fix dma_alloc_coherent on m68knommu / coldfire
Hi all,
this series ensures dma_alloc_coherent returns NULL for m68knommu and coldfire given that these platforms can't actually allocate dma coherent memory after fixing what is according to an old mail from Greg the only driver that actually uses memory returned by dma_alloc_coherent on those platforms.
PATCH 1/2 net: fec: use dma_alloc_noncoherent for m532x
The m532x coldfire platforms can't properly implement dma_alloc_coherent and currently just return noncoherent memory from it. The fec driver than works around this with a flush of all caches in the receive path. Make this hack a little less bad by using the explicit dma_alloc_noncoherent API and documenting the hacky cache flushes.
--- drivers/net/ethernet/freescale/fec_main.c | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+)
drivers/net/ethernet/freescale/fec_main.c +19 -0
@@ -1580,6 +1580,10 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)struct page *page;#ifdef CONFIG_M532x +/* +* Hacky flush of all caches instead of using the DMA API for the TSO +* headers. +*/ flush_cache_all ();#endif rxq = fep->rx_queue[queue_id];
@@ -3123,10 +3127,17 @@ static void fec_enet_free_queue(struct net_device *ndev)for (i =0 ; i < fep->num_tx_queues; i++)if (fep->tx_queue[i] && fep->tx_queue[i]->tso_hdrs) { txq = fep->tx_queue[i]; +#ifdef CONFIG_M532xdma_free_coherent (&fep->pdev->dev, txq->bd.ring_size *TSO_HEADER_SIZE , txq->tso_hdrs, txq->tso_hdrs_dma); +#else +dma_free_noncoherent (&fep->pdev->dev, + txq->bd.ring_size *TSO_HEADER_SIZE , + txq->tso_hdrs, txq->tso_hdrs_dma, +DMA_BIDIRECTIONAL ); +#endif }for (i =0 ; i < fep->num_rx_queues; i++)
@@ -3157,10 +3168,18 @@ static int fec_enet_alloc_queue(struct net_device *ndev) txq->tx_wake_threshold = (txq->bd.ring_size - txq->tx_stop_threshold) /2 ; +#ifdef CONFIG_M532x txq->tso_hdrs =dma_alloc_coherent (&fep->pdev->dev, txq->bd.ring_size *TSO_HEADER_SIZE , &txq->tso_hdrs_dma,GFP_KERNEL ); +#else +/* m68knommu manually flushes all caches in fec_enet_rx_queue */ + txq->tso_hdrs =dma_alloc_noncoherent (&fep->pdev->dev, + txq->bd.ring_size *TSO_HEADER_SIZE , + &txq->tso_hdrs_dma,DMA_BIDIRECTIONAL , +GFP_KERNEL ); +#endif if (!txq->tso_hdrs) { ret = -ENOMEM ; goto alloc_failed;
Signature · 1 line
2.30.2
Geert Uytterhoeven
maintainerRe: [PATCH 1/2] net: fec: use dma_alloc_noncoherent for m532x
Hi Christoph,
Quoted 7 lines · Christoph Hellwig
The m532x coldfire platforms can't properly implement
dma_alloc_coherentand currently just return noncoherent memory from it. The fec driver than works around this with a flush of all caches in the receive path. Make this hack a little less bad by using the explicitdma_alloc_noncoherentAPI and documenting the hacky cache flushes.Signed-off-by Christoph Hellwig <[email protected]>
Thanks for your patch!
Quoted 14 lines
drivers/net/ethernet/freescale/fec_main.c +4
-0@@ -1580,6 +1580,10 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)struct page *page;#ifdef CONFIG_M532x +/* +* Hacky flush of all caches instead of using the DMA API for the TSO +* headers. +*/ flush_cache_all ();#endif rxq = fep->rx_queue[queue_id];
drivers/net/ethernet/freescale/fec_main.c static void fec_enet_free_queue(struct net_device *ndev) · line 3130
Quoted 4 lines
drivers/net/ethernet/freescale/fec_main.c +1
-0@@ -3123,10 +3127,17 @@ static void fec_enet_free_queue(struct net_device *ndev)for (i =0 ; i < fep->num_tx_queues; i++)if (fep->tx_queue[i] && fep->tx_queue[i]->tso_hdrs) { txq = fep->tx_queue[i]; +#ifdef CONFIG_M532x
Shouldn't this be the !CONFIG_M532x path?
