Re: [PATCH v3 1/1] dma: actions: Fix lockdep splat for owl-dma

From: Cristian Ciocaltea
Date: Sat May 02 2020 - 13:35:07 EST

On Sat, May 02, 2020 at 05:53:33PM +0530, Vinod Koul wrote:
> Hi Cristian,
>
> On 29-04-20, 18:28, Cristian Ciocaltea wrote:
> > When the kernel is built with lockdep support and the owl-dma driver is
> > used, the following message is shown:
>
> First the patch title needs upate, we describe the patch in the title
> and not the cause. So use correct lock, or use od lock might be better
> titles, pls revise.
>
> Second, the susbsystem is named dmaengine:... not dma:.. You can always
> check that by using git log on the respective file
>
> Pls do add fixes and further acks received on next iteration.
>

Hi Vinod,

Thank you for reviewing and sorry for the mistakes! I'll be more careful
next time with all those details.

I've submitted the updated patch: [PATCH v4 1/1] dmaengine: owl: Use
correct lock in owl_dma_get_pchan()

Kind regards,
Cristi

> >
> > [ 2.496939] INFO: trying to register non-static key.
> > [ 2.501889] the code is fine but needs lockdep annotation.
> > [ 2.507357] turning off the locking correctness validator.
> > [ 2.512834] CPU: 0 PID: 12 Comm: kworker/0:1 Not tainted 5.6.3+ #15
> > [ 2.519084] Hardware name: Generic DT based system
> > [ 2.523878] Workqueue: events_freezable mmc_rescan
> > [ 2.528681] [<801127f0>] (unwind_backtrace) from [<8010da58>] (show_stack+0x10/0x14)
> > [ 2.536420] [<8010da58>] (show_stack) from [<8080fbe8>] (dump_stack+0xb4/0xe0)
> > [ 2.543645] [<8080fbe8>] (dump_stack) from [<8017efa4>] (register_lock_class+0x6f0/0x718)
> > [ 2.551816] [<8017efa4>] (register_lock_class) from [<8017b7d0>] (__lock_acquire+0x78/0x25f0)
> > [ 2.560330] [<8017b7d0>] (__lock_acquire) from [<8017e5e4>] (lock_acquire+0xd8/0x1f4)
> > [ 2.568159] [<8017e5e4>] (lock_acquire) from [<80831fb0>] (_raw_spin_lock_irqsave+0x3c/0x50)
> > [ 2.576589] [<80831fb0>] (_raw_spin_lock_irqsave) from [<8051b5fc>] (owl_dma_issue_pending+0xbc/0x120)
> > [ 2.585884] [<8051b5fc>] (owl_dma_issue_pending) from [<80668cbc>] (owl_mmc_request+0x1b0/0x390)
> > [ 2.594655] [<80668cbc>] (owl_mmc_request) from [<80650ce0>] (mmc_start_request+0x94/0xbc)
> > [ 2.602906] [<80650ce0>] (mmc_start_request) from [<80650ec0>] (mmc_wait_for_req+0x64/0xd0)
> > [ 2.611245] [<80650ec0>] (mmc_wait_for_req) from [<8065aa10>] (mmc_app_send_scr+0x10c/0x144)
> > [ 2.619669] [<8065aa10>] (mmc_app_send_scr) from [<80659b3c>] (mmc_sd_setup_card+0x4c/0x318)
> > [ 2.628092] [<80659b3c>] (mmc_sd_setup_card) from [<80659f0c>] (mmc_sd_init_card+0x104/0x430)
> > [ 2.636601] [<80659f0c>] (mmc_sd_init_card) from [<8065a3e0>] (mmc_attach_sd+0xcc/0x16c)
> > [ 2.644678] [<8065a3e0>] (mmc_attach_sd) from [<8065301c>] (mmc_rescan+0x3ac/0x40c)
> > [ 2.652332] [<8065301c>] (mmc_rescan) from [<80143244>] (process_one_work+0x2d8/0x780)
> > [ 2.660239] [<80143244>] (process_one_work) from [<80143730>] (worker_thread+0x44/0x598)
> > [ 2.668323] [<80143730>] (worker_thread) from [<8014b5f8>] (kthread+0x148/0x150)
> > [ 2.675708] [<8014b5f8>] (kthread) from [<801010b4>] (ret_from_fork+0x14/0x20)
> > [ 2.682912] Exception stack(0xee8fdfb0 to 0xee8fdff8)
> > [ 2.687954] dfa0: 00000000 00000000 00000000 00000000
> > [ 2.696118] dfc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
> > [ 2.704277] dfe0: 00000000 00000000 00000000 00000000 00000013 00000000
> >
> > The obvious fix would be to use 'spin_lock_init()' on 'pchan->lock'
> > before attempting to call 'spin_lock_irqsave()' in 'owl_dma_get_pchan()'.
> >
> > However, according to Manivannan Sadhasivam, 'pchan->lock' was supposed
> > to only protect 'pchan->vchan' while 'od->lock' does a similar job in
> > 'owl_dma_terminate_pchan'.
> >
> > Therefore, this patch will simply substitute 'pchan->lock' with 'od->lock'
> > and removes the 'lock' attribute in 'owl_dma_pchan' struct.
> >
> > Signed-off-by: Cristian Ciocaltea <cristian.ciocaltea@xxxxxxxxx>
> > Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>
> > ---
> > Changes in v3:
> > * Get rid of the kerneldoc comment for the removed struct attribute
> > * Add the Reviewed-by tag in the commit message
> >
> > Changes in v2:
> > * Improve the fix as suggested by Manivannan Sadhasivam: substitute
> > 'pchan->lock' with 'od->lock' and get rid of the 'lock' attribute in
> > 'owl_dma_pchan' struct
> > * Update the commit message to reflect the changes
> >
> > drivers/dma/owl-dma.c | 8 +++-----
> > 1 file changed, 3 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/dma/owl-dma.c b/drivers/dma/owl-dma.c
> > index c683051257fd..66ef70b00ec0 100644
> > --- a/drivers/dma/owl-dma.c
> > +++ b/drivers/dma/owl-dma.c
> > @@ -175,13 +175,11 @@ struct owl_dma_txd {
> > * @id: physical index to this channel
> > * @base: virtual memory base for the dma channel
> > * @vchan: the virtual channel currently being served by this physical channel
> > - * @lock: a lock to use when altering an instance of this struct
> > */
> > struct owl_dma_pchan {
> > u32 id;
> > void __iomem *base;
> > struct owl_dma_vchan *vchan;
> > - spinlock_t lock;
> > };
> >
> > /**
> > @@ -437,14 +435,14 @@ static struct owl_dma_pchan *owl_dma_get_pchan(struct owl_dma *od,
> > for (i = 0; i < od->nr_pchans; i++) {
> > pchan = &od->pchans[i];
> >
> > - spin_lock_irqsave(&pchan->lock, flags);
> > + spin_lock_irqsave(&od->lock, flags);
> > if (!pchan->vchan) {
> > pchan->vchan = vchan;
> > - spin_unlock_irqrestore(&pchan->lock, flags);
> > + spin_unlock_irqrestore(&od->lock, flags);
> > break;
> > }
> >
> > - spin_unlock_irqrestore(&pchan->lock, flags);
> > + spin_unlock_irqrestore(&od->lock, flags);
> > }
> >
> > return pchan;
> > --
> > 2.26.2
>
> --
> ~Vinod