Message ID | 20180719140955.19444-1-yuehaibing@huawei.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Thu, Jul 19, 2018 at 10:09:55PM +0800, YueHaibing wrote: > Use dma_zalloc_coherent instead of dma_alloc_coherent > followed by memset 0. > > Signed-off-by: YueHaibing <yuehaibing@huawei.com> > --- > drivers/net/ethernet/mediatek/mtk_eth_soc.c | 7 ++----- > 1 file changed, 2 insertions(+), 5 deletions(-) > > diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > index d8ebf0a..fbdb3e3 100644 > --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c > +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > @@ -1221,14 +1221,11 @@ static int mtk_tx_alloc(struct mtk_eth *eth) > if (!ring->buf) > goto no_tx_mem; > > - ring->dma = dma_alloc_coherent(eth->dev, > - MTK_DMA_SIZE * sz, > - &ring->phys, > - GFP_ATOMIC | __GFP_ZERO); > + ring->dma = dma_zalloc_coherent(eth->dev, MTK_DMA_SIZE * sz, > + &ring->phys, GFP_ATOMIC | __GFP_ZERO); > if (!ring->dma) > goto no_tx_mem; > > - memset(ring->dma, 0, MTK_DMA_SIZE * sz); I have to wonder whether this code needs two forms of zeroing... in the original code, __GFP_ZERO _and_ a call to memset() just in case __GFP_ZERO failed to do its job, and in the replacement code, just in case dma_zalloc_coherent() hasn't got the idea... I think you can drop the __GFP_ZERO. ;)
On Thu, 2018-07-19 at 15:17 +0100, Russell King - ARM Linux wrote: > On Thu, Jul 19, 2018 at 10:09:55PM +0800, YueHaibing wrote: > > Use dma_zalloc_coherent instead of dma_alloc_coherent > > followed by memset 0. > > > > Signed-off-by: YueHaibing <yuehaibing@huawei.com> > > --- > > drivers/net/ethernet/mediatek/mtk_eth_soc.c | 7 ++----- > > 1 file changed, 2 insertions(+), 5 deletions(-) > > > > diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > > index d8ebf0a..fbdb3e3 100644 > > --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c > > +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > > @@ -1221,14 +1221,11 @@ static int mtk_tx_alloc(struct mtk_eth *eth) > > if (!ring->buf) > > goto no_tx_mem; > > > > - ring->dma = dma_alloc_coherent(eth->dev, > > - MTK_DMA_SIZE * sz, > > - &ring->phys, > > - GFP_ATOMIC | __GFP_ZERO); > > + ring->dma = dma_zalloc_coherent(eth->dev, MTK_DMA_SIZE * sz, > > + &ring->phys, GFP_ATOMIC | __GFP_ZERO); > > if (!ring->dma) > > goto no_tx_mem; > > > > - memset(ring->dma, 0, MTK_DMA_SIZE * sz); > > I have to wonder whether this code needs two forms of zeroing... in > the original code, __GFP_ZERO _and_ a call to memset() just in case > __GFP_ZERO failed to do its job, and in the replacement code, just > in case dma_zalloc_coherent() hasn't got the idea... > > I think you can drop the __GFP_ZERO. ;) > Just now I did an experiment on 4.14.56 on armv7. I found that dma_zalloc_coherent does not guarantee that the buffer we get is all filled with 0. I really think it's a little bit weird OR what was I missing something for enabling dma_zalloc_coherent ? The result seems to tell that we can't remove freely the memset with 0 at this moment until we get a cause. my test code is ring->dma = dma_zalloc_coherent(eth->dev, MTK_DMA_SIZE * sz, &ring->phys, GFP_ATOMIC); if (!ring->dma) goto no_tx_mem; print_hex_dump(KERN_INFO, "mtk_tx_alloc:", DUMP_PREFIX_OFFSET, 16, 1, ring->dma, MTK_DMA_SIZE * sz, false); memset(ring->dma, 0, MTK_DMA_SIZE * sz); print_hex_dump(KERN_INFO, "mtk_tx_alloc2:", DUMP_PREFIX_OFFSET, 16, 