Quoted 14 lines
drivers/net/ethernet/freescale/fec_main.c +6
-0dma_free_coherent (&fep->pdev->dev, txq->bd.ring_size *TSO_HEADER_SIZE , txq->tso_hdrs, txq->tso_hdrs_dma); +#else +dma_free_noncoherent (&fep->pdev->dev, + txq->bd.ring_size *TSO_HEADER_SIZE , + txq->tso_hdrs, txq->tso_hdrs_dma, +DMA_BIDIRECTIONAL ); +#endif }for (i =0 ; i < fep->num_rx_queues; i++)
drivers/net/ethernet/freescale/fec_main.c static int fec_enet_alloc_queue(struct net_device *ndev) · line 3171
Quoted 4 lines
drivers/net/ethernet/freescale/fec_main.c +1
-0@@ -3157,10 +3168,18 @@ static int fec_enet_alloc_queue(struct net_device *ndev) txq->tx_wake_threshold = (txq->bd.ring_size - txq->tx_stop_threshold) /2 ; +#ifdef CONFIG_M532x
Likewise
Quoted 4 lines
drivers/net/ethernet/freescale/fec_main.c +0
-0txq->tso_hdrs =dma_alloc_coherent (&fep->pdev->dev, txq->bd.ring_size *TSO_HEADER_SIZE , &txq->tso_hdrs_dma,GFP_KERNEL );
drivers/net/ethernet/freescale/fec_main.c static int fec_enet_alloc_queue(struct net_device *ndev) · line 3185
Quoted 10 lines
drivers/net/ethernet/freescale/fec_main.c +7
-0+#else +/* m68knommu manually flushes all caches in fec_enet_rx_queue */ + txq->tso_hdrs =dma_alloc_noncoherent (&fep->pdev->dev, + txq->bd.ring_size *TSO_HEADER_SIZE , + &txq->tso_hdrs_dma,DMA_BIDIRECTIONAL , +GFP_KERNEL ); +#endif if (!txq->tso_hdrs) { ret = -ENOMEM ; goto alloc_failed;
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
Re: [PATCH 1/2] net: fec: use dma_alloc_noncoherent for m532x
Quoted 3 lines · Geert Uytterhoeven
Quoted 1 line
+#ifdef
CONFIG_M532xShouldn't this be the !CONFIG_M532x path?
Yes.
Re: [PATCH 1/2] net: fec: use dma_alloc_noncoherent for m532x
In reply to Christoph HellwigHi Christoph,
Quoted 45 lines · Christoph Hellwig
The m532x coldfire platforms can't properly implement
dma_alloc_coherentand currently just return noncoherent memory from it. The fec driver than works around this with a flush of all caches in the receive path. Make this hack a little less bad by using the explicitdma_alloc_noncoherentAPI and documenting the hacky cache flushes.Signed-off-by Christoph Hellwig <[email protected]>--- drivers/net/ethernet/freescale/fec_main.c | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+)drivers/net/ethernet/freescale/fec_main.c +11
-0@@ -1580,6 +1580,10 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)struct page *page;#ifdef CONFIG_M532x +/* +* Hacky flush of all caches instead of using the DMA API for the TSO +* headers. +*/ flush_cache_all ();#endif rxq = fep->rx_queue[queue_id];@@ -3123,10 +3127,17 @@ static void fec_enet_free_queue(struct net_device *ndev)for (i =0 ; i < fep->num_tx_queues; i++)if (fep->tx_queue[i] && fep->tx_queue[i]->tso_hdrs) { txq = fep->tx_queue[i]; +#ifdef CONFIG_M532xdma_free_coherent (&fep->pdev->dev, txq->bd.ring_size *TSO_HEADER_SIZE , txq->tso_hdrs, txq->tso_hdrs_dma); +#else +dma_free_noncoherent (&fep->pdev->dev, + txq->bd.ring_size *TSO_HEADER_SIZE , + txq->tso_hdrs, txq->tso_hdrs_dma, +DMA_BIDIRECTIONAL ); +#endif }for (i =0 ; i < fep->num_rx_queues; i++)
