26.09.2020 11:07, Nicolin Chen пишет:
...
> + for (pd_index = 0; pd_index < SMMU_NUM_PDE; pd_index++) {
> + struct page *pt;
> + u32 *addr;
> +
> + if (!as->count[pd_index] || !pd[pd_index])
> + continue;
I guess the idea of this patch is to print out the hardware state, isn't
it? Hence the as->count shouldn't be checked here.
> + pde_count++;
> + pte_count += as->count[pd_index];
> + seq_printf(s, "\t[%d] 0x%x (%d)\n",
> + pd_index, pd[pd_index], as->count[pd_index]);
> + pt = as->pts[pd_index];
> + addr = page_address(pt);
> +
> + seq_puts(s, "\t{\n");
> + seq_printf(s, "\t\t%-5s %-4s %12s %12s\n", "PDE", "ATTR", "PHYS", "IOVA");
> + for (pt_index = 0; pt_index < SMMU_NUM_PTE; pt_index++) {
> + u64 iova;
> +
> + if (!addr[pt_index])
> + continue;
> +
> + iova = ((dma_addr_t)pd_index & (SMMU_NUM_PDE - 1)) << SMMU_PDE_SHIFT;
> + iova |= ((dma_addr_t)pt_index & (SMMU_NUM_PTE - 1)) << SMMU_PTE_SHIFT;
> +
> + seq_printf(s, "\t\t#%-4d 0x%-4x 0x%-12llx 0x%-12llx\n",
> + pt_index, addr[pt_index] >> SMMU_PTE_ATTR_SHIFT,
> + SMMU_PFN_PHYS(addr[pt_index] & ~SMMU_PTE_ATTR), iova);
>
Would be nice if you could improve the output formatting by printing out
contiguous ranges that have the same ATTRs, otherwise it could be a bit
too large and unpractical output in a case if lots of pages are mapped.
