Message ID | 20220801144029.57829-3-bhe@redhat.com (mailing list archive) |
---|---|
State | New |
Headers | show |
Series | mm: ioremap: Convert architectures to take GENERIC_IOREMAP way | expand |
On Mon, Aug 01, 2022 at 10:40:20PM +0800, Baoquan He wrote: > This is a preparation patch, no functionality change. There is, please see below. > @@ -3,11 +3,17 @@ > #include <linux/mm.h> > #include <linux/io.h> > > -void __iomem *ioremap_allowed(phys_addr_t phys_addr, size_t size, unsigned long prot) > +void __iomem * > +ioremap_allowed(phys_addr_t *paddr, size_t size, unsigned long *prot_val) > { > - unsigned long last_addr = phys_addr + size - 1; > + unsigned long last_addr, offset, phys_addr = *paddr; > int ret = -EINVAL; > > + offset = phys_addr & (~PAGE_MASK); > + phys_addr -= offset; FWIW, phys_addr &= PAGE_MASK looks much more usual. > @@ -11,13 +11,20 @@ > #include <linux/io.h> > #include <linux/export.h> > > -void __iomem *ioremap_prot(phys_addr_t phys_addr, size_t size, > +void __iomem *ioremap_prot(phys_addr_t paddr, size_t size, > unsigned long prot) > { > unsigned long offset, vaddr; > - phys_addr_t last_addr; > + phys_addr_t last_addr, phys_addr = paddr; > struct vm_struct *area; > void __iomem *base; > + unsigned long prot_val = prot; Why prot_val is needed? > + base = ioremap_allowed(&phys_addr, size, &prot_val); > + if (IS_ERR(base)) > + return NULL; > + else if (base) > + return base; By moving ioremap_allowed() here you allow it to be called before the wrap-around check, including architectures that do not do fixups. And now ioremap_allowed() semantics, prototype and name turn less than obvious. Why not introduce a separate fixup callback? > /* Disallow wrap-around or zero size */ > last_addr = phys_addr + size - 1; > @@ -29,12 +36,6 @@ void __iomem *ioremap_prot(phys_addr_t phys_addr, size_t size, > phys_addr -= offset; > size = PAGE_ALIGN(size + offset); > > - base = ioremap_allowed(phys_addr, size, prot); > - if (IS_ERR(base)) > - return NULL; > - else if (base) > - return base; > - > area = get_vm_area_caller(size, VM_IOREMAP, > __builtin_return_address(0)); > if (!area)
On 08/04/22 at 06:02pm, Alexander Gordeev wrote: > On Mon, Aug 01, 2022 at 10:40:20PM +0800, Baoquan He wrote: > > This is a preparation patch, no functionality change. > > There is, please see below. > > > @@ -3,11 +3,17 @@ > > #include <linux/mm.h> > > #include <linux/io.h> > > > > -void __iomem *ioremap_allowed(phys_addr_t phys_addr, size_t size, unsigned long prot) > > +void __iomem * > > +ioremap_allowed(phys_addr_t *paddr, size_t size, unsigned long *prot_val) > > { > > - unsigned long last_addr = phys_addr + size - 1; > > + unsigned long last_addr, offset, phys_addr = *paddr; > > int ret = -EINVAL; > > > > + offset = phys_addr & (~PAGE_MASK); > > + phys_addr -= offset; > > FWIW, phys_addr &= PAGE_MASK looks much more usual. Sure, will change. > > > @@ -11,13 +11,20 @@ > > #include <linux/io.h> > > #include <linux/export.h> > > > > -void __iomem *ioremap_prot(phys_addr_t phys_addr, size_t size, > > +void __iomem *ioremap_prot(phys_addr_t paddr, size_t size, > > unsigned long prot) > > { > > unsigned long offset, vaddr; > > - phys_addr_t