| Message ID | 1617809456-17693-13-git-send-email-yibin.gong@nxp.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | add ecspi ERR009165 for i.mx6/7 soc family | expand |
On 07-04-21, 23:30, Robin Gong wrote: > Add terminated list for keeping descriptor so that it could be freed in > worker without any potential involving next descriptor raised up before > this descriptor freed, because vchan_get_all_descriptors get all > descriptors including the last terminated descriptor and the next > descriptor, hence, the next descriptor maybe freed unexpectly when it's > done in worker without this patch. > https://www.spinics.net/lists/dmaengine/msg23367.html Sound like you should implement .device_synchronize() and do the actual work there..? > > Signed-off-by: Robin Gong <yibin.gong@nxp.com> > Reported-by: Richard Leitner <richard.leitner@skidata.com> > --- > drivers/dma/imx-sdma.c | 17 ++++++++++------- > 1 file changed, 10 insertions(+), 7 deletions(-) > > diff --git a/drivers/dma/imx-sdma.c b/drivers/dma/imx-sdma.c > index 9519b41..4174580 100644 > --- a/drivers/dma/imx-sdma.c > +++ b/drivers/dma/imx-sdma.c > @@ -381,6 +381,7 @@ struct sdma_channel { > enum dma_status status; > struct imx_dma_data data; > struct work_struct terminate_worker; > + struct list_head terminated; > bool is_ram_script; > }; > > @@ -1041,9 +1042,6 @@ static void sdma_channel_terminate_work(struct work_struct *work) > { > struct sdma_channel *sdmac = container_of(work, struct sdma_channel, > terminate_worker); > - unsigned long flags; > - LIST_HEAD(head); > - > /* > * According to NXP R&D team a delay of one BD SDMA cost time > * (maximum is 1ms) should be added after disable of the channel > @@ -1052,10 +1050,7 @@ static void sdma_channel_terminate_work(struct work_struct *work) > */ > usleep_range(1000, 2000); > > - spin_lock_irqsave(&sdmac->vc.lock, flags); > - vchan_get_all_descriptors(&sdmac->vc, &head); > - spin_unlock_irqrestore(&sdmac->vc.lock, flags); > - vchan_dma_desc_free_list(&sdmac->vc, &head); > + vchan_dma_desc_free_list(&sdmac->vc, &sdmac->terminated); > } > > static int sdma_terminate_all(struct dma_chan *chan) > @@ -1069,6 +1064,13 @@ static int sdma_terminate_all(struct dma_chan *chan) > > if (sdmac->desc) { > vchan_terminate_vdesc(&sdmac->desc->vd); > + /* > + * move out current descriptor into terminated list so that > + * it could be free in sdma_channel_terminate_work alone > + * later without potential involving next descriptor raised > + * up before the last descriptor terminated. > + */ > + vchan_get_all_descriptors(&sdmac->vc, &sdmac->terminated); > sdmac->desc = NULL; > schedule_work(&sdmac->terminate_worker); > } > @@ -2075,6 +2077,7 @@ static int sdma_probe(struct platform_device *pdev) > > sdmac->channel = i; > sdmac->vc.desc_free = sdma_desc_free; > + INIT_LIST_HEAD(&sdmac->terminated); > INIT_WORK(&sdmac->terminate_worker, > sdma_channel_terminate_work); > /* > -- > 2.7.4
On 07-04-21, 23:30, Robin Gong wrote: > Add terminated list for keeping descriptor so that it could be freed in > worker without any potential involving next descriptor raised up before > this descriptor freed, because vchan_get_all_descriptors get all > descriptors including the last terminated descriptor and the next > descriptor, hence, the next descriptor maybe freed unexpectly when it's > done in worker without this patch. > https://www.spinics.net/lists/dmaengine/msg23367.html Acked-By: Vinod Koul <vkoul@kernel.org> -- ~Vinod
diff --git a/drivers/dma/imx-sdma.c b/drivers/dma/imx-sdma.c index 9519b41..4174580 100644 --- a/drivers/dma/imx-sdma.c +++ b/drivers/dma/imx-sdma.c @@ -381,6 +381,7 @@ struct sdma_channel { enum dma_status status; struct imx_dma_data data; struct work_struct terminate_worker; + struct list_head terminated; bool is_ram_script; }; @@ -1041,9 +1042,6 @@ static void sdma_channel_terminate_work(struct work_struct *work) { struct sdma_channel *sdmac = container_of(work, struct sdma_channel, terminate_worker); - unsigned long flags; - LIST_HEAD(head); - /* * According to NXP R&D team a delay of one BD SDMA cost time * (maximum is 1ms) should be added after disable of the channel @@ -1052,10 +1050,7 @@ static void sdma_channel_terminate_work(struct work_struct *work) */ usleep_range(1000, 2000); - spin_lock_irqsave(&sdmac->vc.lock, flags); - vchan_get_all_descriptors(&sdmac->vc, &head); - spin_unlock_irqrestore(&sdmac->vc.lock, flags); - vchan_dma_desc_free_list(&sdmac->vc, &head); + vchan_dma_desc_free_list(&sdmac->vc, &sdmac->terminated); } static int sdma_terminate_all(struct dma_chan *chan) @@ -1069,6 +1064,13 @@ static int sdma_terminate_all(struct dma_chan *chan) if (sdmac->desc) { vchan_terminate_vdesc(&sdmac->desc->vd); + /* + * move out current descriptor into terminated list so that + * it could be free in sdma_channel_terminate_work alone + * later without potential involving next descriptor raised + * up before the last descriptor terminated. + */ + vchan_get_all_descriptors(&sdmac->vc, &sdmac->terminated); sdmac->desc = NULL; schedule_work(&sdmac->terminate_worker); } @@ -2075,6 +2077,7 @@ static int sdma_probe(struct platform_device *pdev) sdmac->channel = i; sdmac->vc.desc_free = sdma_desc_free; + INIT_LIST_HEAD(&sdmac->terminated); INIT_WORK(&sdmac->terminate_worker, sdma_channel_terminate_work); /*
Add terminated list for keeping descriptor so that it could be freed in worker without any potential involving next descriptor raised up before this descriptor freed, because vchan_get_all_descriptors get all descriptors including the last terminated descriptor and the next descriptor, hence, the next descriptor maybe freed unexpectly when it's done in worker without this patch. https://www.spinics.net/lists/dmaengine/msg23367.html Signed-off-by: Robin Gong <yibin.gong@nxp.com> Reported-by: Richard Leitner <richard.leitner@skidata.com> --- drivers/dma/imx-sdma.c | 17 ++++++++++------- 1 file changed, 10 insertions(+), 7 deletions(-)