Re: [oe] initramfs only supported in x86?
Hi Khem, thanks for your pointers. Just gave that a shot, and my initramfs uImage actually takes longer to boot (at about 9 seconds from cold to user space vs 7 seconds before). Good data point! Adam On Thu, May 15, 2014 at 10:41 PM, Khem Raj raj.k...@gmail.com wrote: On Thu, May 15, 2014 at 1:33 PM, Nicolas Dechesne nicolas.deche...@linaro.org wrote: On Thu, May 15, 2014 at 7:13 PM, Adam Lee adam.yh@gmail.com wrote: Hello everyone, I thought I'd compare the performance of booting rootfs from MMC vs ramdisk (the way Android is doing it). Some says it may possibly take longer in ramdisk because the rootfs has to be first loaded into memory. I'd like to test it out myself. So I discovered core-image-minimal-initramfs, and thought it was exactly what I needed. However the build doesn't go far, because its dependency initramfs-live-install [1] only seems to support x86: live images are for x86 only because it uses syslinux and friends. However you can choose any other image to become initramfs image and bunsle it into kernel you have to use INITRAMFS_IMAGE = your-initramfs-image INITRAMFS_IMAGE_BUNDLE = 1 COMPATIBLE_HOST = (i.86|x86_64).*-linux Is this the limitation of current OE? or am I just looking at the wrong corner? i recently had to build an initrd too, and notice that as well... so i am not sure exactly what that means. I am glad you asked... however if you want to test an initrd, you can simply build *any* image and make sure to build the 'cpio' from IMAGE_FSTYPES. you can then use the generated cpio archive as an inited. -- ___ Openembedded-devel mailing list Openembedded-devel@lists.openembedded.org http://lists.openembedded.org/mailman/listinfo/openembedded-devel -- ___ Openembedded-devel mailing list Openembedded-devel@lists.openembedded.org http://lists.openembedded.org/mailman/listinfo/openembedded-devel
[oe] initramfs only supported in x86?
Hello everyone, I thought I'd compare the performance of booting rootfs from MMC vs ramdisk (the way Android is doing it). Some says it may possibly take longer in ramdisk because the rootfs has to be first loaded into memory. I'd like to test it out myself. So I discovered core-image-minimal-initramfs, and thought it was exactly what I needed. However the build doesn't go far, because its dependency initramfs-live-install [1] only seems to support x86: COMPATIBLE_HOST = (i.86|x86_64).*-linux Is this the limitation of current OE? or am I just looking at the wrong corner? Thanks, Adam [1]http://cgit.openembedded.org/openembedded-core/tree/meta/recipes-core/initrdscripts/initramfs-live-install_1.0.bb?h=daisy -- ___ Openembedded-devel mailing list Openembedded-devel@lists.openembedded.org http://lists.openembedded.org/mailman/listinfo/openembedded-devel
Re: [oe] initramfs only supported in x86?
On Thu, May 15, 2014 at 7:13 PM, Adam Lee adam.yh@gmail.com wrote: Hello everyone, I thought I'd compare the performance of booting rootfs from MMC vs ramdisk (the way Android is doing it). Some says it may possibly take longer in ramdisk because the rootfs has to be first loaded into memory. I'd like to test it out myself. So I discovered core-image-minimal-initramfs, and thought it was exactly what I needed. However the build doesn't go far, because its dependency initramfs-live-install [1] only seems to support x86: COMPATIBLE_HOST = (i.86|x86_64).*-linux Is this the limitation of current OE? or am I just looking at the wrong corner? i recently had to build an initrd too, and notice that as well... so i am not sure exactly what that means. I am glad you asked... however if you want to test an initrd, you can simply build *any* image and make sure to build the 'cpio' from IMAGE_FSTYPES. you can then use the generated cpio archive as an initrd. -- ___ Openembedded-devel mailing list Openembedded-devel@lists.openembedded.org http://lists.openembedded.org/mailman/listinfo/openembedded-devel
Re: [oe] initramfs only supported in x86?
