Re: [PATCH v4 05/27] cxl: Add a function-scoped reset entry for vfio-pci

[Alex Williamson <alex@shazbot.org>]

On Thu, 13 Aug 2026 15:06:09 +0530
<mhonap@nvidia.com> wrote:

> From: Manish Honap <mhonap@nvidia.com>
> 
> vfio-pci needs to run a CXL reset on a passthroughed Type-2 device
> without the host-memory handling in cxl_reset_function(): the memory
> behind the decoder belongs to the guest.
> 
> Add cxl_reset_dvsec_sequence(), which runs the DVSEC reset and HDM decoder
> restore under pci_dev_lock and cxl_rwsem, and cxl_reset_capable() to gate
> it on a function-scoped reset.
> 
> Parameterize Memory Clear in cxl_reset_execute() so the guest owns the
> choice; the host path keeps it disabled.
> 
> Signed-off-by: Manish Honap <mhonap@nvidia.com>
> ---
>  drivers/cxl/core/resource.c | 69 ++++++++++++++++++++++++++++++++++---
>  include/cxl/cxl.h           | 12 +++++++
>  2 files changed, 76 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/cxl/core/resource.c b/drivers/cxl/core/resource.c
> index 492ffe8e3576..cd15bd9171e6 100644
> --- a/drivers/cxl/core/resource.c
> +++ b/drivers/cxl/core/resource.c
> @@ -1164,7 +1164,7 @@ static int cxl_reset_enable_cache(struct pci_dev *pdev, int dvsec)
>  				     PCI_DVSEC_CXL_DISABLE_CACHING);
>  }
>  
> -static int cxl_reset_initiate(struct pci_dev *pdev, int dvsec)
> +static int cxl_reset_initiate(struct pci_dev *pdev, int dvsec, bool mem_clr_en)
>  {
>  	u16 ctrl2;
>  	int rc;
> @@ -1173,7 +1173,10 @@ static int cxl_reset_initiate(struct pci_dev *pdev, int dvsec)
>  	if (rc)
>  		return rc;
>  
> -	ctrl2 &= ~PCI_DVSEC_CXL_RST_MEM_CLR_EN;
> +	if (mem_clr_en)
> +		ctrl2 |= PCI_DVSEC_CXL_RST_MEM_CLR_EN;
> +	else
> +		ctrl2 &= ~PCI_DVSEC_CXL_RST_MEM_CLR_EN;
>  	ctrl2 |= PCI_DVSEC_CXL_INIT_CXL_RST;
>  	return cxl_reset_write_ctrl2(pdev, dvsec, ctrl2);
>  }

I'd again avoid a bool arg here, an enum type might be a better option.

I see in the spec though that CXL Reset Mem Clr Enable is only
supported when the DVSEC CXL Capability register reports CXL Reset Mem
Clr Capable though.  Should this return -ENOTTY if memory clearing is
requested when the device is not capable?

