Re: [patch 2.6.29-davinci1 git] dm6446evm CF support
David Brownell wrote: This provides basic support for the CF slot on the dm6446 EVM. Insert/remove events are not (currently) supported. On a very small sample of cards, using hdparm -tT to measure (and correct for buffer access), throughput is about 1.5 MB/sec. That's a third what an ATA drive gives on the same system; the CF cards do not support DMA access modes. Root on an SD card remains the fastest choice for non-NFS root (over 8 MB/sec). (Unrelated: remove obsolete ref to RTC.) Signed-off-by: David Brownell dbrown...@users.sourceforge.net --- a/arch/arm/mach-davinci/Kconfig +++ b/arch/arm/mach-davinci/Kconfig @@ -26,6 +26,17 @@ config MACH_DAVINCI_EVM Configure this option to specify the whether the board used for development is a DM644x EVM +config MACH_DAVINCI_EVM_CF + bool Use CompactFlash, not IDE drive + depends on MACH_DAVINCI_EVM BLK_DEV_PALMCHIP_BK3710 + help + Enable this if you want to use a CompactFlash card instead of + an IDE/ATA drive. The CF card will be used in True IDE mode. + + At this writing, CF insert and remove events are not supported. + Insert the CF at boot time, and leave it there until you power + the system off. + This instruction is already obsolete. As I've already noted, IDE core now supports warm plug via sysfs, see Documentation/ide/warm-plug-howto.txt. One can plug and unplug any time, this just won't be automatically noticed by the kernel, so one will need to notify it. MBR, Sergei ___ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
Re: [patch 2.6.29-davinci1 git] dm6446evm CF support
On Wednesday 01 April 2009, Sergei Shtylyov wrote: +config MACH_DAVINCI_EVM_CF + bool Use CompactFlash, not IDE drive + depends on MACH_DAVINCI_EVM BLK_DEV_PALMCHIP_BK3710 + help + Enable this if you want to use a CompactFlash card instead of + an IDE/ATA drive. The CF card will be used in True IDE mode. + + At this writing, CF insert and remove events are not supported. + Insert the CF at boot time, and leave it there until you power + the system off. + This instruction is already obsolete. As I've already noted, IDE core now supports warm plug via sysfs, see Documentation/ide/warm-plug-howto.txt. One can plug and unplug any time, this just won't be automatically noticed by the kernel, so one will need to notify it. CF insert and remove events are very different from writing some sysfs files. They are handled through a pcmcia_socket abstraction ... and are completely hands-off. Socket events include card insert/removal detection, and socket operations include power on/off and reset. I don't know that true-IDE links work through those sockets, though. That said ... I tried that manual workaround; it didn't work. So: no, that instruction is far from obsolete. ___ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
Re: [patch 2.6.29-davinci1 git] dm6446evm CF support
Hello. David Brownell wrote: +config MACH_DAVINCI_EVM_CF + bool Use CompactFlash, not IDE drive + depends on MACH_DAVINCI_EVM BLK_DEV_PALMCHIP_BK3710 + help + Enable this if you want to use a CompactFlash card instead of + an IDE/ATA drive. The CF card will be used in True IDE mode. + + At this writing, CF insert and remove events are not supported. + Insert the CF at boot time, and leave it there until you power + the system off. + This instruction is already obsolete. As I've already noted, IDE core now supports warm plug via sysfs, see Documentation/ide/warm-plug-howto.txt. One can plug and unplug any time, this just won't be automatically noticed by the kernel, so one will need to notify it. CF insert and remove events are very different from writing some sysfs files. They are handled through a pcmcia_socket abstraction ... and are completely hands-off. Socket events include card insert/removal detection, and socket operations include power on/off and reset. I don't know that true-IDE links work through those sockets, though. That said ... I tried that manual workaround; it didn't work. So: no, that instruction is far from obsolete. So, it would only work with real hard disks... sigh. Sorry for disinformation then -- I hoped warm plug could've been helpful for CF. MBR, Sergei ___ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
Re: [patch 2.6.29-davinci1 git] dm6446evm CF support