last_addr; > > + phys_addr_t last_addr, phys_addr = paddr; > > struct vm_struct *area; > > void __iomem *base; > > + unsigned long prot_val = prot; > > Why prot_val is needed? I will remove it and pass &prot to ioremap_allowed(). I could think too much when I made change here. Thanks. > > > + base = ioremap_allowed(&phys_addr, size, &prot_val); > > + if (IS_ERR(base)) > > + return NULL; > > + else if (base) > > + return base; > > By moving ioremap_allowed() here you allow it to be called > before the wrap-around check, including architectures that > do not do fixups. Yes, just as you say. > > And now ioremap_allowed() semantics, prototype and name turn > less than obvious. Why not introduce a separate fixup callback? I can do that, while a little worried if too many hooks introduced. I will introduce another fixup hook in v2 if no other suggestion or concern. Thanks. > > > /* Disallow wrap-around or zero size */ > > last_addr = phys_addr + size - 1; > > @@ -29,12 +36,6 @@ void __iomem *ioremap_prot(phys_addr_t phys_addr, size_t size, > > phys_addr -= offset; > > size = PAGE_ALIGN(size + offset); > > > > - base = ioremap_allowed(phys_addr, size, prot); > > - if (IS_ERR(base)) > > - return NULL; > > - else if (base) > > - return base; > > - > > area = get_vm_area_caller(size, VM_IOREMAP, > > __builtin_return_address(0)); > > if (!area) >
On 08/04/22 at 06:02pm, Alexander Gordeev wrote: > On Mon, Aug 01, 2022 at 10:40:20PM +0800, Baoquan He wrote: > > This is a preparation patch, no functionality change. > > There is, please see below. > > > @@ -3,11 +3,17 @@ > > #include <linux/mm.h> > > #include <linux/io.h> > > > > -void __iomem *ioremap_allowed(phys_addr_t phys_addr, size_t size, unsigned long prot) > > +void __iomem * > > +ioremap_allowed(phys_addr_t *paddr, size_t size, unsigned long *prot_val) > > { > > - unsigned long last_addr = phys_addr + size - 1; > > + unsigned long last_addr, offset, phys_addr = *paddr; > > int ret = -EINVAL; > > > > + offset = phys_addr & (~PAGE_MASK); > > + phys_addr -= offset; > > FWIW, phys_addr &= PAGE_MASK looks much more usual. > > > @@ -11,13 +11,20 @@ > > #include <linux/io.h> > > #include <linux/export.h> > > > > -void __iomem *ioremap_prot(phys_addr_t phys_addr, size_t size, > > +void __iomem *ioremap_prot(phys_addr_t paddr, size_t size, > > unsigned long prot) > > { > > unsigned long offset, vaddr; > > - phys_addr_t last_addr; > > + phys_addr_t last_addr, phys_addr = paddr; > > struct vm_struct *area; > > void __iomem *base; > > + unsigned long prot_val = prot; > > Why prot_val is needed? > > > + base = ioremap_allowed(&phys_addr, size, &prot_val); > > + if (IS_ERR(base)) > > + return NULL; > > + else if (base) > > + return base; > > By moving ioremap_allowed() here you allow it to be called > before the wrap-around check, including architectures that > do not do fixups. > > And now ioremap_allowed() semantics, prototype and name turn > less than obvious. Why not introduce a separate fixup callback? I finally renamed ioremap_allowed()/iounmap_allowed() to arch_ioremap() and arch_iounmap(). I didn't introduce a separate fixup callback, and have added more explanation to log of patch 1~2, please check that.
