resource: shared I/O region support
in [Kernel]
|
Prev: module: fix __module_ref_addr()
Next: oprofile, perf, x86: introduce new functions to reserve perfctrs by index
From: Alan Cox on 25 Mar 2010 12:00 On Thu, 25 Mar 2010 14:17:41 +0100 Giel van Schijndel <me(a)mortis.eu> wrote: > From: Alan Cox <alan(a)lxorguk.ukuu.org.uk> > > This patch was originally written by Alan Cox [1]. I only updated it to > apply correctly to Linus' current tree and changed IORESOURCE_MUXED's > value from 0x00200000 to 0x00400000 because the former has already been > taken in use since. > > [1] https://patchwork.kernel.org/patch/32397/ > > Patch after this line: > ======================================================================== > SuperIO devices share regions and use lock/unlock operations to chip > select. We therefore need to be able to request a resource and wait for > it to be freed by whichever other SuperIO device currently hogs it. > Right now you have to poll which is horrible. > > Add a MUXED field to IO port resources. If the MUXED field is set on the > resource and on the request (via request_muxed_region) then we block > until the previous owner of the muxed resource releases their region. > > This allows us to implement proper resource sharing and locking for > superio chips using code of the form > > enable_my_superio_dev() { > request_muxed_region(0x44, 0x02, "superio:watchdog"); > outb() ..sequence to enable chip > } > > disable_my_superio_dev() { > outb() .. sequence of disable chip > release_region(0x44, 0x02); > } > > Signed-off-by: Giel van Schijndel <me(a)mortis.eu> > --- > include/linux/ioport.h | 4 +++- > kernel/resource.c | 14 +++++++++++++- > 2 files changed, 16 insertions(+), 2 deletions(-) > > diff --git a/include/linux/ioport.h b/include/linux/ioport.h > index 71ab79d..604fd29 100644 > --- a/include/linux/ioport.h > +++ b/include/linux/ioport.h > @@ -52,6 +52,7 @@ struct resource_list { > > #define IORESOURCE_MEM_64 0x00100000 > #define IORESOURCE_WINDOW 0x00200000 /* forwarded by bridge */ > +#define IORESOURCE_MUXED 0x00400000 /* Resource is software muxed */ > > #define IORESOURCE_EXCLUSIVE 0x08000000 /* Userland may not map this resource */ > #define IORESOURCE_DISABLED 0x10000000 > @@ -141,7 +142,8 @@ static inline unsigned long resource_type(const struct resource *res) > } > > /* Convenience shorthand with allocation */ > -#define request_region(start,n,name) __request_region(&ioport_resource, (start), (n), (name), 0) > +#define request_region(start,n,name) __request_region(&ioport_resource, (start), (n), (name), 0) > +#define request_muxed_region(start,n,name) __request_region(&ioport_resource, (start), (n), (name), IORESOURCE_MUXED) > #define __request_mem_region(start,n,name, excl) __request_region(&iomem_resource, (start), (n), (name), excl) > #define request_mem_region(start,n,name) __request_region(&iomem_resource, (start), (n), (name), 0) > #define request_mem_region_exclusive(start,n,name) \ > diff --git a/kernel/resource.c b/kernel/resource.c > index 2d5be5d..1484180 100644 > --- a/kernel/resource.c > +++ b/kernel/resource.c > @@ -651,6 +651,8 @@ resource_size_t resource_alignment(struct resource *res) > * release_region releases a matching busy region. > */ > > +static DECLARE_WAIT_QUEUE_HEAD(muxed_resource_wait); > + > /** > * __request_region - create a new busy resource region > * @parent: parent resource descriptor > @@ -663,6 +665,7 @@ struct resource * __request_region(struct resource *parent, > resource_size_t start, resource_size_t n, > const char *name, int flags) > { > + DECLARE_WAITQUEUE(wait, current); > struct resource *res = kzalloc(sizeof(*res), GFP_KERNEL); > > if (!res) > @@ -687,7 +690,14 @@ struct resource * __request_region(struct resource *parent, > if (!