| Message ID | 20171113102049.9342-1-johan@kernel.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, Nov 13, 2017 at 02:16:09PM +0000, Daniel Thompson wrote: > On 13/11/17 10:20, Johan Hovold wrote: > > Fix child-node lookup during probe, which ended up searching the whole > > device tree depth-first starting at the parent rather than just matching > > on its children. > > > > To make things worse, the parent mfd node was also prematurely freed. > > > > Note that the nodes returned from the two calls to of_parse_phandle() > > are also leaking, but fixing that is a bit more involved as pointers to > > node fields are being stored for later use. > > Is using a devm_kstrdup() to remember the full_name sufficient so get > each of the FIXMEs cleaned up as well? Yeah, that may be sufficient, but looking closer at this now, it seems the name pointers (su1_fb and su2_fb) are only used as booleans, and the fb_name pointer in struct as3711_bl_data is never used at all. So cleaning that up somehow (e.g. and maybe even dropping non-dt probing) would also work. But since this is a separate, and less critical issue, I think it needs to be done as a follow up to this one. Thanks, Johan
On 14/11/17 18:05, Johan Hovold wrote: > On Mon, Nov 13, 2017 at 02:16:09PM +0000, Daniel Thompson wrote: >> On 13/11/17 10:20, Johan Hovold wrote: >>> Fix child-node lookup during probe, which ended up searching the whole >>> device tree depth-first starting at the parent rather than just matching >>> on its children. >>> >>> To make things worse, the parent mfd node was also prematurely freed. >>> >>> Note that the nodes returned from the two calls to of_parse_phandle() >>> are also leaking, but fixing that is a bit more involved as pointers to >>> node fields are being stored for later use. >> >> Is using a devm_kstrdup() to remember the full_name sufficient so get >> each of the FIXMEs cleaned up as well? > > Yeah, that may be sufficient, but looking closer at this now, it seems > the name pointers (su1_fb and su2_fb) are only used as booleans, and the > fb_name pointer in struct as3711_bl_data is never used at all. > > So cleaning that up somehow (e.g. and maybe even dropping non-dt > probing) would also work. > > But since this is a separate, and less critical issue, I think it needs > to be done as a follow up to this one. To be honest it was adding the separate and less critical FIXMEs into the patches that attracted my attention in the first place. ;-) Daniel.
On Tue, Nov 14, 2017 at 07:48:09PM +0000, Daniel Thompson wrote: > On 14/11/17 18:05, Johan Hovold wrote: > > On Mon, Nov 13, 2017 at 02:16:09PM +0000, Daniel Thompson wrote: > >> On 13/11/17 10:20, Johan Hovold wrote: > >>> Fix child-node lookup during probe, which ended up searching the whole > >>> device tree depth-first starting at the parent rather than just matching > >>> on its children. > >>> > >>> To make things worse, the parent mfd node was also prematurely freed. > >>> > >>> Note that the nodes returned from the two calls to of_parse_phandle() > >>> are also leaking, but fixing that is a bit more involved as pointers to > >>> node fields are being stored for later use. > >> > >> Is using a devm_kstrdup() to remember the full_name sufficient so get > >> each of the FIXMEs cleaned up as well? > > > > Yeah, that may be sufficient, but looking closer at this now, it seems > > the name pointers (su1_fb and su2_fb) are only used as booleans, and the > > fb_name pointer in struct as3711_bl_data is never used at all. > > > > So cleaning that up somehow (e.g. and maybe even dropping non-dt > > probing) would also work. > > > > But since this is a separate, and less critical issue, I think it needs > > to be done as a follow up to this one. > > To be honest it was adding the separate and less critical FIXMEs into > the patches that attracted my attention in the first place. ;-) Heh. Since I was touching those error paths, I at least wanted to record somehow there were further issues to be addressed. But feel free to drop the FIXMEs if you prefer. Thanks, Johan
