[PATCH 13/14] powerpc/eeh: Add a eeh_dev_break debugfs interface

[Sam Bobroff]

On Tue, Sep 03, 2019 at 08:16:04PM +1000, Oliver O'Halloran wrote:
> Add an interface to debugfs for generating an EEH event on a given device.
> This works by disabling memory accesses to and from the device by setting
> the PCI_COMMAND register (or the VF Memory Space Enable on the parent PF).
> 
> This is a somewhat portable alternative to using the platform specific
> error injection mechanisms since those tend to be either hard to use, or
> straight up broken. For pseries the interfaces also requires the use of
> /dev/mem which is probably going to go away in a post-LOCKDOWN world
> (and it's a horrific hack to begin with) so moving to a kernel-provided
> interface makes sense and provides a sane, cross-platform interface for
> userspace so we can write more generic testing scripts.
> 
> Signed-off-by: Oliver O'Halloran <oohall at gmail.com>

Looks good to me. Tested with the previous patch.

Tested-by: Sam Bobroff <sbobroff at linux.ibm.com>
Reviewed-by: Sam Bobroff <sbobroff at linux.ibm.com>

> ---
>  arch/powerpc/kernel/eeh.c | 139 +++++++++++++++++++++++++++++++++++++-
>  1 file changed, 138 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/kernel/eeh.c b/arch/powerpc/kernel/eeh.c
> index ace1c5a6b8ed..a55d2f01da1d 100644
> --- a/arch/powerpc/kernel/eeh.c
> +++ b/arch/powerpc/kernel/eeh.c
> @@ -1894,7 +1894,8 @@ static ssize_t eeh_dev_check_write(struct file *filp,
>  	char buf[20];
>  	int ret;
>  
> -	ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count);
> +	memset(buf, 0, sizeof(buf));
> +	ret = simple_write_to_buffer(buf, sizeof(buf)-1, ppos, user_buf, count);
>  	if (!ret)
>  		return -EFAULT;
>  
> @@ -1931,6 +1932,139 @@ static const struct file_operations eeh_dev_check_fops = {
>  	.read   = eeh_debugfs_dev_usage,
>  };
>  
> +static int eeh_debugfs_break_device(struct pci_dev *pdev)
> +{
> +	struct resource *bar = NULL;
> +	void __iomem *mapped;
> +	u16 old, bit;
> +	int i, pos;
> +
> +	/* Do we have an MMIO BAR to disable? */
> +	for (i = 0; i <= PCI_STD_RESOURCE_END; i++) {
> +		struct resource *r = &pdev->resource[i];
> +
> +		if (!r->flags || !r->start)
> +			continue;
> +		if (r->flags & IORESOURCE_IO)
> +			continue;
> +		if (r->flags & IORESOURCE_UNSET)
> +			continue;
> +
> +		bar = r;
> +		break;
> +	}
> +
> +	if (!bar) {
> +		pci_err(pdev, "Unable to find Memory BAR to cause EEH with\n");
> +		return -ENXIO;
> +	}
> +
> +	pci_err(pdev, "Going to break: %pR\n", bar);
> +
> +	if (pdev->is_virtfn) {
> +#ifndef CONFIG_IOV
> +		return -ENXIO;
> +#else
> +		/*
> +		 * VFs don't have a per-function COMMAND register, so the best
> +		 * we can do is clear the Memory Space Enable bit in the PF's
> +		 * SRIOV control reg.
> +		 *
> +		 * Unfortunately, this requires that we have a PF (i.e doesn't
> +		 * work for a passed-through VF) and it has the potential side
> +		 * effect of also causing an EEH on every other VF under the
> +		 * PF. Oh well.
> +		 */
> +		pdev = pdev->physfn;
> +		if (!pdev)
> +			return -ENXIO; /* passed through VFs have no PF */