Hello. David Brownell wrote: From: David Brownell dbrown...@users.sourceforge.net This provides basic support for the CF slot on the dm6446 EVM. Oops, I was going to submit that months ago, and still haven't. :- Insert/remove events are not (currently) supported. IDE now has warm plug support anyway. On a very small sample of cards, using hdparm -tT to measure (and correct for buffer access), throughput is about 1.5 MB/sec. That's damn slow... That's a third what an ATA drive gives on the same system; the CF cards do not support DMA access modes. Those CF cards probably don't even support PIO3/4 because using DMA ATA drive's speed whould've been order of magnitude better than 4.5 MB/sec (which should correspond to PIO4). (Unrelated: remove obsolete ref to RTC.) Why do it in the same patch? :-/ Signed-off-by: David Brownell dbrown...@users.sourceforge.net --- Just a quick hack, to get my brain off some other issues. ;) arch/arm/mach-davinci/Kconfig| 11 + arch/arm/mach-davinci/board-dm644x-evm.c | 33 ++--- 2 files changed, 32 insertions(+), 12 deletions(-) --- a/arch/arm/mach-davinci/Kconfig +++ b/arch/arm/mach-davinci/Kconfig @@ -26,6 +26,17 @@ config MACH_DAVINCI_EVM Configure this option to specify the whether the board used for development is a DM644x EVM +config MACH_DAVINCI_EVM_CF + bool Use CompactFlash, not IDE drive + depends on MACH_DAVINCI_EVM BLK_DEV_PALMCHIP_BK3710 I was thinking of making this a choice instead. And I don't see why it should depend on BLK_DEV_PALMCHIP_BK3710. WBR, Sergei ___ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
Re: [patch 2.6.29-davinci1 git] dm6446evm CF support
On Monday 30 March 2009, Sergei Shtylyov wrote: That's a third what an ATA drive gives on the same system; the CF cards do not support DMA access modes. Those CF cards probably don't even support PIO3/4 because using DMA ATA drive's speed whould've been order of magnitude better than 4.5 MB/sec (which should correspond to PIO4). I'm not sure that the the IDE controller (Palmchip BK3710) supports transfers that fast even when using DMA. Taking a laptop drive and hooking it up to the IDE link gave lower throughput (much lower!) than the laptop had. It's either the controller or its driver which is the bottleneck. +config MACH_DAVINCI_EVM_CF + bool Use CompactFlash, not IDE drive + depends on MACH_DAVINCI_EVM BLK_DEV_PALMCHIP_BK3710 I was thinking of making this a choice instead. And I don't see why it should depend on BLK_DEV_PALMCHIP_BK3710. There's no point in offering the choice if its IDE driver isn't configured. ___ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
Re: [patch 2.6.29-davinci1 git] dm6446evm CF support
David Brownell wrote: That's a third what an ATA drive gives on the same system; the CF cards do not support DMA access modes. Those CF cards probably don't even support PIO3/4 because using DMA ATA drive's speed whould've been order of magnitude better than 4.5 MB/sec (which should correspond to PIO4). I'm not sure that the the IDE controller (Palmchip BK3710) supports transfers that fast even when using DMA. Taking a laptop drive and hooking it up to the IDE link gave lower throughput (much lower!) than the laptop had. It's either the controller or its driver which is the bottleneck. Russing 'hdrparm -t /dev/hda' gives me about up to 30 MB/s on DM6467 --that's not 45 but not 4.5 either. +config MACH_DAVINCI_EVM_CF + bool Use CompactFlash, not IDE drive + depends on MACH_DAVINCI_EVM BLK_DEV_PALMCHIP_BK3710 I was thinking of making this a choice instead. And I don't see why it should depend on BLK_DEV_PALMCHIP_BK3710. There's no point in offering the choice if its IDE driver isn't configured. I'm generally not happy about the inverse dependencies like that (AKA ARM #ifdef hell). WBR, Sergei ___ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
Re: [patch 2.6.29-davinci1 git] dm6446evm CF support
On Monday 30 March 2009, Sergei Shtylyov wrote: Russing 'hdrparm -t /dev/hda' gives me about up to 30 MB/s on DM6467 --that's not 45 but not 4.5 either. And on dm6446evm, matching $SUBJECT ... ? ___ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
Re: [patch 2.6.29-davinci1 git] dm6446evm CF support
David Brownell wrote: Russing 'hdrparm -t /dev/hda' gives me about up to 30 MB/s on DM6467 --that's not 45 but not 4.5 either. And on dm6446evm, matching $SUBJECT ... ? *Slightly* less. However, it's not with a recent kernel. With a 2.6.29-rc kernel, I din't have my HDD recognized at all. I've already seen such sympthoms with other board (MIPS). :-/ MBR, Sergei ___ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
Re: [patch 2.6.29-davinci1 git] dm6446evm CF support
On Monday 30 March 2009, Sergei Shtylyov wrote: Russing 'hdrparm -t /dev/hda' gives me about up to 30 MB/s on DM6467 --that's not 45 but not 4.5 either. And on dm6446evm, matching $SUBJECT ... ? *Slightly* less. Curious. As I said: moving a drive from a laptop to a dm6446evm gave me a substantial decrease in throughput, as measured by hdparm -tT -- at least 20 MByte/s less. I don't think I've ever seen the IDE interface give me even 6 MB/sec on dm6446evm. Different EVM, IDE claiming UDMA/66 ... 5 MB/sec. However, it's not with a recent kernel. With a 2.6.29-rc kernel, I din't have my HDD recognized at all. Worked fine for me. Unchanged with 2.6.29-final. ___ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source