On Wed, 15 Nov 2017, Johan Hovold wrote: > On Tue, Nov 14, 2017 at 07:48:09PM +0000, Daniel Thompson wrote: > > On 14/11/17 18:05, Johan Hovold wrote: > > > On Mon, Nov 13, 2017 at 02:16:09PM +0000, Daniel Thompson wrote: > > >> On 13/11/17 10:20, Johan Hovold wrote: > > >>> Fix child-node lookup during probe, which ended up searching the whole > > >>> device tree depth-first starting at the parent rather than just matching > > >>> on its children. > > >>> > > >>> To make things worse, the parent mfd node was also prematurely freed. > > >>> > > >>> Note that the nodes returned from the two calls to of_parse_phandle() > > >>> are also leaking, but fixing that is a bit more involved as pointers to > > >>> node fields are being stored for later use. > > >> > > >> Is using a devm_kstrdup() to remember the full_name sufficient so get > > >> each of the FIXMEs cleaned up as well? > > > > > > Yeah, that may be sufficient, but looking closer at this now, it seems > > > the name pointers (su1_fb and su2_fb) are only used as booleans, and the > > > fb_name pointer in struct as3711_bl_data is never used at all. > > > > > > So cleaning that up somehow (e.g. and maybe even dropping non-dt > > > probing) would also work. > > > > > > But since this is a separate, and less critical issue, I think it needs > > > to be done as a follow up to this one. > > > > To be honest it was adding the separate and less critical FIXMEs into > > the patches that attracted my attention in the first place. ;-) > > Heh. Since I was touching those error paths, I at least wanted to record > somehow there were further issues to be addressed. But feel free to drop > the FIXMEs if you prefer. In my experience FIXME's tend not to get addressed: $ git grep -i fixme | wc -l 4431 Submit patches instead. :)
On Wed, Nov 15, 2017 at 02:32:11PM +0000, Lee Jones wrote: > On Wed, 15 Nov 2017, Johan Hovold wrote: > > > On Tue, Nov 14, 2017 at 07:48:09PM +0000, Daniel Thompson wrote: > > > On 14/11/17 18:05, Johan Hovold wrote: > > > > On Mon, Nov 13, 2017 at 02:16:09PM +0000, Daniel Thompson wrote: > > > >> On 13/11/17 10:20, Johan Hovold wrote: > > > >>> Fix child-node lookup during probe, which ended up searching the whole > > > >>> device tree depth-first starting at the parent rather than just matching > > > >>> on its children. > > > >>> > > > >>> To make things worse, the parent mfd node was also prematurely freed. > > > >>> > > > >>> Note that the nodes returned from the two calls to of_parse_phandle() > > > >>> are also leaking, but fixing that is a bit more involved as pointers to > > > >>> node fields are being stored for later use. > > > >> > > > >> Is using a devm_kstrdup() to remember the full_name sufficient so get > > > >> each of the FIXMEs cleaned up as well? > > > > > > > > Yeah, that may be sufficient, but looking closer at this now, it seems > > > > the name pointers (su1_fb and su2_fb) are only used as booleans, and the > > > > fb_name pointer in struct as3711_bl_data is never used at all. > > > > > > > > So cleaning that up somehow (e.g. and maybe even dropping non-dt > > > > probing) would also work. > > > > > > > > But since this is a separate, and less critical issue, I think it needs > > > > to be done as a follow up to this one. > > > > > > To be honest it was adding the separate and less critical FIXMEs into > > > the patches that attracted my attention in the first place. ;-) > > > > Heh. Since I was touching those error paths, I at least wanted to record > > somehow there were further issues to be addressed. But feel free to drop > > the FIXMEs if you prefer. > > In my experience FIXME's tend not to get addressed: > > $ git grep -i fixme | wc -l > 4431 > > Submit patches instead. :) There may be some truth to that, but I still think it's better to mark what is broken (especially since a leaked node is no big deal in this case) than to just ignore and forget about it. Johan
On Wed, Nov 15, 2017 at 03:39:09PM +0100, Johan Hovold wrote: > On Wed, Nov 15, 2017 at 02:32:11PM +0000, Lee Jones wrote: > > On Wed, 15 Nov 2017, Johan Hovold wrote: > > > > > On Tue, Nov 14, 2017 at 07:48:09PM +0000, Daniel Thompson wrote: > > > > On 14/11/17 18:05, Johan Hovold wrote: > > > > > On Mon, Nov 13, 2017 at 02:16:09PM +0000, Daniel Thompson wrote: > > > > >> On 13/11/17 10:20, Johan Hovold wrote: > > > > >>> Fix child-node lookup during probe, which ended up searching the whole > > > > >>> device tree depth-first starting at the parent rather than just matching > > > > >>> on its children. > > > > >>> > > > > >>> To make things worse, the parent mfd node was also prematurely freed. > > > > >>> > > > > >>> Note that the nodes