Hi Nicolas, thanks for your response. I built core-image-minimal in cpio, but the size is whopping 69MB. It will take a couple seconds just to load the image into the memory, which defeats the purpose of using the ramdisk to boot. After watching 300 ms from reset-to-shell boot video [1] using 1.5MB large kernel + rootfs combined, the size of the core-image-minimal feels gargantuan and my boot time of ~7 seconds feels like an eternity. That said, my goal is to reduce the boot time through optimization of the current image. Writing custom boot loader or severely crippling the rootfs is out of scope. So I thought the tested and proven core-image-minimal-initramfs would be a great start. Adam [1] http://www.makelinux.com/emb/fastboot/omap On Thu, May 15, 2014 at 1:33 PM, Nicolas Dechesne nicolas.deche...@linaro.org wrote: On Thu, May 15, 2014 at 7:13 PM, Adam Lee adam.yh@gmail.com wrote: Hello everyone, I thought I'd compare the performance of booting rootfs from MMC vs ramdisk (the way Android is doing it). Some says it may possibly take longer in ramdisk because the rootfs has to be first loaded into memory. I'd like to test it out myself. So I discovered core-image-minimal-initramfs, and thought it was exactly what I needed. However the build doesn't go far, because its dependency initramfs-live-install [1] only seems to support x86: COMPATIBLE_HOST = (i.86|x86_64).*-linux Is this the limitation of current OE? or am I just looking at the wrong corner? i recently had to build an initrd too, and notice that as well... so i am not sure exactly what that means. I am glad you asked... however if you want to test an initrd, you can simply build *any* image and make sure to build the 'cpio' from IMAGE_FSTYPES. you can then use the generated cpio archive as an initrd. -- ___ Openembedded-devel mailing list Openembedded-devel@lists.openembedded.org http://lists.openembedded.org/mailman/listinfo/openembedded-devel -- ___ Openembedded-devel mailing list Openembedded-devel@lists.openembedded.org http://lists.openembedded.org/mailman/listinfo/openembedded-devel
Re: [oe] initramfs only supported in x86?
ah I spoke too soon. I had extra DISTRO_FEATURES. I removed them and the root is ~2.5MB. Now it's worth a try. Adam On Thu, May 15, 2014 at 3:58 PM, Adam Lee adam.yh@gmail.com wrote: Hi Nicolas, thanks for your response. I built core-image-minimal in cpio, but the size is whopping 69MB. It will take a couple seconds just to load the image into the memory, which defeats the purpose of using the ramdisk to boot. After watching 300 ms from reset-to-shell boot video [1] using 1.5MB large kernel + rootfs combined, the size of the core-image-minimal feels gargantuan and my boot time of ~7 seconds feels like an eternity. That said, my goal is to reduce the boot time through optimization of the current image. Writing custom boot loader or severely crippling the rootfs is out of scope. So I thought the tested and proven core-image-minimal-initramfs would be a great start. Adam [1] http://www.makelinux.com/emb/fastboot/omap On Thu, May 15, 2014 at 1:33 PM, Nicolas Dechesne nicolas.deche...@linaro.org wrote: On Thu, May 15, 2014 at 7:13 PM, Adam Lee adam.yh@gmail.com wrote: Hello everyone, I thought I'd compare the performance of booting rootfs from MMC vs ramdisk (the way Android is doing it). Some says it may possibly take longer in ramdisk because the rootfs has to be first loaded into memory. I'd like to test it out myself. So I discovered core-image-minimal-initramfs, and thought it was exactly what I needed. However the build doesn't go far, because its dependency initramfs-live-install [1] only seems to support x86: COMPATIBLE_HOST = (i.86|x86_64).*-linux Is this the limitation of current OE? or am I just looking at the wrong corner? i recently had to build an initrd too, and notice that as well... so i am not sure exactly what that means. I am glad you asked... however if you want to test an initrd, you can simply build *any* image and make sure to build the 'cpio' from IMAGE_FSTYPES. you can then use the generated cpio archive as an initrd. -- ___ Openembedded-devel mailing list Openembedded-devel@lists.openembedded.org http://lists.openembedded.org/mailman/listinfo/openembedded-devel -- ___ Openembedded-devel mailing list Openembedded-devel@lists.openembedded.org http://lists.openembedded.org/mailman/listinfo/openembedded-devel
Re: [oe] initramfs only supported in x86?
On Thu, May 15, 2014 at 1:33 PM, Nicolas Dechesne nicolas.deche...@linaro.org wrote: On Thu, May 15, 2014 at 7:13 PM, Adam Lee adam.yh@gmail.com wrote: Hello everyone, I thought I'd compare the performance of booting rootfs from MMC vs ramdisk (the way Android is doing it). Some says it may possibly take longer in ramdisk because the rootfs has to be first loaded into memory. I'd like to test it out myself. So I discovered core-image-minimal-initramfs, and thought it was exactly what I needed. However the build doesn't go far, because its dependency initramfs-live-install [1] only seems to support x86: live images are for x86 only because it uses syslinux and friends. However you can choose any other image to become initramfs image and bunsle it into kernel you have to use INITRAMFS_IMAGE = your-initramfs-image INITRAMFS_IMAGE_BUNDLE = 1 COMPATIBLE_HOST = (i.86|x86_64).*-linux Is this the limitation of current OE? or am I just looking at the wrong corner? i recently had to build an initrd too, and notice that as well... so i am not sure exactly what that means. I am glad you asked... however if you want to test an initrd, you can simply build *any* image and make sure to build the 'cpio' from IMAGE_FSTYPES. you can then use the generated cpio archive as an inited. -- ___ Openembedded-devel mailing list Openembedded-devel@lists.openembedded.org http://lists.openembedded.org/mailman/listinfo/openembedded-devel