| Message ID | 1524238793-24481-1-git-send-email-jared.bents@rockwellcollins.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Jared Bents <jared.bents@rockwellcollins.com> writes: > Update to convert the use of kmemdup to dma_alloc_coherent as > dma_alloc_coherent will consider DMA region limits such as > those seen with CONFIG_FSL_PCI && CONFIG_ZONE_DMA32 whereas > kmemdup does not take those limitations into account. > > Signed-off-by: Jared Bents <jared.bents@rockwellcollins.com> Does this fix a real bug or is it just a theoretical issue you noticed while looking at the code? If this fixes a real bug please give a description of it. https://wireless.wiki.kernel.org/en/users/drivers/ath10k/submittingpatches#guidelines
Hi Kalle, On Sat, Apr 21, 2018 at 2:10 AM, Kalle Valo <kvalo@codeaurora.org> wrote: > Jared Bents <jared.bents@rockwellcollins.com> writes: > >> Update to convert the use of kmemdup to dma_alloc_coherent as >> dma_alloc_coherent will consider DMA region limits such as >> those seen with CONFIG_FSL_PCI && CONFIG_ZONE_DMA32 whereas >> kmemdup does not take those limitations into account. >> >> Signed-off-by: Jared Bents <jared.bents@rockwellcollins.com> > > Does this fix a real bug or is it just a theoretical issue you noticed > while looking at the code? If this fixes a real bug please give a > description of it. It fixes a real bug that I have experienced. My processor is a qoriq T1042 powerpc and I am running the fsl-sdk-v2.0-1703 kernel (mainline v4.1.35-rt41). In 32 bit, the ath10k and ath9k work without issue. After migrating to 64 bit, we experienced dma mapping errors with both devices. This patch is the only one that applies to a driver, the rest of the updates performed to fix the dma mapping issues deal with the sk_buffs are handled in skbuff, mac80211, and netlink. We worked with NXP to resolve the issue and I believe they will be handling the ones that don't apply to the ath10k driver through their process. If we don't see them after a while, we will be sending them to the kernel as well. Here is the output from the dma mapping problem: ath10k_pci 0000:01:00.0: unable to get target info from device ath10k_pci 0000:01:00.0: could not get target info (-5) ath10k_pci 0000:01:00.0: could not probe fw (-5) > > https://wireless.wiki.kernel.org/en/users/drivers/ath10k/submittingpatches#guidelines > > -- > Kalle Valo Jared Bents
Jared Bents <jared.bents@rockwellcollins.com> writes: > On Sat, Apr 21, 2018 at 2:10 AM, Kalle Valo <kvalo@codeaurora.org> wrote: >> Jared Bents <jared.bents@rockwellcollins.com> writes: >> >>> Update to convert the use of kmemdup to dma_alloc_coherent as >>> dma_alloc_coherent will consider DMA region limits such as >>> those seen with CONFIG_FSL_PCI && CONFIG_ZONE_DMA32 whereas >>> kmemdup does not take those limitations into account. >>> >>> Signed-off-by: Jared Bents <jared.bents@rockwellcollins.com> >> >> Does this fix a real bug or is it just a theoretical issue you noticed >> while looking at the code? If this fixes a real bug please give a >> description of it. > > It fixes a real bug that I have experienced. My processor is a qoriq > T1042 powerpc and I am running the fsl-sdk-v2.0-1703 kernel (mainline > v4.1.35-rt41). In 32 bit, the ath10k and ath9k work without issue. > After migrating to 64 bit, we experienced dma mapping errors with both > devices. Ok, please add this to the commit log. > Here is the output from the dma mapping problem: > > ath10k_pci 0000:01:00.0: unable to get target info from device > ath10k_pci 0000:01:00.0: could not get target info (-5) > ath10k_pci 0000:01:00.0: could not probe fw (-5) Please add this also. And the patch doesn't seem to apply to at all, not sure why. But please submit v2 and make sure that it applies to my ath.git master branch. patching file drivers/net/wireless/ath/ath10k/pci.c Hunk #1 FAILED at 1372. Hunk #2 FAILED at 1422. 2 out of 2 hunks FAILED -- saving rejects to file drivers/net/wireless/ath/ath10k/pci.c.rej More info: https://wireless.wiki.kernel.org/en/users/drivers/ath10k/submittingpatches
diff --git a/drivers/net/wireless/ath/ath10k/pci.c b/drivers/net/wireless/ath/ath10k/pci.c index 3c4c800..8637bfe 100644 --- a/drivers/net/wireless/ath/ath10k/pci.c +++ b/drivers/net/wireless/ath/ath10k/pci.c @@ -1372,28 +1372,16 @@ static int ath10k_pci_hif_exchange_bmi_msg(struct ath10k *ar, if (resp && resp_len && *resp_len == 0) return -EINVAL; - treq = kmemdup(req, req_len, GFP_KERNEL); - if (!treq) - return -ENOMEM; - - req_paddr = dma_map_single(ar->dev, treq, req_len, DMA_TO_DEVICE); - ret = dma_mapping_error(ar->dev, req_paddr); - if (ret) { + treq = dma_alloc_coherent(ar->dev, req_len, &req_paddr, GFP_KERNEL); + if (!treq) { ret = -EIO; goto err_dma; } + memcpy(treq, req, req_len); if (resp && resp_len) { - tresp = kzalloc(*resp_len, GFP_KERNEL); + tresp = dma_alloc_coherent(ar->dev, *resp_len, &resp_paddr, GFP_KERNEL); if (!tresp) { - ret = -ENOMEM; - goto err_req; - } - - resp_paddr = dma_map_single(ar->dev, tresp, *resp_len, - DMA_FROM_DEVICE); - ret = dma_mapping_error(ar->dev, resp_paddr); - if (ret) { ret = EIO; goto err_req; } @@ -1422,23 +1410,19 @@ static int ath10k_pci_hif_exchange_bmi_msg(struct ath10k *ar, } err_resp: + if (ret == 0 && resp_len) { + *resp_len = min(*resp_len, xfer.resp_len); + memcpy(resp, tresp, xfer.resp_len); + } if (resp) { u32 unused_buffer; ath10k_ce_revoke_recv_next(ce_rx, NULL, &unused_buffer); - dma_unmap_single(ar->dev, resp_paddr, - *resp_len, DMA_FROM_DEVICE); + dma_free_coherent(ar->dev, *resp_len, tresp, resp_paddr); } err_req: - dma_unmap_single(ar->dev, req_paddr, req_len, DMA_TO_DEVICE); - - if (ret == 0 && resp_len) { - *resp_len = min(*resp_len, xfer.resp_len); - memcpy(resp, tresp, xfer.resp_len); - } + dma_free_coherent(ar->dev, req_len, treq, req_paddr); err_dma: - kfree(treq); - kfree(tresp); return ret; }
Update to convert the use of kmemdup to dma_alloc_coherent as dma_alloc_coherent will consider DMA region limits such as those seen with CONFIG_FSL_PCI && CONFIG_ZONE_DMA32 whereas kmemdup does not take those limitations into account. Signed-off-by: Jared Bents <jared.bents@rockwellcollins.com> --- drivers/net/wireless/ath/ath10k/pci.c | 36 ++++++++++------------------------- 1 file changed, 10 insertions(+), 26 deletions(-)