| Message ID | E1Yb5t5-0006I8-Oi@rmk-PC.arm.linux.org.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Russell King <rmk+kernel@arm.linux.org.uk> writes: > Convert the pxa2xx socket driver memory allocation to use devm_kzalloc() > to simplify the cleanup path. > > Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> > --- > drivers/pcmcia/pxa2xx_base.c | 8 +++----- > 1 file changed, 3 insertions(+), 5 deletions(-) > > diff --git a/drivers/pcmcia/pxa2xx_base.c b/drivers/pcmcia/pxa2xx_base.c > index 197c77a64ce0..dfcf07828605 100644 > --- a/drivers/pcmcia/pxa2xx_base.c > +++ b/drivers/pcmcia/pxa2xx_base.c > @@ -302,7 +302,8 @@ static int pxa2xx_drv_pcmcia_probe(struct platform_device *dev) > > pxa2xx_drv_pcmcia_ops(ops); > > - sinfo = kzalloc(SKT_DEV_INFO_SIZE(ops->nr), GFP_KERNEL); > + sinfo = devm_kzalloc(&pdev->dev, SKT_DEV_INFO_SIZE(ops->nr), There is a typo here : &pdev->dev should be &dev->dev. Now given that fix, you can add my : Tested-by: Robert Jarzmik <robert.jarzmik@free.fr> This test of mine is rather cheap : - it relies on 4.0-rc4 - I don't have any PCMCIA card to insert in the lubbock, hence the "cheap" - I judge that pcmcia is working because : - cat /proc/iomem gives: 10000000-10001fff : sa1111 10000400-100005ff : 0400 10000800-100009ff : 0800 10000a00-10000bff : 0a00 10000a00-10000bff : sa1111-ps2 10000c00-10000dff : 0c00 10000c00-10000dff : sa1111-ps2 10001600-100017ff : irq 10001800-100019ff : 1800 10001800-100019ff : sa1111-pcmcia - I don't see any error message in kernel log As a side note mainly to myself, the pcmcia is broken on pxa2xx architectures and panics the kernel at boot time because : - in drivers/pcmcia/sa1111_lubbock.c I see : pcmcia_lubbock_init() pxa2xx_configure_sockets(&sadev->dev); here it is expected that platform_data contains the pcmcia ops, but it's NULL, and the dereferencing in pxa2xx_configure_sockets() of ops->first panics. Cheers.
On Sat, Mar 28, 2015 at 10:38:24PM +0100, Robert Jarzmik wrote: > Russell King <rmk+kernel@arm.linux.org.uk> writes: > > > Convert the pxa2xx socket driver memory allocation to use devm_kzalloc() > > to simplify the cleanup path. > > > > Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> > > --- > > drivers/pcmcia/pxa2xx_base.c | 8 +++----- > > 1 file changed, 3 insertions(+), 5 deletions(-) > > > > diff --git a/drivers/pcmcia/pxa2xx_base.c b/drivers/pcmcia/pxa2xx_base.c > > index 197c77a64ce0..dfcf07828605 100644 > > --- a/drivers/pcmcia/pxa2xx_base.c > > +++ b/drivers/pcmcia/pxa2xx_base.c > > @@ -302,7 +302,8 @@ static int pxa2xx_drv_pcmcia_probe(struct platform_device *dev) > > > > pxa2xx_drv_pcmcia_ops(ops); > > > > - sinfo = kzalloc(SKT_DEV_INFO_SIZE(ops->nr), GFP_KERNEL); > > + sinfo = devm_kzalloc(&pdev->dev, SKT_DEV_INFO_SIZE(ops->nr), > There is a typo here : &pdev->dev should be &dev->dev. > > Now given that fix, you can add my : > Tested-by: Robert Jarzmik <robert.jarzmik@free.fr> > > This test of mine is rather cheap : > - it relies on 4.0-rc4 > - I don't have any PCMCIA card to insert in the lubbock, hence the "cheap" > - I judge that pcmcia is working because : > - cat /proc/iomem gives: > 10000000-10001fff : sa1111 > 10000400-100005ff : 0400 > 10000800-100009ff : 0800 > 10000a00-10000bff : 0a00 > 10000a00-10000bff : sa1111-ps2 > 10000c00-10000dff : 0c00 > 10000c00-10000dff : sa1111-ps2 > 10001600-100017ff : irq > 10001800-100019ff : 1800 > 10001800-100019ff : sa1111-pcmcia > - I don't see any error message in kernel log > > As a side note mainly to myself, the pcmcia is broken on pxa2xx architectures > and panics the kernel at boot time because : > - in drivers/pcmcia/sa1111_lubbock.c I see : > pcmcia_lubbock_init() > pxa2xx_configure_sockets(&sadev->dev); > here it is expected that platform_data contains the pcmcia ops, but it's > NULL, and the dereferencing in pxa2xx_configure_sockets() of ops->first > panics. I guess it's broken as a result of 5b703683b6cc3cb97bbe6b1b14898b273eb59279 which added that call without really understanding the call path. I don't see any solution other than partially reverting that commit. As can be seen from the commit message, this was done blindly with no testing - so it's no surprise that it broke.
diff --git a/drivers/pcmcia/pxa2xx_base.c b/drivers/pcmcia/pxa2xx_base.c index 197c77a64ce0..dfcf07828605 100644 --- a/drivers/pcmcia/pxa2xx_base.c +++ b/drivers/pcmcia/pxa2xx_base.c @@ -302,7 +302,8 @@ static int pxa2xx_drv_pcmcia_probe(struct platform_device *dev) pxa2xx_drv_pcmcia_ops(ops); - sinfo = kzalloc(SKT_DEV_INFO_SIZE(ops->nr), GFP_KERNEL); + sinfo = devm_kzalloc(&pdev->dev, SKT_DEV_INFO_SIZE(ops->nr), + GFP_KERNEL); if (!sinfo) return -ENOMEM; @@ -330,7 +331,7 @@ static int pxa2xx_drv_pcmcia_probe(struct platform_device *dev) err1: while (--i >= 0) soc_pcmcia_remove_one(&sinfo->skt[i]); - kfree(sinfo); + err0: return ret; } @@ -340,12 +341,9 @@ static int pxa2xx_drv_pcmcia_remove(struct platform_device *dev) struct skt_dev_info *sinfo = platform_get_drvdata(dev); int i; - platform_set_drvdata(dev, NULL); - for (i = 0; i < sinfo->nskt; i++) soc_pcmcia_remove_one(&sinfo->skt[i]); - kfree(sinfo); return 0; }
Convert the pxa2xx socket driver memory allocation to use devm_kzalloc() to simplify the cleanup path. Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> --- drivers/pcmcia/pxa2xx_base.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-)