| Message ID | 35FD53F367049845BC99AC72306C23D103D6DB4915FC@CNBJMBX05.corpusers.net (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, Sep 15, 2014 at 01:11:14PM +0800, Wang, Yalin wrote: > this patch extend the start and end address of initrd to be page aligned, > so that we can free all memory including the un-page aligned head or tail > page of initrd, if the start or end address of initrd are not page > aligned, the page can't be freed by free_initrd_mem() function. Have you tested this patch? If so, how thorough was your testing?
Hi I tested it on my phone, From log: <4>[ 0.000000] INITRD unalign phys address:0x02000000+0x0022fb0e <4>[ 0.000000] INITRD aligned phys address:0x02000000+0x00230000 <4>[ 0.579474] free_initrd_mem: free pfn:8192---8752 The tail address is not aligned for most initrd image, This page will not be freed and lost . This patch have a limitation that the tail page's not used Part should not be reserved by any other driver, And must be memory . This is true for most bootloaders , And we will print error if it is false . Thanks -----Original Message----- From: Russell King - ARM Linux [mailto:linux@arm.linux.org.uk] Sent: Monday, September 15, 2014 4:46 PM To: Wang, Yalin Cc: 'Will Deacon'; 'linux-kernel@vger.kernel.org'; 'linux-arm-kernel@lists.infradead.org'; 'linux-mm@kvack.org'; linux-arm-msm@vger.kernel.org Subject: Re: [RFC] arm:extend the reserved mrmory for initrd to be page aligned On Mon, Sep 15, 2014 at 01:11:14PM +0800, Wang, Yalin wrote: > this patch extend the start and end address of initrd to be page > aligned, so that we can free all memory including the un-page aligned > head or tail page of initrd, if the start or end address of initrd are > not page aligned, the page can't be freed by free_initrd_mem() function. Have you tested this patch? If so, how thorough was your testing? -- FTTC broadband for 0.8mile line: currently at 9.5Mbps down 400kbps up according to speedtest.net.
On Mon, Sep 15, 2014 at 05:07:53PM +0800, Wang, Yalin wrote: > Hi > > I tested it on my phone, > >From log: > <4>[ 0.000000] INITRD unalign phys address:0x02000000+0x0022fb0e > <4>[ 0.000000] INITRD aligned phys address:0x02000000+0x00230000 > > <4>[ 0.579474] free_initrd_mem: free pfn:8192---8752 > > The tail address is not aligned for most initrd image, > This page will not be freed and lost . Right, so from this I can assume that you only tested it by seeing what the addresses were, and the values used in free_initrd_mem(). What you haven't tested is whether the initrd actually gets used with your changes, which is more what I was interested in given what I found when reading your patch.
Hi Add more log: <4>[ 0.000000] INITRD unalign phys address:0x02000000+0x0022fb0e <4>[ 0.000000] INITRD aligned phys address:0x02000000+0x00230000 <4>[ 0.574868] free_initrd: free initrd 0xc2000000+0xc222fb0e <4>[ 0.579398] free_initrd_mem: free pfn:8192---8752 The inird used memory is still the same as the one passed by bootloads, I don't change it. It should be safe. -----Original Message----- From: Russell King - ARM Linux [mailto:linux@arm.linux.org.uk] Sent: Monday, September 15, 2014 5:30 PM To: Wang, Yalin Cc: 'Will Deacon'; 'linux-kernel@vger.kernel.org'; 'linux-arm-kernel@lists.infradead.org'; 'linux-mm@kvack.org'; linux-arm-msm@vger.kernel.org Subject: Re: [RFC] arm:extend the reserved mrmory for initrd to be page aligned On Mon, Sep 15, 2014 at 05:07:53PM +0800, Wang, Yalin wrote: > Hi > > I tested it on my phone, > >From log: > <4>[ 0.000000] INITRD unalign phys address:0x02000000+0x0022fb0e > <4>[ 0.000000] INITRD aligned phys address:0x02000000+0x00230000 > > <4>[ 0.579474] free_initrd_mem: free pfn:8192---8752 > > The tail address is not aligned for most initrd image, This page will > not be freed and lost . Right, so from this I can assume that you only tested it by seeing what the addresses were, and the values used in free_initrd_mem(). What you haven't tested is whether the initrd actually gets used with your changes, which is more what I was interested in given what I found when reading your patch. -- FTTC broadband for 0.8mile line: currently at 9.5Mbps down 400kbps up according to speedtest.net.
