Re: [RFC PATCH 01/11] pci: endpoint: add EP core layer to enable EP controller and EP functions

From: Christoph Hellwig
Date: Wed Oct 12 2016 - 09:12:04 EST


> +/**
> + * pci_epc_stop() - stop the PCI link
> + * @epc: the link of the EPC device that has to be stopped
> + *
> + * Invoke to stop the PCI link
> + */
> +void pci_epc_stop(struct pci_epc *epc)
> +{
> + if (IS_ERR(epc) || !epc->ops->stop)
> + return;
> +
> + spin_lock_irq(&epc->irq_lock);
> + epc->ops->stop(epc);
> + spin_unlock_irq(&epc->irq_lock);
> +}
> +EXPORT_SYMBOL_GPL(pci_epc_stop);

Can you elaborate on the synchronization strategy here? It seems
like irq_lock is generally taken irq save and just around method
calls. Wou;dn't it be better to leave locking to the methods
themselves?

> +/**
> + * struct pci_epc - represents the PCI EPC device
> + * @dev: PCI EPC device
> + * @ops: function pointers for performing endpoint operations
> + * @mutex: mutex to protect pci_epc ops
> + */
> +struct pci_epc {
> + struct device dev;
> + /* support only single function PCI device for now */
> + struct pci_epf *epf;
> + const struct pci_epc_ops *ops;
> + spinlock_t irq_lock;
> +};

And this still documentes a mutex instead of the irq save spinlock,
while we're at it..

> +/**
> + * struct pci_epf_bar - represents the BAR of EPF device
> + * @phys_addr: physical address that should be mapped to the BAR
> + * @size: the size of the address space present in BAR
> + */
> +struct pci_epf_bar {
> + dma_addr_t phys_addr;
> + size_t size;
> +};

Just curious: shouldn't this be a phys_addr_t instead of a dma_addr_t?


Otherwise this looks like a nice little framework to get started!