> +
> +		pos  = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
> +		pos += PCI_SRIOV_CTRL;
> +		bit  = PCI_SRIOV_CTRL_MSE;
> +#endif /* !CONFIG_IOV */
> +	} else {
> +		bit = PCI_COMMAND_MEMORY;
> +		pos = PCI_COMMAND;
> +	}
> +
> +	/*
> +	 * Process here is:
> +	 *
> +	 * 1. Disable Memory space.
> +	 *
> +	 * 2. Perform an MMIO to the device. This should result in an error
> +	 *    (CA  / UR) being raised by the device which results in an EEH
> +	 *    PE freeze. Using the in_8() accessor skips the eeh detection hook
> +	 *    so the freeze hook so the EEH Detection machinery won't be
> +	 *    triggered here. This is to match the usual behaviour of EEH
> +	 *    where the HW will asyncronously freeze a PE and it's up to
> +	 *    the kernel to notice and deal with it.
> +	 *
> +	 * 3. Turn Memory space back on. This is more important for VFs
> +	 *    since recovery will probably fail if we don't. For normal
> +	 *    the COMMAND register is reset as a part of re-initialising
> +	 *    the device.
> +	 *
> +	 * Breaking stuff is the point so who cares if it's racy ;)
> +	 */
> +	pci_read_config_word(pdev, pos, &old);
> +
> +	mapped = ioremap(bar->start, PAGE_SIZE);
> +	if (!mapped) {
> +		pci_err(pdev, "Unable to map MMIO BAR %pR\n", bar);
> +		return -ENXIO;
> +	}
> +
> +	pci_write_config_word(pdev, pos, old & ~bit);
> +	in_8(mapped);
> +	pci_write_config_word(pdev, pos, old);
> +
> +	iounmap(mapped);
> +
> +	return 0;
> +}
> +
> +static ssize_t eeh_dev_break_write(struct file *filp,
> +				const char __user *user_buf,
> +				size_t count, loff_t *ppos)
> +{
> +	uint32_t domain, bus, dev, fn;
> +	struct pci_dev *pdev;
> +	char buf[20];
> +	int ret;
> +
> +	memset(buf, 0, sizeof(buf));
> +	ret = simple_write_to_buffer(buf, sizeof(buf)-1, ppos, user_buf, count);
> +	if (!ret)
> +		return -EFAULT;
> +
> +	ret = sscanf(buf, "%x:%x:%x.%x", &domain, &bus, &dev, &fn);
> +	if (ret != 4) {
> +		pr_err("%s: expected 4 args, got %d\n", __func__, ret);
> +		return -EINVAL;
> +	}
> +
> +	pdev = pci_get_domain_bus_and_slot(domain, bus, (dev << 3) | fn);
> +	if (!pdev)
> +		return -ENODEV;
> +
> +	ret = eeh_debugfs_break_device(pdev);
> +	pci_dev_put(pdev);
> +
> +	if (ret < 0)
> +		return ret;
> +
> +	return count;
> +}
> +
> +static const struct file_operations eeh_dev_break_fops = {
> +	.open	= simple_open,
> +	.llseek	= no_llseek,
> +	.write	= eeh_dev_break_write,
> +	.read   = eeh_debugfs_dev_usage,
> +};
> +
>  #endif
>  
>  static int __init eeh_init_proc(void)
> @@ -1949,6 +2083,9 @@ static int __init eeh_init_proc(void)
>  		debugfs_create_file_unsafe("eeh_dev_check", 0600,
>  				powerpc_debugfs_root, NULL,
>  				&eeh_dev_check_fops);
> +		debugfs_create_file_unsafe("eeh_dev_break", 0600,
> +				powerpc_debugfs_root, NULL,
> +				&eeh_dev_break_fops);
>  		debugfs_create_file_unsafe("eeh_force_recover", 0600,
>  				powerpc_debugfs_root, NULL,
>  				&eeh_force_recover_fops);
> -- 
> 2.21.0
> 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: not available
URL: <http://lists.ozlabs.org/pipermail/linuxppc-dev/attachments/20190917/9706283f/attachment-0001.sig>