1, ring->dma, MTK_DMA_SIZE * sz, false); and the output ... [ 259.610413] mtk_tx_alloc:00000f40: 00 00 00 00 50 1f 00 bc 00 00 00 c0 00 00 00 00 [ 259.617934] mtk_tx_alloc:00000f50: 00 00 00 00 60 1f 00 bc 00 00 00 c0 00 00 00 00 [ 259.625470] mtk_tx_alloc:00000f60: 00 00 00 00 70 1f 00 bc 00 00 00 c0 00 00 00 00 [ 259.633005] mtk_tx_alloc:00000f70: 00 00 00 00 80 1f 00 bc 00 00 00 c0 00 00 00 00 [ 259.640539] mtk_tx_alloc:00000f80: 00 00 00 00 90 1f 00 bc 00 00 00 c0 00 00 00 00 [ 259.648073] mtk_tx_alloc:00000f90: 00 00 00 00 a0 1f 00 bc 00 00 00 c0 00 00 00 00 [ 259.655590] mtk_tx_alloc:00000fa0: 00 00 00 00 b0 1f 00 bc 00 00 00 c0 00 00 00 00 [ 259.663124] mtk_tx_alloc:00000fb0: 00 00 00 00 c0 1f 00 bc 00 00 00 c0 00 00 00 00 [ 259.670660] mtk_tx_alloc:00000fc0: 00 00 00 00 d0 1f 00 bc 00 00 00 c0 00 00 00 00 [ 259.678196] mtk_tx_alloc:00000fd0: 00 00 00 00 e0 1f 00 bc 00 00 00 c0 00 00 00 00 [ 259.685713] mtk_tx_alloc:00000fe0: 00 00 00 00 f0 1f 00 bc 00 00 00 c0 00 00 00 00 [ 259.693247] mtk_tx_alloc:00000ff0: 02 c0 90 ad 00 10 00 bc 00 40 5a c0 00 00 00 02 [ 259.700782] mtk_tx_alloc2:00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 259.708405] mtk_tx_alloc2:00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 259.716013] mtk_tx_alloc2:00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 259.723634] mtk_tx_alloc2:00000030: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 259.731253] mtk_tx_alloc2:00000040: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 259.738875] mtk_tx_alloc2:00000050: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 259.746481] mtk_tx_alloc2:00000060: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 259.754103] mtk_tx_alloc2:00000070: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 259.761723] mtk_tx_alloc2:00000080: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 259.769344] mtk_tx_alloc2:00000090: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 259.776951] mtk_tx_alloc2:000000a0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 .... Sean
On 2018/7/20 1:02, Sean Wang wrote: > On Thu, 2018-07-19 at 15:17 +0100, Russell King - ARM Linux wrote: >> On Thu, Jul 19, 2018 at 10:09:55PM +0800, YueHaibing wrote: >>> Use dma_zalloc_coherent instead of dma_alloc_coherent >>> followed by memset 0. >>> >>> Signed-off-by: YueHaibing <yuehaibing@huawei.com> >>> --- >>> drivers/net/ethernet/mediatek/mtk_eth_soc.c | 7 ++----- >>> 1 file changed, 2 insertions(+), 5 deletions(-) >>> >>> diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c >>> index d8ebf0a..fbdb3e3 100644 >>> --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c >>> +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c >>> @@ -1221,14 +1221,11 @@ static int mtk_tx_alloc(struct mtk_eth *eth) >>> if (!ring->buf) >>> goto no_tx_mem; >>> >>> - ring->dma = dma_alloc_coherent(eth->dev, >>> - MTK_DMA_SIZE * sz, >>> - &ring->phys, >>> - GFP_ATOMIC | __GFP_ZERO); >>> + ring->dma = dma_zalloc_coherent(eth->dev, MTK_DMA_SIZE * sz, >>> + &ring->phys, GFP_ATOMIC | __GFP_ZERO); >>> if (!ring->dma) >>> goto no_tx_mem; >>> >>> - memset(ring->dma, 0, MTK_DMA_SIZE * sz); >> >> I have to wonder whether this code needs two forms of zeroing... in >> the original code, __GFP_ZERO _and_ a call to memset() just in case >> __GFP_ZERO failed to do its job, and in the replacement code, just >> in case dma_zalloc_coherent() hasn't got the idea... >> >> I