diff options
author | Jerome Glisse <jglisse@redhat.com> | 2010-02-17 16:54:29 -0500 |
---|---|---|
committer | Dave Airlie <airlied@redhat.com> | 2010-02-17 23:49:35 -0500 |
commit | d594e46ace22afa1621254f6f669e65430048153 (patch) | |
tree | befd5b54ce1b8284acc4ee450d085a7d2c7b01fd /drivers/gpu/drm/radeon/radeon_device.c | |
parent | 44ca7478d46aaad488d916f7262253e000ee60f9 (diff) |
drm/radeon/kms: simplify memory controller setup V2
Get rid of _location and use _start/_end also simplify the
computation of vram_start|end & gtt_start|end. For R1XX-R2XX
we place VRAM at the same address of PCI aperture, those GPU
shouldn't have much memory and seems to behave better when
setup that way. For R3XX and newer we place VRAM at 0. For
R6XX-R7XX AGP we place VRAM before or after AGP aperture this
might limit to limit the VRAM size but it's very unlikely.
For IGP we don't change the VRAM placement.
Tested on (compiz,quake3,suspend/resume):
PCI/PCIE:RV280,R420,RV515,RV570,RV610,RV710
AGP:RV100,RV280,R420,RV350,RV620(RPB*),RV730
IGP:RS480(RPB*),RS690,RS780(RPB*),RS880
RPB: resume previously broken
V2 correct commit message to reflect more accurately the bug
and move VRAM placement to 0 for most of the GPU to avoid
limiting VRAM.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Diffstat (limited to 'drivers/gpu/drm/radeon/radeon_device.c')
-rw-r--r-- | drivers/gpu/drm/radeon/radeon_device.c | 157 |
1 files changed, 90 insertions, 67 deletions
diff --git a/drivers/gpu/drm/radeon/radeon_device.c b/drivers/gpu/drm/radeon/radeon_device.c index 7be3a6968463..91a9b966238e 100644 --- a/drivers/gpu/drm/radeon/radeon_device.c +++ b/drivers/gpu/drm/radeon/radeon_device.c | |||
@@ -100,80 +100,103 @@ void radeon_scratch_free(struct radeon_device *rdev, uint32_t reg) | |||
100 | } | 100 | } |
101 | } | 101 | } |
102 | 102 | ||
103 | /* | 103 | /** |
104 | * MC common functions | 104 | * radeon_vram_location - try to find VRAM location |
105 | * @rdev: radeon device structure holding all necessary informations | ||
106 | * @mc: memory controller structure holding memory informations | ||
107 | * @base: base address at which to put VRAM | ||
108 | * | ||
109 | * Function will place try to place VRAM at base address provided | ||
110 | * as parameter (which is so far either PCI aperture address or | ||
111 | * for IGP TOM base address). | ||
112 | * | ||
113 | * If there is not enough space to fit the unvisible VRAM in the 32bits | ||
114 | * address space then we limit the VRAM size to the aperture. | ||
115 | * | ||
116 | * If we are using AGP and if the AGP aperture doesn't allow us to have | ||
117 | * room for all the VRAM than we restrict the VRAM to the PCI aperture | ||
118 | * size and print a warning. | ||
119 | * | ||
120 | * This function will never fails, worst case are limiting VRAM. | ||
121 | * | ||
122 | * Note: GTT start, end, size should be initialized before calling this | ||
123 | * function on AGP platform. | ||
124 | * | ||
125 | * Note: We don't explictly enforce VRAM start to be aligned on VRAM size, | ||
126 | * this shouldn't be a problem as we are using the PCI aperture as a reference. | ||
127 | * Otherwise this would be needed for rv280, all r3xx, and all r4xx, but | ||
