From: Liu Qiang-B32616 Subject: RE: [PATCH v5 3/6] fsl-dma: change release process of dma descriptor for supporting async_tx Date: Thu, 2 Aug 2012 11:13:26 +0000 Message-ID: References: <1343810957-25378-1-git-send-email-qiang.liu@freescale.com> <20120801172526.GD11359@ovro.caltech.edu> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Cc: "herbert@gondor.apana.org.au" , Vinod Koul , "linux-kernel@vger.kernel.org" , "dan.j.williams@gmail.com" , "linux-crypto@vger.kernel.org" , Dan Williams , "linuxppc-dev@lists.ozlabs.org" , "davem@davemloft.net" To: "Ira W. Snyder" Return-path: In-Reply-To: <20120801172526.GD11359@ovro.caltech.edu> Content-Language: en-US List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: linuxppc-dev-bounces+glppe-linuxppc-embedded-2=m.gmane.org@lists.ozlabs.org Sender: "Linuxppc-dev" List-Id: linux-crypto.vger.kernel.org Hi Ira, Here, I want to talk about the issue of dma lock when use dmatest with original patch. I do some tests on p1022ds, 2 cores, 6 dma channels (actually is 4 channels, I am investigating why is 6, but it doesn't matter). I would like to share with you to find something:) First, it is easy to reproduce your issue with 6 channels after up to 200,000 iterations per channel, there is not any response from terminal, then I run top to see the 2 cores load, I found cpu load is 100%, terminal won't update any info after a long time; Second, I only enable 4 channels to run dmatest, I found cpu load is 50% per channel, terminal also update info after a long time; Last, I only enable 1 channel to run dmatest, serial interrupt is attached to core1, and dma channel interrupt is attached to core0, I found terminal is very slow after boot up, but it's only slow whatever how many interrupts raised by the only dma channel (over 900,000 iterations). So I want to know, our test results are same? How do you know dma engine is self locked but not for heavy cpu load, there is no chance to schedule? I also read dmatest.c, there is not any sleep to release cpu (msleep(100) if tx is null). If our test results are same, I think I can explain why your original patch can "fix" the issue of dma lock, because your patch avoid mass cpu data operation. 83xx is old, and P1 series is low end chip of Freescale, this maybe the reason why p4080 cannot reproduce this issue. You can have a test to verify my point if you are available. I will follow this issue until it's resolved. I will let you know if any other new clues. Thanks. > -----Original Message----- > From: Ira W. Snyder [mailto:iws@ovro.caltech.edu] > Sent: Thursday, August 02, 2012 1:25 AM > To: Liu Qiang-B32616 > Cc: linux-crypto@vger.kernel.org; linuxppc-dev@lists.ozlabs.org; linux- > kernel@vger.kernel.org; dan.j.williams@gmail.com; Vinod Koul; > herbert@gondor.hengli.com.au; Dan Williams; davem@davemloft.net > Subject: Re: [PATCH v5 3/6] fsl-dma: change release process of dma > descriptor for supporting async_tx > > On Wed, Aug 01, 2012 at 04:49:17PM +0800, qiang.liu@freescale.com wrote: > > From: Qiang Liu > > > > Fix the potential risk when enable config NET_DMA and ASYNC_TX. > > Async_tx is lack of support in current release process of dma > descriptor, > > all descriptors will be released whatever is acked or no-acked by > async_tx, > > so there is a potential race condition when dma engine is uesd by > others > > clients (e.g. when enable NET_DMA to offload TCP). > > > > In our case, a race condition which is raised when use both of talitos > > and dmaengine to offload xor is because napi scheduler will sync all > > pending requests in dma channels, it affects the process