drivers/net/ethernet/freescale/fec_main.c static int fec_enet_alloc_queue(struct net_device *ndev) · line 3175
Quoted 8 lines
drivers/net/ethernet/freescale/fec_main.c +1
-0@@ -3157,10 +3168,18 @@ static int fec_enet_alloc_queue(struct net_device *ndev) txq->tx_wake_threshold = (txq->bd.ring_size - txq->tx_stop_threshold) /2 ; +#ifdef CONFIG_M532x txq->tso_hdrs =dma_alloc_coherent (&fep->pdev->dev, txq->bd.ring_size *TSO_HEADER_SIZE , &txq->tso_hdrs_dma,GFP_KERNEL );
Even with this corrected this will now end up failing on all other ColdFire types with the FEC hardware module (all the non-M532x types) once the arch_dma_alloc() returns NULL.
Did you mean "ifndef CONFIG_COLDFIRE" here?
drivers/net/ethernet/freescale/fec_main.c static int fec_enet_alloc_queue(struct net_device *ndev) · line 3185
Quoted 10 lines
drivers/net/ethernet/freescale/fec_main.c +7
-0+#else +/* m68knommu manually flushes all caches in fec_enet_rx_queue */ + txq->tso_hdrs =dma_alloc_noncoherent (&fep->pdev->dev, + txq->bd.ring_size *TSO_HEADER_SIZE , + &txq->tso_hdrs_dma,DMA_BIDIRECTIONAL , +GFP_KERNEL ); +#endif if (!txq->tso_hdrs) { ret = -ENOMEM ; goto alloc_failed;
And what about the dmam_alloc_coherent() call in fec_enet_init()? Does that need changing too?
Regards
Greg
Re: [PATCH 1/2] net: fec: use dma_alloc_noncoherent for m532x
Quoted 11 lines · Greg Ungerer
drivers/net/ethernet/freescale/fec_main.c fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id) · line 1587
Quoted 6 lines
#ifdef CONFIG_M532xdrivers/net/ethernet/freescale/fec_main.c +4
-0+/* +* Hacky flush of all caches instead of using the DMA API for the TSO +* headers. +*/ flush_cache_all ();Even with this corrected this will now end up failing on all other ColdFire types with the FEC hardware module (all the non-M532x types) once the
arch_dma_alloc()returns NULL.
Quoted 1 line
Did you mean "ifndef
CONFIG_COLDFIRE" here?
How did these work before given that the cache flush is conditional on CONFIG_M532x?
Quoted 5 lines
Quoted 5 lines
drivers/net/ethernet/freescale/fec_main.c +5
-0+#else +/* m68knommu manually flushes all caches in fec_enet_rx_queue */ + txq->tso_hdrs = dma_alloc_noncoherent(&fep->pdev->dev, + txq->bd.ring_size * TSO_HEADER_SIZE, + &txq->tso_hdrs_dma, DMA_BIDIRECTIONAL,
Quoted 8 lines
drivers/net/ethernet/freescale/fec_main.c static int fec_enet_alloc_queue(struct net_device *ndev) · line 3185
Quoted 5 lines
drivers/net/ethernet/freescale/fec_main.c +2
-0+ GFP_KERNEL); +#endif if (!txq->tso_hdrs) { ret = -ENOMEM ; goto alloc_failed;And what about the
dmam_alloc_coherent()call infec_enet_init()? Does that need changing too?
