From mboxrd@z Thu Jan 1 00:00:00 1970 From: Johannes Weiner Subject: Re: [PATCH 02/24] x86: Make sure free_init_pages() free pages in boundary Date: Sat, 27 Mar 2010 01:07:23 +0100 Message-ID: <20100327000723.GF29222@cmpxchg.org> References: <1269642114-16588-1-git-send-email-yinghai@kernel.org> <1269642114-16588-3-git-send-email-yinghai@kernel.org> <20100326230615.GB29222@cmpxchg.org> <4BAD4734.8020503@kernel.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <4BAD4734.8020503@kernel.org> Sender: linux-kernel-owner@vger.kernel.org To: Yinghai Lu Cc: Ingo Molnar , Thomas Gleixner , "H. Peter Anvin" , Andrew Morton , David Miller , Benjamin Herrenschmidt , Linus Torvalds , linux-kernel@vger.kernel.org, linux-arch@vger.kernel.org List-Id: linux-arch.vger.kernel.org On Fri, Mar 26, 2010 at 04:45:56PM -0700, Yinghai Lu wrote: > On 03/26/2010 04:06 PM, Johannes Weiner wrote: > > On Fri, Mar 26, 2010 at 03:21:32PM -0700, Yinghai Lu wrote: > >> diff --git a/arch/x86/kernel/head32.c b/arch/x86/kernel/head32.c > >> index adedeef..fe3d953 100644 > >> --- a/arch/x86/kernel/head32.c > >> +++ b/arch/x86/kernel/head32.c > >> @@ -47,6 +47,7 @@ void __init i386_start_kernel(void) > >> u64 ramdisk_image = boot_params.hdr.ramdisk_image; > >> u64 ramdisk_size = boot_params.hdr.ramdisk_size; > >> u64 ramdisk_end = ramdisk_image + ramdisk_size; > >> + ramdisk_end = PFN_UP(ramdisk_end) << PAGE_SHIFT; > > > > Why not PAGE_ALIGN()? > > looks like PFN_UP is more clear. Why would you convert to a pfn just to go back to a physical address, when all you wanted is align at the next page boundary? It looks rather clumsy. PAGE_ALIGN() is pretty wide spread and I think it's clear to most people what it does. > >> + size_aligned = ramdisk_end - ramdisk_image; > >> + > >> /* We need to move the initrd down into lowmem */ > >> - ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size, > >> + ramdisk_here = find_e820_area(0, end_of_lowmem, size_aligned, > >> PAGE_SIZE); > >> > >> if (ramdisk_here == -1ULL) > >> panic("Cannot find place for new RAMDISK of size %lld\n", > >> - ramdisk_size); > >> + size_aligned); > >> > >> /* Note: this includes all the lowmem currently occupied by > >> the initrd, we rely on that fact to keep the data intact. */ > >> - reserve_early(ramdisk_here, ramdisk_here + ramdisk_size, > >> + reserve_early(ramdisk_here, ramdisk_here + size_aligned, > >> "NEW RAMDISK"); > >> initrd_start = ramdisk_here + PAGE_OFFSET; > >> initrd_end = initrd_start + ramdisk_size; > >> printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n", > >> - ramdisk_here, ramdisk_here + ramdisk_size); > >> + ramdisk_here, ramdisk_here + size_aligned); > >> > >> q = (char *)initrd_start; > >> > >> /* Copy any lowmem portion of the initrd */ > >> - if (ramdisk_image < end_of_lowmem) { > >> - clen = end_of_lowmem - ramdisk_image; > >> - p = (char *)__va(ramdisk_image); > >> + image = ramdisk_image; > >> + if (image < end_of_lowmem) { > >> + clen = end_of_lowmem - image; > >> + p = (char *)__va(image); > >> memcpy(q, p, clen); > >> q += clen; > >> - ramdisk_image += clen; > >> - ramdisk_size -= clen; > >> + image += clen; > >> + size_aligned -= clen; > >> } > >> > >> /* Copy the highmem portion of the initrd */ > >> - while (ramdisk_size) { > >> - slop = ramdisk_image & ~PAGE_MASK; > >> - clen = ramdisk_size; > >> + while (size_aligned) { > >> + slop = image & ~PAGE_MASK; > >> + clen = size_aligned; > >> if (clen > MAX_MAP_CHUNK-slop) > >> clen = MAX_MAP_CHUNK-slop; > >> - mapaddr = ramdisk_image & PAGE_MASK; > >> + mapaddr = image & PAGE_MASK; > >> p = early_memremap(mapaddr, clen+slop); > >> memcpy(q, p+slop, clen); > >> early_iounmap(p, clen+slop); > >> q += clen; > >> - ramdisk_image += clen; > >> - ramdisk_size -= clen; > >> + image += clen; > >> + size_aligned -= clen; > > > > Those appear to be a lot of spurious changes. We don't need to > > copy the alignment padding as well, so it only matters that the > > arguments to find_e820_area() and reserve_early() are aligned, no? > > we need to reserve that whole page, otherwise other user may use the same page. I know that. And you would know that I know that, if you actually read what I wrote. > >> --- a/arch/x86/mm/init.c > >> +++ b/arch/x86/mm/init.c > >> @@ -332,6 +332,16 @@ int devmem_is_allowed(unsigned long pagenr) > >> void free_init_pages(char *what, unsigned long begin, unsigned long end) > >> { > >> unsigned long addr = begin; > >> + unsigned long addr_aligned, end_aligned; > >> + > >> + /* Make sure boundaries are page aligned */ > >> + addr_aligned = PFN_UP(addr) << PAGE_SHIFT; > >> + end_aligned = PFN_DOWN(end) << PAGE_SHIFT; > >> + > >> + if (WARN(addr_aligned != addr || end_aligned != end, "free_init_pages: range [%#lx, %#lx] is not aligned\n", addr, end)) { > >> + addr = addr_aligned; > >> + end = end_aligned; > >> + } > > > > Maybe realign only for when it is not aligned? So to keep the fixup > > out of line. > > > > I suppose WARN_ON() is enough as it will print a stack trace which > > should be enough clue of what went wrong. > > we need to PFN_DOWN the end, and don't free the partial page. > otherwise could crash the system. Dito. if (WARN_ON(not_aligned)) { align } is what I meant. > So just print out the trace, and system still can be used. reporter to get dmesg if > he don't have serial console. /me goes arguing with M-x doctor, that thing actually acts on the supplied input.