Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider: - [Medium] drm/imagination: wrong page count and loop bound in pvr_vm_get_pt_phys_addrs() - [Low] drm/imagination: missing types.h include and prefix in pvr_vm_mips.h --
commit ba830ed70a6d676494fb621937678d6996067bfb Author: Alessio Belle <[email protected]> drm/imagination: Limit access to the MIPS firmware interface This commit splits struct pvr_fw_mips_data into separate internal structures for the main MIPS firmware logic and the MIPS VM logic, removes pvr_fw_mips.h, renames the MIPS firmware processor headers with an _internal suffix, and adds pvr_vm_get_pt_phys_addrs() to retrieve the MIPS firmware page table addresses. > diff --git a/drivers/gpu/drm/imagination/pvr_vm_mips.c > b/drivers/gpu/drm/imagination/pvr_vm_mips.c > index 5847a1c92bea8..264ea0fccf9c1 100644 > --- a/drivers/gpu/drm/imagination/pvr_vm_mips.c > +++ b/drivers/gpu/drm/imagination/pvr_vm_mips.c > @@ -16,6 +17,31 @@ > #include <linux/types.h> > #include <linux/vmalloc.h> > > +#define PVR_MIPS_PT_PAGE_COUNT \ > + DIV_ROUND_UP(ROGUE_MIPSFW_MAX_NUM_PAGETABLE_PAGES * > ROGUE_MIPSFW_PAGE_SIZE_4K, PAGE_SIZE) > + > +/** > + * struct pvr_vm_mips_data - MIPS-VM-specific data > + */ > +struct pvr_vm_mips_data { > + /** > + * @pt_pages: Pages containing MIPS pagetable. > + */ > + struct page *pt_pages[PVR_MIPS_PT_PAGE_COUNT]; > + > + /** @pt: Pointer to CPU mapping of MIPS pagetable. */ > + u32 *pt; > + > + /** @pt_dma_addr: DMA mappings of MIPS pagetable. */ > + dma_addr_t pt_dma_addr[PVR_MIPS_PT_PAGE_COUNT]; [ ... ] > @@ -230,8 +256,38 @@ pvr_vm_mips_unmap(struct pvr_device *pvr_dev, struct > pvr_fw_object *fw_obj) > ROGUE_MIPSFW_LOG2_PAGE_SIZE_4K; > > for (u32 pfn = start_pfn; pfn < end_pfn; pfn++) > - WRITE_ONCE(mips_data->pt[pfn], 0); > + WRITE_ONCE(vm_data->pt[pfn], 0); > > pvr_mmu_flush_request_all(pvr_dev); > WARN_ON(pvr_mmu_flush_exec(pvr_dev, true)); > } > + > +/** > + * pvr_vm_get_pt_phys_addrs() - Acquire the physical addresses of the MIPS > FW page table pages > + * @pvr_dev: Target PowerVR device. > + * @phys_addrs: Output array for physical addresses of MIPS FW page table > pages. > + * @max_num_addrs: Maximum number of addresses that can be written to > @phys_addrs. > + * > + * Returns: > + * * 0 on success, or > + * * -%EINVAL if @phys_addrs cannot accommodate all MIPS FW page table > pages. > + */ > +int > +pvr_vm_get_pt_phys_addrs(struct pvr_device *pvr_dev, u64 *phys_addrs, int > max_num_addrs) > +{ > + struct pvr_fw_device *fw_dev = &pvr_dev->fw_dev; > + struct pvr_vm_mips_data *vm_data = > fw_dev->processor_data.mips_data.vm_data; > + > + if (ARRAY_SIZE(vm_data->pt_pages) > max_num_addrs) > + return -EINVAL; > + > + for (u32 page_nr = 0; page_nr < max_num_addrs; page_nr++) { > + /* Firmware expects 4k pages, but host page size might be > different. */ > + u32 src_page_nr = (page_nr * ROGUE_MIPSFW_PAGE_SIZE_4K) >> > PAGE_SHIFT; > + u32 page_offset = (page_nr * ROGUE_MIPSFW_PAGE_SIZE_4K) & > ~PAGE_MASK; > + > + phys_addrs[page_nr] = vm_data->pt_dma_addr[src_page_nr] + > page_offset; > + } > + > + return 0; > +} [Severity: Medium] As reported by Sashiko on v4 [1], in pvr_vm_get_pt_phys_addrs(), does checking ARRAY_SIZE(vm_data->pt_pages) and looping up to max_num_addrs use the wrong page count? vm_data->pt_pages and vm_data->pt_dma_addr are sized by PVR_MIPS_PT_PAGE_COUNT, which is the number of host PAGE_SIZE pages (for example, 1 page on 16K or 64K page kernels), whereas the loop steps by 4K firmware pages and requires ROGUE_MIPSFW_MAX_NUM_PAGETABLE_PAGES (4) entries. On hosts with PAGE_SIZE > 4K, ARRAY_SIZE(vm_data->pt_pages) is 1, so the check fails to reject a phys_addrs buffer with fewer than 4 entries. In addition, if a caller passes max_num_addrs greater than ROGUE_MIPSFW_MAX_NUM_PAGETABLE_PAGES, the loop reads past the end of vm_data->pt_dma_addr[]. Also, because max_num_addrs is a signed int, comparing it against ARRAY_SIZE() and u32 page_nr triggers -Wsign-compare warnings and would treat negative values as large unsigned numbers. Should max_num_addrs be u32 (or size_t), and should both the check and loop bound use ROGUE_MIPSFW_MAX_NUM_PAGETABLE_PAGES instead? > diff --git a/drivers/gpu/drm/imagination/pvr_vm_mips.h > b/drivers/gpu/drm/imagination/pvr_vm_mips.h > index 0fd59f68fb5b8..3db0606c50374 100644 > --- a/drivers/gpu/drm/imagination/pvr_vm_mips.h > +++ b/drivers/gpu/drm/imagination/pvr_vm_mips.h > @@ -18,5 +18,7 @@ int > pvr_vm_mips_map(struct pvr_device *pvr_dev, struct pvr_fw_object *fw_obj); > void > pvr_vm_mips_unmap(struct pvr_device *pvr_dev, struct pvr_fw_object *fw_obj); > +int > +pvr_vm_get_pt_phys_addrs(struct pvr_device *pvr_dev, u64 *phys_addrs, int > max_num_addrs); > > #endif /* PVR_VM_MIPS_H */ [Severity: Low] In drivers/gpu/drm/imagination/pvr_vm_mips.h, pvr_vm_get_pt_phys_addrs() uses u64 in its prototype, but pvr_vm_mips.h does not include <linux/types.h>. Could this fail to compile if pvr_vm_mips.h is included before <linux/types.h>? Also, this isn't a bug, but all other functions in pvr_vm_mips.h use the pvr_vm_mips_ prefix (pvr_vm_mips_init(), pvr_vm_mips_fini(), pvr_vm_mips_map(), pvr_vm_mips_unmap()). Should this function be named pvr_vm_mips_get_pt_phys_addrs() for consistency? [1] https://lore.kernel.org/r/[email protected] -- Sashiko AI review ยท https://sashiko.dev/#/patchset/[email protected]?part=1
