nvme: invalidate correct memory range after read
authorStefan Agner <stefan@agner.ch>
Mon, 4 Oct 2021 09:24:51 +0000 (11:24 +0200)
committerTom Rini <trini@konsulko.com>
Tue, 12 Oct 2021 20:49:21 +0000 (16:49 -0400)
The current code invalidates the range after the read buffer since the
buffer pointer gets incremented in the read loop. Use a temporary
pointer to make sure we have a pristine pointer to invalidate the
correct memory range after read.

Fixes: 704e040a51d2 ("nvme: Apply cache operations on the DMA buffers")
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Stefan Agner <stefan@agner.ch>
drivers/nvme/nvme.c

index f6465ea..3c529a2 100644 (file)
@@ -743,6 +743,7 @@ static ulong nvme_blk_rw(struct udevice *udev, lbaint_t blknr,
        u64 prp2;
        u64 total_len = blkcnt << desc->log2blksz;
        u64 temp_len = total_len;
+       uintptr_t temp_buffer = (uintptr_t)buffer;
 
        u64 slba = blknr;
        u16 lbas = 1 << (dev->max_transfer_shift - ns->lba_shift);
@@ -770,19 +771,19 @@ static ulong nvme_blk_rw(struct udevice *udev, lbaint_t blknr,
                }
 
                if (nvme_setup_prps(dev, &prp2,
-                                   lbas << ns->lba_shift, (ulong)buffer))
+                                   lbas << ns->lba_shift, temp_buffer))
                        return -EIO;
                c.rw.slba = cpu_to_le64(slba);
                slba += lbas;
                c.rw.length = cpu_to_le16(lbas - 1);
-               c.rw.prp1 = cpu_to_le64((ulong)buffer);
+               c.rw.prp1 = cpu_to_le64(temp_buffer);
                c.rw.prp2 = cpu_to_le64(prp2);
                status = nvme_submit_sync_cmd(dev->queues[NVME_IO_Q],
                                &c, NULL, IO_TIMEOUT);
                if (status)
                        break;
                temp_len -= (u32)lbas << ns->lba_shift;
-               buffer += lbas << ns->lba_shift;
+               temp_buffer += lbas << ns->lba_shift;
        }
 
        if (read)