ARM TCM (Tightly-Coupled Memory) support v1 For details about TCM and this specific interface for Linux, see the patch, which includes a document in Documentation/arm/tcm.txt
I expect a few iterations on this, so it's not like this is a merge candidate. Wait for v5 or so... However it works. Result of test run using the example program in tcm.txt on U335: CPU: found DTCM 8k @ 00010000, enabled CPU: moved DTCM 8k to ffff4000, enabled CPU DTCM: copied data from ffff4000 - ffff4008 CPU: found ITCM 8k @ 00000000, enabled CPU: moved ITCM 8k to ffff2000, enabled CPU ITCM: copied code from ffff2000 - ffff201c Setting up TCM memory pool CPU DTCM: Added 00001ff8 bytes @ ffff4008 to the TCM memory pool CPU ITCM: Added 00001fe4 bytes @ ffff201c to the TCM memory pool Hello TCM executed from ITCM RAM TCM variable from testrun: 200 @ ffff4000 TCM variable: 0xdeadbeef @ ffff4000 TCM assigned variable: 0x2badbabe @ ffff4004 TCM constant: 0xcafebabe @ ffff2018 TCM Allocated 20 bytes of TCM @ ffff201c TCM tcmem[0] = deadbeef TCM tcmem[1] = 2badbabe TCM tcmem[2] = cafebabe TCM tcmem[3] = deadbeef TCM tcmem[4] = 2badbabe Other implementations: I've noticed that OMAP1 has TCM, it is however only ever used from the assembly code found in mach-omap1/sleep.S. Using this on the OMAP too it's possible to write C and assembly alike for the TCM. C is nice. I expect it'd be easy to switch over to using this. System RAM: SystemRAM is a generic on-chip memory (also called retention RAM, backup RAM or just plain on-die/chip RAM) which is used in som SoC designs for much the same purposes as TCM memories. TCM is not used much in newer designs in favor of System RAM. SystemRAM is different in that it is not Harvard-style so like in common RAM instructions and data can be mixed. It is also often below L1 or even L2 caches, so it behaves just like normal memory. Following the style of the TCM memory implementation it is easy to add a similar mechanism för System RAM with some interface in asm/sysram.h or so, but it is distincly different due to the Harvard nature of TCM and the fact that TCMs can be arbitrarily moved about in the physical memory map. TI:s DaVinci seems to have something called "sysRAM" instead in mach-davinci/sram.c perhaps this can be treated exactly as an ITCM, what do I know, however in difference from ITCM I believe that is outside the CPU and thus code running from that RAM can "see" the MMU remappings. davinci creates a special memory pool for the sysram on that platform and currently that seems to be the only way to use it. I stronly believe that the DaVinci solution should be made generic and moved out of mach-davinci, and modeled to be able to compile code into as with TCM, but I have no suitable machine to test that on so that will be on my TODO. Yours, Linus Walleij -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
