[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [Xen-devel] [PATCH v4 05/10] acpi: Refactor acpi_os_map_memory to be architecturally independent
On 2016/1/22 18:15, Jan Beulich wrote: You really want that? Even though the way here I use is not too many dunplicated codes (and I think it looks clearer).On 22.01.16 at 10:37,<zhaoshenglong@xxxxxxxxxx> wrote:> >On 2016/1/22 16:47, Jan Beulich wrote:>>>>>On 22.01.16 at 09:38,<zhaoshenglong@xxxxxxxxxx> wrote:>>> > >>> >On 2016/1/18 21:33, Jan Beulich wrote:>>>>>>> >>>>>On 16.01.16 at 06:01,<zhaoshenglong@xxxxxxxxxx> wrote:>>>>>> >>> >--- a/xen/drivers/acpi/osl.c >>>>>> >>> >+++ b/xen/drivers/acpi/osl.c >>>>>> >>> >@@ -86,17 +86,7 @@ acpi_physical_address __init>>> >acpi_os_get_root_pointer(void)>>>>>> >>> > void __iomem * >>>>>> >>> > acpi_os_map_memory(acpi_physical_address phys, acpi_size size) >>>>>> >>> > { >>>>>> >>> >- if (system_state >= SYS_STATE_active) { >>>>>> >>> >- mfn_t mfn = _mfn(PFN_DOWN(phys)); >>>>>> >>> >- unsigned int offs = phys & (PAGE_SIZE - 1); >>>>>> >>> >- >>>>>> >>> >- /* The low first Mb is always mapped. */ >>>>>> >>> >- if ( !((phys + size - 1) >> 20) ) >>>>>> >>> >- return __va(phys); >>>>>> >>> >- return __vmap(&mfn, PFN_UP(offs + size), 1, 1, >>>>>> >>> >- PAGE_HYPERVISOR_NOCACHE) + offs; >>>>>> >>> >- } >>>>>> >>> >- return __acpi_map_table(phys, size); >>>>>> >>> >+ return arch_acpi_os_map_memory(phys, size); >>>>>> >>> > }>>>> >>I'm quite sure I've said before that this goes too far: The __vmap() >>>> >>part and likely also the early-boot __acpi_map_table() one already >>>> >>are architecture independent and hence should stay. The factoring >>>> >>hence should only concern the first Mb handling and maybe the >>>> >>the mapping attributes passed to __vmap().>>> > >>> >Yes, the first MB handling and __vmap() should refactor. So if it only >>> >moves them to an architecture function, how about below patch? >>> > >>> >diff --git a/xen/arch/x86/acpi/lib.c b/xen/arch/x86/acpi/lib.c >>> >index cc15ea3..5885a3a 100644 >>> >--- a/xen/arch/x86/acpi/lib.c >>> >+++ b/xen/arch/x86/acpi/lib.c >>> >@@ -33,6 +33,19 @@ u8 __read_mostly acpi_disable_value; >>> > u32 __read_mostly x86_acpiid_to_apicid[MAX_MADT_ENTRIES] = >>> > {[0 ... MAX_MADT_ENTRIES - 1] = BAD_APICID }; >>> > >>> >+void __iomem * >>> >+arch_acpi_os_map_memory(acpi_physical_address phys, acpi_size size) >>> >+{ >>> >+ mfn_t mfn = _mfn(PFN_DOWN(phys)); >>> >+ unsigned int offs = phys & (PAGE_SIZE - 1); >>> >+ >>> >+ /* The low first Mb is always mapped. */ >>> >+ if ( !((phys + size - 1) >> 20) ) >>> >+ return __va(phys); >>> >+ return __vmap(&mfn, PFN_UP(offs + size), 1, 1, >>> >+ PAGE_HYPERVISOR_NOCACHE) + offs; >>> >+}>>Well, I had clearly said the vmap() part is generic; if there's >>anything architecture dependent here, then the mapping >>attributes (and hence only_those_ should be factored out, >>not the entire function invocation).>I know what you said. But how can we change the attribute for ARM in >acpi_os_map_memory() without moving these codes out?By having each arch #define their value, and use that constant here? -- Shannon _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxx http://lists.xen.org/xen-devel
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |