From: Frank Li <Frank.li@oss.nxp.com>
To: Rosen Penev <rosenp@gmail.com>
Cc: dmaengine@vger.kernel.org, Vinod Koul <vkoul@kernel.org>,
Frank Li <Frank.Li@kernel.org>, Stefan Roese <sr@denx.de>,
open list <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] dmaengine: mv_xor: protect MBUS window access with a spinlock
Date: Wed, 16 Sep 2026 14:46:06 -0500 [thread overview]
Message-ID: <aqrx_qKwUCCNxRda@SMW015318> (raw)
In-Reply-To: <CAKxU2N_bJJijZ7QLx=ubaVWA6Mn=q7tswKj=mE8NFJXz=Uf6RQ@mail.gmail.com>
On Wed, Sep 16, 2026 at 12:34:31PM -0700, Rosen Penev wrote:
> On Wed, Sep 16, 2026 at 12:07 PM Frank Li <Frank.li@oss.nxp.com> wrote:
> >
> > On Wed, Sep 16, 2026 at 11:26:08AM -0700, Rosen Penev wrote:
> > > mv_xor_add_io_win() reads and writes shared MBUS window registers
> > > and updates the shared win_start/win_end arrays in mv_xor_device.
> > > Multiple DMA channels can call this function concurrently via
> > > mv_xor_prep_dma_xor(), leading to races where two threads can
> > > select the same free window slot, corrupt the registers, or produce
> > > inconsistent cached state.
> > >
> > > Add a spinlock to struct mv_xor_device and hold it across the
> > > register read-modify-write and cache update in mv_xor_add_io_win().
> > >
> > > Fixes: 77ff7a706f01 ("mv_xor: Add support for IO (PCIe) src/dst areas")
> > > Assisted-by: LLM
> > > Signed-off-by: Rosen Penev <rosenp@gmail.com>
> > > ---
> > > drivers/dma/mv_xor.c | 15 +++++++++++++--
> > > drivers/dma/mv_xor.h | 1 +
> > > 2 files changed, 14 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/dma/mv_xor.c b/drivers/dma/mv_xor.c
> > > index 0de2b1ad5c30..e7c6ec54ce25 100644
> > > --- a/drivers/dma/mv_xor.c
> > > +++ b/drivers/dma/mv_xor.c
> > > @@ -497,11 +497,14 @@ static int mv_xor_add_io_win(struct mv_xor_chan *mv_chan, u32 addr)
> > > u8 target, attr;
> > > int ret;
> > > int i;
> > > + unsigned long flags;
> > >
> > > /* Nothing needs to get done for the Armada 3700 */
> > > if (xordev->xor_type == XOR_ARMADA_37XX)
> > > return 0;
> > >
> > > + spin_lock_irqsave(&xordev->win_lock, flags);
> > > +
> >
> > use auto clean up guard()
> Could that be backported though?
I think it should be fine now. guard() already use quick popular.
Frank
> >
> > Frank
> > > /*
> > > * Loop over the cached windows to check, if the requested area
> > > * is already mapped. If this the case, nothing needs to be done
> > > @@ -511,6 +514,7 @@ static int mv_xor_add_io_win(struct mv_xor_chan *mv_chan, u32 addr)
> > > if (addr >= xordev->win_start[i] &&
> > > addr <= xordev->win_end[i]) {
> > > /* Window is already mapped */
> > > + spin_unlock_irqrestore(&xordev->win_lock, flags);
> > > return 0;
> > > }
> > > }
> > > @@ -521,8 +525,10 @@ static int mv_xor_add_io_win(struct mv_xor_chan *mv_chan, u32 addr)
> > >
> > > /* If no IO window is found that addr has to be located in SDRAM */
> > > ret = mvebu_mbus_get_io_win_info(addr, &size, &target, &attr);
> > > - if (ret < 0)
> > > + if (ret < 0) {
> > > + spin_unlock_irqrestore(&xordev->win_lock, flags);
> > > return 0;
> > > + }
> > >
> > > /*
> > > * Mask the base addr 'addr' according to 'size' read back from the
> > > @@ -540,8 +546,10 @@ static int mv_xor_add_io_win(struct mv_xor_chan *mv_chan, u32 addr)
> > >
> > > /* Set 'i' to the first free window to write the new values to */
> > > i = ffs(~win_enable) - 1;
> > > - if (i >= WINDOW_COUNT)
> > > + if (i >= WINDOW_COUNT) {
> > > + spin_unlock_irqrestore(&xordev->win_lock, flags);
> > > return -ENOMEM;
> > > + }
> > >
> > > writel((addr & 0xffff0000) | (attr << 8) | target,
> > > base + WINDOW_BASE(i));
> > > @@ -556,6 +564,8 @@ static int mv_xor_add_io_win(struct mv_xor_chan *mv_chan, u32 addr)
> > > writel(win_enable, base + WINDOW_BAR_ENABLE(0));
> > > writel(win_enable, base + WINDOW_BAR_ENABLE(1));
> > >
> > > + spin_unlock_irqrestore(&xordev->win_lock, flags);
> > > +
> > > return 0;
> > > }
> > >
> > > @@ -1353,6 +1363,7 @@ static int mv_xor_probe(struct platform_device *pdev)
> > >
> > > platform_set_drvdata(pdev, xordev);
> > >
> > > + spin_lock_init(&xordev->win_lock);
> > >
> > > /*
> > > * We need to know which type of XOR device we use before
> > > diff --git a/drivers/dma/mv_xor.h b/drivers/dma/mv_xor.h
> > > index c87cefd38a07..034db4d0bfb1 100644
> > > --- a/drivers/dma/mv_xor.h
> > > +++ b/drivers/dma/mv_xor.h
> > > @@ -80,6 +80,7 @@ struct mv_xor_device {
> > > struct clk *clk;
> > > struct mv_xor_chan *channels[MV_XOR_MAX_CHANNELS];
> > > int xor_type;
> > > + spinlock_t win_lock;
> > >
> > > u32 win_start[WINDOW_COUNT];
> > > u32 win_end[WINDOW_COUNT];
> > > --
> > > 2.55.0
> > >
prev parent reply other threads:[~2026-09-16 19:46 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-09-16 18:26 Rosen Penev
2026-09-16 19:07 ` Frank Li
2026-09-16 19:34 ` Rosen Penev
2026-09-16 19:46 ` Frank Li [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=aqrx_qKwUCCNxRda@SMW015318 \
--to=frank.li@oss.nxp.com \
--cc=Frank.Li@kernel.org \
--cc=dmaengine@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=rosenp@gmail.com \
--cc=sr@denx.de \
--cc=vkoul@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®