128 | * not IGP. | ||
129 | * | ||
130 | * Note: we use mc_vram_size as on some board we need to program the mc to | ||
131 | * cover the whole aperture even if VRAM size is inferior to aperture size | ||
132 | * Novell bug 204882 + along with lots of ubuntu ones | ||
133 | * | ||
134 | * Note: when limiting vram it's safe to overwritte real_vram_size because | ||
135 | * we are not in case where real_vram_size is inferior to mc_vram_size (ie | ||
136 | * note afected by bogus hw of Novell bug 204882 + along with lots of ubuntu | ||
137 | * ones) | ||
138 | * | ||
139 | * Note: IGP TOM addr should be the same as the aperture addr, we don't | ||
140 | * explicitly check for that thought. | ||
141 | * | ||
142 | * FIXME: when reducing VRAM size align new size on power of 2. | ||
105 | */ | 143 | */ |
106 | int radeon_mc_setup(struct radeon_device *rdev) | 144 | void radeon_vram_location(struct radeon_device *rdev, struct radeon_mc *mc, u64 base) |
145 | { | ||
146 | mc->vram_start = base; | ||
147 | if (mc->mc_vram_size > (0xFFFFFFFF - base + 1)) { | ||
148 | dev_warn(rdev->dev, "limiting VRAM to PCI aperture size\n"); | ||
149 | mc->real_vram_size = mc->aper_size; | ||
150 | mc->mc_vram_size = mc->aper_size; | ||
151 | } | ||
152 | mc->vram_end = mc->vram_start + mc->mc_vram_size - 1; | ||
153 | if (rdev->flags & RADEON_IS_AGP && mc->vram_end > mc->gtt_start && mc->vram_end <= mc->gtt_end) { | ||
154 | dev_warn(rdev->dev, "limiting VRAM to PCI aperture size\n"); | ||
155 | mc->real_vram_size = mc->aper_size; | ||
156 | mc->mc_vram_size = mc->aper_size; | ||
157 | } | ||
158 | mc->vram_end = mc->vram_start + mc->mc_vram_size - 1; | ||
159 | dev_info(rdev->dev, "VRAM: %lluM 0x%08llX - 0x%08llX (%lluM used)\n", | ||
160 | mc->mc_vram_size >> 20, mc->vram_start, | ||
161 | mc->vram_end, mc->real_vram_size >> 20); | ||
162 | } | ||
163 | |||
164 | /** | ||
165 | * radeon_gtt_location - try to find GTT location | ||
166 | * @rdev: radeon device structure holding all necessary informations | ||
167 | * @mc: memory controller structure holding memory informations | ||
168 | * | ||
169 | * Function will place try to place GTT before or after VRAM. | ||
170 | * | ||
171 | * If GTT size is bigger than space left then we ajust GTT size. | ||
172 | * Thus function will never fails. | ||
173 | * | ||
174 | * FIXME: when reducing GTT size align new size on power of 2. | ||
175 | */ | ||
176 | void radeon_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc) | ||
107 | { | 177 | { |
108 | uint32_t tmp; | 178 | u64 size_af, size_bf; |
109 | 179 | ||
110 | /* Some chips have an "issue" with the memory controller, the | 180 | size_af = 0xFFFFFFFF - mc->vram_end; |
111 | * location must be aligned to the size. We just align it down, | 181 | size_bf = mc->vram_start; |
112 | * too bad if we walk over the top of system memory, we don't | 182 | if (size_bf > size_af) { |
113 | * use DMA without a remapped anyway. | 183 | if (mc->gtt_size > size_bf) { |
114 | * Affected chips are rv280, all r3xx, and all r4xx, but not IGP | 184 | dev_warn(rdev->dev, "limiting GTT\n"); |
115 | */ | 185 | mc->gtt_size = size_bf; |
116 | /* FGLRX seems to setup like this, VRAM a 0, then GART. | ||
