lkml.org 
[lkml]   [2017]   [Jul]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: [PATCH v8 6/6] drm/amdgpu: resize VRAM BAR for CPU access v4
    On Mon, Jul 10, 2017 at 04:59:54PM +0200, Christian König wrote:
    > From: Christian König <christian.koenig@amd.com>
    >
    > Try to resize BAR0 to let CPU access all of VRAM.
    >
    > v2: rebased, style cleanups, disable mem decode before resize,
    > handle gmc_v9 as well, round size up to power of two.
    > v3: handle gmc_v6 as well, release and reassign all BARs in the driver.
    > v4: rename new function to amdgpu_device_resize_fb_bar,
    > reenable mem decoding only if all resources are assigned.
    >
    > Signed-off-by: Christian König <christian.koenig@amd.com>
    > ---
    > drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 +
    > drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 46 ++++++++++++++++++++++++++++++
    > drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c | 8 ++++--
    > drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 8 ++++--
    > drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 8 ++++--
    > drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 10 ++++---
    > 6 files changed, 68 insertions(+), 13 deletions(-)
    >
    > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
    > index ff7bf1a..5809e0e 100644
    > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
    > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
    > @@ -1944,6 +1944,7 @@ uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
    > struct ttm_mem_reg *mem);
    > void amdgpu_vram_location(struct amdgpu_device *adev, struct amdgpu_mc *mc, u64 base);
    > void amdgpu_gtt_location(struct amdgpu_device *adev, struct amdgpu_mc *mc);
    > +void amdgpu_device_resize_fb_bar(struct amdgpu_device *adev);
    > void amdgpu_ttm_set_active_vram_size(struct amdgpu_device *adev, u64 size);
    > int amdgpu_ttm_init(struct amdgpu_device *adev);
    > void amdgpu_ttm_fini(struct amdgpu_device *adev);
    > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
    > index 7d1a04a..a5123f5 100644
    > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
    > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
    > @@ -428,6 +428,9 @@ static int amdgpu_doorbell_init(struct amdgpu_device *adev)
    > return 0;
    > }
    >
    > + if (pci_resource_flags(adev->pdev, 2) & IORESOURCE_UNSET)
    > + return -EINVAL;
    > +
    > /* doorbell bar mapping */
    > adev->doorbell.base = pci_resource_start(adev->pdev, 2);
    > adev->doorbell.size = pci_resource_len(adev->pdev, 2);
    > @@ -709,6 +712,49 @@ void amdgpu_gtt_location(struct amdgpu_device *adev, struct amdgpu_mc *mc)
    > mc->gtt_size >> 20, mc->gtt_start, mc->gtt_end);
    > }
    >
    > +/**
    > + * amdgpu_device_resize_fb_bar - try to resize FB BAR
    > + *
    > + * @adev: amdgpu_device pointer
    > + *
    > + * Try to resize FB BAR to make all VRAM CPU accessible.
    > + */
    > +void amdgpu_device_resize_fb_bar(struct amdgpu_device *adev)
    > +{
    > + u64 space_needed = roundup_pow_of_two(adev->mc.real_vram_size);
    > + u32 rbar_size = order_base_2(((space_needed >> 20) | 1)) - 1;
    > + u16 cmd;
    > + int r;
    > +
    > + /* Disable memory decoding while we change the BAR addresses and size */
    > + pci_read_config_word(adev->pdev, PCI_COMMAND, &cmd);
    > + pci_write_config_word(adev->pdev, PCI_COMMAND,
    > + cmd & ~PCI_COMMAND_MEMORY);
    > +
    > + /* Free the VRAM and doorbell BAR, we most likely need to move both. */
    > + amdgpu_doorbell_fini(adev);
    > + pci_release_resource(adev->pdev, 0);
    > + if (adev->asic_type >= CHIP_BONAIRE)
    > + pci_release_resource(adev->pdev, 2);

    The related pieces should be together, e.g.,

    amdgpu_doorbell_fini(adev);
    if (adev->asic_type >= CHIP_BONAIRE)
    pci_release_resource(adev->pdev, 2);

    pci_release_resource(adev->pdev, 0);

    > + r = pci_resize_resource(adev->pdev, 0, rbar_size);
    > + if (r == -ENOSPC)
    > + DRM_INFO("Not enough PCI address space for a large BAR.");
    > + else if (r && r != -ENOTSUPP)
    > + DRM_ERROR("Problem resizing BAR0 (%d).", r);
    > +
    > + pci_assign_unassigned_bus_resources(adev->pdev->bus);
    > +
    > + /* When the doorbell or fb BAR isn't available we have no chance of
    > + * using the device.
    > + */
    > + r = amdgpu_doorbell_init(adev);
    > + if ((pci_resource_flags(adev->pdev, 0) & IORESOURCE_UNSET) || r)
    > + BUG();

    Seems extreme to panic the system in this case.

    I don't think IORESOURCE_UNSET should appear in a driver. I think that's
    an internal property that should be handled in the core.

    What if you added something like this:

    int pci_enable_mem_decoding(struct pci_dev *dev)
    {
    unsigned int bars = 0;

    for (i = 0; i <= PCI_ROM_RESOURCE; i++)
    if (dev->resource[i].flags & IORESOURCE_MEM)
    bars |= (1 << i);

    return pci_enable_resources(pdev, bars);
    }

    > + else
    > + pci_write_config_word(adev->pdev, PCI_COMMAND, cmd);
    > +}

    \
     
     \ /
      Last update: 2017-07-11 00:24    [W:3.769 / U:0.296 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site