think you can drop the __GFP_ZERO. ;) >> > > Just now I did an experiment on 4.14.56 on armv7. I found that > dma_zalloc_coherent does not guarantee that the buffer we get > is all filled with 0. > > > I really think it's a little bit weird OR what was I missing something > for enabling dma_zalloc_coherent ? The result seems to tell that we > can't remove freely the memset with 0 at this moment until we get a > cause. > That means dma_zalloc_coherent doesn't work as expect on armv7? > > my test code is > > ring->dma = dma_zalloc_coherent(eth->dev, > MTK_DMA_SIZE * sz, > &ring->phys, > GFP_ATOMIC); > if (!ring->dma) > goto no_tx_mem; > > print_hex_dump(KERN_INFO, "mtk_tx_alloc:", > DUMP_PREFIX_OFFSET, 16, 1, > ring->dma, MTK_DMA_SIZE * sz, false); > > memset(ring->dma, 0, MTK_DMA_SIZE * sz); > > print_hex_dump(KERN_INFO, "mtk_tx_alloc2:", > DUMP_PREFIX_OFFSET, 16, 1, > ring->dma, MTK_DMA_SIZE * sz, false); > > > and the output > > ... > > [ 259.610413] mtk_tx_alloc:00000f40: 00 00 00 00 50 1f 00 bc 00 00 00 c0 00 00 00 00 > [ 259.617934] mtk_tx_alloc:00000f50: 00 00 00 00 60 1f 00 bc 00 00 00 c0 00 00 00 00 > [ 259.625470] mtk_tx_alloc:00000f60: 00 00 00 00 70 1f 00 bc 00 00 00 c0 00 00 00 00 > [ 259.633005] mtk_tx_alloc:00000f70: 00 00 00 00 80 1f 00 bc 00 00 00 c0 00 00 00 00 > [ 259.640539] mtk_tx_alloc:00000f80: 00 00 00 00 90 1f 00 bc 00 00 00 c0 00 00 00 00 > [ 259.648073] mtk_tx_alloc:00000f90: 00 00 00 00 a0 1f 00 bc 00 00 00 c0 00 00 00 00 > [ 259.655590] mtk_tx_alloc:00000fa0: 00 00 00 00 b0 1f 00 bc 00 00 00 c0 00 00 00 00 > [ 259.663124] mtk_tx_alloc:00000fb0: 00 00 00 00 c0 1f 00 bc 00 00 00 c0 00 00 00 00 > [ 259.670660] mtk_tx_alloc:00000fc0: 00 00 00 00 d0 1f 00 bc 00 00 00 c0 00 00 00 00 > [ 259.678196] mtk_tx_alloc:00000fd0: 00 00 00 00 e0 1f 00 bc 00 00 00 c0 00 00 00 00 > [ 259.685713] mtk_tx_alloc:00000fe0: 00 00 00 00 f0 1f 00 bc 00 00 00 c0 00 00 00 00 > [ 259.693247] mtk_tx_alloc:00000ff0: 02 c0 90 ad 00 10 00 bc 00 40 5a c0 00 00 00 02 > [ 259.700782] mtk_tx_alloc2:00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > [ 259.708405] mtk_tx_alloc2:00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > [ 259.716013] mtk_tx_alloc2:00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > [ 259.723634] mtk_tx_alloc2:00000030: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > [ 259.731253] mtk_tx_alloc2:00000040: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > [ 259.738875] mtk_tx_alloc2:00000050: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > [ 259.746481] mtk_tx_alloc2:00000060: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > [ 259.754103] mtk_tx_alloc2:00000070: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > [ 259.761723] mtk_tx_alloc2:00000080: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > [ 259.769344] mtk_tx_alloc2:00000090: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > [ 259.776951] mtk_tx_alloc2:000000a0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > .... > > > Sean > > > > . >