On Mon, Sep 15, 2014 at 05:59:27PM +0800, Wang, Yalin wrote: > Hi > > Add more log: > <4>[ 0.000000] INITRD unalign phys address:0x02000000+0x0022fb0e > <4>[ 0.000000] INITRD aligned phys address:0x02000000+0x00230000 > <4>[ 0.574868] free_initrd: free initrd 0xc2000000+0xc222fb0e > <4>[ 0.579398] free_initrd_mem: free pfn:8192---8752 > > The inird used memory is still the same as the one passed by bootloads, > I don't change it. It should be safe. This tells me nothing about whether the initrd is actually /used/. What it tells me is that it's being freed. The function of an initrd is not to be a chunk of memory which gets freed later on in the boot process. It is there to provide an "initial ramdisk" (whether it be a filesystem image, or a CPIO compressed archive) for userspace to run. So, have you checked that initrd is still functional after this patch?
Hi Oh, I see your meaning, Yeah , my initrd is a cpio image, And it can still work after apply this patch. -----Original Message----- From: Russell King - ARM Linux [mailto:linux@arm.linux.org.uk] Sent: Monday, September 15, 2014 6:17 PM To: Wang, Yalin Cc: 'Will Deacon'; 'linux-kernel@vger.kernel.org'; 'linux-arm-kernel@lists.infradead.org'; 'linux-mm@kvack.org'; linux-arm-msm@vger.kernel.org Subject: Re: [RFC] arm:extend the reserved mrmory for initrd to be page aligned On Mon, Sep 15, 2014 at 05:59:27PM +0800, Wang, Yalin wrote: > Hi > > Add more log: > <4>[ 0.000000] INITRD unalign phys address:0x02000000+0x0022fb0e > <4>[ 0.000000] INITRD aligned phys address:0x02000000+0x00230000 > <4>[ 0.574868] free_initrd: free initrd 0xc2000000+0xc222fb0e > <4>[ 0.579398] free_initrd_mem: free pfn:8192---8752 > > The inird used memory is still the same as the one passed by > bootloads, I don't change it. It should be safe. This tells me nothing about whether the initrd is actually /used/. What it tells me is that it's being freed. The function of an initrd is not to be a chunk of memory which gets freed later on in the boot process. It is there to provide an "initial ramdisk" (whether it be a filesystem image, or a CPIO compressed archive) for userspace to run. So, have you checked that initrd is still functional after this patch? -- FTTC broadband for 0.8mile line: currently at 9.5Mbps down 400kbps up according to speedtest.net.
On Mon, Sep 15, 2014 at 06:22:12PM +0800, Wang, Yalin wrote: > Oh, I see your meaning, > Yeah , my initrd is a cpio image, > And it can still work after apply this patch. Okay, that's what I wanted to know. However, I believe your patch to be incorrect. You delete the assignments to initrd_start and initrd_end in arm_memblock_init(), which will result in non-OF platforms having no initrd. The reason is that OF platforms set initrd_start and initrd_size from the OF code (drivers/of/fdt.c), but ATAG platforms only set our private phys_* versions. The reason I went with phys_* stuff was to permit better verification of the addresses passed - that the addresses were indeed memory locations before passing them through something like __va().
Oh, I see, I don't consider non-of platform kernels, I will send V2 patch for this . Thanks -----Original Message----- From: Russell King - ARM Linux [mailto:linux@arm.linux.org.uk] Sent: Monday, September 15, 2014 6:30 PM To: Wang, Yalin Cc: 'Will Deacon'; 'linux-kernel@vger.kernel.org'; 'linux-arm-kernel@lists.infradead.org'; 'linux-mm@kvack.org'; linux-arm-msm@vger.kernel.org Subject: Re: [RFC] arm:extend the reserved mrmory for initrd to be page aligned On Mon, Sep 15, 2014 at 06:22:12PM +0800, Wang, Yalin wrote: > Oh, I see your meaning, > Yeah , my initrd is a cpio image, > And it can still work after apply this patch. Okay, that's what I wanted to know. However, I believe your patch to be incorrect. You delete the assignments to initrd_start and initrd_end in arm_memblock_init(), which will result in non-OF platforms having no initrd. The reason is that OF platforms set initrd_start and initrd_size from the OF code (drivers/of/fdt.c), but ATAG platforms only set our private phys_* versions. The reason I went with phys_* stuff was to permit better verification of the addresses passed - that the addresses were indeed memory locations before passing them through something like __va(). -- FTTC broadband for 0.8mile line: currently at 9.5Mbps down 400kbps up according to speedtest.net.
