Re: [PATCH] edac: sifive: Add EDAC platform driver for SiFive SoCs

[James Morse]

Hi Yash,

Sorry for the delay on the earlier version of this - I was trying to work out what happens
when multiple edac drivers probe based on DT...


On 02/05/2019 12:16, Yash Shah wrote:
> The initial ver of EDAC driver supports:
> - ECC event monitoring and reporting through the EDAC framework for SiFive
>   L2 cache controller.
> 

You probably don't want this bit preserved in the kernel log:
{

> This patch depends on patch
> 'RISC-V: sifive_l2_cache: Add L2 cache controller driver for SiFive SoCs'
> https://lkml.org/lkml/2019/5/2/309

}

> The EDAC driver registers for notifier events from the L2 cache controller
> driver (arch/riscv/mm/sifive_l2_cache.c) for L2 ECC events
> 
> Signed-off-by: Yash Shah <yash.shah@xxxxxxxxxx>
> ---

(if you put it here, it gets discarded when the patch is applied)

Having an separately posted dependency like this is tricky, as this code can't be
used/tested until the other bits are merged.


>  MAINTAINERS                |   6 +++
>  arch/riscv/Kconfig         |   1 +
>  drivers/edac/Kconfig       |   6 +++
>  drivers/edac/Makefile      |   1 +
>  drivers/edac/sifive_edac.c | 121 +++++++++++++++++++++++++++++++++++++++++++++
>  5 files changed, 135 insertions(+)
>  create mode 100644 drivers/edac/sifive_edac.c

> diff --git a/drivers/edac/sifive_edac.c b/drivers/edac/sifive_edac.c
> new file mode 100644
> index 0000000..eb7a9b9
> --- /dev/null
> +++ b/drivers/edac/sifive_edac.c
> @@ -0,0 +1,121 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * SiFive Platform EDAC Driver
> + *
> + * Copyright (C) 2018-2019 SiFive, Inc.
> + *
> + * This driver is partially based on octeon_edac-pc.c
> + *
> + */
> +#include <linux/edac.h>
> +#include <linux/platform_device.h>
> +#include "edac_module.h"
> +
> +#define DRVNAME "sifive_edac"
> +
> +extern int register_sifive_l2_error_notifier(struct notifier_block *nb);
> +extern int unregister_sifive_l2_error_notifier(struct notifier_block *nb);

Ideally these would live in some header file.


> +struct sifive_edac_priv {
> +	struct notifier_block notifier;
> +	struct edac_device_ctl_info *dci;
> +};
> +
> +/**
> + * EDAC error callback
> + *
> + * @event: non-zero if unrecoverable.
> + */
> +static
> +int ecc_err_event(struct notifier_block *this, unsigned long event, void *ptr)
> +{
> +	const char *msg = (char *)ptr;
> +	struct sifive_edac_priv *p;
> +
> +	p = container_of(this, struct sifive_edac_priv, notifier);
> +
> +	if (event)
> +		edac_device_handle_ue(p->dci, 0, 0, msg);
> +	else
> +		edac_device_handle_ce(p->dci, 0, 0, msg);

This would be easier to read if your SIFIVE_L2_ERR_TYPE_UE were exposed via some header file.


> +
> +	return NOTIFY_STOP;

Your notifier register calls are EXPORT_SYMBOL()d, but Kconfig forbids building this as a
module, so its not for this driver. If there is another user of this notifier-chain, won't
NOTIFY_STOP here break it?


> +}
> +
> +static int ecc_register(struct platform_device *pdev)
> +{
> +	struct sifive_edac_priv *p;
> +
> +	p = devm_kzalloc(&pdev->dev, sizeof(*p), GFP_KERNEL);
> +	if (!p)
> +		return -ENOMEM;
> +
> +	p->notifier.notifier_call = ecc_err_event;
> +	platform_set_drvdata(pdev, p);
> +
> +	p->dci = edac_device_alloc_ctl_info(sizeof(*p), "sifive_ecc", 1,

sizeof(*p) here is how much space in struct edac_device_ctl_info you need for private
storage... but you never touch p->dci->pvt_info, so you aren't using it.

0?


> +					    "sifive_ecc", 1, 1, NULL, 0,
> +					    edac_device_alloc_index());
> +	if (IS_ERR(p->dci))
> +		return PTR_ERR(p->dci);
> +
> +	p->dci->dev = &pdev->dev;
> +	p->dci->mod_name = "Sifive ECC Manager";
> +	p->dci->ctl_name = dev_name(&pdev->dev);
> +	p->dci->dev_name = dev_name(&pdev->dev);
> +
> +	if (edac_device_add_device(p->dci)) {
> +		dev_err(p->dci->dev, "failed to register with EDAC core\n");
> +		goto err;
> +	}
> +
> +	register_sifive_l2_error_notifier(&p->notifier);
> +
> +	return 0;
> +
> +err:
> +	edac_device_free_ctl_info(p->dci);
> +
> +	return -ENXIO;
> +}

> +struct platform_device *sifive_pdev;

static?


> +static int __init sifive_edac_init(void)
> +{
> +	int ret;
> +
> +	sifive_pdev = platform_device_register_simple(DRVNAME, 0, NULL, 0);
> +	if (IS_ERR(sifive_pdev))
> +		return PTR_ERR(sifive_pdev);
> +
> +	ret = ecc_register(sifive_pdev);
> +	if (ret)
> +		platform_device_unregister(sifive_pdev);
> +
> +	return ret;
> +}
> +
> +static void __exit sifive_edac_exit(void)
> +{
> +	ecc_unregister(sifive_pdev);
> +	platform_device_unregister(sifive_pdev);
> +}

Looks good to me. I think this patch should go with its two dependencies, I'm not sure why
it got split off...

Reviewed-by: James Morse <james.morse@xxxxxxx>


Thanks,

James