diff --git a/arch/arm64/include/asm/io.h b/arch/arm64/include/asm/io.h index 9c56a077b687..7f0c5f60d946 100644 --- a/arch/arm64/include/asm/io.h +++ b/arch/arm64/include/asm/io.h @@ -164,7 +164,8 @@ extern void __memset_io(volatile void __iomem *, int, size_t); * I/O memory mapping functions. */ -void __iomem *ioremap_allowed(phys_addr_t phys_addr, size_t size, unsigned long prot); +void __iomem * +ioremap_allowed(phys_addr_t *paddr, size_t size, unsigned long *prot_val); #define ioremap_allowed ioremap_allowed #define _PAGE_IOREMAP PROT_DEVICE_nGnRE diff --git a/arch/arm64/mm/ioremap.c b/arch/arm64/mm/ioremap.c index 49467c781283..b9febcf0f5f4 100644 --- a/arch/arm64/mm/ioremap.c +++ b/arch/arm64/mm/ioremap.c @@ -3,11 +3,17 @@ #include <linux/mm.h> #include <linux/io.h> -void __iomem *ioremap_allowed(phys_addr_t phys_addr, size_t size, unsigned long prot) +void __iomem * +ioremap_allowed(phys_addr_t *paddr, size_t size, unsigned long *prot_val) { - unsigned long last_addr = phys_addr + size - 1; + unsigned long last_addr, offset, phys_addr = *paddr; int ret = -EINVAL; + offset = phys_addr & (~PAGE_MASK); + phys_addr -= offset; + size = PAGE_ALIGN(size + offset); + last_addr = phys_addr + size - 1; + /* Don't allow outside PHYS_MASK */ if (last_addr & ~PHYS_MASK) return IOMEM_ERR_PTR(ret); diff --git a/include/asm-generic/io.h b/include/asm-generic/io.h index d72eb310fb3c..0b5cd3cef99d 100644 --- a/include/asm-generic/io.h +++ b/include/asm-generic/io.h @@ -976,8 +976,8 @@ static inline void iounmap(volatile void __iomem *addr) */ #ifndef ioremap_allowed #define ioremap_allowed ioremap_allowed -static inline void __iomem *ioremap_allowed(phys_addr_t phys_addr, size_t size, - unsigned long prot) +static inline void __iomem *ioremap_allowed(phys_addr_t *paddr, size_t size, + unsigned long *prot_val) { return NULL; } @@ -991,7 +991,7 @@ static inline int iounmap_allowed(void *addr) } #endif -void __iomem *ioremap_prot(phys_addr_t phys_addr, size_t size, +void __iomem *ioremap_prot(phys_addr_t paddr, size_t size, unsigned long prot); void iounmap(volatile void __iomem *addr); diff --git a/mm/ioremap.c b/mm/ioremap.c index b16ee9debea2..01439882112e 100644 --- a/mm/ioremap.c +++ b/mm/ioremap.c @@ -11,13 +11,20 @@ #include <linux/io.h> #include <linux/export.h> -void __iomem *ioremap_prot(phys_addr_t phys_addr, size_t size, +void __iomem *ioremap_prot(phys_addr_t paddr, size_t size, unsigned long prot) { unsigned long offset, vaddr; - phys_addr_t last_addr; + phys_addr_t last_addr, phys_addr = paddr; struct vm_struct *area; void __iomem *base; + unsigned long prot_val = prot; + + base = ioremap_allowed(&phys_addr, size, &prot_val); + if (IS_ERR(base)) + return NULL; + else if (base) + return base; /* Disallow wrap-around or zero size */ last_addr = phys_addr + size - 1; @@ -29,12 +36,6 @@ void __iomem *ioremap_prot(phys_addr_t phys_addr, size_t size, phys_addr -= offset; size = PAGE_ALIGN(size + offset); - base = ioremap_allowed(phys_addr, size, prot); - if (IS_ERR(base)) - return NULL; - else if (base) - return base; - area = get_vm_area_caller(size, VM_IOREMAP, __builtin_return_address(0)); if (!area) @@ -43,7 +44,7 @@ void __iomem *ioremap_prot(phys_addr_t phys_addr, size_t size, area->phys_addr = phys_addr; if (ioremap_page_range(vaddr, vaddr + size, phys_addr, - __pgprot(prot))) { + __pgprot(prot_val))) { free_vm_area(area); return NULL; }
In some architectures, the io address will be fixed up before doing mapping, e.g, parisc, mips. In oder to convert them to take GENERIC_IOREMAP method, we need wrap the address fixing up code into ioremap_allowed() and pass the new address out for ioremap_prot() handling. This is a preparation patch, no functionality change. Signed-off-by: Baoquan He <bhe@redhat.com> --- arch/arm64/include/asm/io.h | 3 ++- arch/arm64/mm/ioremap.c | 10 ++++++++-- include/asm-generic/io.h | 6 +++--- mm/ioremap.c | 19 ++++++++++--------- 4 files changed, 23 insertions(+), 15 deletions(-)