Messages in this thread Patch in this message | | | Date | Mon, 28 Apr 2008 18:34:57 +0200 | From | Ingo Molnar <> | Subject | Re: [Bug 10328] New: [regression] performance drop for glx |
| |
* Suresh Siddha <suresh.b.siddha@intel.com> wrote:
> Alexey, Sorry. I missed posting a fix in the post PAT series for this > issue. Can you please try the appended patch and Ack? > > copying X folks get their attention and their Ack on the patch and > next steps outlined.
i queued up the patch below - hand-merged to x86.git.
Ingo
-----------> Subject: x86 PAT: fix performance drop for glx, use UC minus for ioremap(), ioremap_nocache() and pci_mmap_page_range() From: Suresh Siddha <suresh.b.siddha@intel.com> Date: Fri, 25 Apr 2008 17:07:22 -0700
Use UC_MINUS for ioremap(), ioremap_nocache() instead of strong UC. Once all the X drivers move to ioremap_wc(), we can go back to strong UC semantics for ioremap() and ioremap_nocache().
To avoid attribute aliasing issues, pci_mmap_page_range() will also use UC_MINUS for default non write-combining mapping request.
Next steps: a) change all the video drivers using ioremap() or ioremap_nocache() and adding WC MTTR using mttr_add() to ioremap_wc()
b) for strict usage, we can go back to strong uc semantics for ioremap() and ioremap_nocache() after some grace period for completing step-a.
c) user level X server needs to use the appropriate method for setting up WC mapping (like using resourceX_wc sysfs file instead of adding MTRR for WC and using /dev/mem or resourceX under /sys)
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> --- arch/x86/mm/ioremap.c | 20 ++++++++++++++++---- arch/x86/mm/pageattr.c | 10 ++++++++-- arch/x86/pci/i386.c | 12 +++++++++--- 3 files changed, 33 insertions(+), 9 deletions(-)
Index: linux-x86.q/arch/x86/mm/ioremap.c =================================================================== --- linux-x86.q.orig/arch/x86/mm/ioremap.c +++ linux-x86.q/arch/x86/mm/ioremap.c @@ -176,11 +176,11 @@ static void __iomem *__ioremap(resource_ /* * Do not fallback to certain memory types with certain * requested type: - * - request is uncached, return cannot be write-back - * - request is uncached, return cannot be write-combine + * - request is uc-, return cannot be write-back + * - request is uc-, return cannot be write-combine * - request is write-combine, return cannot be write-back */ - if ((prot_val == _PAGE_CACHE_UC && + if ((prot_val == _PAGE_CACHE_UC_MINUS && (new_prot_val == _PAGE_CACHE_WB || new_prot_val == _PAGE_CACHE_WC)) || (prot_val == _PAGE_CACHE_WC && @@ -201,6 +201,9 @@ static void __iomem *__ioremap(resource_ default: prot = PAGE_KERNEL_NOCACHE; break; + case _PAGE_CACHE_UC_MINUS: + prot = PAGE_KERNEL_UC_MINUS; + break; case _PAGE_CACHE_WC: prot = PAGE_KERNEL_WC; break; @@ -255,7 +258,16 @@ static void __iomem *__ioremap(resource_ */ void __iomem *ioremap_nocache(resource_size_t phys_addr, unsigned long size) { - return __ioremap(phys_addr, size, _PAGE_CACHE_UC); + /* + * Ideally, this should be: + * pat_wc_enabled ? _PAGE_CACHE_UC : _PAGE_CACHE_UC_MINUS; + * + * Till we fix all X drivers to use ioremap_wc(), we will use + * UC MINUS. + */ + unsigned long val = _PAGE_CACHE_UC_MINUS; + + return __ioremap(phys_addr, size, val); } EXPORT_SYMBOL(ioremap_nocache); Index: linux-x86.q/arch/x86/mm/pageattr.c =================================================================== --- linux-x86.q.orig/arch/x86/mm/pageattr.c +++ linux-x86.q/arch/x86/mm/pageattr.c @@ -829,14 +829,20 @@ static inline int change_page_attr_clear int _set_memory_uc(unsigned long addr, int numpages) { + /* + * for now UC MINUS. see comments in ioremap_nocache() + */ return change_page_attr_set(&addr, numpages, - __pgprot(_PAGE_CACHE_UC), 0); + __pgprot(_PAGE_CACHE_UC_MINUS), 0); } int set_memory_uc(unsigned long addr, int numpages) { + /* + * for now UC MINUS. see comments in ioremap_nocache() + */ if (reserve_memtype(addr, addr + numpages * PAGE_SIZE, - _PAGE_CACHE_UC, NULL)) + _PAGE_CACHE_UC_MINUS, NULL)) return -EINVAL; return _set_memory_uc(addr, numpages); Index: linux-x86.q/arch/x86/pci/i386.c =================================================================== --- linux-x86.q.orig/arch/x86/pci/i386.c +++ linux-x86.q/arch/x86/pci/i386.c @@ -301,6 +301,13 @@ int pci_mmap_page_range(struct pci_dev * prot = pgprot_val(vma->vm_page_prot); if (pat_wc_enabled && write_combine) prot |= _PAGE_CACHE_WC; + else if (pat_wc_enabled) + /* + * ioremap() and ioremap_nocache() defaults to UC MINUS for now. + * To avoid attribute conflicts, request UC MINUS here + * aswell. + */ + prot |= _PAGE_CACHE_UC_MINUS; else if (boot_cpu_data.x86 > 3) prot |= _PAGE_CACHE_UC; @@ -319,9 +326,8 @@ int pci_mmap_page_range(struct pci_dev * * - request is uncached, return cannot be write-combine * - request is write-combine, return cannot be write-back */ - if ((flags == _PAGE_CACHE_UC && - (new_flags == _PAGE_CACHE_WB || - new_flags == _PAGE_CACHE_WC)) || + if ((flags == _PAGE_CACHE_UC_MINUS && + (new_flags == _PAGE_CACHE_WB)) || (flags == _PAGE_CACHE_WC && new_flags == _PAGE_CACHE_WB)) { free_memtype(addr, addr+len);
| |