(conflict->flags & IORESOURCE_BUSY)) > continue; > } > - > + if (conflict->flags & flags & IORESOURCE_MUXED) { > + add_wait_queue(&muxed_resource_wait, &wait); > + write_unlock(&resource_lock); needs a set_current_state(TASK_UNINTERRUPTIBLE); here (error from my original test patch that I noticed re-reviewing it) -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Giel van Schijndel on 25 Mar 2010 14:10 On Thu, Mar 25, 2010 at 03:57:05PM +0000, Alan Cox wrote: > On Thu, 25 Mar 2010 14:17:41 +0100 Giel van Schijndel wrote: >> From: Alan Cox <alan(a)lxorguk.ukuu.org.uk> >> >> This patch was originally written by Alan Cox [1]. I only updated it to >> apply correctly to Linus' current tree and changed IORESOURCE_MUXED's >> value from 0x00200000 to 0x00400000 because the former has already been >> taken in use since. >> >> [1] https://patchwork.kernel.org/patch/32397/ >> >> Patch after this line: >> ======================================================================== >> ... snip ... >> @@ -687,7 +690,14 @@ struct resource * __request_region(struct resource *parent, >> if (!(conflict->flags & IORESOURCE_BUSY)) >> continue; >> } >> - >> + if (conflict->flags & flags & IORESOURCE_MUXED) { >> + add_wait_queue(&muxed_resource_wait, &wait); >> + write_unlock(&resource_lock); > > needs a set_current_state(TASK_UNINTERRUPTIBLE); here > > (error from my original test patch that I noticed re-reviewing it) Always fun reviewing your own code ;-) New patch attached. One question though. I did some reading in Documentation/memory-barriers.txt and think it is correct *not* to call set_current_state(TASK_INTERRUPTIBLE) after schedule() finishes. Could you confirm or deny that? Patch after this line: resource: shared I/O region support SuperIO devices share regions and use lock/unlock operations to chip select. We therefore need to be able to request a resource and wait for it to be freed by whichever other SuperIO device currently hogs it. Right now you have to poll which is horrible. Add a MUXED field to IO port resources. If the MUXED field is set on the resource and on the request (via request_muxed_region) then we block until the previous owner of the muxed resource releases their region. This allows us to implement proper resource sharing and locking for superio chips using code of the form enable_my_superio_dev() { request_muxed_region(0x44, 0x02, "superio:watchdog"); outb() ..sequence to enable chip } disable_my_superio_dev() { outb() .. sequence of disable chip release_region(0x44, 0x02); } Signed-off-by: Giel van Schijndel <me(a)mortis.eu> --- include/linux/ioport.h | 4 +++- kernel/resource.c | 16 +++++++++++++++- 2 files changed, 18 insertions(+), 2 deletions(-) diff --git a/include/linux/ioport.h b/include/linux/ioport.h index 71ab79d..604fd29 100644 --- a/include/linux/ioport.h +++ b/include/linux/ioport.h @@ -52,6 +52,7 @@ struct resource_list { #define IORESOURCE_MEM_64 0x00100000 #define IORESOURCE_WINDOW 0x00200000 /* forwarded by bridge */ +#define IORESOURCE_MUXED 0x00400000 /* Resource is software muxed */ #define IORESOURCE_EXCLUSIVE 0x08000000 /* Userland may not map this resource */ #define IORESOURCE_DISABLED 0x10000000 @@ -141,7 +142,8 @@ static inline unsigned long resource_type(const struct resource *res) } /* Convenience shorthand with allocation */ -#define request_region(start,n,name) __request_region(&ioport_resource, (start), (n), (name), 0) +#define request_region(start,n,name) __request_region(&ioport_resource, (start), (n), (name), 0) +#define request_muxed_region(start,n,name) __request_region(&ioport_resource, (start), (n), (name), IORESOURCE_MUXED) #define __request_mem_region(start,n,name, excl) __request_region(&iomem_resource, (start), (n), (name), excl) #define request_mem_region(start,n,name) __request_region(&iomem_resource, (start), (n), (name), 0) #define