On Fri, 2018-07-20 at 14:30 +0800, YueHaibing wrote: > On 2018/7/20 1:02, Sean Wang wrote: > > On Thu, 2018-07-19 at 15:17 +0100, Russell King - ARM Linux wrote: > >> On Thu, Jul 19, 2018 at 10:09:55PM +0800, YueHaibing wrote: > >>> Use dma_zalloc_coherent instead of dma_alloc_coherent > >>> followed by memset 0. > >>> > >>> Signed-off-by: YueHaibing <yuehaibing@huawei.com> > >>> --- > >>> drivers/net/ethernet/mediatek/mtk_eth_soc.c | 7 ++----- > >>> 1 file changed, 2 insertions(+), 5 deletions(-) > >>> > >>> diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > >>> index d8ebf0a..fbdb3e3 100644 > >>> --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c > >>> +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > >>> @@ -1221,14 +1221,11 @@ static int mtk_tx_alloc(struct mtk_eth *eth) > >>> if (!ring->buf) > >>> goto no_tx_mem; > >>> > >>> - ring->dma = dma_alloc_coherent(eth->dev, > >>> - MTK_DMA_SIZE * sz, > >>> - &ring->phys, > >>> - GFP_ATOMIC | __GFP_ZERO); > >>> + ring->dma = dma_zalloc_coherent(eth->dev, MTK_DMA_SIZE * sz, > >>> + &ring->phys, GFP_ATOMIC | __GFP_ZERO); > >>> if (!ring->dma) > >>> goto no_tx_mem; > >>> > >>> - memset(ring->dma, 0, MTK_DMA_SIZE * sz); > >> > >> I have to wonder whether this code needs two forms of zeroing... in > >> the original code, __GFP_ZERO _and_ a call to memset() just in case > >> __GFP_ZERO failed to do its job, and in the replacement code, just > >> in case dma_zalloc_coherent() hasn't got the idea... > >> > >> I think you can drop the __GFP_ZERO. ;) > >> > > > > Just now I did an experiment on 4.14.56 on armv7. I found that > > dma_zalloc_coherent does not guarantee that the buffer we get > > is all filled with 0. > > > > > > I really think it's a little bit weird OR what was I missing something > > for enabling dma_zalloc_coherent ? The result seems to tell that we > > can't remove freely the memset with 0 at this moment until we get a > > cause. > > > > That means dma_zalloc_coherent doesn't work as expect on armv7? > I'm not sure if it's true for every armv7. or it's only happening on my device. anyway, i think we can replace all occurrences in the driver for dma_alloc_coherent with __GFP_ZERO by dma_zalloc_coherent, and but keep the extra memset as is. Sean > > > > my test code is > > > > ring->dma = dma_zalloc_coherent(eth->dev, > > MTK_DMA_SIZE * sz, > > &ring->phys, > > GFP_ATOMIC); > > if (!ring->dma) > > goto no_tx_mem; > > > > print_hex_dump(KERN_INFO, "mtk_tx_alloc:", > > DUMP_PREFIX_OFFSET, 16, 1, > > ring->dma, MTK_DMA_SIZE * sz, false); > > > > memset(ring->dma, 0, MTK_DMA_SIZE * sz); > > > > print_hex_dump(KERN_INFO, "mtk_tx_alloc2:", > > DUMP_PREFIX_OFFSET, 16, 1, > > ring->dma, MTK_DMA_SIZE * sz, false); > > > > > > and the output > > > > ... > > > > [ 259.610413] mtk_tx_alloc:00000f40: 00 00 00 00 50 1f 00 bc 00 00 00 c0 00 00 00 00 > > [ 259.617934] mtk_tx_alloc:00000f50: 00 00 00 00 60 1f 00 bc 00 00 00 c0 00 00 00 00 > > [ 259.625470] mtk_tx_alloc:00000f60: 00 00 00 00 70 1f 00 bc 00 00 00 c0 00 00 00 00 > > [ 259.633005] mtk_tx_alloc:00000f70: 00 00 00 00 80 1f 00 bc 00 00 00 c0 00 00 00 00 > > [ 259.640539] mtk_tx_alloc:00000f80: 00 00 00 00 90 1f 00 bc 00 00 00 c0 00 00 00 00 > > [ 259.648073] mtk_tx_alloc:00000f90: 00 00 00 00 a0 1f 00 bc 00 00 00 c0 00 00 00 00 > > [ 259.655590] mtk_tx_alloc:00000fa0: 00 00 00 00 b0 1f 00 bc 00 00 00 c0 00 00 00 00 > > [ 259.663124] mtk_tx_alloc:00000fb0: 00 00 00 00 c0 1f 00 bc 00 00 00 c0 00 00 00 00 > > [ 259.670660] mtk_tx_alloc:00000fc0: 00 00 00 00 d0 1f 00 bc 00 00 00 c0 00 00 00 00 > > [ 259.678196] mtk_tx_alloc:00000fd0: 00 00 00 00 e0 1f 00 bc 00 00 00 c0 00 00 00 00 > > [ 259.685713] mtk_tx_alloc:00000fe0: 00 00 00 00 f0 1f 00 bc 00 00 00 c0 00 00 00 00 > > [ 259.693247] mtk_tx_alloc:00000ff0: 02 c0 90 ad 00 10 00 bc 00 