returned from the two calls to of_parse_phandle() > > > > >>> are also leaking, but fixing that is a bit more involved as pointers to > > > > >>> node fields are being stored for later use. > > > > >> > > > > >> Is using a devm_kstrdup() to remember the full_name sufficient so get > > > > >> each of the FIXMEs cleaned up as well? > > > > > > > > > > Yeah, that may be sufficient, but looking closer at this now, it seems > > > > > the name pointers (su1_fb and su2_fb) are only used as booleans, and the > > > > > fb_name pointer in struct as3711_bl_data is never used at all. > > > > > > > > > > So cleaning that up somehow (e.g. and maybe even dropping non-dt > > > > > probing) would also work. > > > > > > > > > > But since this is a separate, and less critical issue, I think it needs > > > > > to be done as a follow up to this one. > > > > > > > > To be honest it was adding the separate and less critical FIXMEs into > > > > the patches that attracted my attention in the first place. ;-) > > > > > > Heh. Since I was touching those error paths, I at least wanted to record > > > somehow there were further issues to be addressed. But feel free to drop > > > the FIXMEs if you prefer. > > > > In my experience FIXME's tend not to get addressed: > > > > $ git grep -i fixme | wc -l > > 4431 > > > > Submit patches instead. :) > > There may be some truth to that, but I still think it's better to mark > what is broken (especially since a leaked node is no big deal in this > case) than to just ignore and forget about it. I just sent a v2 including a new patch fixing these node leaks instead of just flagging them. The driver really had no business storing those node full_name fields in the first place. Johan
On Mon, 20 Nov 2017, Johan Hovold wrote: > On Wed, Nov 15, 2017 at 03:39:09PM +0100, Johan Hovold wrote: > > On Wed, Nov 15, 2017 at 02:32:11PM +0000, Lee Jones wrote: > > > On Wed, 15 Nov 2017, Johan Hovold wrote: > > > > > > > On Tue, Nov 14, 2017 at 07:48:09PM +0000, Daniel Thompson wrote: > > > > > On 14/11/17 18:05, Johan Hovold wrote: > > > > > > On Mon, Nov 13, 2017 at 02:16:09PM +0000, Daniel Thompson wrote: > > > > > >> On 13/11/17 10:20, Johan Hovold wrote: > > > > > >>> Fix child-node lookup during probe, which ended up searching the whole > > > > > >>> device tree depth-first starting at the parent rather than just matching > > > > > >>> on its children. > > > > > >>> > > > > > >>> To make things worse, the parent mfd node was also prematurely freed. > > > > > >>> > > > > > >>> Note that the nodes returned from the two calls to of_parse_phandle() > > > > > >>> are also leaking, but fixing that is a bit more involved as pointers to > > > > > >>> node fields are being stored for later use. > > > > > >> > > > > > >> Is using a devm_kstrdup() to remember the full_name sufficient so get > > > > > >> each of the FIXMEs cleaned up as well? > > > > > > > > > > > > Yeah, that may be sufficient, but looking closer at this now, it seems > > > > > > the name pointers (su1_fb and su2_fb) are only used as booleans, and the > > > > > > fb_name pointer in struct as3711_bl_data is never used at all. > > > > > > > > > > > > So cleaning that up somehow (e.g. and maybe even dropping non-dt > > > > > > probing) would also work. > > > > > > > > > > > > But since this is a separate, and less critical issue, I think it needs > > > > > > to be done as a follow up to this one. > > > > > > > > > > To be honest it was adding the separate and less critical FIXMEs into > > > > > the patches that attracted my attention in the first place. ;-) > > > > > > > > Heh. Since I was touching those error paths, I at least wanted to record > > > > somehow there were further issues to be addressed. But feel free to drop > > > > the FIXMEs if you prefer. > > > > > > In my experience FIXME's tend not to get addressed: > > > > > > $ git grep -i fixme | wc -l > > > 4431 > > > > > > Submit patches instead. :) > > > > There may be some truth to that, but I still think it's better to mark > > what is broken (especially since a leaked node is no big deal in this > > case) than to just ignore and forget about it. > > I just sent a v2 including a new patch fixing these node leaks instead > of just flagging them. The driver really had no business storing those > node full_name fields in the first place. That's more like it. :) You're a star, thanks.
