Message ID | 1354420465-29315-1-git-send-email-broonie@opensource.wolfsonmicro.com (mailing list archive) |
---|---|
State | Superseded, archived |
Headers | show |
On Sun, 2 Dec 2012 12:54:25 +0900, Mark Brown <broonie@opensource.wolfsonmicro.com> wrote: > In order to avoid constantly allocating and deallocating there is a fixed > buffer which spi_write_then_read() uses for transfers, with an early error > check to ensure that the transfer fits within the buffer. This limits the > size of transfers to this size, currently max(32, SMP_CACHE_BYTES). > > Since we can dynamically allocate and in fact already have a fallback > to do so when there is contention for the fixed buffer remove this > restriction and instead dynamically allocate a suitably sized buffer if > the transfer won't fit. > > Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Looks good to me. Probably 3.9 material though. Acked-by: Grant Likely <grant.likely@secretlab.ca> > --- > drivers/spi/spi.c | 24 +++++++++++------------- > 1 file changed, 11 insertions(+), 13 deletions(-) > > diff --git a/drivers/spi/spi.c b/drivers/spi/spi.c > index c4f7d71..224b7bc 100644 > --- a/drivers/spi/spi.c > +++ b/drivers/spi/spi.c > @@ -1646,12 +1646,18 @@ int spi_write_then_read(struct spi_device *spi, > struct spi_transfer x[2]; > u8 *local_buf; > > - /* Use preallocated DMA-safe buffer. We can't avoid copying here, > - * (as a pure convenience thing), but we can keep heap costs > - * out of the hot path ... > + /* Use preallocated DMA-safe buffer if we can. We can't avoid > + * copying here, (as a pure convenience thing), but we can > + * keep heap costs out of the hot path unless someone else is > + * using the pre-allocated buffer or the transfer is too large. > */ > - if ((n_tx + n_rx) > SPI_BUFSIZ) > - return -EINVAL; > + if ((n_tx + n_rx) > SPI_BUFSIZ || !mutex_trylock(&lock)) { > + local_buf = kmalloc(max(SPI_BUFSIZ, n_tx + n_rx), GFP_KERNEL); > + if (!local_buf) > + return -ENOMEM; > + } else { > + local_buf = buf; > + } > > spi_message_init(&message); > memset(x, 0, sizeof x); > @@ -1664,14 +1670,6 @@ int spi_write_then_read(struct spi_device *spi, > spi_message_add_tail(&x[1], &message); > } > > - /* ... unless someone else is using the pre-allocated buffer */ > - if (!mutex_trylock(&lock)) { > - local_buf = kmalloc(SPI_BUFSIZ, GFP_KERNEL); > - if (!local_buf) > - return -ENOMEM; > - } else > - local_buf = buf; > - > memcpy(local_buf, txbuf, n_tx); > x[0].tx_buf = local_buf; > x[1].rx_buf = local_buf + n_tx; > -- > 1.7.10.4 >
On Thu, Dec 06, 2012 at 12:00:26AM +0000, Grant Likely wrote: > Looks good to me. Probably 3.9 material though. > Acked-by: Grant Likely <grant.likely@secretlab.ca> Hrm, I'd be inclined to apply it now - it's isolated enough that it can be reverted if it explodes and we have the -rc cycle to notice any problems. ------------------------------------------------------------------------------ LogMeIn Rescue: Anywhere, Anytime Remote support for IT. Free Trial Remotely access PCs and mobile devices and provide instant support Improve your efficiency, and focus on delivering more value-add services Discover what IT Professionals Know. Rescue delivers http://p.sf.net/sfu/logmein_12329d2d
On Thu, 6 Dec 2012 14:37:33 +0900, Mark Brown <broonie@opensource.wolfsonmicro.com> wrote: > On Thu, Dec 06, 2012 at 12:00:26AM +0000, Grant Likely wrote: > > > Looks good to me. Probably 3.9 material though. > > > Acked-by: Grant Likely <grant.likely@secretlab.ca> > > Hrm, I'd be inclined to apply it now - it's isolated enough that it can > be reverted if it explodes and we have the -rc cycle to notice any > problems. Alright, applied. I've also merged in your spi-next tree. Let me know if that causes problems because it needs to be rebased. g.