40 5a c0 00 00 00 02 > > [ 259.700782] mtk_tx_alloc2:00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > [ 259.708405] mtk_tx_alloc2:00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > [ 259.716013] mtk_tx_alloc2:00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > [ 259.723634] mtk_tx_alloc2:00000030: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > [ 259.731253] mtk_tx_alloc2:00000040: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > [ 259.738875] mtk_tx_alloc2:00000050: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > [ 259.746481] mtk_tx_alloc2:00000060: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > [ 259.754103] mtk_tx_alloc2:00000070: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > [ 259.761723] mtk_tx_alloc2:00000080: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > [ 259.769344] mtk_tx_alloc2:00000090: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > [ 259.776951] mtk_tx_alloc2:000000a0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > > > > .... > > > > > > Sean > > > > > > > > . > > >
On Fri, Jul 20, 2018 at 02:30:53PM +0800, YueHaibing wrote: > On 2018/7/20 1:02, Sean Wang wrote: > > On Thu, 2018-07-19 at 15:17 +0100, Russell King - ARM Linux wrote: > >> On Thu, Jul 19, 2018 at 10:09:55PM +0800, YueHaibing wrote: > >>> Use dma_zalloc_coherent instead of dma_alloc_coherent > >>> followed by memset 0. > >>> > >>> Signed-off-by: YueHaibing <yuehaibing@huawei.com> > >>> --- > >>> drivers/net/ethernet/mediatek/mtk_eth_soc.c | 7 ++----- > >>> 1 file changed, 2 insertions(+), 5 deletions(-) > >>> > >>> diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > >>> index d8ebf0a..fbdb3e3 100644 > >>> --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c > >>> +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > >>> @@ -1221,14 +1221,11 @@ static int mtk_tx_alloc(struct mtk_eth *eth) > >>> if (!ring->buf) > >>> goto no_tx_mem; > >>> > >>> - ring->dma = dma_alloc_coherent(eth->dev, > >>> - MTK_DMA_SIZE * sz, > >>> - &ring->phys, > >>> - GFP_ATOMIC | __GFP_ZERO); > >>> + ring->dma = dma_zalloc_coherent(eth->dev, MTK_DMA_SIZE * sz, > >>> + &ring->phys, GFP_ATOMIC | __GFP_ZERO); > >>> if (!ring->dma) > >>> goto no_tx_mem; > >>> > >>> - memset(ring->dma, 0, MTK_DMA_SIZE * sz); > >> > >> I have to wonder whether this code needs two forms of zeroing... in > >> the original code, __GFP_ZERO _and_ a call to memset() just in case > >> __GFP_ZERO failed to do its job, and in the replacement code, just > >> in case dma_zalloc_coherent() hasn't got the idea... > >> > >> I think you can drop the __GFP_ZERO. ;) > >> > > > > Just now I did an experiment on 4.14.56 on armv7. I found that > > dma_zalloc_coherent does not guarantee that the buffer we get > > is all filled with 0. > > > > > > I really think it's a little bit weird OR what was I missing something > > for enabling dma_zalloc_coherent ? The result seems to tell that we > > can't remove freely the memset with 0 at this moment until we get a > > cause. > > > > That means dma_zalloc_coherent doesn't work as expect on armv7? Can someone work out which underlying allocator is being used - the possibilities are: - dma_alloc_from_dev_coherent - cma - simple - remap - pool Looking at the code, I'd guess it's the pool allocator, as I don't see anything which zeros memory there, and it doesn't honor the __GFP_ZERO flag. This is definitely an allocator bug.
