>  #define  PCI_DVSEC_CXL_RANGE_SIZE_LOW(i)		(0x1C + (i * 0x10))
>  #define   PCI_DVSEC_CXL_MEM_INFO_VALID			_BITUL(0)
>  #define   PCI_DVSEC_CXL_MEM_ACTIVE			_BITUL(1)
> +#define   PCI_DVSEC_CXL_MEM_ACTIVE_TIMEOUT		__GENMASK(15, 13)
>  #define   PCI_DVSEC_CXL_MEM_SIZE_LOW			__GENMASK(31, 28)
>  #define  PCI_DVSEC_CXL_RANGE_BASE_HIGH(i)		(0x20 + (i * 0x10))
>  #define  PCI_DVSEC_CXL_RANGE_BASE_LOW(i)		(0x24 + (i * 0x10))