diff --git a/arch/arm/mm/init.c b/arch/arm/mm/init.c index 659c75d..6c1db07 100644 --- a/arch/arm/mm/init.c +++ b/arch/arm/mm/init.c @@ -288,7 +288,12 @@ void __init arm_memblock_init(const struct machine_desc *mdesc) phys_initrd_start = __virt_to_phys(initrd_start); phys_initrd_size = initrd_end - initrd_start; } - initrd_start = initrd_end = 0; + + /* make sure the start and end address are page aligned */ + phys_initrd_size = round_up(phys_initrd_start + phys_initrd_size, PAGE_SIZE); + phys_initrd_start = round_down(phys_initrd_start, PAGE_SIZE); + phys_initrd_size -= phys_initrd_start; + if (phys_initrd_size && !memblock_is_region_memory(phys_initrd_start, phys_initrd_size)) { pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n", @@ -301,13 +306,11 @@ void __init arm_memblock_init(const struct machine_desc *mdesc) (u64)phys_initrd_start, phys_initrd_size); phys_initrd_start = phys_initrd_size = 0; } - if (phys_initrd_size) { + if (phys_initrd_size) memblock_reserve(phys_initrd_start, phys_initrd_size); + else + initrd_start = initrd_end = 0; - /* Now convert initrd to virtual addresses */ - initrd_start = __phys_to_virt(phys_initrd_start); - initrd_end = initrd_start + phys_initrd_size; - } #endif arm_mm_memblock_reserve(); @@ -636,6 +639,11 @@ static int keep_initrd; void free_initrd_mem(unsigned long start, unsigned long end) { if (!keep_initrd) { + if (start == initrd_start) + start = round_down(start, PAGE_SIZE); + if (end == initrd_end) + end = round_up(end, PAGE_SIZE); + poison_init_mem((void *)start, PAGE_ALIGN(end) - start); free_reserved_area((void *)start, (void *)end, -1, "initrd"); } diff --git a/arch/arm64/mm/init.c b/arch/arm64/mm/init.c index 5472c24..9dfd9a6 100644 --- a/arch/arm64/mm/init.c +++ b/arch/arm64/mm/init.c @@ -138,15 +138,38 @@ static void arm64_memory_present(void) void __init arm64_memblock_init(void) { phys_addr_t dma_phys_limit = 0; - + phys_addr_t phys_initrd_start; + phys_addr_t phys_initrd_size; /* * Register the kernel text, kernel data, initrd, and initial * pagetables with memblock. */ memblock_reserve(__pa(_text), _end - _text); #ifdef CONFIG_BLK_DEV_INITRD - if (initrd_start) - memblock_reserve(__virt_to_phys(initrd_start), initrd_end - initrd_start); + if (initrd_start) { + phys_initrd_start = __virt_to_phys(initrd_start); + phys_initrd_size = initrd_end - initrd_start; + /* make sure the start and end address are page aligned */ + phys_initrd_size = round_up(phys_initrd_start + phys_initrd_size, PAGE_SIZE); + phys_initrd_start = round_down(phys_initrd_start, PAGE_SIZE); + phys_initrd_size -= phys_initrd_start; + if (phys_initrd_size && + !memblock_is_region_memory(phys_initrd_start, phys_initrd_size)) { + pr_err("INITRD: %pa+%pa is not a memory region - disabling initrd\n", + &phys_initrd_start, &phys_initrd_size); + phys_initrd_start = phys_initrd_size = 0; + } + if (phys_initrd_size && + memblock_is_region_reserved(phys_initrd_start, phys_initrd_size)) { + pr_err("INITRD: %pa+%pa overlaps in-use memory region - disabling initrd\n", + &phys_initrd_start, &phys_initrd_size); + phys_initrd_start = phys_initrd_size = 0; + } + if (phys_initrd_size) + memblock_reserve(phys_initrd_start, phys_initrd_size); + else + initrd_start = initrd_end = 0; + } #endif if (!efi_enabled(EFI_MEMMAP)) @@ -334,8 +357,14 @@ static int keep_initrd; void free_initrd_mem(unsigned long start, unsigned long end) { - if (!keep_initrd) + if (!keep_initrd) { + if (start == initrd_start) + start = round_down(start, PAGE_SIZE); + if (end == initrd_end) + end = round_up(end, PAGE_SIZE); + free_reserved_area((void *)start, (void *)end, 0, "initrd"); + } } static int __init keepinitrd_setup(char *__unused)
this patch extend the start and end address of initrd to be page aligned, so that we can free all memory including the un-page aligned head or tail page of initrd, if the start or end address of initrd are not page aligned, the page can't be freed by free_initrd_mem() function. Signed-off-by: Yalin Wang <yalin.wang@sonymobile.com> --- arch/arm/mm/init.c | 20 ++++++++++++++------ arch/arm64/mm/init.c | 37 +++++++++++++++++++++++++++++++++---- 2 files changed, 47 insertions(+), 10 deletions(-)