Hi Lior,
On Thu, Dec 14, 2023 at 04:04:31PM +0000, Lior Weintraub wrote:
> Hi,
>
> The below patch fixes the mmu_early_enable function to correctly map 40bits of virtual address into physical address with a 1:1 mapping.
> It uses the init_range function to sets 2 table entries on TTB level0 and then fill level1 with the correct 1:1 mapping.
>
> This patch was merged from an older Barebox version into the most resent master (Commit: 975acf1bafba2366eb40c5e8d8cb732b53f27aa1).
> Since it wasn't tested on the most recent master branch (lack of resources) I would appreciate if someone can test it on a 64bit ARM v8 platform.
>
> IMHO, the old implementation is wrong because:
> 1. It tries to map the full range of VA (48bits) with 1:1 mapping but there are only maximum of 40 PA bits.
> As a result, there is a wraparound that causes wrong mapping.
> 2. TTB Level0 cannot have a block descriptor. Only table descriptor.
> According "Learn the architecture - AArch64 memory management", Figure 6-1: Translation table format:
> "Each entry is 64 bits and the bottom two bits determine the type of entry.
> Notice that some of the table entries are only valid at specific levels. The maximum number of
> levels of tables is four, which is why there is no table descriptor for level 3 (or the fourth level),
> tables. Similarly, there are no Block descriptors or Page descriptors for level 0. Because level 0
> entry covers a large region of virtual address space, it does not make sense to allow blocks."
>
> Cheers,
> Lior.
>
> From a98fa2bad05721fd4c3ceae4f63eedd90c29c244 Mon Sep 17 00:00:00 2001
> From: Lior Weintraub <liorw@xxxxxxxxxx>
> Date: Thu, 14 Dec 2023 17:05:04 +0200
> Subject: [PATCH] [ARM64][MMU] Fix mmu_early_enable VA->PA mapping
>
> ---
> arch/arm/cpu/mmu_64.c | 17 ++++++++++++++++-
> arch/arm/cpu/mmu_64.h | 19 +++++++++++++++++--
> arch/arm/include/asm/pgtable64.h | 1 +
> 3 files changed, 34 insertions(+), 3 deletions(-)
>
> diff --git a/arch/arm/cpu/mmu_64.c b/arch/arm/cpu/mmu_64.c
> index c6ea63e655..f35c1b5937 100644
> --- a/arch/arm/cpu/mmu_64.c
> +++ b/arch/arm/cpu/mmu_64.c
> @@ -294,6 +294,19 @@ void dma_flush_range(void *ptr, size_t size)
> v8_flush_dcache_range(start, end);
> }
>
> +void init_range(void *virt_addr, size_t size)
> +{
> + uint64_t *ttb = get_ttb();
> + uint64_t addr = (uint64_t)virt_addr;
> + while(size) {
> + remap_range((void *)addr, L0_XLAT_SIZE, MAP_UNCACHED);
This should be early_remap_range(). remap_range() is not safe to be
called at this early stage and breaks running with qemu:
https://github.com/barebox/barebox/actions/runs/7246849041/job/19739568067
When sending a correct version, could you please add your signed-off-by
to the patch?
Also, please send the patch as a new thread, not in response to another
mail.
Sascha
--
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 |
