On Tue, Jul 28, 2020 at 07:50:04AM +0000, Peng Fan wrote: > > Subject: Re: [PATCH 00/10] remoteproc: imx_rproc: support iMX8M and early > > boot > > > > On Mon, Jul 27, 2020 at 09:18:31AM +0000, Peng Fan wrote: > > > > Subject: Re: [PATCH 00/10] remoteproc: imx_rproc: support iMX8M and > > > > early boot > > > > > > > > On Mon, Jul 27, 2020 at 06:44:32AM +0000, Peng Fan wrote: > > > > > Hi Oleksij, > > > > > > > > > > > Subject: Re: [PATCH 00/10] remoteproc: imx_rproc: support iMX8M > > > > > > and early boot > > > > > > > > > > > > Hi, > > > > > > > > > > > > On Fri, Jul 24, 2020 at 04:08:03PM +0800, Peng Fan wrote: > > > > > > > This patchset is to support i.MX8MQ/M coproc booted before linux. > > > > > > > Since i.MX8MQ/M was not supported, several patches are needed > > > > > > > to first support the platform, then support early boot case. > > > > > > > > > > > > > > I intended to included i.MX8QM/QXP, but that would introduce a > > > > > > > large patchset, so not included. But the clk/syscon optional > > > > > > > patch for i.MX8QM/QXP was still kept here to avoid rebase error. > > > > > > > > > > > > Thank you for your work. > > > > > > > > > > > > Can you please provide more information about big picture of this > > work. > > > > > > > > > > > > If I see it correctly, we have here support for i.MX8MM, which > > > > > > seems to be able to fully control Cortex M4 (enable CPU core, etc...). > > > > > > > > > > Yes. > > > > > > > > In this case, I would recommend to mainline the i.MX8MM part > > > > first/separately. > > > > > > Only the last patch is to support earlyboot, all others is imx8mm part. > > > > ok > > > > > > > > > > > > > > > > > > And other case, where remoteproc is running on application > > > > > > processor and can't or should not touch M4 (i.MX7ULP, > > > > > > i.MX8QM/QXP..). Since M4 provides some functionality, you are > > > > > > trying to reuse remoteproc framework to get resource table > > > > > > present in ELF header and to dynamically load things. For some > > > > > > reasons this header provides more information then needed, so > > > > > > you are changing the ELF parser in the kernel > > > > to workaround it. > > > > > > > > > > Not exactly. > > > > > > > > > > For i.MX8MM, we support two cases. M4 kicked by U-Boot, M4 kicked > > > > > by > > > > Linux remoteproc. > > > > > For i.MX8QM/QXP, the typical usecase is M4 kicked by SCFW, but we > > > > > will also add M4 kicked by Linux remoteproc. > > > > > For i.MX7ULP, I would only support M4 dual boot case, M4 control > > > > everything. > > > > > > > > From current state of discussion, i'm not sure what role plays > > > > remoteproc in the scenario where M4 is started before linux. > > > > Especially if we are not using resource table. > > > > > > We are using resource table from an address, not in elf file. > > > This is the new feature in Linux-next to support coproc booted early. > > > > > > > > > > > > The reason the change the elf parser is that when M4 elf is loaded > > > > > by Linux remoteproc, It use memset to clear area. > > > > > > > > The use of memset, depends on ELF format. Fix/change the linker > > > > script on your firmware and memset will be never called. > > > > > > > > > However we use ioremap, memset on ARM64 will report crash to > > > > > device nGnRE memory. And we could not use ioremap_wc to TCM area, > > > > > since it could have data correctly written into TCM. > > > > > > > > I have strong feeling, that we are talking about badly or not > > > > properly formatted ELF binary. I would prefer to double check it, > > > > before we will apply fixes on wrong place. > > > > > > > > > Maintainer not wanna to drop memset in common code, and TI guys > > > > > suggest add i.MX specific elf stuff. So I add elf handler in i.MX code. > > > > > > > > I think, removing memset may damage current users of imx_rproc driver. > > > > Since, like I said: the use of memset depends on ELF format. > > > > > > In my elf file, the last PT_LOAD contains data/bss/heap/stack. I'll > > > check with our MCU guys, we only need the specific data loaded. > > > > > > Elf file type is EXEC (Executable file) Entry point 0x1ffe0355 There > > > are 3 program headers, starting at offset 52 > > > > > > Program Headers: > > > Type Offset VirtAddr PhysAddr FileSiz MemSiz > > Flg Align > > > LOAD 0x010000 0x1ffe0000 0x1ffe0000 0x00240 0x00240 > > R 0x10000 > > > LOAD 0x010240 0x1ffe0240 0x1ffe0240 0x03e90 0x03e90 > > RWE 0x10000 > > > LOAD 0x020000 0x20000000 0x1ffe40d0 0x00068 0x0ad00 > > RW 0x10000 > > > > > > Section to Segment mapping: > > > Segment Sections... > > > 00 .interrupts > > > 01 .resource_table .text .ARM .init_array .fini_array > > > 02 .data .bss .heap .stack > > > > Here is an example of formatting ELF for remoteproc: > > https://git.pengutronix.de/cgit/ore/OSELAS.BSP-Pengutronix-DualKit/tree/loc > > al_src/remoteproc-elf/linker.ld > > https://git.pengutronix.de/cgit/ore/OSELAS.BSP-Pengutronix-DualKit/tree/loc > > al_src/remoteproc-elf/imx7m4.S > > > > In this example I pack linux in to remoteproc elf image and start linux on > > imx7d-m4 part. > > Will be interesting if you can do the same on imx8* SoCs ;) > > In NXP release, the m4 elf files have data/bss/heap/stack in the same > data area, so the linker merged them into one segment and cause > memsz > filesz. > > I think I need to propose platform specific elf memset/memcpy, > such as rproc_elf_memcpy, rproc_elf_memset, > > To i.MX, need use memset_io and memcpy_toio, taking TCM > as device memory. > > Note: memset without io will cause abort when memsz>filesz. > So use memset_io is safe. Sounds good, i would prefer this way. -- Pengutronix e.K. | | Steuerwalder Str. 21 | http://www.pengutronix.de/ | 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |