[Qemu-devel] Data bus error with redeonfb on mips_fulong2e

[BALATON Zoltan]

Hello,

Trying to debug the Linux kernel oops with radeonfb I've added some more 
debug logs and got this:

radeonfb_pci_register BEGIN
pci_host_data: pci_data_read: ati-vga: addr=04 val=00000000 len=4
PCI: Enabling device 0000:00:06.0 (0000 -> 0003)
pci_host_data: pci_data_read: ati-vga: addr=04 val=00000000 len=4
pci_host_data: pci_data_write: ati-vga: addr=04 val=00000003 len=4
pci_update_mappings: adding bar 0 to pci.mem @ 0x14000000
pci_update_mappings: adding bar 1 to io @ 0x4000
pci_update_mappings: adding bar 2 to pci.mem @ 0x15050000
pci_host_data: pci_data_read: ati-vga: addr=3c val=00000000 len=4
radeonfb_pci_register fb_base_phys=14000000
radeonfb_pci_register mmio_base_phys=15050000
radeonfb_pci_register request mem regions
radeonfb_pci_register map regions
radeonfb_pci_register mmio_base=9000000015050000
Data bus error, epc == ffffffff80418e4c, ra == ffffffff80418e4c
(full oops dump in previous message but I couldn't find anything useful in 
there so omitted here)

I think the problem is around here (I have to guess because Linux prints 
(ptrval) instead of offsets in oops message as a security measure that I 
don't know how to turn off and the mips64 objdump that comes with my 
distro can't interleave source with disassembly so even if I have the 
offset can't find the source line by it so I had to go by debug printfs):

linux-4.15.10/drivers/video/fbdev/aty/radeon_base.c:
   2311         /* Set base addrs */
   2312         rinfo->fb_base_phys = pci_resource_start (pdev, 0);
   2313         rinfo->mmio_base_phys = pci_resource_start (pdev, 2);
   2314         pr_debug("radeonfb_pci_register fb_base_phys=%lx\n", 
rinfo->fb_base_phys);
   2315         pr_debug("radeonfb_pci_register mmio_base_phys=%lx\n", 
rinfo->mmio_base_phys);
   2316
   2317         /* request the mem regions */
   2318         pr_debug("radeonfb_pci_register request mem regions\n");
   2319         ret = pci_request_region(pdev, 0, "radeonfb framebuffer");
   2320         if (ret < 0) {
   2321                 printk( KERN_ERR "radeonfb (%s): cannot request region 
0.\n",
   2322                         pci_name(rinfo->pdev));
   2323                 goto err_release_fb;
   2324         }
   2325
   2326         ret = pci_request_region(pdev, 2, "radeonfb mmio");
   2327         if (ret < 0) {
   2328                 printk( KERN_ERR "radeonfb (%s): cannot request region 
2.\n",
   2329                         pci_name(rinfo->pdev));
   2330                 goto err_release_pci0;
   2331         }
   2332
   2333         /* map the regions */
   2334         pr_debug("radeonfb_pci_register map regions\n");
   2335         rinfo->mmio_base = ioremap(rinfo->mmio_base_phys, 
RADEON_REGSIZE);
   2336         if (!rinfo->mmio_base) {
   2337                 printk(KERN_ERR "radeonfb (%s): cannot map MMIO\n",
   2338                        pci_name(rinfo->pdev));
   2339                 ret = -EIO;
   2340                 goto err_release_pci2;
   2341         }
   2342         pr_debug("radeonfb_pci_register mmio_base=%px\n", 
rinfo->mmio_base);
   2343
   2344         rinfo->fb_local_base = INREG(MC_FB_LOCATION) << 16;
   2345         pr_debug("radeonfb_pci_register fb_local_base=%lx\n", 
rinfo->fb_local_base);

So it fails in INREG on line 2344 which is in radeonfb.h:380

#define INREG(addr)             readl((rinfo->mmio_base)+addr)

This reg read never reaches QEMU so either something's wrong with the 
mmio_base or readl can't handle it correctly. This seems to be something 
Linux MIPS specific that I have no idea about. Does this make sense to 
anyone?

Regards,
BALATON Zoltan