On Mon, 2009-06-22 at 17:46 +0100, Russell King - ARM Linux wrote: > On Mon, Jun 22, 2009 at 05:13:16PM +0100, Catalin Marinas wrote: > > In my case (ext2 over pata_platform), there is no flush_dcache_page() > > call after the page was written with data from the CompactFlash (neither > > the driver nor the VFS layer do this and we used hardware tracing to > > double-check). When the page is mapped into user space, > > update_mmu_cache() is called but the page hasn't been marked as dirty > > and no D-cache flushing occurs. Calling flush_dcache_page() in > > mpage_end_io_read() works around this issue. > > As already covered, there's no chance of adding such a call to the > generic kernel. It's the responsibility of the drivers to ensure that > data they read in hits the underlying page - in the same way that DMA > does. The patch below appears to solve the problem with CompactFlash using pata_platform (I cc'ed linux-ide since the patch changes their code). The patch only handles the read case but similarly it may need to handle the write case if D-cache aliasing between user and kernel mappings exists. For the USB mass storage, I haven't yet figured out the best place to call flush_dcache_page(). Call flush_dcache_page after PIO data transfer in libata-aff.c From: Catalin Marinas <catalin.marinas@xxxxxxx> When reading data from an ATA device using PIO, the kernel dirties the D-cache but there is no flush_dcache_page() call in ata_pio_sector(). Since neither the VFS layer calls this function, a subsequent update_mmu_cache() is not aware of the dirty page which may lead to cache incoherency in user space. Signed-off-by: Catalin Marinas <catalin.marinas@xxxxxxx> --- drivers/ata/libata-sff.c | 3 +++ 1 files changed, 3 insertions(+), 0 deletions(-) diff --git a/drivers/ata/libata-sff.c b/drivers/ata/libata-sff.c index bbbb1fa..2ae15c3 100644 --- a/drivers/ata/libata-sff.c +++ b/drivers/ata/libata-sff.c @@ -893,6 +893,9 @@ static void ata_pio_sector(struct ata_queued_cmd *qc) do_write); } + if (!do_write) + flush_dcache_page(page); + qc->curbytes += qc->sect_size; qc->cursg_ofs += qc->sect_size; -- Catalin -- To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