If that's actually use by the FEC implementations on coldire: yes. But maybe I need even more help on how the cache flushing is suppoѕed to actually work here.
Re: [PATCH 1/2] net: fec: use dma_alloc_noncoherent for m532x
Quoted 12 lines · Christoph Hellwig
Quoted 11 lines · Greg Ungerer
drivers/net/ethernet/freescale/fec_main.c fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id) · line 1587
Quoted 6 lines
#ifdef CONFIG_M532xdrivers/net/ethernet/freescale/fec_main.c +4
-0+/* +* Hacky flush of all caches instead of using the DMA API for the TSO +* headers. +*/ flush_cache_all ();Even with this corrected this will now end up failing on all other ColdFire types with the FEC hardware module (all the non-M532x types) once the
arch_dma_alloc()returns NULL.
Quoted 4 lines
Quoted 1 line
Did you mean "ifndef
CONFIG_COLDFIRE" here?How did these work before given that the cache flush is conditional on
CONFIG_M532x?
The case for the 5272 is that it only has instruction cache, no data cache. That was the first supported part, and the one I have used the most over the years (though not so much recently). So it should be ok.
I am not convinced about the other version 2 cores either. They all have both instruction and data cache - they can be flexibily configured to be all instruction, or a mix of instruction and data - but the default is both.
I don't have a 532x platform, so I have never done any work on that. The flush_cache_all() for that seems dubious to me.
Quoted 14 lines
Quoted 5 lines
Quoted 5 lines
drivers/net/ethernet/freescale/fec_main.c +5
-0+#else +/* m68knommu manually flushes all caches in fec_enet_rx_queue */ + txq->tso_hdrs = dma_alloc_noncoherent(&fep->pdev->dev, + txq->bd.ring_size * TSO_HEADER_SIZE, + &txq->tso_hdrs_dma, DMA_BIDIRECTIONAL,Quoted 8 lines
drivers/net/ethernet/freescale/fec_main.c static int fec_enet_alloc_queue(struct net_device *ndev) · line 3185
Quoted 5 lines
drivers/net/ethernet/freescale/fec_main.c +2
-0+ GFP_KERNEL); +#endif if (!txq->tso_hdrs) { ret = -ENOMEM ; goto alloc_failed;And what about the
dmam_alloc_coherent()call infec_enet_init()? Does that need changing too?
Quoted 1 line
If that's actually use by the FEC implementations on coldire: yes.
It is. Testing these changes failed at boot without changing this one too.
Quoted 2 lines
But maybe I need even more help on how the cache flushing is suppoѕed to actually work here.
Regards
Greg
PATCH 2/2 m68k: return NULL from dma_alloc_coherent for nommu or coldfire
In reply to Christoph Hellwigm68knommu and coldfire platforms can't support allocating coherent DMA memory. Instead of just returning noncoherent memory, just fail the allocation.
--- arch/m68k/kernel/dma.c | 14 ++------------ 1 file changed, 2 insertions(+), 12 deletions(-)
arch/m68k/kernel/dma.c +2 -12
@@ -37,23 +37,13 @@ pgprot_t pgprot_dmacoherent(pgprot_t prot)void *arch_dma_alloc (struct device *dev,size_t size,dma_addr_t *dma_handle,gfp_t gfp,unsigned long attrs) { -void *ret; - -if (dev ==NULL || (*dev->dma_mask <0xffffffff )) - gfp |=GFP_DMA ; - ret = (void *)__get_free_pages (gfp,get_order (size)); - -if (ret !=NULL ) { -memset (ret,0 , size); - *dma_handle =virt_to_phys (ret); - } -return ret; +return NULL ; }void arch_dma_free (struct device *dev,size_t size,void *vaddr,dma_addr_t dma_handle,unsigned long attrs) { -free_pages ((unsigned long )vaddr,get_order (size)); +WARN_ON_ONCE (1 ); }#endif /* CONFIG_MMU && !CONFIG_COLDFIRE */
Signature · 1 line
2.30.2