diff --git a/drivers/video/backlight/as3711_bl.c b/drivers/video/backlight/as3711_bl.c index 734a9158946b..21ce56053c88 100644 --- a/drivers/video/backlight/as3711_bl.c +++ b/drivers/video/backlight/as3711_bl.c @@ -262,15 +262,16 @@ static int as3711_bl_register(struct platform_device *pdev, static int as3711_backlight_parse_dt(struct device *dev) { struct as3711_bl_pdata *pdata = dev_get_platdata(dev); - struct device_node *bl = - of_find_node_by_name(dev->parent->of_node, "backlight"), *fb; + struct device_node *bl, *fb; int ret; + bl = of_get_child_by_name(dev->parent->of_node, "backlight"); if (!bl) { dev_dbg(dev, "backlight node not found\n"); return -ENODEV; } + /* FIXME: need to drop reference to returned node */ fb = of_parse_phandle(bl, "su1-dev", 0); if (fb) { pdata->su1_fb = fb->full_name; @@ -279,9 +280,10 @@ static int as3711_backlight_parse_dt(struct device *dev) if (pdata->su1_max_uA <= 0) ret = -EINVAL; if (ret < 0) - return ret; + goto err_put_bl; } + /* FIXME: need to drop reference to returned node */ fb = of_parse_phandle(bl, "su2-dev", 0); if (fb) { int count = 0; @@ -292,7 +294,7 @@ static int as3711_backlight_parse_dt(struct device *dev) if (pdata->su2_max_uA <= 0) ret = -EINVAL; if (ret < 0) - return ret; + goto err_put_bl; if (of_find_property(bl, "su2-feedback-voltage", NULL)) { pdata->su2_feedback = AS3711_SU2_VOLTAGE; @@ -314,8 +316,10 @@ static int as3711_backlight_parse_dt(struct device *dev) pdata->su2_feedback = AS3711_SU2_CURR_AUTO; count++; } - if (count != 1) - return -EINVAL; + if (count != 1) { + ret = -EINVAL; + goto err_put_bl; + } count = 0; if (of_find_property(bl, "su2-fbprot-lx-sd4", NULL)) { @@ -334,8 +338,10 @@ static int as3711_backlight_parse_dt(struct device *dev) pdata->su2_fbprot = AS3711_SU2_GPIO4; count++; } - if (count != 1) - return -EINVAL; + if (count != 1) { + ret = -EINVAL; + goto err_put_bl; + } count = 0; if (of_find_property(bl, "su2-auto-curr1", NULL)) { @@ -355,11 +361,20 @@ static int as3711_backlight_parse_dt(struct device *dev) * At least one su2-auto-curr* must be specified iff * AS3711_SU2_CURR_AUTO is used */ - if (!count ^ (pdata->su2_feedback != AS3711_SU2_CURR_AUTO)) - return -EINVAL; + if (!count ^ (pdata->su2_feedback != AS3711_SU2_CURR_AUTO)) { + ret = -EINVAL; + goto err_put_bl; + } } + of_node_put(bl); + return 0; + +err_put_bl: + of_node_put(bl); + + return ret; } static int as3711_backlight_probe(struct platform_device *pdev)
Fix child-node lookup during probe, which ended up searching the whole device tree depth-first starting at the parent rather than just matching on its children. To make things worse, the parent mfd node was also prematurely freed. Note that the nodes returned from the two calls to of_parse_phandle() are also leaking, but fixing that is a bit more involved as pointers to node fields are being stored for later use. Fixes: 59eb2b5e57ea ("drivers/video/backlight/as3711_bl.c: add OF support") Cc: stable <stable@vger.kernel.org> # 3.10 Cc: Guennadi Liakhovetski <g.liakhovetski@gmx.de> Signed-off-by: Johan Hovold <johan@kernel.org> --- drivers/video/backlight/as3711_bl.c | 35 +++++++++++++++++++++++++---------- 1 file changed, 25 insertions(+), 10 deletions(-)