On Thu, Dec 06, 2012 at 02:04:27PM +0000, Grant Likely wrote: > Alright, applied. I've also merged in your spi-next tree. Let me know if > that causes problems because it needs to be rebased. No problem - do you just want to take over the SPI tree again or should I carry on applying things? ------------------------------------------------------------------------------ LogMeIn Rescue: Anywhere, Anytime Remote support for IT. Free Trial Remotely access PCs and mobile devices and provide instant support Improve your efficiency, and focus on delivering more value-add services Discover what IT Professionals Know. Rescue delivers http://p.sf.net/sfu/logmein_12329d2d
On Fri, 7 Dec 2012 12:41:41 +0900, Mark Brown <broonie@opensource.wolfsonmicro.com> wrote: > On Thu, Dec 06, 2012 at 02:04:27PM +0000, Grant Likely wrote: > > > Alright, applied. I've also merged in your spi-next tree. Let me know if > > that causes problems because it needs to be rebased. > > No problem - do you just want to take over the SPI tree again or should > I carry on applying things? Yes, please keep doing what you're doing! It's a huge help. Whenever I sit down to do patch maintenance, I'll pull in your tree first. As long as your tree doesn't get rebased then that should work well. I also promise to not rebase my tree without talking to you first so you can pull in mine before applying more patches on top. It's a huge help to be reading through the mailing list and be able to ignore any messages that I've seen you've already replied to or applied. g.
diff --git a/drivers/spi/spi.c b/drivers/spi/spi.c index c4f7d71..224b7bc 100644 --- a/drivers/spi/spi.c +++ b/drivers/spi/spi.c @@ -1646,12 +1646,18 @@ int spi_write_then_read(struct spi_device *spi, struct spi_transfer x[2]; u8 *local_buf; - /* Use preallocated DMA-safe buffer. We can't avoid copying here, - * (as a pure convenience thing), but we can keep heap costs - * out of the hot path ... + /* Use preallocated DMA-safe buffer if we can. We can't avoid + * copying here, (as a pure convenience thing), but we can + * keep heap costs out of the hot path unless someone else is + * using the pre-allocated buffer or the transfer is too large. */ - if ((n_tx + n_rx) > SPI_BUFSIZ) - return -EINVAL; + if ((n_tx + n_rx) > SPI_BUFSIZ || !mutex_trylock(&lock)) { + local_buf = kmalloc(max(SPI_BUFSIZ, n_tx + n_rx), GFP_KERNEL); + if (!local_buf) + return -ENOMEM; + } else { + local_buf = buf; + } spi_message_init(&message); memset(x, 0, sizeof x); @@ -1664,14 +1670,6 @@ int spi_write_then_read(struct spi_device *spi, spi_message_add_tail(&x[1], &message); } - /* ... unless someone else is using the pre-allocated buffer */ - if (!mutex_trylock(&lock)) { - local_buf = kmalloc(SPI_BUFSIZ, GFP_KERNEL); - if (!local_buf) - return -ENOMEM; - } else - local_buf = buf; - memcpy(local_buf, txbuf, n_tx); x[0].tx_buf = local_buf; x[1].rx_buf = local_buf + n_tx;
In order to avoid constantly allocating and deallocating there is a fixed buffer which spi_write_then_read() uses for transfers, with an early error check to ensure that the transfer fits within the buffer. This limits the size of transfers to this size, currently max(32, SMP_CACHE_BYTES). Since we can dynamically allocate and in fact already have a fallback to do so when there is contention for the fixed buffer remove this restriction and instead dynamically allocate a suitably sized buffer if the transfer won't fit. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> --- drivers/spi/spi.c | 24 +++++++++++------------- 1 file changed, 11 insertions(+), 13 deletions(-)