[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAPDyKFr8nkVCQB63rZ6ZcHeTrmQno7aYv2Pa3PmZwzW9Ei3HDQ@mail.gmail.com>
Date: Thu, 23 Sep 2021 20:12:38 +0200
From: Ulf Hansson <ulf.hansson@...aro.org>
To: Xin Xiong <xiongx18@...an.edu.cn>
Cc: linux-mmc <linux-mmc@...r.kernel.org>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
yuanxzhang@...an.edu.cn, Xiyu Yang <xiyuyang19@...an.edu.cn>,
Xin Tan <tanxin.ctf@...il.com>
Subject: Re: [PATCH] drivers/mmc: fix reference count leaks in moxart_probe
On Thu, 16 Sept 2021 at 20:19, Xin Xiong <xiongx18@...an.edu.cn> wrote:
>
> The issue happens in several error handling paths on two refcounted
> object related to the object "host" (dma_chan_rx, dma_chan_tx). In
> these paths, the function forgets to decrement the reference count of
> one or both objects' reference count increased earlier by
> dma_request_chan(), causing reference count leaks.
>
> Fix it by decreasing reference counts of both objects in each path
> separately.
>
> Signed-off-by: Xin Xiong <xiongx18@...an.edu.cn>
> Signed-off-by: Xiyu Yang <xiyuyang19@...an.edu.cn>
> Signed-off-by: Xin Tan <tanxin.ctf@...il.com>
> ---
> drivers/mmc/host/moxart-mmc.c | 38 ++++++++++++++++++++++++++---------
> 1 file changed, 28 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/mmc/host/moxart-mmc.c b/drivers/mmc/host/moxart-mmc.c
> index 6c9d38132..b5aa1010c 100644
> --- a/drivers/mmc/host/moxart-mmc.c
> +++ b/drivers/mmc/host/moxart-mmc.c
> @@ -606,7 +606,28 @@ static int moxart_probe(struct platform_device *pdev)
> host->sysclk = clk_get_rate(clk);
> host->fifo_width = readl(host->base + REG_FEATURE) << 2;
> host->dma_chan_tx = dma_request_chan(dev, "tx");
> + if (IS_ERR(host->dma_chan_tx)) {
> + if (PTR_ERR(host->dma_chan_tx) == -EPROBE_DEFER) {
> + ret = -EPROBE_DEFER;
> + goto out;
> + }
> + }
> +
> host->dma_chan_rx = dma_request_chan(dev, "rx");
> + if (IS_ERR(host->dma_chan_rx)) {
> + if (!IS_ERR(host->dma_chan_tx))
> + dma_release_channel(host->dma_chan_tx);
> + if (PTR_ERR(host->dma_chan_rx) == -EPROBE_DEFER) {
> + ret = -EPROBE_DEFER;
> + goto out;
> + }
> + dev_dbg(dev, "PIO mode transfer enabled\n");
> + host->have_dma = false;
> + } else if (IS_ERR(host->dma_chan_tx)) {
> + dma_release_channel(host->chan_rx);
> + dev_dbg(dev, "PIO mode transfer enabled\n");
> + host->have_dma = false;
> + }
>
> spin_lock_init(&host->lock);
>
> @@ -615,15 +636,7 @@ static int moxart_probe(struct platform_device *pdev)
> mmc->f_min = DIV_ROUND_CLOSEST(host->sysclk, CLK_DIV_MASK * 2);
> mmc->ocr_avail = 0xffff00; /* Support 2.0v - 3.6v power. */
>
> - if (IS_ERR(host->dma_chan_tx) || IS_ERR(host->dma_chan_rx)) {
> - if (PTR_ERR(host->dma_chan_tx) == -EPROBE_DEFER ||
> - PTR_ERR(host->dma_chan_rx) == -EPROBE_DEFER) {
> - ret = -EPROBE_DEFER;
> - goto out;
> - }
> - dev_dbg(dev, "PIO mode transfer enabled\n");
> - host->have_dma = false;
> - } else {
> + if (!IS_ERR(host->dma_chan_tx) && !IS_ERR(host->dma_chan_rx)) {
> dev_dbg(dev, "DMA channels found (%p,%p)\n",
> host->dma_chan_tx, host->dma_chan_rx);
> host->have_dma = true;
> @@ -664,8 +677,13 @@ static int moxart_probe(struct platform_device *pdev)
> }
>
> ret = devm_request_irq(dev, irq, moxart_irq, 0, "moxart-mmc", host);
> - if (ret)
> + if (ret) {
> + if (host->have_dma) {
> + dma_release_channel(host->dma_chan_tx);
> + dma_release_channel(host->dma_chan_rx);
> + }
> goto out;
> + }
You are right, that we should call dma_release_channel in some error
paths. Although, the above looks rather messy to me. May I suggest
that we add the error handling below the "out" label instead. Along
the lines of this:
if (!IS_ERR(host->dma_chan_tx))
dma_release_channel(host->dma_chan_tx);
if (!IS_ERR(host->dma_chan_rx))
dma_release_channel(host->dma_chan_rx);
>
> dev_set_drvdata(dev, mmc);
> mmc_add_host(mmc);
> --
> 2.25.1
>
Kind regards
Uffe
Powered by blists - more mailing lists