117 | */ | ||
118 | /* | ||
119 | * Note: from R6xx the address space is 40bits but here we only | ||
120 | * use 32bits (still have to see a card which would exhaust 4G | ||
121 | * address space). | ||
122 | */ | ||
123 | if (rdev->mc.vram_location != 0xFFFFFFFFUL) { | ||
124 | /* vram location was already setup try to put gtt after | ||
125 | * if it fits */ | ||
126 | tmp = rdev->mc.vram_location + rdev->mc.mc_vram_size; | ||
127 | tmp = (tmp + rdev->mc.gtt_size - 1) & ~(rdev->mc.gtt_size - 1); | ||
128 | if ((0xFFFFFFFFUL - tmp) >= rdev->mc.gtt_size) { | ||
129 | rdev->mc.gtt_location = tmp; | ||
130 | } else { | ||
131 | if (rdev->mc.gtt_size >= rdev->mc.vram_location) { | ||
132 | printk(KERN_ERR "[drm] GTT too big to fit " | ||
133 | "before or after vram location.\n"); | ||
134 | return -EINVAL; | ||
135 | } | ||
136 | rdev->mc.gtt_location = 0; | ||
137 | } | ||
138 | } else if (rdev->mc.gtt_location != 0xFFFFFFFFUL) { | ||
139 | /* gtt location was already setup try to put vram before | ||
140 | * if it fits */ | ||
141 | if (rdev->mc.mc_vram_size < rdev->mc.gtt_location) { | ||
142 | rdev->mc.vram_location = 0; | ||
143 | } else { | ||
144 | tmp = rdev->mc.gtt_location + rdev->mc.gtt_size; | ||
145 | tmp += (rdev->mc.mc_vram_size - 1); | ||
146 | tmp &= ~(rdev->mc.mc_vram_size - 1); | ||
147 | if ((0xFFFFFFFFUL - tmp) >= rdev->mc.mc_vram_size) { | ||
148 | rdev->mc.vram_location = tmp; | ||
149 | } else { | ||
150 | printk(KERN_ERR "[drm] vram too big to fit " | ||
151 | "before or after GTT location.\n"); | ||
152 | return -EINVAL; | ||
153 | } | ||
154 | } | 186 | } |
187 | mc->gtt_start = mc->vram_start - mc->gtt_size; | ||
155 | } else { | 188 | } else { |
156 | rdev->mc.vram_location = 0; | 189 | if (mc->gtt_size > size_af) { |
157 | tmp = rdev->mc.mc_vram_size; | 190 | dev_warn(rdev->dev, "limiting GTT\n"); |
158 | tmp = (tmp + rdev->mc.gtt_size - 1) & ~(rdev->mc.gtt_size - 1); | 191 | mc->gtt_size = size_af; |
159 | rdev->mc.gtt_location = tmp; | 192 | } |
160 | } | 193 | mc->gtt_start = mc->vram_end + 1; |
161 | rdev->mc.vram_start = rdev->mc.vram_location; | 194 | } |
162 | rdev->mc.vram_end = rdev->mc.vram_location + rdev->mc.mc_vram_size - 1; | 195 | mc->gtt_end = mc->gtt_start + mc->gtt_size - 1; |
163 | rdev->mc.gtt_start = rdev->mc.gtt_location; | 196 | dev_info(rdev->dev, "GTT: %lluM 0x%08llX - 0x%08llX\n", |
164 | rdev->mc.gtt_end = rdev->mc.gtt_location + rdev->mc.gtt_size - 1; | 197 | mc->gtt_size >> 20, mc->gtt_start, mc->gtt_end); |
165 | DRM_INFO("radeon: VRAM %uM\n", (unsigned)(rdev->mc.mc_vram_size >> 20)); | ||
166 | DRM_INFO("radeon: VRAM from 0x%08X to 0x%08X\n", | ||
167 | (unsigned)rdev->mc.vram_location, | ||
168 | (unsigned)(rdev->mc.vram_location + rdev->mc.mc_vram_size - 1)); | ||
169 | DRM_INFO("radeon: GTT %uM\n", (unsigned)(rdev->mc.gtt_size >> 20)); | ||
170 | DRM_INFO("radeon: GTT from 0x%08X to 0x%08X\n", | ||
171 | (unsigned)rdev->mc.gtt_location, | ||
172 | (unsigned)(rdev->mc.gtt_location + rdev->mc.gtt_size - 1)); | ||
173 | return 0; | ||
174 | } | 198 | } |
175 | 199 | ||
176 | |||
177 | /* | 200 | /* |
178 | * GPU helpers function. | 201 | * GPU helpers function. |
179 | */ | 202 | */ |