request_mem_region_exclusive(start,n,name) \ diff --git a/kernel/resource.c b/kernel/resource.c index 2d5be5d..ab44c7f 100644 --- a/kernel/resource.c +++ b/kernel/resource.c @@ -15,6 +15,7 @@ #include <linux/spinlock.h> #include <linux/fs.h> #include <linux/proc_fs.h> +#include <linux/sched.h> #include <linux/seq_file.h> #include <linux/device.h> #include <linux/pfn.h> @@ -651,6 +652,8 @@ resource_size_t resource_alignment(struct resource *res) * release_region releases a matching busy region. */ +static DECLARE_WAIT_QUEUE_HEAD(muxed_resource_wait); + /** * __request_region - create a new busy resource region * @parent: parent resource descriptor @@ -663,6 +666,7 @@ struct resource * __request_region(struct resource *parent, resource_size_t start, resource_size_t n, const char *name, int flags) { + DECLARE_WAITQUEUE(wait, current); struct resource *res = kzalloc(sizeof(*res), GFP_KERNEL); if (!res) @@ -687,7 +691,15 @@ struct resource * __request_region(struct resource *parent, if (!(conflict->flags & IORESOURCE_BUSY)) continue; } - + if (conflict->flags & flags & IORESOURCE_MUXED) { + add_wait_queue(&muxed_resource_wait, &wait); + write_unlock(&resource_lock); + set_current_state(TASK_UNINTERRUPTIBLE); + schedule(); + remove_wait_queue(&muxed_resource_wait, &wait); + write_lock(&resource_lock); + continue; + } /* Uhhuh, that didn't work out.. */ kfree(res); res = NULL; @@ -761,6 +773,8 @@ void __release_region(struct resource *parent, resource_size_t start, break; *p = res->sibling; write_unlock(&resource_lock); + if (res->flags & IORESOURCE_MUXED) + wake_up(&muxed_resource_wait); kfree(res); return; } -- 1.6.4.4 -- Met vriendelijke groet, With kind regards, Giel van Schijndel
From: Alan Cox on 25 Mar 2010 14:20 > Always fun reviewing your own code ;-) > > New patch attached. > > One question though. I did some reading in > Documentation/memory-barriers.txt and think it is correct *not* to call > set_current_state(TASK_INTERRUPTIBLE) after schedule() finishes. Could > you confirm or deny that? After the schedule() returns we will be in TASK_RUNNING which is where we want to be ... > + if (res->flags & IORESOURCE_MUXED) > + wake_up(&muxed_resource_wait); > kfree(res); > return; And you'll want a Signed-off-by: Alan Cox <alan(a)linux.intel.com> > } -- -- "Alan, I'm getting a bit worried about you." -- Linus Torvalds -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Giel van Schijndel on 29 Mar 2010 04:20 On Thu, Mar 25, 2010 at 06:16:33PM +0000, Alan Cox wrote: >> + if (res->flags & IORESOURCE_MUXED) >> + wake_up(&muxed_resource_wait); >> kfree(res); >> return; > > And you'll want a > > Signed-off-by: Alan Cox <alan(a)linux.intel.com> Yes, thanks. Anyone specific I should mail to get this patch applied? I don't see any entry in MAINTAINERS for kernel/resource.c or include/linux/resource.h. -- Met vriendelijke groet, With kind regards, Giel van Schijndel
From: Jesse Barnes on 29 Mar 2010 12:10 On Mon, 29 Mar 2010 10:18:34 +0200 Giel van Schijndel <me(a)mortis.eu> wrote: > On Thu, Mar 25, 2010 at 06:16:33PM +0000, Alan Cox wrote: > >> + if (res->flags & IORESOURCE_MUXED) > >> + wake_up(&muxed_resource_wait); > >> kfree(res); > >> return; > > > > And you'll want a > > > > Signed-off-by: Alan Cox <alan(a)linux.intel.com> > > Yes, thanks. > > Anyone specific I should mail to get this patch applied? I don't see any > entry in MAINTAINERS for kernel/resource.c or include/linux/resource.h. I can take it; sometimes Linus applies resource stuff directly though. Can you resend with the appropriate s-o-bs etc? Thanks, -- Jesse Barnes, Intel Open Source Technology Center -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
|
Next
|
Last
Pages: 1 2 3 Prev: module: fix __module_ref_addr() Next: oprofile, perf, x86: introduce new functions to reserve perfctrs by index |