of raid > operations > > due to ack_tx is not checked in fsl dma. The no-acked descriptor is > freed > > which is submitted just now, as a dependent tx, this freed descriptor > trigger > > BUG_ON(async_tx_test_ack(depend_tx)) in async_tx_submit(). > > > > TASK = ee1a94a0[1390] 'md0_raid5' THREAD: ecf40000 CPU: 0 > > GPR00: 00000001 ecf41ca0 ee44/921a94a0 0000003f 00000001 c00593e4 > 00000000 00000001 > > GPR08: 00000000 a7a7a7a7 00000001 045/920000002 42028042 100a38d4 > ed576d98 00000000 > > GPR16: ed5a11b0 00000000 2b162000 00000200 046/920000000 2d555000 > ed3015e8 c15a7aa0 > > GPR24: 00000000 c155fc40 00000000 ecb63220 ecf41d28 e47/92f640bb0 > ef640c30 ecf41ca0 > > NIP [c02b048c] async_tx_submit+0x6c/0x2b4 > > LR [c02b068c] async_tx_submit+0x26c/0x2b4 > > Call Trace: > > [ecf41ca0] [c02b068c] async_tx_submit+0x26c/0x2b448/92 (unreliable) > > [ecf41cd0] [c02b0a4c] async_memcpy+0x240/0x25c > > [ecf41d20] [c0421064] async_copy_data+0xa0/0x17c > > [ecf41d70] [c0421cf4] __raid_run_ops+0x874/0xe10 > > [ecf41df0] [c0426ee4] handle_stripe+0x820/0x25e8 > > [ecf41e90] [c0429080] raid5d+0x3d4/0x5b4 > > [ecf41f40] [c04329b8] md_thread+0x138/0x16c > > [ecf41f90] [c008277c] kthread+0x8c/0x90 > > [ecf41ff0] [c0011630] kernel_thread+0x4c/0x68 > > > > Another major modification in this patch is the change to completed > descriptors, > > there is a potential risk which caused by exception interrupt, all > descriptors > > in ld_running list are seemed completed when an interrupt raised, it > works fine > > under normal condition, but if there is an exception occured, it cannot > work > > as our excepted. Hardware should not depend on s/w list, the right way > is > > to read current descriptor address register to find the last completed > > descriptor. If an interrupt is raised by an error, all descriptors in > ld_running > > should not be seemed finished, or these unfinished descriptors in > ld_running > > will be released wrongly. > > > > A simple way to reproduce, > > Enable dmatest first, then insert some bad descriptors which can > trigger > > Programming Error interrupts before the good descriptors. Last, the > good > > descriptors will be freed before they are processsed because of the > exception > > intrerrupt. > > > > Note: the bad descriptors are only for simulating an exception > interrupt. > > This case can illustrate the potential risk in current fsl-dma very > well. > > > > I've never managed to trigger a PE (programming error) interrupt on the > 83xx hardware. Any time I intentionally caused an error, the hardware > wedged itself. The CB (channel busy) bit is stuck high, and cannot be > cleared without a hard reset of the board. > > I agree the "snoop on the hardware" technique works. As far as I can > tell, you have implemented the code correctly. > > The MPC8349EARM.pdf from Freescale indicates that the hardware will halt > in response to a programming error, and generate a PE interrupt. See > section 12.5.3.3 (pg 568). > > The driver, as it is written, will never recover from such a condition. > Since you are complaining about this situation, do you intend to fix it? > > > Cc: Dan Williams > > Cc: Dan Williams > > Cc: Vinod Koul > > Cc: Li Yang > > Cc: Ira W. Snyder > > Signed-off-by: Qiang Liu > > --- > > drivers/dma/fsldma.c | 242 +++++++++++++++++++++++++++++++++++------- > -------- > > drivers/dma/fsldma.h | 1 + > > 2 files changed, 172 insertions(+), 71 deletions(-) > > > > diff --git a/drivers/dma/fsldma.c b/drivers/dma/fsldma.c > > index 4f2f212..87f52c0 100644 > > --- a/drivers/dma/fsldma.c > > +++ b/drivers/dma/fsldma.c > > @@ -400,6 +400,125 @@ out_splice: > > list_splice_tail_init(&desc->tx_list, &chan->ld_pending); > > } > > > > +static void fsldma_cleanup_descriptor(struct fsldma_chan *chan); > > +static void fsl_chan_xfer_ld_queue(struct fsldma_chan *chan); > > + > > As noted in my reply to patch 4/6, please swap the order of this patch > and the following patch. > > These lines should not be added or removed in either patch. > > > +/** > > + * fsldma_clean_completed_descriptor - free all descriptors which > > + * has been completed and acked > > + * @chan: Freescale DMA channel > > + * > > + * This function is used on all completed and acked descriptors. > > + * All descriptors should only be freed in this function. > > + */ > > +static int > > +fsldma_clean_completed_descriptor(struct fsldma_chan *chan) > > This should be 'static void'. It does not return an error code. > > > +{ > > + struct fsl_desc_sw *desc, *_desc; > > + > > + /* Run the callback for each descriptor, in order */ > > + list_for_each_entry_safe(desc, _desc, &chan->ld_completed, node) { > > + > > + if (async_tx_test_ack(&desc->async_tx)) { > > + /* Remove from the list of transactions */ > > + list_del(&desc->node); > > +#ifdef FSL_DMA_LD_DEBUG > > + chan_dbg(chan, "LD %p free\n", desc); > > +#endif > > + dma_pool_free(chan->desc_pool, desc, > > + desc->async_tx.phys); > > This code appears in multiple places in the driver. Please consider > adding my patch 3/7 titled "[PATCH 3/7] fsl-dma: add > fsl_dma_free_descriptor() to reduce code duplication" to your patch > series. > > > + } > > + } > > + > > + return 0; > > +} > > + > > +/** > > + * fsldma_run_tx_complete_actions - cleanup and free a single link > descriptor > > This documentation is incorrect. This code NEVER frees a descriptor. > > > + * @chan: Freescale DMA channel > > + * @desc: descriptor to cleanup and free > > + * @cookie: Freescale DMA transaction identifier > > + * > > + * This function is used on a descriptor which has been executed by > the DMA > > + * controller. It will run any callbacks, submit any dependencies. > > + */ > > +static dma_cookie_t fsldma_run_tx_complete_actions(struct fsl_desc_sw > *desc, > > + struct fsldma_chan *chan, dma_cookie_t cookie) > > Please change the parameter order to: > > static dma_cookie_t fsldma_run_tx_complete_actions(struct fsldma_chan > *chan, > struct fsl_desc_sw *desc, dma_cookie_t cookie) > > Every other function in the driver uses this parameter order. Channel > comes first, then descriptor. > > > +{ > > + struct dma_async_tx_descriptor *txd = &desc->async_tx; > > + struct device *dev = chan->common.device->dev; > > + dma_addr_t src = get_desc_src(chan, desc); > > + dma_addr_t dst = get_desc_dst(chan, desc); > > + u32 len = get_desc_cnt(chan, desc); > > + > > + BUG_ON(txd->cookie < 0); > > + > > + if (txd->cookie > 0) { > > It will significantly reduce your patch size if you move this if > statement to the function which calls this one. I've provided an example > down below, in the one place where this code is used. > > > + cookie = txd->cookie; > > + > > + /* Run the link descriptor callback function */ > > + if (txd->callback) { > > +#ifdef FSL_DMA_LD_DEBUG > > + chan_dbg(chan, "LD %p callback\n", desc); > > +#endif > > + txd->callback(txd->callback_param); > > + } > > + > > + /* Unmap the dst buffer, if requested */ > > + if (!(txd->flags & DMA_COMPL_SKIP_DEST_UNMAP)) { > > + if (txd->flags & DMA_COMPL_DEST_UNMAP_SINGLE) > > + dma_unmap_single(dev, dst, len, DMA_FROM_DEVICE); > > + else > > + dma_unmap_page(dev, dst, len, DMA_FROM_DEVICE); > > + } > > + > > + /* Unmap the src buffer, if requested */ > > + if (!