On Fri, Jul 20, 2018 at 02:54:23PM +0800, Sean Wang wrote: > On Fri, 2018-07-20 at 14:30 +0800, YueHaibing wrote: > > On 2018/7/20 1:02, Sean Wang wrote: > > > On Thu, 2018-07-19 at 15:17 +0100, Russell King - ARM Linux wrote: > > >> On Thu, Jul 19, 2018 at 10:09:55PM +0800, YueHaibing wrote: > > >>> Use dma_zalloc_coherent instead of dma_alloc_coherent > > >>> followed by memset 0. > > >>> > > >>> Signed-off-by: YueHaibing <yuehaibing@huawei.com> > > >>> --- > > >>> drivers/net/ethernet/mediatek/mtk_eth_soc.c | 7 ++----- > > >>> 1 file changed, 2 insertions(+), 5 deletions(-) > > >>> > > >>> diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > > >>> index d8ebf0a..fbdb3e3 100644 > > >>> --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c > > >>> +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > > >>> @@ -1221,14 +1221,11 @@ static int mtk_tx_alloc(struct mtk_eth *eth) > > >>> if (!ring->buf) > > >>> goto no_tx_mem; > > >>> > > >>> - ring->dma = dma_alloc_coherent(eth->dev, > > >>> - MTK_DMA_SIZE * sz, > > >>> - &ring->phys, > > >>> - GFP_ATOMIC | __GFP_ZERO); > > >>> + ring->dma = dma_zalloc_coherent(eth->dev, MTK_DMA_SIZE * sz, > > >>> + &ring->phys, GFP_ATOMIC | __GFP_ZERO); > > >>> if (!ring->dma) > > >>> goto no_tx_mem; > > >>> > > >>> - memset(ring->dma, 0, MTK_DMA_SIZE * sz); > > >> > > >> I have to wonder whether this code needs two forms of zeroing... in > > >> the original code, __GFP_ZERO _and_ a call to memset() just in case > > >> __GFP_ZERO failed to do its job, and in the replacement code, just > > >> in case dma_zalloc_coherent() hasn't got the idea... > > >> > > >> I think you can drop the __GFP_ZERO. ;) > > >> > > > > > > Just now I did an experiment on 4.14.56 on armv7. I found that > > > dma_zalloc_coherent does not guarantee that the buffer we get > > > is all filled with 0. > > > > > > > > > I really think it's a little bit weird OR what was I missing something > > > for enabling dma_zalloc_coherent ? The result seems to tell that we > > > can't remove freely the memset with 0 at this moment until we get a > > > cause. > > > > > > > That means dma_zalloc_coherent doesn't work as expect on armv7? > > > > I'm not sure if it's true for every armv7. or it's only happening on my > device. > > anyway, i think we can replace all occurrences in the driver for > dma_alloc_coherent with __GFP_ZERO by dma_zalloc_coherent, and but > keep the extra memset as is. No, a bug in the allocator has been found that needs fixing. The right solution is to fix the allocator and remove what should be unnecessary memset()s. This should have been reported when the __GFP_ZERO flag was not being honoured and memset() was initially found to be required - all that dma_zalloc_coherent() does is set the __GFP_ZERO flag before calling dma_alloc_coherent().
