[PATCH] EEH: remove eeh device from OF node

[Benjamin Herrenschmidt]

On Wed, 2012-03-14 at 18:31 +0800, Gavin Shan wrote:
> Originally, the PCI sensitive OF node is tracing the eeh device
> through struct device_node::edev. However, it was regarded as
> bad idea.
> 
> The patch removes struct device_node::edev. In addition, the
> global list of eeh devices is introduced, and do retrival of
> eeh device according to the given OF node through the global
> list.

So I'm not too happy with that. The main problem I see is that
the code -constantly- calls of_node_to_eeh_dev.

IE. On any MMIO that happens to return all 1's, we'll call
eeh_check_failure() for example, which does that. In fact for a hot path
it's pretty horrid, it will:

 - Do an address cache lookup to get the pci_dev
 - Get the device-node frokm the pci_dev
 - Lookup your list to get the eeh_dev

Shouldn't we instead change the address cache to contain eeh_dev
instead ? And if you prefer keeping pci_dev, then it shouldn't be hard
to stick a pointer to the eeh_dev in there via either struct archdata or
maybe platform_data.

The EEH code still, even after your rework, constantly pass "dn"'s as
argument to functions just to convert it back to a struct eeh_dev rather
than trying to pass the eeh_dev.

So turning that into a list lookup will slow things down to a crawl.
Also your patch never seems to remove anything from your list, which
doesn't look right vs. hotplug.

I suggest we fix that with a two phase approach:

 1- ASAP so we can still get that into 3.4, move the eeh_dev pointer to
struct pci_dn instead of struct device_node. This structure is
accessible directly via dn->data and is ppc specific, that will be a
better temporary solution and and adding stuff to the generic struct
device_node.

 2- Then, what we need to do is generalize the use of eeh_dev rather
than device_node as the main object being worked on in the eeh layer and
thus as the argument to most functions.

Cheers,
Ben.

> Signed-off-by: Gavin Shan <shangw at linux.vnet.ibm.com>
> ---
>  arch/powerpc/include/asm/eeh.h           |    7 +++++++
>  arch/powerpc/platforms/pseries/eeh_dev.c |   29 ++++++++++++++++++++++++++++-
>  include/linux/of.h                       |   10 ----------
>  3 files changed, 35 insertions(+), 11 deletions(-)
> 
> diff --git a/arch/powerpc/include/asm/eeh.h b/arch/powerpc/include/asm/eeh.h
> index d60f998..591e0a1 100644
> --- a/arch/powerpc/include/asm/eeh.h
> +++ b/arch/powerpc/include/asm/eeh.h
> @@ -56,6 +56,7 @@ struct eeh_dev {
>  	struct pci_controller *phb;	/* Associated PHB		*/
>  	struct device_node *dn;		/* Associated device node	*/
>  	struct pci_dev *pdev;		/* Associated PCI device	*/
> +	struct list_head list;		/* Form the global link list	*/
>  };
>  
>  static inline struct device_node *eeh_dev_to_of_node(struct eeh_dev *edev)
> @@ -115,6 +116,7 @@ extern int eeh_subsystem_enabled;
>   */
>  #define EEH_MAX_ALLOWED_FREEZES 5
>  
> +struct eeh_dev *eeh_dev_from_of_node(struct device_node *dn);
>  void * __devinit eeh_dev_init(struct device_node *dn, void *data);
>  void __devinit eeh_dev_phb_init_dynamic(struct pci_controller *phb);
>  void __init eeh_dev_phb_init(void);
> @@ -132,6 +134,11 @@ void eeh_add_device_tree_early(struct device_node *);
>  void eeh_add_device_tree_late(struct pci_bus *);
>  void eeh_remove_bus_device(struct pci_dev *);
>  
> +static inline struct eeh_dev *of_node_to_eeh_dev(struct device_node *dn)
> +{
> +	return eeh_dev_from_of_node(dn);
> +}
> +
>  /**
>   * EEH_POSSIBLE_ERROR() -- test for possible MMIO failure.
>   *
> diff --git a/arch/powerpc/platforms/pseries/eeh_dev.c b/arch/powerpc/platforms/pseries/eeh_dev.c
> index f3aed7d..925d3a3 100644
> --- a/arch/powerpc/platforms/pseries/eeh_dev.c
> +++ b/arch/powerpc/platforms/pseries/eeh_dev.c
> @@ -34,6 +34,7 @@
>  #include <linux/export.h>
>  #include <linux/gfp.h>
>  #include <linux/init.h>
> +#include <linux/list.h>
>  #include <linux/kernel.h>
>  #include <linux/pci.h>
>  #include <linux/string.h>
> @@ -41,6 +42,30 @@
>  #include <asm/pci-bridge.h>
>  #include <asm/ppc-pci.h>
>  
> +/* eeh device list */
> +static LIST_HEAD(eeh_dev_list);
> +
> +/**
> + * eeh_dev_from_of_node - Retrieve EEH device according to OF node
> + * @dn: OF node
> + *
> + * All existing eeh devices have been put into the global list.
> + * In addition, the eeh device is tracing the corresponding
> + * OF node. The function is used to retrieve the corresponding
> + * eeh device according to the given OF node.
> + */
> +struct eeh_dev *eeh_dev_from_of_node(struct device_node *dn)
> +{
> +	struct eeh_dev *edev = NULL;
> +
> +	list_for_each_entry(edev, &eeh_dev_list, list) {
> +		if (edev->dn && edev->dn == dn)
> +			return edev;
> +	}
> +
> +	return NULL;
> +}
> +
>  /**
>   * eeh_dev_init - Create EEH device according to OF node
>   * @dn: device node
> @@ -62,10 +87,12 @@ void * __devinit eeh_dev_init(struct device_node *dn, void *data)
>  	}
>  
>  	/* Associate EEH device with OF node */
> -	dn->edev  = edev;
>  	edev->dn  = dn;
>  	edev->phb = phb;
>  
> +	/* Add to global list */
> +	list_add_tail(&edev->list, &eeh_dev_list);
> +
>  	return NULL;
>  }
>  
> diff --git a/include/linux/of.h b/include/linux/of.h
> index 3e710d8..a75a831 100644
> --- a/include/linux/of.h
> +++ b/include/linux/of.h
> @@ -58,9 +58,6 @@ struct device_node {
>  	struct	kref kref;
>  	unsigned long _flags;
>  	void	*data;
> -#if defined(CONFIG_EEH)
> -	struct eeh_dev *edev;
> -#endif
>  #if defined(CONFIG_SPARC)
>  	char	*path_component_name;
>  	unsigned int unique_id;
> @@ -75,13 +72,6 @@ struct of_phandle_args {
>  	uint32_t args[MAX_PHANDLE_ARGS];
>  };
>  
> -#if defined(CONFIG_EEH)
> -static inline struct eeh_dev *of_node_to_eeh_dev(struct device_node *dn)
> -{
> -	return dn->edev;
> -}
> -#endif
> -
>  #if defined(CONFIG_SPARC) || !defined(CONFIG_OF)
>  /* Dummy ref counting routines - to be implemented later */
>  static inline struct device_node *of_node_get(struct device_node *node)