(txd->flags & DMA_COMPL_SKIP_SRC_UNMAP)) { > > + if (txd->flags & DMA_COMPL_SRC_UNMAP_SINGLE) > > + dma_unmap_single(dev, src, len, DMA_TO_DEVICE); > > + else > > + dma_unmap_page(dev, src, len, DMA_TO_DEVICE); > > + } > > + } > > + > > + /* Run any dependencies */ > > + dma_run_dependencies(txd); > > + > > + return cookie; > > +} > > + > > +/** > > + * fsldma_clean_running_descriptor - move the completed descriptor > from > > + * ld_running to ld_completed > > + * @chan: Freescale DMA channel > > + * @desc: the descriptor which is completed > > + * > > + * Free the descriptor directly if acked by async_tx api, or move it > to > > + * queue ld_completed. > > + */ > > +static int > > This code never returns an error code. It should be 'static void'. > > > +fsldma_clean_running_descriptor(struct fsldma_chan *chan, > > + struct fsl_desc_sw *desc) > > +{ > > + /* Remove from the list of transactions */ > > + list_del(&desc->node); > > + /* > > + * the client is allowed to attach dependent operations > > + * until 'ack' is set > > + */ > > + if (!async_tx_test_ack(&desc->async_tx)) { > > + /* > > + * Move this descriptor to the list of descriptors which is > > + * completed, but still awaiting the 'ack' bit to be set. > > + */ > > + list_add_tail(&desc->node, &chan->ld_completed); > > + return 0; > > + } > > + > > + dma_pool_free(chan->desc_pool, desc, desc->async_tx.phys); > > + return 0; > > +} > > + > > static dma_cookie_t fsl_dma_tx_submit(struct dma_async_tx_descriptor > *tx) > > { > > struct fsldma_chan *chan = to_fsl_chan(tx->chan); > > @@ -534,8 +653,10 @@ static void fsl_dma_free_chan_resources(struct > dma_chan *dchan) > > > > chan_dbg(chan, "free all channel resources\n"); > > spin_lock_irqsave(&chan->desc_lock, flags); > > + fsldma_cleanup_descriptor(chan); > > fsldma_free_desc_list(chan, &chan->ld_pending); > > fsldma_free_desc_list(chan, &chan->ld_running); > > + fsldma_free_desc_list(chan, &chan->ld_completed); > > spin_unlock_irqrestore(&chan->desc_lock, flags); > > > > dma_pool_destroy(chan->desc_pool); > > @@ -819,46 +940,53 @@ static int fsl_dma_device_control(struct dma_chan > *dchan, > > * controller. It will run any callbacks, submit any dependencies, and > then > > * free the descriptor. > > */ > > This documentation is now wrong. This function no longer operates on a > single descriptor. It operates on all descriptors in ld_running and > ld_completed. > > Please fix the documentation, and add locking notes. > > > -static void fsldma_cleanup_descriptor(struct fsldma_chan *chan, > > - struct fsl_desc_sw *desc) > > +static void fsldma_cleanup_descriptor(struct fsldma_chan *chan) > > I think the name should change to fsldma_cleanup_descriptors(). It > cleans up one or more descriptors now. > > > { > > - struct dma_async_tx_descriptor *txd = &desc->async_tx; > > - struct device *dev = chan->common.device->dev; > > - dma_addr_t src = get_desc_src(chan, desc); > > - dma_addr_t dst = get_desc_dst(chan, desc); > > - u32 len = get_desc_cnt(chan, desc); > > + struct fsl_desc_sw *desc, *_desc; > > + dma_cookie_t cookie = 0; > > + dma_addr_t curr_phys = get_cdar(chan); > > + int idle = dma_is_idle(chan); > > + int seen_current = 0; > > > > The hardware can advance quite a bit between here, where you save the > current descriptor address and idle status. > > > - /* Run the link descriptor callback function */ > > - if (txd->callback) { > > -#ifdef FSL_DMA_LD_DEBUG > > - chan_dbg(chan, "LD %p callback\n", desc); > > -#endif > > - txd->callback(txd->callback_param); > > - } > > + fsldma_clean_completed_descriptor(chan); > > > > - /* Run any dependencies */ > > - dma_run_dependencies(txd); > > + /* Run the callback for each descriptor, in order */ > > + list_for_each_entry_safe(desc, _desc, &chan->ld_running, node) { > > + /* > > + * do not advance past the current descriptor loaded into the > > + * hardware channel, subsequent descriptors are either in > > + * process or have not been submitted > > + */ > > + if (seen_current) > > + break; > > > > - /* Unmap the dst buffer, if requested */ > > - if (!