On 2018/7/20 17:25, Russell King - ARM Linux wrote: > On Fri, Jul 20, 2018 at 02:54:23PM +0800, Sean Wang wrote: >> On Fri, 2018-07-20 at 14:30 +0800, YueHaibing wrote: >>> On 2018/7/20 1:02, Sean Wang wrote: >>>> On Thu, 2018-07-19 at 15:17 +0100, Russell King - ARM Linux wrote: >>>>> On Thu, Jul 19, 2018 at 10:09:55PM +0800, YueHaibing wrote: >>>>>> Use dma_zalloc_coherent instead of dma_alloc_coherent >>>>>> followed by memset 0. >>>>>> >>>>>> Signed-off-by: YueHaibing <yuehaibing@huawei.com> >>>>>> --- >>>>>> drivers/net/ethernet/mediatek/mtk_eth_soc.c | 7 ++----- >>>>>> 1 file changed, 2 insertions(+), 5 deletions(-) >>>>>> >>>>>> diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c >>>>>> index d8ebf0a..fbdb3e3 100644 >>>>>> --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c >>>>>> +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c >>>>>> @@ -1221,14 +1221,11 @@ static int mtk_tx_alloc(struct mtk_eth *eth) >>>>>> if (!ring->buf) >>>>>> goto no_tx_mem; >>>>>> >>>>>> - ring->dma = dma_alloc_coherent(eth->dev, >>>>>> - MTK_DMA_SIZE * sz, >>>>>> - &ring->phys, >>>>>> - GFP_ATOMIC | __GFP_ZERO); >>>>>> + ring->dma = dma_zalloc_coherent(eth->dev, MTK_DMA_SIZE * sz, >>>>>> + &ring->phys, GFP_ATOMIC | __GFP_ZERO); >>>>>> if (!ring->dma) >>>>>> goto no_tx_mem; >>>>>> >>>>>> - memset(ring->dma, 0, MTK_DMA_SIZE * sz); >>>>> >>>>> I have to wonder whether this code needs two forms of zeroing... in >>>>> the original code, __GFP_ZERO _and_ a call to memset() just in case >>>>> __GFP_ZERO failed to do its job, and in the replacement code, just >>>>> in case dma_zalloc_coherent() hasn't got the idea... >>>>> >>>>> I think you can drop the __GFP_ZERO. ;) >>>>> >>>> >>>> Just now I did an experiment on 4.14.56 on armv7. I found that >>>> dma_zalloc_coherent does not guarantee that the buffer we get >>>> is all filled with 0. >>>> >>>> >>>> I really think it's a little bit weird OR what was I missing something >>>> for enabling dma_zalloc_coherent ? The result seems to tell that we >>>> can't remove freely the memset with 0 at this moment until we get a >>>> cause. >>>> >>> >>> That means dma_zalloc_coherent doesn't work as expect on armv7? >>> >> >> I'm not sure if it's true for every armv7. or it's only happening on my >> device. >> >> anyway, i think we can replace all occurrences in the driver for >> dma_alloc_coherent with __GFP_ZERO by dma_zalloc_coherent, and but >> keep the extra memset as is. > > No, a bug in the allocator has been found that needs fixing. The > right solution is to fix the allocator and remove what should be > unnecessary memset()s. Agree, will send v2. > > This should have been reported when the __GFP_ZERO flag was not > being honoured and memset() was initially found to be required - > all that dma_zalloc_coherent() does is set the __GFP_ZERO flag > before calling dma_alloc_coherent(). >
On Fri, 2018-07-20 at 09:13 +0100, Russell King - ARM Linux wrote: > On Fri, Jul 20, 2018 at 02:30:53PM +0800, YueHaibing wrote: > > On 2018/7/20 1:02, Sean Wang wrote: > > > On Thu, 2018-07-19 at 15:17 +0100, Russell King - ARM Linux wrote: > > >> On Thu, Jul 19, 2018 at 10:09:55PM +0800, YueHaibing wrote: > > >>> Use dma_zalloc_coherent instead of dma_alloc_coherent > > >>> followed by memset 0. > > >>> > > >>> Signed-off-by: YueHaibing <yuehaibing@huawei.com> > > >>> --- > > >>> drivers/net/ethernet/mediatek/mtk_eth_soc.c | 7 ++----- > > >>> 1 file changed, 2 insertions(+), 5 deletions(-) > > >>> > > >>> diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > > >>> index d8ebf0a..fbdb3e3 100644 > > >>> --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c > > >>> +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c > > >>> @@ -1221,14 +1221,11 @@ static int mtk_tx_alloc(struct mtk_eth *eth) > > >>> if (!ring->buf) > > >>> goto no_tx_mem; > > >>> > > >>> - ring->dma = dma_alloc_coherent(eth->dev, > > >>> - MTK_DMA_SIZE * sz, > > >>> - &ring->phys, > > >>> - GFP_ATOMIC | __GFP_ZERO); > > >>> + ring->dma = dma_zalloc_coherent(eth->dev, MTK_DMA_SIZE * sz, > > >>> + &ring->phys, GFP_ATOMIC | __GFP_ZERO); > > >>> if (!ring->dma) > > >>> goto no_tx_mem; > > >>> > > >>> - memset(ring->dma, 0, MTK_DMA_SIZE * sz); > > >> > > >> I have to wonder whether this code needs two forms of zeroing... in > > >> the original code, __GFP_ZERO _and_ a call to memset() just in case > > >> __GFP_ZERO failed to do its job, and in the replacement code, just > > >> in case dma_zalloc_coherent() hasn't got the idea... > > >> > > >> I think you can drop the __GFP_ZERO. ;) > > >> > > > > > > Just now I did an experiment on 4.14.56 on armv7. I found that > > > dma_zalloc_coherent does not guarantee that the buffer we get > > > is all filled with 0. > > > > > > > > > I really think it's a little bit weird OR what was I missing something > > > for enabling dma_zalloc_coherent ? The result seems to tell that we > > > can't remove freely the memset with 0 at this moment until we get a > > > cause. > > > > > > > That means dma_zalloc_coherent doesn't work as expect on armv7? > > Can someone work out which underlying allocator is being used - the > possibilities are: > > - dma_alloc_from_dev_coherent > - cma > - simple > - remap > - pool > > Looking at the code, I'd guess it's the pool allocator, as I don't see > anything which zeros memory there, and it doesn't honor the __GFP_ZERO > flag. This is definitely an allocator bug. > Yes, your guess is right. the allocator is from pool. I show the full stack as below when calling dma_zalloc_coherent [ 54.358310] [<c0113ca8>] (unwind_backtrace) from [<c010e558>] (show_stack+0x20/0x24) [ 54.366012] [<c010e558>] (show_stack) from [<c0927f58>] (dump_stack+0x98/0xac) [ 54.373196] [<c0927f58>] (dump_stack) from [<c011c1f8>] (pool_allocator_alloc+0x20/0x30) [ 54.381238] [<c011c1f8>] (pool_allocator_alloc) from [<c011a238>] (__dma_alloc+0x1b8/0x344) [ 54.389538] [<c011a238>] (__dma_alloc) from [<c011a45c>] (arm_dma_alloc+0x50/0x58) [ 54.397063] [<c011a45c>] (arm_dma_alloc) from [<c05dc208>] (mtk_open+0xf4/0x710) [ 54.404416] [<c05dc208>] (mtk_open) from [<c076d110>] (__dev_open+0xdc/0x160) [ 54.411507] [<c076d110>] (__dev_open) from [<c076d530>] (__dev_change_flags+0x178/0x1c4) [ 54.419547] [<c076d530>] (__dev_change_flags) from [<c076d5a4>] (dev_change_flags+0x28/0x58) [ 54.427934] [<c076d5a4>] (dev_change_flags) from [<c07ea7c0>] (devinet_ioctl+0x630/0x720) [ 54.436062] [<c07ea7c0>] (devinet_ioctl) from [<c07ed070>] (inet_ioctl+0x210/0x3a8) [ 54.443674] [<c07ed070>] (inet_ioctl) from [<c0747094>] (sock_ioctl+0x234/0x4dc) [ 54.451029] [<c0747094>] (sock_ioctl) from [<c028206c>] (do_vfs_ioctl+0xc0/0x914) [ 54.458468] [<c028206c>] (do_vfs_ioctl) from [<c0282904>] (SyS_ioctl+0x44/0x6c) [ 54.465733] [<c0282904>] (SyS_ioctl) from [<c0101000>] (ret_fast_syscall+0x0/0x54)
diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c index d8ebf0a..fbdb3e3 100644 --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c @@ -1221,14 +1221,11 @@ static int mtk_tx_alloc(struct mtk_eth *eth) if (!ring->buf) goto no_tx_mem; - ring->dma = dma_alloc_coherent(eth->dev, - MTK_DMA_SIZE * sz, - &ring->phys, - GFP_ATOMIC | __GFP_ZERO); + ring->dma = dma_zalloc_coherent(eth->dev, MTK_DMA_SIZE * sz, + &ring->phys, GFP_ATOMIC | __GFP_ZERO); if (!ring->dma) goto no_tx_mem; - memset(ring->dma, 0, MTK_DMA_SIZE * sz); for (i = 0; i < MTK_DMA_SIZE; i++) { int next = (i + 1) % MTK_DMA_SIZE; u32 next_ptr = ring->phys + next * sz;
Use dma_zalloc_coherent instead of dma_alloc_coherent followed by memset 0. Signed-off-by: YueHaibing <yuehaibing@huawei.com> --- drivers/net/ethernet/mediatek/mtk_eth_soc.c | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-)