> @@ -1273,7 +1276,7 @@ static int cxl_reset_wait_done(struct pci_dev *pdev, int dvsec, u16 cap)
>  }
>  
>  static int cxl_reset_execute(struct pci_dev *pdev, bool *target_prepared,
> -			     int dvsec, u16 cap)
> +			     int dvsec, u16 cap, bool mem_clr_en)
>  {
>  	int rc, rc2;
>  
> @@ -1283,7 +1286,7 @@ static int cxl_reset_execute(struct pci_dev *pdev, bool *target_prepared,
>  
>  	rc = cxl_pci_target_reset_prepare(pdev, target_prepared);
>  	if (!rc)
> -		rc = cxl_reset_initiate(pdev, dvsec);
> +		rc = cxl_reset_initiate(pdev, dvsec, mem_clr_en);
>  	if (!rc)
>  		rc = cxl_reset_wait_done(pdev, dvsec, cap);
>  
> @@ -1322,7 +1325,8 @@ int cxl_reset_function(struct pci_dev *pdev, bool probe)
>  	scoped_guard(rwsem_write, &cxl_rwsem.region) {
>  		rc = cxl_hdm_ranges_prepare(&range_ctx, pdev);
>  		if (!rc)
> -			rc = cxl_reset_execute(pdev, &target_prepared, dvsec, cap);
> +			rc = cxl_reset_execute(pdev, &target_prepared, dvsec,
> +					       cap, false);
>  		if (!rc) {
>  			u16 command;
>  
> @@ -1340,3 +1344,58 @@ int cxl_reset_function(struct pci_dev *pdev, bool probe)
>  	cxl_pci_target_reset_done(pdev, &target_prepared);
>  	return rc;
>  }
> +
> +/* True when a function-scoped CXL reset is available for @pdev. */
> +bool cxl_reset_capable(struct pci_dev *pdev)
> +{
> +	u16 cap;
> +
> +	if (cxl_reset_dvsec(pdev, &cap) < 0)
> +		return false;
> +
> +	if (pdev->multifunction)
> +		return false;
> +
> +	return cxl_reset_hdm_available(pdev);
> +}
> +EXPORT_SYMBOL_NS_GPL(cxl_reset_capable, "CXL");
> +
> +/*
> + * Run the DVSEC reset sequence and restore HDM state for a caller that owns
> + * device quiesce and PCI config save/restore, such as vfio-pci. The HDM range
> + * collection and CPU cache flush that cxl_reset_function() does for host-owned
> + * memory are skipped; that memory belongs to the guest here.

It's worth noting in the comment that a device memory clear without a
preceding cache invalidation risks that the cache writes back some of
the memory we just cleared.  The use case later in the series does that
cache invalidation, but the requirement is a bit subtle here.  Thanks,

Alex


> + */
> +int cxl_reset_dvsec_sequence(struct pci_dev *pdev, bool mem_clr_en)
> +{
> +	bool target_prepared = false;
> +	int dvsec;
> +	int rc;
> +	u16 cap;
> +
> +	dvsec = cxl_reset_dvsec(pdev, &cap);
> +	if (dvsec < 0)
> +		return dvsec;
> +
> +	if (pdev->multifunction)
> +		return -ENOTTY;
> +
> +	/*
> +	 * Trylock rather than block: This follows the trylock convention of
> +	 * pci_reset_bus().
> +	 */
> +	if (!pci_dev_trylock(pdev))
> +		return -EBUSY;
> +
> +	scoped_guard(rwsem_write, &cxl_rwsem.region) {
> +		rc = cxl_reset_execute(pdev, &target_prepared, dvsec, cap,
> +				       mem_clr_en);
> +		if (!rc)
> +			rc = cxl_restore_hdm_after_pci_reset(pdev);
> +	}
> +
> +	cxl_pci_target_reset_done(pdev, &target_prepared);
> +	pci_dev_unlock(pdev);
> +	return rc;
> +}
> +EXPORT_SYMBOL_NS_GPL(cxl_reset_dvsec_sequence, "CXL");
> diff --git a/include/cxl/cxl.h b/include/cxl/cxl.h
> index f8e8fddba152..541ed6de75a6 100644
> --- a/include/cxl/cxl.h
> +++ b/include/cxl/cxl.h
> @@ -164,6 +164,8 @@ void pci_cxl_hdm_init(struct pci_dev *pdev);
>  void pci_cxl_hdm_release(struct pci_dev *pdev);
>  int cxl_restore_hdm_after_pci_reset(struct pci_dev *pdev);
>  int cxl_reset_function(struct pci_dev *pdev, bool probe);
> +bool cxl_reset_capable(struct pci_dev *pdev);
> +int cxl_reset_dvsec_sequence(struct pci_dev *pdev, bool mem_clr_en);
>  #else
>  static inline void pci_cxl_hdm_init(struct pci_dev *pdev)
>  {
> @@ -182,6 +184,16 @@ static inline int cxl_reset_function(struct pci_dev *pdev, bool probe)
>  {
>  	return -ENOTTY;
>  }
> +
> +static inline bool cxl_reset_capable(struct pci_dev *pdev)
> +{
> +	return false;
> +}
> +
> +static inline int cxl_reset_dvsec_sequence(struct pci_dev *pdev, bool mem_clr_en)
> +{
> +	return -ENOTTY;
> +}
>  #endif
>  
>  struct cxl_reg_map {