(txd->flags & DMA_COMPL_SKIP_DEST_UNMAP)) { > > - if (txd->flags & DMA_COMPL_DEST_UNMAP_SINGLE) > > - dma_unmap_single(dev, dst, len, DMA_FROM_DEVICE); > > - else > > - dma_unmap_page(dev, dst, len, DMA_FROM_DEVICE); > > - } > > + /* > > + * stop the search if we reach the current descriptor and the > > + * channel is busy > > + */ > > + if (desc->async_tx.phys == curr_phys) { > > + seen_current = 1; > > + if (!idle) > > + break; > > + } > > And here, where you check the current descriptor address and idle > status. > > Should this change to: > > if (desc->async_tx.phys == get_cdar(chan)) { > seen_current = 1; > if (!dma_is_idle(chan)) > break; > } > > > + > > + cookie = fsldma_run_tx_complete_actions(desc, chan, cookie); > > + > > I would prefer if the code just kept track of the cookie here, rather > than passing it through this function call. This code also illustrates > how you can remove the "if (txd->cookie > 0)" check from > fsldma_run_tx_complete_actions() to reduce the patch size. > > /* > * Only descriptors with non-zero cookies need their completion > * actions run. > */ > if (desc->async_tx.cookie > 0) { > cookie = desc->async_tx.cookie; > fsldma_run_tx_complete_actions(chan, desc); > desc->async_tx.cookie = 0; > } > > /* This descriptor has been ACKed, free it */ > if (async_tx_test_ack(&desc->async_tx)) { > fsl_dma_free_descriptor(chan, desc); > continue; > } > > /* > * This descriptor was not ACKed, add it to the ld_completed > * list, to be freed after the ACK bit is set. > */ > list_del(&desc->node); > list_add_tail(&desc->node, &chan->ld_completed); > > > > + if (fsldma_clean_running_descriptor(chan, desc)) > > + break; > > > > This if statement will never trigger. fsldma_clean_running_descriptor() > only returns 0. It is useless. > > > - /* Unmap the src buffer, if requested */ > > - if (!(txd->flags & DMA_COMPL_SKIP_SRC_UNMAP)) { > > - if (txd->flags & DMA_COMPL_SRC_UNMAP_SINGLE) > > - dma_unmap_single(dev, src, len, DMA_TO_DEVICE); > > - else > > - dma_unmap_page(dev, src, len, DMA_TO_DEVICE); > > } > > > > -#ifdef FSL_DMA_LD_DEBUG > > - chan_dbg(chan, "LD %p free\n", desc); > > -#endif > > - dma_pool_free(chan->desc_pool, desc, txd->phys); > > + /* > > + * Start any pending transactions automatically > > + * > > + * In the ideal case, we keep the DMA controller busy while we go > > + * ahead and free the descriptors below. > > + */ > > + fsl_chan_xfer_ld_queue(chan); > > + > > + if (cookie > 0) > > + chan->common.completed_cookie = cookie; > > } > > > > /** > > @@ -954,11 +1082,15 @@ static enum dma_status fsl_tx_status(struct > dma_chan *dchan, > > enum dma_status ret; > > unsigned long flags; > > > > - spin_lock_irqsave(&chan->desc_lock, flags); > > ret = dma_cookie_status(dchan, cookie, txstate); > > + if (ret == DMA_SUCCESS) > > + return ret; > > + > > + spin_lock_irqsave(&chan->desc_lock, flags); > > + fsldma_cleanup_descriptor(chan); > > spin_unlock_irqrestore(&chan->desc_lock, flags); > > > > - return ret; > > + return dma_cookie_status(dchan, cookie, txstate); > > } > > > > /*-------------------------------------------------------------------- > --------*/ > > @@ -1035,52 +1167,19 @@ static irqreturn_t fsldma_chan_irq(int irq, > void *data) > > static void dma_do_tasklet(unsigned long data) > > { > > struct fsldma_chan *chan = (struct fsldma_chan *)data; > > - struct fsl_desc_sw *desc, *_desc; > > - LIST_HEAD(ld_cleanup); > > unsigned long flags; > > > > chan_dbg(chan, "tasklet entry\n"); > > > > spin_lock_irqsave(&chan->desc_lock, flags); > > > > - /* update the cookie if we have some descriptors to cleanup */ > > - if (!list_empty(&chan->ld_running)) { > > - dma_cookie_t cookie; > > - > > - desc = to_fsl_desc(chan->ld_running.prev); > > - cookie = desc->async_tx.cookie; > > - dma_cookie_complete(&desc->async_tx); > > - > > - chan_dbg(chan, "completed_cookie=%d\n", cookie); > > - } > > - > > - /* > > - * move the descriptors to a temporary list so we can drop the lock > > - * during the entire cleanup operation > > - */ > > - list_splice_tail_init(&chan->ld_running, &ld_cleanup); > > - > > /* the hardware is now idle and ready for more */ > > chan->idle = true; > > > > - /* > > - * Start any pending transactions automatically > > - * > > - * In the ideal case, we keep the DMA controller busy while we go > > - * ahead and free the descriptors below. > > - */ > > - fsl_chan_xfer_ld_queue(chan); > > - spin_unlock_irqrestore(&chan->desc_lock, flags); > > - > > - /* Run the callback for each descriptor, in order */ > > - list_for_each_entry_safe(desc, _desc, &ld_cleanup, node) { > > + /* Run all cleanup for this descriptor */ > > Nitpick. This should be: > > /* Run cleanup for all descriptors */ > > > + fsldma_cleanup_descriptor(chan); > > > > - /* Remove from the list of transactions */ > > - list_del(&desc->node); > > - > > - /* Run all cleanup for this descriptor */ > > - fsldma_cleanup_descriptor(chan, desc); > > - } > > + spin_unlock_irqrestore(&chan->desc_lock, flags); > > > > chan_dbg(chan, "tasklet exit\n"); > > } > > @@ -1262,6 +1361,7 @@ static int __devinit fsl_dma_chan_probe(struct > fsldma_device *fdev, > > spin_lock_init(&chan->desc_lock); > > INIT_LIST_HEAD(&chan->ld_pending); > > INIT_LIST_HEAD(&chan->ld_running); > > + INIT_LIST_HEAD(&chan->ld_completed); > > chan->idle = true; > > > > chan->common.device = &fdev->common; > > diff --git a/drivers/dma/fsldma.h b/drivers/dma/fsldma.h > > index f5c3879..7ede908 100644 > > --- a/drivers/dma/fsldma.h > > +++ b/drivers/dma/fsldma.h > > @@ -140,6 +140,7 @@ struct fsldma_chan { > > spinlock_t desc_lock; /* Descriptor operation lock */ > > struct list_head ld_pending; /* Link descriptors queue */ > > struct list_head ld_running; /* Link descriptors queue */ > > + struct list_head ld_completed; /* Link descriptors queue */ > > It may help to add some documentation here. It would have helped me to > review this patch. Something like this: > > /* > * Descriptors which are queued to run, but have not yet been handed > * to the hardware for execution > */ > struct list_head ld_pending; > > /* > * Descriptors which are currently being executed by the hardware > */ > struct list_head ld_running; > > /* > * Descriptors which have finished execution by the hardware. These > * descriptors have already had their cleanup actions run. They are > * waiting for the ACK bit to be set by the async_tx API. > */ > struct list_head ld_completed; > > > struct dma_chan common; /* DMA common channel */ > > struct dma_pool *desc_pool; /* Descriptors pool */ > > struct device *dev; /* Channel device */ > > -- > > 1.7.5.1 > > > > > > _______________________________________________ > > Linuxppc-dev mailing list > > Linuxppc-dev@lists.ozlabs.org > > https://lists.ozlabs.org/listinfo/linuxppc-dev